?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: docs: fix typo in btrfs-quota
[btrfs-progs-unstable/devel.git] / volumes.c
blob59670c007fc782ba351255136390ac23bd0a32b1
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
32 struct stripe {
33 struct btrfs_device *dev;
34 u64 physical;
35 };
36
37 static inline int nr_parity_stripes(struct map_lookup *map)
38 {
39 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
40 return 1;
41 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
42 return 2;
43 else
44 return 0;
45 }
46
47 static inline int nr_data_stripes(struct map_lookup *map)
48 {
49 return map->num_stripes - nr_parity_stripes(map);
50 }
51
52 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
53
54 static LIST_HEAD(fs_uuids);
55
56 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
57 u8 *uuid)
58 {
59 struct btrfs_device *dev;
60 struct list_head *cur;
61
62 list_for_each(cur, head) {
63 dev = list_entry(cur, struct btrfs_device, dev_list);
64 if (dev->devid == devid &&
65 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
66 return dev;
67 }
68 }
69 return NULL;
70 }
71
72 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
73 {
74 struct list_head *cur;
75 struct btrfs_fs_devices *fs_devices;
76
77 list_for_each(cur, &fs_uuids) {
78 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
79 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
80 return fs_devices;
81 }
82 return NULL;
83 }
84
85 static int device_list_add(const char *path,
86 struct btrfs_super_block *disk_super,
87 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
88 {
89 struct btrfs_device *device;
90 struct btrfs_fs_devices *fs_devices;
91 u64 found_transid = btrfs_super_generation(disk_super);
92
93 fs_devices = find_fsid(disk_super->fsid);
94 if (!fs_devices) {
95 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
96 if (!fs_devices)
97 return -ENOMEM;
98 INIT_LIST_HEAD(&fs_devices->devices);
99 list_add(&fs_devices->list, &fs_uuids);
100 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
101 fs_devices->latest_devid = devid;
102 fs_devices->latest_trans = found_transid;
103 fs_devices->lowest_devid = (u64)-1;
104 device = NULL;
105 } else {
106 device = __find_device(&fs_devices->devices, devid,
107 disk_super->dev_item.uuid);
108 }
109 if (!device) {
110 device = kzalloc(sizeof(*device), GFP_NOFS);
111 if (!device) {
112 /* we can safely leave the fs_devices entry around */
113 return -ENOMEM;
114 }
115 device->fd = -1;
116 device->devid = devid;
117 device->generation = found_transid;
118 memcpy(device->uuid, disk_super->dev_item.uuid,
119 BTRFS_UUID_SIZE);
120 device->name = kstrdup(path, GFP_NOFS);
121 if (!device->name) {
122 kfree(device);
123 return -ENOMEM;
124 }
125 device->label = kstrdup(disk_super->label, GFP_NOFS);
126 if (!device->label) {
127 kfree(device->name);
128 kfree(device);
129 return -ENOMEM;
130 }
131 device->total_devs = btrfs_super_num_devices(disk_super);
132 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
133 device->total_bytes =
134 btrfs_stack_device_total_bytes(&disk_super->dev_item);
135 device->bytes_used =
136 btrfs_stack_device_bytes_used(&disk_super->dev_item);
137 list_add(&device->dev_list, &fs_devices->devices);
138 device->fs_devices = fs_devices;
139 } else if (!device->name || strcmp(device->name, path)) {
140 char *name = strdup(path);
141 if (!name)
142 return -ENOMEM;
143 kfree(device->name);
144 device->name = name;
145 }
146
147
148 if (found_transid > fs_devices->latest_trans) {
149 fs_devices->latest_devid = devid;
150 fs_devices->latest_trans = found_transid;
151 }
152 if (fs_devices->lowest_devid > devid) {
153 fs_devices->lowest_devid = devid;
154 }
155 *fs_devices_ret = fs_devices;
156 return 0;
157 }
158
159 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
160 {
161 struct btrfs_fs_devices *seed_devices;
162 struct btrfs_device *device;
163
164 again:
165 if (!fs_devices)
166 return 0;
167 while (!list_empty(&fs_devices->devices)) {
168 device = list_entry(fs_devices->devices.next,
169 struct btrfs_device, dev_list);
170 if (device->fd != -1) {
171 fsync(device->fd);
172 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
173 fprintf(stderr, "Warning, could not drop caches\n");
174 close(device->fd);
175 device->fd = -1;
176 }
177 device->writeable = 0;
178 list_del(&device->dev_list);
179 /* free the memory */
180 free(device->name);
181 free(device->label);
182 free(device);
183 }
184
185 seed_devices = fs_devices->seed;
186 fs_devices->seed = NULL;
187 if (seed_devices) {
188 struct btrfs_fs_devices *orig;
189
190 orig = fs_devices;
191 fs_devices = seed_devices;
192 list_del(&orig->list);
193 free(orig);
194 goto again;
195 } else {
196 list_del(&fs_devices->list);
197 free(fs_devices);
198 }
199
200 return 0;
201 }
202
203 void btrfs_close_all_devices(void)
204 {
205 struct btrfs_fs_devices *fs_devices;
206
207 while (!list_empty(&fs_uuids)) {
208 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
209 list);
210 btrfs_close_devices(fs_devices);
211 }
212 }
213
214 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
215 {
216 int fd;
217 struct list_head *head = &fs_devices->devices;
218 struct list_head *cur;
219 struct btrfs_device *device;
220 int ret;
221
222 list_for_each(cur, head) {
223 device = list_entry(cur, struct btrfs_device, dev_list);
224 if (!device->name) {
225 printk("no name for device %llu, skip it now\n", device->devid);
226 continue;
227 }
228
229 fd = open(device->name, flags);
230 if (fd < 0) {
231 ret = -errno;
232 error("cannot open device '%s': %s", device->name,
233 strerror(errno));
234 goto fail;
235 }
236
237 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
238 fprintf(stderr, "Warning, could not drop caches\n");
239
240 if (device->devid == fs_devices->latest_devid)
241 fs_devices->latest_bdev = fd;
242 if (device->devid == fs_devices->lowest_devid)
243 fs_devices->lowest_bdev = fd;
244 device->fd = fd;
245 if (flags & O_RDWR)
246 device->writeable = 1;
247 }
248 return 0;
249 fail:
250 btrfs_close_devices(fs_devices);
251 return ret;
252 }
253
254 int btrfs_scan_one_device(int fd, const char *path,
255 struct btrfs_fs_devices **fs_devices_ret,
256 u64 *total_devs, u64 super_offset, unsigned sbflags)
257 {
258 struct btrfs_super_block *disk_super;
259 char buf[BTRFS_SUPER_INFO_SIZE];
260 int ret;
261 u64 devid;
262
263 disk_super = (struct btrfs_super_block *)buf;
264 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
265 if (ret < 0)
266 return -EIO;
267 devid = btrfs_stack_device_id(&disk_super->dev_item);
268 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
269 *total_devs = 1;
270 else
271 *total_devs = btrfs_super_num_devices(disk_super);
272
273 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
274
275 return ret;
276 }
277
278 /*
279 * find_free_dev_extent_start - find free space in the specified device
280 * @device: the device which we search the free space in
281 * @num_bytes: the size of the free space that we need
282 * @search_start: the position from which to begin the search
283 * @start: store the start of the free space.
284 * @len: the size of the free space. that we find, or the size
285 * of the max free space if we don't find suitable free space
286 *
287 * this uses a pretty simple search, the expectation is that it is
288 * called very infrequently and that a given device has a small number
289 * of extents
290 *
291 * @start is used to store the start of the free space if we find. But if we
292 * don't find suitable free space, it will be used to store the start position
293 * of the max free space.
294 *
295 * @len is used to store the size of the free space that we find.
296 * But if we don't find suitable free space, it is used to store the size of
297 * the max free space.
298 */
299 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
300 struct btrfs_device *device, u64 num_bytes,
301 u64 search_start, u64 *start, u64 *len)
302 {
303 struct btrfs_key key;
304 struct btrfs_root *root = device->dev_root;
305 struct btrfs_dev_extent *dev_extent;
306 struct btrfs_path *path;
307 u64 hole_size;
308 u64 max_hole_start;
309 u64 max_hole_size;
310 u64 extent_end;
311 u64 search_end = device->total_bytes;
312 int ret;
313 int slot;
314 struct extent_buffer *l;
315 u64 min_search_start;
316
317 /*
318 * We don't want to overwrite the superblock on the drive nor any area
319 * used by the boot loader (grub for example), so we make sure to start
320 * at an offset of at least 1MB.
321 */
322 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
323 search_start = max(search_start, min_search_start);
324
325 path = btrfs_alloc_path();
326 if (!path)
327 return -ENOMEM;
328
329 max_hole_start = search_start;
330 max_hole_size = 0;
331
332 if (search_start >= search_end) {
333 ret = -ENOSPC;
334 goto out;
335 }
336
337 path->reada = 2;
338
339 key.objectid = device->devid;
340 key.offset = search_start;
341 key.type = BTRFS_DEV_EXTENT_KEY;
342
343 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
344 if (ret < 0)
345 goto out;
346 if (ret > 0) {
347 ret = btrfs_previous_item(root, path, key.objectid, key.type);
348 if (ret < 0)
349 goto out;
350 }
351
352 while (1) {
353 l = path->nodes[0];
354 slot = path->slots[0];
355 if (slot >= btrfs_header_nritems(l)) {
356 ret = btrfs_next_leaf(root, path);
357 if (ret == 0)
358 continue;
359 if (ret < 0)
360 goto out;
361
362 break;
363 }
364 btrfs_item_key_to_cpu(l, &key, slot);
365
366 if (key.objectid < device->devid)
367 goto next;
368
369 if (key.objectid > device->devid)
370 break;
371
372 if (key.type != BTRFS_DEV_EXTENT_KEY)
373 goto next;
374
375 if (key.offset > search_start) {
376 hole_size = key.offset - search_start;
377
378 /*
379 * Have to check before we set max_hole_start, otherwise
380 * we could end up sending back this offset anyway.
381 */
382 if (hole_size > max_hole_size) {
383 max_hole_start = search_start;
384 max_hole_size = hole_size;
385 }
386
387 /*
388 * If this free space is greater than which we need,
389 * it must be the max free space that we have found
390 * until now, so max_hole_start must point to the start
391 * of this free space and the length of this free space
392 * is stored in max_hole_size. Thus, we return
393 * max_hole_start and max_hole_size and go back to the
394 * caller.
395 */
396 if (hole_size >= num_bytes) {
397 ret = 0;
398 goto out;
399 }
400 }
401
402 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
403 extent_end = key.offset + btrfs_dev_extent_length(l,
404 dev_extent);
405 if (extent_end > search_start)
406 search_start = extent_end;
407 next:
408 path->slots[0]++;
409 cond_resched();
410 }
411
412 /*
413 * At this point, search_start should be the end of
414 * allocated dev extents, and when shrinking the device,
415 * search_end may be smaller than search_start.
416 */
417 if (search_end > search_start) {
418 hole_size = search_end - search_start;
419
420 if (hole_size > max_hole_size) {
421 max_hole_start = search_start;
422 max_hole_size = hole_size;
423 }
424 }
425
426 /* See above. */
427 if (max_hole_size < num_bytes)
428 ret = -ENOSPC;
429 else
430 ret = 0;
431
432 out:
433 btrfs_free_path(path);
434 *start = max_hole_start;
435 if (len)
436 *len = max_hole_size;
437 return ret;
438 }
439
440 int find_free_dev_extent(struct btrfs_trans_handle *trans,
441 struct btrfs_device *device, u64 num_bytes,
442 u64 *start)
443 {
444 /* FIXME use last free of some kind */
445 return find_free_dev_extent_start(trans, device,
446 num_bytes, 0, start, NULL);
447 }
448
449 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
450 struct btrfs_device *device,
451 u64 chunk_tree, u64 chunk_objectid,
452 u64 chunk_offset,
453 u64 num_bytes, u64 *start, int convert)
454 {
455 int ret;
456 struct btrfs_path *path;
457 struct btrfs_root *root = device->dev_root;
458 struct btrfs_dev_extent *extent;
459 struct extent_buffer *leaf;
460 struct btrfs_key key;
461
462 path = btrfs_alloc_path();
463 if (!path)
464 return -ENOMEM;
465
466 /*
467 * For convert case, just skip search free dev_extent, as caller
468 * is responsible to make sure it's free.
469 */
470 if (!convert) {
471 ret = find_free_dev_extent(trans, device, num_bytes,
472 start);
473 if (ret)
474 goto err;
475 }
476
477 key.objectid = device->devid;
478 key.offset = *start;
479 key.type = BTRFS_DEV_EXTENT_KEY;
480 ret = btrfs_insert_empty_item(trans, root, path, &key,
481 sizeof(*extent));
482 BUG_ON(ret);
483
484 leaf = path->nodes[0];
485 extent = btrfs_item_ptr(leaf, path->slots[0],
486 struct btrfs_dev_extent);
487 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
488 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
489 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
490
491 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
492 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
493 BTRFS_UUID_SIZE);
494
495 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
496 btrfs_mark_buffer_dirty(leaf);
497 err:
498 btrfs_free_path(path);
499 return ret;
500 }
501
502 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
503 {
504 struct btrfs_path *path;
505 int ret;
506 struct btrfs_key key;
507 struct btrfs_chunk *chunk;
508 struct btrfs_key found_key;
509
510 path = btrfs_alloc_path();
511 if (!path)
512 return -ENOMEM;
513
514 key.objectid = objectid;
515 key.offset = (u64)-1;
516 key.type = BTRFS_CHUNK_ITEM_KEY;
517
518 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
519 if (ret < 0)
520 goto error;
521
522 BUG_ON(ret == 0);
523
524 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
525 if (ret) {
526 *offset = 0;
527 } else {
528 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
529 path->slots[0]);
530 if (found_key.objectid != objectid)
531 *offset = 0;
532 else {
533 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
534 struct btrfs_chunk);
535 *offset = found_key.offset +
536 btrfs_chunk_length(path->nodes[0], chunk);
537 }
538 }
539 ret = 0;
540 error:
541 btrfs_free_path(path);
542 return ret;
543 }
544
545 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
546 u64 *objectid)
547 {
548 int ret;
549 struct btrfs_key key;
550 struct btrfs_key found_key;
551
552 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
553 key.type = BTRFS_DEV_ITEM_KEY;
554 key.offset = (u64)-1;
555
556 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
557 if (ret < 0)
558 goto error;
559
560 BUG_ON(ret == 0);
561
562 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
563 BTRFS_DEV_ITEM_KEY);
564 if (ret) {
565 *objectid = 1;
566 } else {
567 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
568 path->slots[0]);
569 *objectid = found_key.offset + 1;
570 }
571 ret = 0;
572 error:
573 btrfs_release_path(path);
574 return ret;
575 }
576
577 /*
578 * the device information is stored in the chunk root
579 * the btrfs_device struct should be fully filled in
580 */
581 int btrfs_add_device(struct btrfs_trans_handle *trans,
582 struct btrfs_root *root,
583 struct btrfs_device *device)
584 {
585 int ret;
586 struct btrfs_path *path;
587 struct btrfs_dev_item *dev_item;
588 struct extent_buffer *leaf;
589 struct btrfs_key key;
590 unsigned long ptr;
591 u64 free_devid = 0;
592
593 root = root->fs_info->chunk_root;
594
595 path = btrfs_alloc_path();
596 if (!path)
597 return -ENOMEM;
598
599 ret = find_next_devid(root, path, &free_devid);
600 if (ret)
601 goto out;
602
603 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
604 key.type = BTRFS_DEV_ITEM_KEY;
605 key.offset = free_devid;
606
607 ret = btrfs_insert_empty_item(trans, root, path, &key,
608 sizeof(*dev_item));
609 if (ret)
610 goto out;
611
612 leaf = path->nodes[0];
613 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
614
615 device->devid = free_devid;
616 btrfs_set_device_id(leaf, dev_item, device->devid);
617 btrfs_set_device_generation(leaf, dev_item, 0);
618 btrfs_set_device_type(leaf, dev_item, device->type);
619 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
620 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
621 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
622 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
623 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
624 btrfs_set_device_group(leaf, dev_item, 0);
625 btrfs_set_device_seek_speed(leaf, dev_item, 0);
626 btrfs_set_device_bandwidth(leaf, dev_item, 0);
627 btrfs_set_device_start_offset(leaf, dev_item, 0);
628
629 ptr = (unsigned long)btrfs_device_uuid(dev_item);
630 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
631 ptr = (unsigned long)btrfs_device_fsid(dev_item);
632 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
633 btrfs_mark_buffer_dirty(leaf);
634 ret = 0;
635
636 out:
637 btrfs_free_path(path);
638 return ret;
639 }
640
641 int btrfs_update_device(struct btrfs_trans_handle *trans,
642 struct btrfs_device *device)
643 {
644 int ret;
645 struct btrfs_path *path;
646 struct btrfs_root *root;
647 struct btrfs_dev_item *dev_item;
648 struct extent_buffer *leaf;
649 struct btrfs_key key;
650
651 root = device->dev_root->fs_info->chunk_root;
652
653 path = btrfs_alloc_path();
654 if (!path)
655 return -ENOMEM;
656
657 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
658 key.type = BTRFS_DEV_ITEM_KEY;
659 key.offset = device->devid;
660
661 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
662 if (ret < 0)
663 goto out;
664
665 if (ret > 0) {
666 ret = -ENOENT;
667 goto out;
668 }
669
670 leaf = path->nodes[0];
671 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
672
673 btrfs_set_device_id(leaf, dev_item, device->devid);
674 btrfs_set_device_type(leaf, dev_item, device->type);
675 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
676 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
677 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
678 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
679 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
680 btrfs_mark_buffer_dirty(leaf);
681
682 out:
683 btrfs_free_path(path);
684 return ret;
685 }
686
687 int btrfs_add_system_chunk(struct btrfs_root *root,
688 struct btrfs_key *key,
689 struct btrfs_chunk *chunk, int item_size)
690 {
691 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
692 struct btrfs_disk_key disk_key;
693 u32 array_size;
694 u8 *ptr;
695
696 array_size = btrfs_super_sys_array_size(super_copy);
697 if (array_size + item_size + sizeof(disk_key)
698 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
699 return -EFBIG;
700
701 ptr = super_copy->sys_chunk_array + array_size;
702 btrfs_cpu_key_to_disk(&disk_key, key);
703 memcpy(ptr, &disk_key, sizeof(disk_key));
704 ptr += sizeof(disk_key);
705 memcpy(ptr, chunk, item_size);
706 item_size += sizeof(disk_key);
707 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
708 return 0;
709 }
710
711 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
712 int sub_stripes)
713 {
714 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
715 return calc_size;
716 else if (type & BTRFS_BLOCK_GROUP_RAID10)
717 return calc_size * (num_stripes / sub_stripes);
718 else if (type & BTRFS_BLOCK_GROUP_RAID5)
719 return calc_size * (num_stripes - 1);
720 else if (type & BTRFS_BLOCK_GROUP_RAID6)
721 return calc_size * (num_stripes - 2);
722 else
723 return calc_size * num_stripes;
724 }
725
726
727 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
728 {
729 /* TODO, add a way to store the preferred stripe size */
730 return BTRFS_STRIPE_LEN;
731 }
732
733 /*
734 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
735 *
736 * It is not equal to "device->total_bytes - device->bytes_used".
737 * We do not allocate any chunk in 1M at beginning of device, and not
738 * allowed to allocate any chunk before alloc_start if it is specified.
739 * So search holes from max(1M, alloc_start) to device->total_bytes.
740 */
741 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
742 struct btrfs_device *device,
743 u64 *avail_bytes)
744 {
745 struct btrfs_path *path;
746 struct btrfs_root *root = device->dev_root;
747 struct btrfs_key key;
748 struct btrfs_dev_extent *dev_extent = NULL;
749 struct extent_buffer *l;
750 u64 search_start = root->fs_info->alloc_start;
751 u64 search_end = device->total_bytes;
752 u64 extent_end = 0;
753 u64 free_bytes = 0;
754 int ret;
755 int slot = 0;
756
757 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
758
759 path = btrfs_alloc_path();
760 if (!path)
761 return -ENOMEM;
762
763 key.objectid = device->devid;
764 key.offset = root->fs_info->alloc_start;
765 key.type = BTRFS_DEV_EXTENT_KEY;
766
767 path->reada = 2;
768 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
769 if (ret < 0)
770 goto error;
771 ret = btrfs_previous_item(root, path, 0, key.type);
772 if (ret < 0)
773 goto error;
774
775 while (1) {
776 l = path->nodes[0];
777 slot = path->slots[0];
778 if (slot >= btrfs_header_nritems(l)) {
779 ret = btrfs_next_leaf(root, path);
780 if (ret == 0)
781 continue;
782 if (ret < 0)
783 goto error;
784 break;
785 }
786 btrfs_item_key_to_cpu(l, &key, slot);
787
788 if (key.objectid < device->devid)
789 goto next;
790 if (key.objectid > device->devid)
791 break;
792 if (key.type != BTRFS_DEV_EXTENT_KEY)
793 goto next;
794 if (key.offset > search_end)
795 break;
796 if (key.offset > search_start)
797 free_bytes += key.offset - search_start;
798
799 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
800 extent_end = key.offset + btrfs_dev_extent_length(l,
801 dev_extent);
802 if (extent_end > search_start)
803 search_start = extent_end;
804 if (search_start > search_end)
805 break;
806 next:
807 path->slots[0]++;
808 cond_resched();
809 }
810
811 if (search_start < search_end)
812 free_bytes += search_end - search_start;
813
814 *avail_bytes = free_bytes;
815 ret = 0;
816 error:
817 btrfs_free_path(path);
818 return ret;
819 }
820
821 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
822 - sizeof(struct btrfs_item) \
823 - sizeof(struct btrfs_chunk)) \
824 / sizeof(struct btrfs_stripe) + 1)
825
826 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
827 - 2 * sizeof(struct btrfs_disk_key) \
828 - 2 * sizeof(struct btrfs_chunk)) \
829 / sizeof(struct btrfs_stripe) + 1)
830
831 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
832 struct btrfs_root *extent_root, u64 *start,
833 u64 *num_bytes, u64 type)
834 {
835 u64 dev_offset;
836 struct btrfs_fs_info *info = extent_root->fs_info;
837 struct btrfs_root *chunk_root = info->chunk_root;
838 struct btrfs_stripe *stripes;
839 struct btrfs_device *device = NULL;
840 struct btrfs_chunk *chunk;
841 struct list_head private_devs;
842 struct list_head *dev_list = &info->fs_devices->devices;
843 struct list_head *cur;
844 struct map_lookup *map;
845 int min_stripe_size = SZ_1M;
846 u64 calc_size = SZ_8M;
847 u64 min_free;
848 u64 max_chunk_size = 4 * calc_size;
849 u64 avail = 0;
850 u64 max_avail = 0;
851 u64 percent_max;
852 int num_stripes = 1;
853 int max_stripes = 0;
854 int min_stripes = 1;
855 int sub_stripes = 0;
856 int looped = 0;
857 int ret;
858 int index;
859 int stripe_len = BTRFS_STRIPE_LEN;
860 struct btrfs_key key;
861 u64 offset;
862
863 if (list_empty(dev_list)) {
864 return -ENOSPC;
865 }
866
867 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
868 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
869 BTRFS_BLOCK_GROUP_RAID10 |
870 BTRFS_BLOCK_GROUP_DUP)) {
871 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
872 calc_size = SZ_8M;
873 max_chunk_size = calc_size * 2;
874 min_stripe_size = SZ_1M;
875 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
876 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
877 calc_size = SZ_1G;
878 max_chunk_size = 10 * calc_size;
879 min_stripe_size = SZ_64M;
880 max_stripes = BTRFS_MAX_DEVS(chunk_root);
881 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
882 calc_size = SZ_1G;
883 max_chunk_size = 4 * calc_size;
884 min_stripe_size = SZ_32M;
885 max_stripes = BTRFS_MAX_DEVS(chunk_root);
886 }
887 }
888 if (type & BTRFS_BLOCK_GROUP_RAID1) {
889 num_stripes = min_t(u64, 2,
890 btrfs_super_num_devices(info->super_copy));
891 if (num_stripes < 2)
892 return -ENOSPC;
893 min_stripes = 2;
894 }
895 if (type & BTRFS_BLOCK_GROUP_DUP) {
896 num_stripes = 2;
897 min_stripes = 2;
898 }
899 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
900 num_stripes = btrfs_super_num_devices(info->super_copy);
901 if (num_stripes > max_stripes)
902 num_stripes = max_stripes;
903 min_stripes = 2;
904 }
905 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
906 num_stripes = btrfs_super_num_devices(info->super_copy);
907 if (num_stripes > max_stripes)
908 num_stripes = max_stripes;
909 if (num_stripes < 4)
910 return -ENOSPC;
911 num_stripes &= ~(u32)1;
912 sub_stripes = 2;
913 min_stripes = 4;
914 }
915 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
916 num_stripes = btrfs_super_num_devices(info->super_copy);
917 if (num_stripes > max_stripes)
918 num_stripes = max_stripes;
919 if (num_stripes < 2)
920 return -ENOSPC;
921 min_stripes = 2;
922 stripe_len = find_raid56_stripe_len(num_stripes - 1,
923 btrfs_super_stripesize(info->super_copy));
924 }
925 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
926 num_stripes = btrfs_super_num_devices(info->super_copy);
927 if (num_stripes > max_stripes)
928 num_stripes = max_stripes;
929 if (num_stripes < 3)
930 return -ENOSPC;
931 min_stripes = 3;
932 stripe_len = find_raid56_stripe_len(num_stripes - 2,
933 btrfs_super_stripesize(info->super_copy));
934 }
935
936 /* we don't want a chunk larger than 10% of the FS */
937 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
938 max_chunk_size = min(percent_max, max_chunk_size);
939
940 again:
941 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
942 max_chunk_size) {
943 calc_size = max_chunk_size;
944 calc_size /= num_stripes;
945 calc_size /= stripe_len;
946 calc_size *= stripe_len;
947 }
948 /* we don't want tiny stripes */
949 calc_size = max_t(u64, calc_size, min_stripe_size);
950
951 calc_size /= stripe_len;
952 calc_size *= stripe_len;
953 INIT_LIST_HEAD(&private_devs);
954 cur = dev_list->next;
955 index = 0;
956
957 if (type & BTRFS_BLOCK_GROUP_DUP)
958 min_free = calc_size * 2;
959 else
960 min_free = calc_size;
961
962 /* build a private list of devices we will allocate from */
963 while(index < num_stripes) {
964 device = list_entry(cur, struct btrfs_device, dev_list);
965 ret = btrfs_device_avail_bytes(trans, device, &avail);
966 if (ret)
967 return ret;
968 cur = cur->next;
969 if (avail >= min_free) {
970 list_move_tail(&device->dev_list, &private_devs);
971 index++;
972 if (type & BTRFS_BLOCK_GROUP_DUP)
973 index++;
974 } else if (avail > max_avail)
975 max_avail = avail;
976 if (cur == dev_list)
977 break;
978 }
979 if (index < num_stripes) {
980 list_splice(&private_devs, dev_list);
981 if (index >= min_stripes) {
982 num_stripes = index;
983 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
984 num_stripes /= sub_stripes;
985 num_stripes *= sub_stripes;
986 }
987 looped = 1;
988 goto again;
989 }
990 if (!looped && max_avail > 0) {
991 looped = 1;
992 calc_size = max_avail;
993 goto again;
994 }
995 return -ENOSPC;
996 }
997 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
998 &offset);
999 if (ret)
1000 return ret;
1001 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1002 key.type = BTRFS_CHUNK_ITEM_KEY;
1003 key.offset = offset;
1004
1005 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1006 if (!chunk)
1007 return -ENOMEM;
1008
1009 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1010 if (!map) {
1011 kfree(chunk);
1012 return -ENOMEM;
1013 }
1014
1015 stripes = &chunk->stripe;
1016 *num_bytes = chunk_bytes_by_type(type, calc_size,
1017 num_stripes, sub_stripes);
1018 index = 0;
1019 while(index < num_stripes) {
1020 struct btrfs_stripe *stripe;
1021 BUG_ON(list_empty(&private_devs));
1022 cur = private_devs.next;
1023 device = list_entry(cur, struct btrfs_device, dev_list);
1024
1025 /* loop over this device again if we're doing a dup group */
1026 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1027 (index == num_stripes - 1))
1028 list_move_tail(&device->dev_list, dev_list);
1029
1030 ret = btrfs_alloc_dev_extent(trans, device,
1031 info->chunk_root->root_key.objectid,
1032 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1033 calc_size, &dev_offset, 0);
1034 BUG_ON(ret);
1035
1036 device->bytes_used += calc_size;
1037 ret = btrfs_update_device(trans, device);
1038 BUG_ON(ret);
1039
1040 map->stripes[index].dev = device;
1041 map->stripes[index].physical = dev_offset;
1042 stripe = stripes + index;
1043 btrfs_set_stack_stripe_devid(stripe, device->devid);
1044 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1045 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1046 index++;
1047 }
1048 BUG_ON(!list_empty(&private_devs));
1049
1050 /* key was set above */
1051 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1052 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1053 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1054 btrfs_set_stack_chunk_type(chunk, type);
1055 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1056 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1057 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1058 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1059 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1060 map->sector_size = extent_root->sectorsize;
1061 map->stripe_len = stripe_len;
1062 map->io_align = stripe_len;
1063 map->io_width = stripe_len;
1064 map->type = type;
1065 map->num_stripes = num_stripes;
1066 map->sub_stripes = sub_stripes;
1067
1068 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1069 btrfs_chunk_item_size(num_stripes));
1070 BUG_ON(ret);
1071 *start = key.offset;;
1072
1073 map->ce.start = key.offset;
1074 map->ce.size = *num_bytes;
1075
1076 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1077 BUG_ON(ret);
1078
1079 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1080 ret = btrfs_add_system_chunk(chunk_root, &key,
1081 chunk, btrfs_chunk_item_size(num_stripes));
1082 BUG_ON(ret);
1083 }
1084
1085 kfree(chunk);
1086 return ret;
1087 }
1088
1089 /*
1090 * Alloc a DATA chunk with SINGLE profile.
1091 *
1092 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1093 * (btrfs logical bytenr == on-disk bytenr)
1094 * For that case, caller must make sure the chunk and dev_extent are not
1095 * occupied.
1096 */
1097 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1098 struct btrfs_root *extent_root, u64 *start,
1099 u64 num_bytes, u64 type, int convert)
1100 {
1101 u64 dev_offset;
1102 struct btrfs_fs_info *info = extent_root->fs_info;
1103 struct btrfs_root *chunk_root = info->chunk_root;
1104 struct btrfs_stripe *stripes;
1105 struct btrfs_device *device = NULL;
1106 struct btrfs_chunk *chunk;
1107 struct list_head *dev_list = &info->fs_devices->devices;
1108 struct list_head *cur;
1109 struct map_lookup *map;
1110 u64 calc_size = SZ_8M;
1111 int num_stripes = 1;
1112 int sub_stripes = 0;
1113 int ret;
1114 int index;
1115 int stripe_len = BTRFS_STRIPE_LEN;
1116 struct btrfs_key key;
1117
1118 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1119 key.type = BTRFS_CHUNK_ITEM_KEY;
1120 if (convert) {
1121 if (*start != round_down(*start, extent_root->sectorsize)) {
1122 error("DATA chunk start not sectorsize aligned: %llu",
1123 (unsigned long long)*start);
1124 return -EINVAL;
1125 }
1126 key.offset = *start;
1127 dev_offset = *start;
1128 } else {
1129 u64 tmp;
1130
1131 ret = find_next_chunk(chunk_root,
1132 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1133 &tmp);
1134 key.offset = tmp;
1135 if (ret)
1136 return ret;
1137 }
1138
1139 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1140 if (!chunk)
1141 return -ENOMEM;
1142
1143 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1144 if (!map) {
1145 kfree(chunk);
1146 return -ENOMEM;
1147 }
1148
1149 stripes = &chunk->stripe;
1150 calc_size = num_bytes;
1151
1152 index = 0;
1153 cur = dev_list->next;
1154 device = list_entry(cur, struct btrfs_device, dev_list);
1155
1156 while (index < num_stripes) {
1157 struct btrfs_stripe *stripe;
1158
1159 ret = btrfs_alloc_dev_extent(trans, device,
1160 info->chunk_root->root_key.objectid,
1161 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1162 calc_size, &dev_offset, convert);
1163 BUG_ON(ret);
1164
1165 device->bytes_used += calc_size;
1166 ret = btrfs_update_device(trans, device);
1167 BUG_ON(ret);
1168
1169 map->stripes[index].dev = device;
1170 map->stripes[index].physical = dev_offset;
1171 stripe = stripes + index;
1172 btrfs_set_stack_stripe_devid(stripe, device->devid);
1173 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1174 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1175 index++;
1176 }
1177
1178 /* key was set above */
1179 btrfs_set_stack_chunk_length(chunk, num_bytes);
1180 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1181 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1182 btrfs_set_stack_chunk_type(chunk, type);
1183 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1184 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1185 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1186 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1187 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1188 map->sector_size = extent_root->sectorsize;
1189 map->stripe_len = stripe_len;
1190 map->io_align = stripe_len;
1191 map->io_width = stripe_len;
1192 map->type = type;
1193 map->num_stripes = num_stripes;
1194 map->sub_stripes = sub_stripes;
1195
1196 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1197 btrfs_chunk_item_size(num_stripes));
1198 BUG_ON(ret);
1199 if (!convert)
1200 *start = key.offset;
1201
1202 map->ce.start = key.offset;
1203 map->ce.size = num_bytes;
1204
1205 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1206 BUG_ON(ret);
1207
1208 kfree(chunk);
1209 return ret;
1210 }
1211
1212 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1213 {
1214 struct cache_extent *ce;
1215 struct map_lookup *map;
1216 int ret;
1217
1218 ce = search_cache_extent(&map_tree->cache_tree, logical);
1219 if (!ce) {
1220 fprintf(stderr, "No mapping for %llu-%llu\n",
1221 (unsigned long long)logical,
1222 (unsigned long long)logical+len);
1223 return 1;
1224 }
1225 if (ce->start > logical || ce->start + ce->size < logical) {
1226 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1227 "%llu-%llu\n", (unsigned long long)logical,
1228 (unsigned long long)logical+len,
1229 (unsigned long long)ce->start,
1230 (unsigned long long)ce->start + ce->size);
1231 return 1;
1232 }
1233 map = container_of(ce, struct map_lookup, ce);
1234
1235 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1236 ret = map->num_stripes;
1237 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1238 ret = map->sub_stripes;
1239 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1240 ret = 2;
1241 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1242 ret = 3;
1243 else
1244 ret = 1;
1245 return ret;
1246 }
1247
1248 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1249 u64 *size, u64 type)
1250 {
1251 struct cache_extent *ce;
1252 struct map_lookup *map;
1253 u64 cur = *logical;
1254
1255 ce = search_cache_extent(&map_tree->cache_tree, cur);
1256
1257 while (ce) {
1258 /*
1259 * only jump to next bg if our cur is not 0
1260 * As the initial logical for btrfs_next_bg() is 0, and
1261 * if we jump to next bg, we skipped a valid bg.
1262 */
1263 if (cur) {
1264 ce = next_cache_extent(ce);
1265 if (!ce)
1266 return -ENOENT;
1267 }
1268
1269 cur = ce->start;
1270 map = container_of(ce, struct map_lookup, ce);
1271 if (map->type & type) {
1272 *logical = ce->start;
1273 *size = ce->size;
1274 return 0;
1275 }
1276 }
1277
1278 return -ENOENT;
1279 }
1280
1281 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1282 u64 chunk_start, u64 physical, u64 devid,
1283 u64 **logical, int *naddrs, int *stripe_len)
1284 {
1285 struct cache_extent *ce;
1286 struct map_lookup *map;
1287 u64 *buf;
1288 u64 bytenr;
1289 u64 length;
1290 u64 stripe_nr;
1291 u64 rmap_len;
1292 int i, j, nr = 0;
1293
1294 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1295 BUG_ON(!ce);
1296 map = container_of(ce, struct map_lookup, ce);
1297
1298 length = ce->size;
1299 rmap_len = map->stripe_len;
1300 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1301 length = ce->size / (map->num_stripes / map->sub_stripes);
1302 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1303 length = ce->size / map->num_stripes;
1304 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1305 BTRFS_BLOCK_GROUP_RAID6)) {
1306 length = ce->size / nr_data_stripes(map);
1307 rmap_len = map->stripe_len * nr_data_stripes(map);
1308 }
1309
1310 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1311
1312 for (i = 0; i < map->num_stripes; i++) {
1313 if (devid && map->stripes[i].dev->devid != devid)
1314 continue;
1315 if (map->stripes[i].physical > physical ||
1316 map->stripes[i].physical + length <= physical)
1317 continue;
1318
1319 stripe_nr = (physical - map->stripes[i].physical) /
1320 map->stripe_len;
1321
1322 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1323 stripe_nr = (stripe_nr * map->num_stripes + i) /
1324 map->sub_stripes;
1325 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1326 stripe_nr = stripe_nr * map->num_stripes + i;
1327 } /* else if RAID[56], multiply by nr_data_stripes().
1328 * Alternatively, just use rmap_len below instead of
1329 * map->stripe_len */
1330
1331 bytenr = ce->start + stripe_nr * rmap_len;
1332 for (j = 0; j < nr; j++) {
1333 if (buf[j] == bytenr)
1334 break;
1335 }
1336 if (j == nr)
1337 buf[nr++] = bytenr;
1338 }
1339
1340 *logical = buf;
1341 *naddrs = nr;
1342 *stripe_len = rmap_len;
1343
1344 return 0;
1345 }
1346
1347 static inline int parity_smaller(u64 a, u64 b)
1348 {
1349 return a > b;
1350 }
1351
1352 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1353 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1354 {
1355 struct btrfs_bio_stripe s;
1356 int i;
1357 u64 l;
1358 int again = 1;
1359
1360 while (again) {
1361 again = 0;
1362 for (i = 0; i < bbio->num_stripes - 1; i++) {
1363 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1364 s = bbio->stripes[i];
1365 l = raid_map[i];
1366 bbio->stripes[i] = bbio->stripes[i+1];
1367 raid_map[i] = raid_map[i+1];
1368 bbio->stripes[i+1] = s;
1369 raid_map[i+1] = l;
1370 again = 1;
1371 }
1372 }
1373 }
1374 }
1375
1376 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1377 u64 logical, u64 *length,
1378 struct btrfs_multi_bio **multi_ret, int mirror_num,
1379 u64 **raid_map_ret)
1380 {
1381 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1382 multi_ret, mirror_num, raid_map_ret);
1383 }
1384
1385 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1386 u64 logical, u64 *length, u64 *type,
1387 struct btrfs_multi_bio **multi_ret, int mirror_num,
1388 u64 **raid_map_ret)
1389 {
1390 struct cache_extent *ce;
1391 struct map_lookup *map;
1392 u64 offset;
1393 u64 stripe_offset;
1394 u64 stripe_nr;
1395 u64 *raid_map = NULL;
1396 int stripes_allocated = 8;
1397 int stripes_required = 1;
1398 int stripe_index;
1399 int i;
1400 struct btrfs_multi_bio *multi = NULL;
1401
1402 if (multi_ret && rw == READ) {
1403 stripes_allocated = 1;
1404 }
1405 again:
1406 ce = search_cache_extent(&map_tree->cache_tree, logical);
1407 if (!ce) {
1408 kfree(multi);
1409 *length = (u64)-1;
1410 return -ENOENT;
1411 }
1412 if (ce->start > logical) {
1413 kfree(multi);
1414 *length = ce->start - logical;
1415 return -ENOENT;
1416 }
1417
1418 if (multi_ret) {
1419 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1420 GFP_NOFS);
1421 if (!multi)
1422 return -ENOMEM;
1423 }
1424 map = container_of(ce, struct map_lookup, ce);
1425 offset = logical - ce->start;
1426
1427 if (rw == WRITE) {
1428 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1429 BTRFS_BLOCK_GROUP_DUP)) {
1430 stripes_required = map->num_stripes;
1431 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1432 stripes_required = map->sub_stripes;
1433 }
1434 }
1435 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1436 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1437 /* RAID[56] write or recovery. Return all stripes */
1438 stripes_required = map->num_stripes;
1439
1440 /* Only allocate the map if we've already got a large enough multi_ret */
1441 if (stripes_allocated >= stripes_required) {
1442 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1443 if (!raid_map) {
1444 kfree(multi);
1445 return -ENOMEM;
1446 }
1447 }
1448 }
1449
1450 /* if our multi bio struct is too small, back off and try again */
1451 if (multi_ret && stripes_allocated < stripes_required) {
1452 stripes_allocated = stripes_required;
1453 kfree(multi);
1454 multi = NULL;
1455 goto again;
1456 }
1457 stripe_nr = offset;
1458 /*
1459 * stripe_nr counts the total number of stripes we have to stride
1460 * to get to this block
1461 */
1462 stripe_nr = stripe_nr / map->stripe_len;
1463
1464 stripe_offset = stripe_nr * map->stripe_len;
1465 BUG_ON(offset < stripe_offset);
1466
1467 /* stripe_offset is the offset of this block in its stripe*/
1468 stripe_offset = offset - stripe_offset;
1469
1470 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1471 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1472 BTRFS_BLOCK_GROUP_RAID10 |
1473 BTRFS_BLOCK_GROUP_DUP)) {
1474 /* we limit the length of each bio to what fits in a stripe */
1475 *length = min_t(u64, ce->size - offset,
1476 map->stripe_len - stripe_offset);
1477 } else {
1478 *length = ce->size - offset;
1479 }
1480
1481 if (!multi_ret)
1482 goto out;
1483
1484 multi->num_stripes = 1;
1485 stripe_index = 0;
1486 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1487 if (rw == WRITE)
1488 multi->num_stripes = map->num_stripes;
1489 else if (mirror_num)
1490 stripe_index = mirror_num - 1;
1491 else
1492 stripe_index = stripe_nr % map->num_stripes;
1493 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1494 int factor = map->num_stripes / map->sub_stripes;
1495
1496 stripe_index = stripe_nr % factor;
1497 stripe_index *= map->sub_stripes;
1498
1499 if (rw == WRITE)
1500 multi->num_stripes = map->sub_stripes;
1501 else if (mirror_num)
1502 stripe_index += mirror_num - 1;
1503
1504 stripe_nr = stripe_nr / factor;
1505 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1506 if (rw == WRITE)
1507 multi->num_stripes = map->num_stripes;
1508 else if (mirror_num)
1509 stripe_index = mirror_num - 1;
1510 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1511 BTRFS_BLOCK_GROUP_RAID6)) {
1512
1513 if (raid_map) {
1514 int rot;
1515 u64 tmp;
1516 u64 raid56_full_stripe_start;
1517 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1518
1519 /*
1520 * align the start of our data stripe in the logical
1521 * address space
1522 */
1523 raid56_full_stripe_start = offset / full_stripe_len;
1524 raid56_full_stripe_start *= full_stripe_len;
1525
1526 /* get the data stripe number */
1527 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1528 stripe_nr = stripe_nr / nr_data_stripes(map);
1529
1530 /* Work out the disk rotation on this stripe-set */
1531 rot = stripe_nr % map->num_stripes;
1532
1533 /* Fill in the logical address of each stripe */
1534 tmp = stripe_nr * nr_data_stripes(map);
1535
1536 for (i = 0; i < nr_data_stripes(map); i++)
1537 raid_map[(i+rot) % map->num_stripes] =
1538 ce->start + (tmp + i) * map->stripe_len;
1539
1540 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1541 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1542 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1543
1544 *length = map->stripe_len;
1545 stripe_index = 0;
1546 stripe_offset = 0;
1547 multi->num_stripes = map->num_stripes;
1548 } else {
1549 stripe_index = stripe_nr % nr_data_stripes(map);
1550 stripe_nr = stripe_nr / nr_data_stripes(map);
1551
1552 /*
1553 * Mirror #0 or #1 means the original data block.
1554 * Mirror #2 is RAID5 parity block.
1555 * Mirror #3 is RAID6 Q block.
1556 */
1557 if (mirror_num > 1)
1558 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1559
1560 /* We distribute the parity blocks across stripes */
1561 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1562 }
1563 } else {
1564 /*
1565 * after this do_div call, stripe_nr is the number of stripes
1566 * on this device we have to walk to find the data, and
1567 * stripe_index is the number of our device in the stripe array
1568 */
1569 stripe_index = stripe_nr % map->num_stripes;
1570 stripe_nr = stripe_nr / map->num_stripes;
1571 }
1572 BUG_ON(stripe_index >= map->num_stripes);
1573
1574 for (i = 0; i < multi->num_stripes; i++) {
1575 multi->stripes[i].physical =
1576 map->stripes[stripe_index].physical + stripe_offset +
1577 stripe_nr * map->stripe_len;
1578 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1579 stripe_index++;
1580 }
1581 *multi_ret = multi;
1582
1583 if (type)
1584 *type = map->type;
1585
1586 if (raid_map) {
1587 sort_parity_stripes(multi, raid_map);
1588 *raid_map_ret = raid_map;
1589 }
1590 out:
1591 return 0;
1592 }
1593
1594 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1595 u8 *uuid, u8 *fsid)
1596 {
1597 struct btrfs_device *device;
1598 struct btrfs_fs_devices *cur_devices;
1599
1600 cur_devices = root->fs_info->fs_devices;
1601 while (cur_devices) {
1602 if (!fsid ||
1603 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1604 root->fs_info->ignore_fsid_mismatch)) {
1605 device = __find_device(&cur_devices->devices,
1606 devid, uuid);
1607 if (device)
1608 return device;
1609 }
1610 cur_devices = cur_devices->seed;
1611 }
1612 return NULL;
1613 }
1614
1615 struct btrfs_device *
1616 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1617 u64 devid, int instance)
1618 {
1619 struct list_head *head = &fs_devices->devices;
1620 struct btrfs_device *dev;
1621 int num_found = 0;
1622
1623 list_for_each_entry(dev, head, dev_list) {
1624 if (dev->devid == devid && num_found++ == instance)
1625 return dev;
1626 }
1627 return NULL;
1628 }
1629
1630 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1631 {
1632 struct cache_extent *ce;
1633 struct map_lookup *map;
1634 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1635 int readonly = 0;
1636 int i;
1637
1638 /*
1639 * During chunk recovering, we may fail to find block group's
1640 * corresponding chunk, we will rebuild it later
1641 */
1642 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1643 if (!root->fs_info->is_chunk_recover)
1644 BUG_ON(!ce);
1645 else
1646 return 0;
1647
1648 map = container_of(ce, struct map_lookup, ce);
1649 for (i = 0; i < map->num_stripes; i++) {
1650 if (!map->stripes[i].dev->writeable) {
1651 readonly = 1;
1652 break;
1653 }
1654 }
1655
1656 return readonly;
1657 }
1658
1659 static struct btrfs_device *fill_missing_device(u64 devid)
1660 {
1661 struct btrfs_device *device;
1662
1663 device = kzalloc(sizeof(*device), GFP_NOFS);
1664 device->devid = devid;
1665 device->fd = -1;
1666 return device;
1667 }
1668
1669 /*
1670 * slot == -1: SYSTEM chunk
1671 * return -EIO on error, otherwise return 0
1672 */
1673 int btrfs_check_chunk_valid(struct btrfs_root *root,
1674 struct extent_buffer *leaf,
1675 struct btrfs_chunk *chunk,
1676 int slot, u64 logical)
1677 {
1678 u64 length;
1679 u64 stripe_len;
1680 u16 num_stripes;
1681 u16 sub_stripes;
1682 u64 type;
1683
1684 length = btrfs_chunk_length(leaf, chunk);
1685 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1686 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1687 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1688 type = btrfs_chunk_type(leaf, chunk);
1689
1690 /*
1691 * These valid checks may be insufficient to cover every corner cases.
1692 */
1693 if (!IS_ALIGNED(logical, root->sectorsize)) {
1694 error("invalid chunk logical %llu", logical);
1695 return -EIO;
1696 }
1697 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1698 error("invalid chunk sectorsize %llu",
1699 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1700 return -EIO;
1701 }
1702 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1703 error("invalid chunk length %llu", length);
1704 return -EIO;
1705 }
1706 if (stripe_len != BTRFS_STRIPE_LEN) {
1707 error("invalid chunk stripe length: %llu", stripe_len);
1708 return -EIO;
1709 }
1710 /* Check on chunk item type */
1711 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1712 error("invalid chunk type %llu", type);
1713 return -EIO;
1714 }
1715 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1716 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1717 error("unrecognized chunk type: %llu",
1718 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1719 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1720 return -EIO;
1721 }
1722 /*
1723 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1724 * it can't exceed the system chunk array size
1725 * For normal chunk, it should match its chunk item size.
1726 */
1727 if (num_stripes < 1 ||
1728 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1729 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1730 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1731 btrfs_item_size_nr(leaf, slot))) {
1732 error("invalid num_stripes: %u", num_stripes);
1733 return -EIO;
1734 }
1735 /*
1736 * Device number check against profile
1737 */
1738 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1739 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1740 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1741 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1742 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1743 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1744 num_stripes != 1)) {
1745 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1746 num_stripes, sub_stripes,
1747 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1748 return -EIO;
1749 }
1750
1751 return 0;
1752 }
1753
1754 /*
1755 * Slot is used to verify the chunk item is valid
1756 *
1757 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1758 */
1759 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1760 struct extent_buffer *leaf,
1761 struct btrfs_chunk *chunk, int slot)
1762 {
1763 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1764 struct map_lookup *map;
1765 struct cache_extent *ce;
1766 u64 logical;
1767 u64 length;
1768 u64 devid;
1769 u8 uuid[BTRFS_UUID_SIZE];
1770 int num_stripes;
1771 int ret;
1772 int i;
1773
1774 logical = key->offset;
1775 length = btrfs_chunk_length(leaf, chunk);
1776 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1777 /* Validation check */
1778 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1779 if (ret) {
1780 error("%s checksums match, but it has an invalid chunk, %s",
1781 (slot == -1) ? "Superblock" : "Metadata",
1782 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1783 return ret;
1784 }
1785
1786 ce = search_cache_extent(&map_tree->cache_tree, logical);
1787
1788 /* already mapped? */
1789 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1790 return 0;
1791 }
1792
1793 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1794 if (!map)
1795 return -ENOMEM;
1796
1797 map->ce.start = logical;
1798 map->ce.size = length;
1799 map->num_stripes = num_stripes;
1800 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1801 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1802 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1803 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1804 map->type = btrfs_chunk_type(leaf, chunk);
1805 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1806
1807 for (i = 0; i < num_stripes; i++) {
1808 map->stripes[i].physical =
1809 btrfs_stripe_offset_nr(leaf, chunk, i);
1810 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1811 read_extent_buffer(leaf, uuid, (unsigned long)
1812 btrfs_stripe_dev_uuid_nr(chunk, i),
1813 BTRFS_UUID_SIZE);
1814 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1815 NULL);
1816 if (!map->stripes[i].dev) {
1817 map->stripes[i].dev = fill_missing_device(devid);
1818 printf("warning, device %llu is missing\n",
1819 (unsigned long long)devid);
1820 list_add(&map->stripes[i].dev->dev_list,
1821 &root->fs_info->fs_devices->devices);
1822 }
1823
1824 }
1825 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1826 BUG_ON(ret);
1827
1828 return 0;
1829 }
1830
1831 static int fill_device_from_item(struct extent_buffer *leaf,
1832 struct btrfs_dev_item *dev_item,
1833 struct btrfs_device *device)
1834 {
1835 unsigned long ptr;
1836
1837 device->devid = btrfs_device_id(leaf, dev_item);
1838 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1839 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1840 device->type = btrfs_device_type(leaf, dev_item);
1841 device->io_align = btrfs_device_io_align(leaf, dev_item);
1842 device->io_width = btrfs_device_io_width(leaf, dev_item);
1843 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1844
1845 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1846 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1847
1848 return 0;
1849 }
1850
1851 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1852 {
1853 struct btrfs_fs_devices *fs_devices;
1854 int ret;
1855
1856 fs_devices = root->fs_info->fs_devices->seed;
1857 while (fs_devices) {
1858 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1859 ret = 0;
1860 goto out;
1861 }
1862 fs_devices = fs_devices->seed;
1863 }
1864
1865 fs_devices = find_fsid(fsid);
1866 if (!fs_devices) {
1867 /* missing all seed devices */
1868 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1869 if (!fs_devices) {
1870 ret = -ENOMEM;
1871 goto out;
1872 }
1873 INIT_LIST_HEAD(&fs_devices->devices);
1874 list_add(&fs_devices->list, &fs_uuids);
1875 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1876 }
1877
1878 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1879 if (ret)
1880 goto out;
1881
1882 fs_devices->seed = root->fs_info->fs_devices->seed;
1883 root->fs_info->fs_devices->seed = fs_devices;
1884 out:
1885 return ret;
1886 }
1887
1888 static int read_one_dev(struct btrfs_root *root,
1889 struct extent_buffer *leaf,
1890 struct btrfs_dev_item *dev_item)
1891 {
1892 struct btrfs_device *device;
1893 u64 devid;
1894 int ret = 0;
1895 u8 fs_uuid[BTRFS_UUID_SIZE];
1896 u8 dev_uuid[BTRFS_UUID_SIZE];
1897
1898 devid = btrfs_device_id(leaf, dev_item);
1899 read_extent_buffer(leaf, dev_uuid,
1900 (unsigned long)btrfs_device_uuid(dev_item),
1901 BTRFS_UUID_SIZE);
1902 read_extent_buffer(leaf, fs_uuid,
1903 (unsigned long)btrfs_device_fsid(dev_item),
1904 BTRFS_UUID_SIZE);
1905
1906 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1907 ret = open_seed_devices(root, fs_uuid);
1908 if (ret)
1909 return ret;
1910 }
1911
1912 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1913 if (!device) {
1914 device = kzalloc(sizeof(*device), GFP_NOFS);
1915 if (!device)
1916 return -ENOMEM;
1917 device->fd = -1;
1918 list_add(&device->dev_list,
1919 &root->fs_info->fs_devices->devices);
1920 }
1921
1922 fill_device_from_item(leaf, dev_item, device);
1923 device->dev_root = root->fs_info->dev_root;
1924 return ret;
1925 }
1926
1927 int btrfs_read_sys_array(struct btrfs_root *root)
1928 {
1929 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1930 struct extent_buffer *sb;
1931 struct btrfs_disk_key *disk_key;
1932 struct btrfs_chunk *chunk;
1933 u8 *array_ptr;
1934 unsigned long sb_array_offset;
1935 int ret = 0;
1936 u32 num_stripes;
1937 u32 array_size;
1938 u32 len = 0;
1939 u32 cur_offset;
1940 struct btrfs_key key;
1941
1942 sb = btrfs_find_create_tree_block(root->fs_info,
1943 BTRFS_SUPER_INFO_OFFSET,
1944 BTRFS_SUPER_INFO_SIZE);
1945 if (!sb)
1946 return -ENOMEM;
1947 btrfs_set_buffer_uptodate(sb);
1948 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1949 array_size = btrfs_super_sys_array_size(super_copy);
1950
1951 array_ptr = super_copy->sys_chunk_array;
1952 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1953 cur_offset = 0;
1954
1955 while (cur_offset < array_size) {
1956 disk_key = (struct btrfs_disk_key *)array_ptr;
1957 len = sizeof(*disk_key);
1958 if (cur_offset + len > array_size)
1959 goto out_short_read;
1960
1961 btrfs_disk_key_to_cpu(&key, disk_key);
1962
1963 array_ptr += len;
1964 sb_array_offset += len;
1965 cur_offset += len;
1966
1967 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1968 chunk = (struct btrfs_chunk *)sb_array_offset;
1969 /*
1970 * At least one btrfs_chunk with one stripe must be
1971 * present, exact stripe count check comes afterwards
1972 */
1973 len = btrfs_chunk_item_size(1);
1974 if (cur_offset + len > array_size)
1975 goto out_short_read;
1976
1977 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1978 if (!num_stripes) {
1979 printk(
1980 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1981 num_stripes, cur_offset);
1982 ret = -EIO;
1983 break;
1984 }
1985
1986 len = btrfs_chunk_item_size(num_stripes);
1987 if (cur_offset + len > array_size)
1988 goto out_short_read;
1989
1990 ret = read_one_chunk(root, &key, sb, chunk, -1);
1991 if (ret)
1992 break;
1993 } else {
1994 printk(
1995 "ERROR: unexpected item type %u in sys_array at offset %u\n",
1996 (u32)key.type, cur_offset);
1997 ret = -EIO;
1998 break;
1999 }
2000 array_ptr += len;
2001 sb_array_offset += len;
2002 cur_offset += len;
2003 }
2004 free_extent_buffer(sb);
2005 return ret;
2006
2007 out_short_read:
2008 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2009 len, cur_offset);
2010 free_extent_buffer(sb);
2011 return -EIO;
2012 }
2013
2014 int btrfs_read_chunk_tree(struct btrfs_root *root)
2015 {
2016 struct btrfs_path *path;
2017 struct extent_buffer *leaf;
2018 struct btrfs_key key;
2019 struct btrfs_key found_key;
2020 int ret;
2021 int slot;
2022
2023 root = root->fs_info->chunk_root;
2024
2025 path = btrfs_alloc_path();
2026 if (!path)
2027 return -ENOMEM;
2028
2029 /*
2030 * Read all device items, and then all the chunk items. All
2031 * device items are found before any chunk item (their object id
2032 * is smaller than the lowest possible object id for a chunk
2033 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2034 */
2035 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2036 key.offset = 0;
2037 key.type = 0;
2038 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2039 if (ret < 0)
2040 goto error;
2041 while(1) {
2042 leaf = path->nodes[0];
2043 slot = path->slots[0];
2044 if (slot >= btrfs_header_nritems(leaf)) {
2045 ret = btrfs_next_leaf(root, path);
2046 if (ret == 0)
2047 continue;
2048 if (ret < 0)
2049 goto error;
2050 break;
2051 }
2052 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2053 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2054 struct btrfs_dev_item *dev_item;
2055 dev_item = btrfs_item_ptr(leaf, slot,
2056 struct btrfs_dev_item);
2057 ret = read_one_dev(root, leaf, dev_item);
2058 BUG_ON(ret);
2059 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2060 struct btrfs_chunk *chunk;
2061 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2062 ret = read_one_chunk(root, &found_key, leaf, chunk,
2063 slot);
2064 BUG_ON(ret);
2065 }
2066 path->slots[0]++;
2067 }
2068
2069 ret = 0;
2070 error:
2071 btrfs_free_path(path);
2072 return ret;
2073 }
2074
2075 struct list_head *btrfs_scanned_uuids(void)
2076 {
2077 return &fs_uuids;
2078 }
2079
2080 static int rmw_eb(struct btrfs_fs_info *info,
2081 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2082 {
2083 int ret;
2084 unsigned long orig_off = 0;
2085 unsigned long dest_off = 0;
2086 unsigned long copy_len = eb->len;
2087
2088 ret = read_whole_eb(info, eb, 0);
2089 if (ret)
2090 return ret;
2091
2092 if (eb->start + eb->len <= orig_eb->start ||
2093 eb->start >= orig_eb->start + orig_eb->len)
2094 return 0;
2095 /*
2096 * | ----- orig_eb ------- |
2097 * | ----- stripe ------- |
2098 * | ----- orig_eb ------- |
2099 * | ----- orig_eb ------- |
2100 */
2101 if (eb->start > orig_eb->start)
2102 orig_off = eb->start - orig_eb->start;
2103 if (orig_eb->start > eb->start)
2104 dest_off = orig_eb->start - eb->start;
2105
2106 if (copy_len > orig_eb->len - orig_off)
2107 copy_len = orig_eb->len - orig_off;
2108 if (copy_len > eb->len - dest_off)
2109 copy_len = eb->len - dest_off;
2110
2111 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2112 return 0;
2113 }
2114
2115 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2116 struct extent_buffer *orig_eb,
2117 struct extent_buffer **ebs,
2118 u64 stripe_len, u64 *raid_map,
2119 int num_stripes)
2120 {
2121 struct extent_buffer **tmp_ebs;
2122 u64 start = orig_eb->start;
2123 u64 this_eb_start;
2124 int i;
2125 int ret = 0;
2126
2127 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2128 if (!tmp_ebs)
2129 return -ENOMEM;
2130
2131 /* Alloc memory in a row for data stripes */
2132 for (i = 0; i < num_stripes; i++) {
2133 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2134 break;
2135
2136 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2137 if (!tmp_ebs[i]) {
2138 ret = -ENOMEM;
2139 goto clean_up;
2140 }
2141 }
2142
2143 for (i = 0; i < num_stripes; i++) {
2144 struct extent_buffer *eb = tmp_ebs[i];
2145
2146 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2147 break;
2148
2149 eb->start = raid_map[i];
2150 eb->len = stripe_len;
2151 eb->refs = 1;
2152 eb->flags = 0;
2153 eb->fd = -1;
2154 eb->dev_bytenr = (u64)-1;
2155
2156 this_eb_start = raid_map[i];
2157
2158 if (start > this_eb_start ||
2159 start + orig_eb->len < this_eb_start + stripe_len) {
2160 ret = rmw_eb(info, eb, orig_eb);
2161 if (ret)
2162 goto clean_up;
2163 } else {
2164 memcpy(eb->data, orig_eb->data + eb->start - start,
2165 stripe_len);
2166 }
2167 ebs[i] = eb;
2168 }
2169 free(tmp_ebs);
2170 return ret;
2171 clean_up:
2172 for (i = 0; i < num_stripes; i++)
2173 free(tmp_ebs[i]);
2174 free(tmp_ebs);
2175 return ret;
2176 }
2177
2178 int write_raid56_with_parity(struct btrfs_fs_info *info,
2179 struct extent_buffer *eb,
2180 struct btrfs_multi_bio *multi,
2181 u64 stripe_len, u64 *raid_map)
2182 {
2183 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2184 int i;
2185 int ret;
2186 int alloc_size = eb->len;
2187 void **pointers;
2188
2189 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2190 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2191 if (!ebs || !pointers) {
2192 free(ebs);
2193 free(pointers);
2194 return -ENOMEM;
2195 }
2196
2197 if (stripe_len > alloc_size)
2198 alloc_size = stripe_len;
2199
2200 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2201 multi->num_stripes);
2202 if (ret)
2203 goto out;
2204
2205 for (i = 0; i < multi->num_stripes; i++) {
2206 struct extent_buffer *new_eb;
2207 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2208 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2209 ebs[i]->fd = multi->stripes[i].dev->fd;
2210 multi->stripes[i].dev->total_ios++;
2211 if (ebs[i]->start != raid_map[i]) {
2212 ret = -EINVAL;
2213 goto out_free_split;
2214 }
2215 continue;
2216 }
2217 new_eb = malloc(sizeof(*eb) + alloc_size);
2218 if (!new_eb) {
2219 ret = -ENOMEM;
2220 goto out_free_split;
2221 }
2222 new_eb->dev_bytenr = multi->stripes[i].physical;
2223 new_eb->fd = multi->stripes[i].dev->fd;
2224 multi->stripes[i].dev->total_ios++;
2225 new_eb->len = stripe_len;
2226
2227 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2228 p_eb = new_eb;
2229 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2230 q_eb = new_eb;
2231 }
2232 if (q_eb) {
2233 ebs[multi->num_stripes - 2] = p_eb;
2234 ebs[multi->num_stripes - 1] = q_eb;
2235
2236 for (i = 0; i < multi->num_stripes; i++)
2237 pointers[i] = ebs[i]->data;
2238
2239 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2240 } else {
2241 ebs[multi->num_stripes - 1] = p_eb;
2242 for (i = 0; i < multi->num_stripes; i++)
2243 pointers[i] = ebs[i]->data;
2244 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2245 multi->num_stripes - 1, pointers);
2246 if (ret < 0)
2247 goto out_free_split;
2248 }
2249
2250 for (i = 0; i < multi->num_stripes; i++) {
2251 ret = write_extent_to_disk(ebs[i]);
2252 if (ret < 0)
2253 goto out_free_split;
2254 }
2255
2256 out_free_split:
2257 for (i = 0; i < multi->num_stripes; i++) {
2258 if (ebs[i] != eb)
2259 free(ebs[i]);
2260 }
2261 out:
2262 free(ebs);
2263 free(pointers);
2264
2265 return ret;
2266 }
(deprecated) Btrfs progs developments and patch integration