?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: report I/O errors when closing the filesystem
[btrfs-progs-unstable/devel.git] / volumes.c
blobb350e259dded91c782ce94a67d6fac14df455ae4
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 int ret = 0;
164
165 again:
166 if (!fs_devices)
167 return 0;
168 while (!list_empty(&fs_devices->devices)) {
169 device = list_entry(fs_devices->devices.next,
170 struct btrfs_device, dev_list);
171 if (device->fd != -1) {
172 if (fsync(device->fd) == -1) {
173 warning("fsync on device %llu failed: %s",
174 device->devid, strerror(errno));
175 ret = -errno;
176 }
177 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
178 fprintf(stderr, "Warning, could not drop caches\n");
179 close(device->fd);
180 device->fd = -1;
181 }
182 device->writeable = 0;
183 list_del(&device->dev_list);
184 /* free the memory */
185 free(device->name);
186 free(device->label);
187 free(device);
188 }
189
190 seed_devices = fs_devices->seed;
191 fs_devices->seed = NULL;
192 if (seed_devices) {
193 struct btrfs_fs_devices *orig;
194
195 orig = fs_devices;
196 fs_devices = seed_devices;
197 list_del(&orig->list);
198 free(orig);
199 goto again;
200 } else {
201 list_del(&fs_devices->list);
202 free(fs_devices);
203 }
204
205 return ret;
206 }
207
208 void btrfs_close_all_devices(void)
209 {
210 struct btrfs_fs_devices *fs_devices;
211
212 while (!list_empty(&fs_uuids)) {
213 fs_devices = list_entry(fs_uuids.next, struct btrfs_fs_devices,
214 list);
215 btrfs_close_devices(fs_devices);
216 }
217 }
218
219 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
220 {
221 int fd;
222 struct list_head *head = &fs_devices->devices;
223 struct list_head *cur;
224 struct btrfs_device *device;
225 int ret;
226
227 list_for_each(cur, head) {
228 device = list_entry(cur, struct btrfs_device, dev_list);
229 if (!device->name) {
230 printk("no name for device %llu, skip it now\n", device->devid);
231 continue;
232 }
233
234 fd = open(device->name, flags);
235 if (fd < 0) {
236 ret = -errno;
237 error("cannot open device '%s': %s", device->name,
238 strerror(errno));
239 goto fail;
240 }
241
242 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
243 fprintf(stderr, "Warning, could not drop caches\n");
244
245 if (device->devid == fs_devices->latest_devid)
246 fs_devices->latest_bdev = fd;
247 if (device->devid == fs_devices->lowest_devid)
248 fs_devices->lowest_bdev = fd;
249 device->fd = fd;
250 if (flags & O_RDWR)
251 device->writeable = 1;
252 }
253 return 0;
254 fail:
255 btrfs_close_devices(fs_devices);
256 return ret;
257 }
258
259 int btrfs_scan_one_device(int fd, const char *path,
260 struct btrfs_fs_devices **fs_devices_ret,
261 u64 *total_devs, u64 super_offset, unsigned sbflags)
262 {
263 struct btrfs_super_block *disk_super;
264 char buf[BTRFS_SUPER_INFO_SIZE];
265 int ret;
266 u64 devid;
267
268 disk_super = (struct btrfs_super_block *)buf;
269 ret = btrfs_read_dev_super(fd, disk_super, super_offset, sbflags);
270 if (ret < 0)
271 return -EIO;
272 devid = btrfs_stack_device_id(&disk_super->dev_item);
273 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
274 *total_devs = 1;
275 else
276 *total_devs = btrfs_super_num_devices(disk_super);
277
278 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
279
280 return ret;
281 }
282
283 /*
284 * find_free_dev_extent_start - find free space in the specified device
285 * @device: the device which we search the free space in
286 * @num_bytes: the size of the free space that we need
287 * @search_start: the position from which to begin the search
288 * @start: store the start of the free space.
289 * @len: the size of the free space. that we find, or the size
290 * of the max free space if we don't find suitable free space
291 *
292 * this uses a pretty simple search, the expectation is that it is
293 * called very infrequently and that a given device has a small number
294 * of extents
295 *
296 * @start is used to store the start of the free space if we find. But if we
297 * don't find suitable free space, it will be used to store the start position
298 * of the max free space.
299 *
300 * @len is used to store the size of the free space that we find.
301 * But if we don't find suitable free space, it is used to store the size of
302 * the max free space.
303 */
304 static int find_free_dev_extent_start(struct btrfs_trans_handle *trans,
305 struct btrfs_device *device, u64 num_bytes,
306 u64 search_start, u64 *start, u64 *len)
307 {
308 struct btrfs_key key;
309 struct btrfs_root *root = device->dev_root;
310 struct btrfs_dev_extent *dev_extent;
311 struct btrfs_path *path;
312 u64 hole_size;
313 u64 max_hole_start;
314 u64 max_hole_size;
315 u64 extent_end;
316 u64 search_end = device->total_bytes;
317 int ret;
318 int slot;
319 struct extent_buffer *l;
320 u64 min_search_start;
321
322 /*
323 * We don't want to overwrite the superblock on the drive nor any area
324 * used by the boot loader (grub for example), so we make sure to start
325 * at an offset of at least 1MB.
326 */
327 min_search_start = max(root->fs_info->alloc_start, (u64)SZ_1M);
328 search_start = max(search_start, min_search_start);
329
330 path = btrfs_alloc_path();
331 if (!path)
332 return -ENOMEM;
333
334 max_hole_start = search_start;
335 max_hole_size = 0;
336
337 if (search_start >= search_end) {
338 ret = -ENOSPC;
339 goto out;
340 }
341
342 path->reada = 2;
343
344 key.objectid = device->devid;
345 key.offset = search_start;
346 key.type = BTRFS_DEV_EXTENT_KEY;
347
348 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
349 if (ret < 0)
350 goto out;
351 if (ret > 0) {
352 ret = btrfs_previous_item(root, path, key.objectid, key.type);
353 if (ret < 0)
354 goto out;
355 }
356
357 while (1) {
358 l = path->nodes[0];
359 slot = path->slots[0];
360 if (slot >= btrfs_header_nritems(l)) {
361 ret = btrfs_next_leaf(root, path);
362 if (ret == 0)
363 continue;
364 if (ret < 0)
365 goto out;
366
367 break;
368 }
369 btrfs_item_key_to_cpu(l, &key, slot);
370
371 if (key.objectid < device->devid)
372 goto next;
373
374 if (key.objectid > device->devid)
375 break;
376
377 if (key.type != BTRFS_DEV_EXTENT_KEY)
378 goto next;
379
380 if (key.offset > search_start) {
381 hole_size = key.offset - search_start;
382
383 /*
384 * Have to check before we set max_hole_start, otherwise
385 * we could end up sending back this offset anyway.
386 */
387 if (hole_size > max_hole_size) {
388 max_hole_start = search_start;
389 max_hole_size = hole_size;
390 }
391
392 /*
393 * If this free space is greater than which we need,
394 * it must be the max free space that we have found
395 * until now, so max_hole_start must point to the start
396 * of this free space and the length of this free space
397 * is stored in max_hole_size. Thus, we return
398 * max_hole_start and max_hole_size and go back to the
399 * caller.
400 */
401 if (hole_size >= num_bytes) {
402 ret = 0;
403 goto out;
404 }
405 }
406
407 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
408 extent_end = key.offset + btrfs_dev_extent_length(l,
409 dev_extent);
410 if (extent_end > search_start)
411 search_start = extent_end;
412 next:
413 path->slots[0]++;
414 cond_resched();
415 }
416
417 /*
418 * At this point, search_start should be the end of
419 * allocated dev extents, and when shrinking the device,
420 * search_end may be smaller than search_start.
421 */
422 if (search_end > search_start) {
423 hole_size = search_end - search_start;
424
425 if (hole_size > max_hole_size) {
426 max_hole_start = search_start;
427 max_hole_size = hole_size;
428 }
429 }
430
431 /* See above. */
432 if (max_hole_size < num_bytes)
433 ret = -ENOSPC;
434 else
435 ret = 0;
436
437 out:
438 btrfs_free_path(path);
439 *start = max_hole_start;
440 if (len)
441 *len = max_hole_size;
442 return ret;
443 }
444
445 int find_free_dev_extent(struct btrfs_trans_handle *trans,
446 struct btrfs_device *device, u64 num_bytes,
447 u64 *start)
448 {
449 /* FIXME use last free of some kind */
450 return find_free_dev_extent_start(trans, device,
451 num_bytes, 0, start, NULL);
452 }
453
454 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
455 struct btrfs_device *device,
456 u64 chunk_tree, u64 chunk_objectid,
457 u64 chunk_offset,
458 u64 num_bytes, u64 *start, int convert)
459 {
460 int ret;
461 struct btrfs_path *path;
462 struct btrfs_root *root = device->dev_root;
463 struct btrfs_dev_extent *extent;
464 struct extent_buffer *leaf;
465 struct btrfs_key key;
466
467 path = btrfs_alloc_path();
468 if (!path)
469 return -ENOMEM;
470
471 /*
472 * For convert case, just skip search free dev_extent, as caller
473 * is responsible to make sure it's free.
474 */
475 if (!convert) {
476 ret = find_free_dev_extent(trans, device, num_bytes,
477 start);
478 if (ret)
479 goto err;
480 }
481
482 key.objectid = device->devid;
483 key.offset = *start;
484 key.type = BTRFS_DEV_EXTENT_KEY;
485 ret = btrfs_insert_empty_item(trans, root, path, &key,
486 sizeof(*extent));
487 BUG_ON(ret);
488
489 leaf = path->nodes[0];
490 extent = btrfs_item_ptr(leaf, path->slots[0],
491 struct btrfs_dev_extent);
492 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
493 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
494 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
495
496 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
497 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
498 BTRFS_UUID_SIZE);
499
500 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
501 btrfs_mark_buffer_dirty(leaf);
502 err:
503 btrfs_free_path(path);
504 return ret;
505 }
506
507 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
508 {
509 struct btrfs_path *path;
510 int ret;
511 struct btrfs_key key;
512 struct btrfs_chunk *chunk;
513 struct btrfs_key found_key;
514
515 path = btrfs_alloc_path();
516 if (!path)
517 return -ENOMEM;
518
519 key.objectid = objectid;
520 key.offset = (u64)-1;
521 key.type = BTRFS_CHUNK_ITEM_KEY;
522
523 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
524 if (ret < 0)
525 goto error;
526
527 BUG_ON(ret == 0);
528
529 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
530 if (ret) {
531 *offset = 0;
532 } else {
533 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
534 path->slots[0]);
535 if (found_key.objectid != objectid)
536 *offset = 0;
537 else {
538 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
539 struct btrfs_chunk);
540 *offset = found_key.offset +
541 btrfs_chunk_length(path->nodes[0], chunk);
542 }
543 }
544 ret = 0;
545 error:
546 btrfs_free_path(path);
547 return ret;
548 }
549
550 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
551 u64 *objectid)
552 {
553 int ret;
554 struct btrfs_key key;
555 struct btrfs_key found_key;
556
557 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
558 key.type = BTRFS_DEV_ITEM_KEY;
559 key.offset = (u64)-1;
560
561 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
562 if (ret < 0)
563 goto error;
564
565 BUG_ON(ret == 0);
566
567 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
568 BTRFS_DEV_ITEM_KEY);
569 if (ret) {
570 *objectid = 1;
571 } else {
572 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
573 path->slots[0]);
574 *objectid = found_key.offset + 1;
575 }
576 ret = 0;
577 error:
578 btrfs_release_path(path);
579 return ret;
580 }
581
582 /*
583 * the device information is stored in the chunk root
584 * the btrfs_device struct should be fully filled in
585 */
586 int btrfs_add_device(struct btrfs_trans_handle *trans,
587 struct btrfs_root *root,
588 struct btrfs_device *device)
589 {
590 int ret;
591 struct btrfs_path *path;
592 struct btrfs_dev_item *dev_item;
593 struct extent_buffer *leaf;
594 struct btrfs_key key;
595 unsigned long ptr;
596 u64 free_devid = 0;
597
598 root = root->fs_info->chunk_root;
599
600 path = btrfs_alloc_path();
601 if (!path)
602 return -ENOMEM;
603
604 ret = find_next_devid(root, path, &free_devid);
605 if (ret)
606 goto out;
607
608 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
609 key.type = BTRFS_DEV_ITEM_KEY;
610 key.offset = free_devid;
611
612 ret = btrfs_insert_empty_item(trans, root, path, &key,
613 sizeof(*dev_item));
614 if (ret)
615 goto out;
616
617 leaf = path->nodes[0];
618 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
619
620 device->devid = free_devid;
621 btrfs_set_device_id(leaf, dev_item, device->devid);
622 btrfs_set_device_generation(leaf, dev_item, 0);
623 btrfs_set_device_type(leaf, dev_item, device->type);
624 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
625 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
626 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
627 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
628 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
629 btrfs_set_device_group(leaf, dev_item, 0);
630 btrfs_set_device_seek_speed(leaf, dev_item, 0);
631 btrfs_set_device_bandwidth(leaf, dev_item, 0);
632 btrfs_set_device_start_offset(leaf, dev_item, 0);
633
634 ptr = (unsigned long)btrfs_device_uuid(dev_item);
635 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
636 ptr = (unsigned long)btrfs_device_fsid(dev_item);
637 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
638 btrfs_mark_buffer_dirty(leaf);
639 ret = 0;
640
641 out:
642 btrfs_free_path(path);
643 return ret;
644 }
645
646 int btrfs_update_device(struct btrfs_trans_handle *trans,
647 struct btrfs_device *device)
648 {
649 int ret;
650 struct btrfs_path *path;
651 struct btrfs_root *root;
652 struct btrfs_dev_item *dev_item;
653 struct extent_buffer *leaf;
654 struct btrfs_key key;
655
656 root = device->dev_root->fs_info->chunk_root;
657
658 path = btrfs_alloc_path();
659 if (!path)
660 return -ENOMEM;
661
662 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
663 key.type = BTRFS_DEV_ITEM_KEY;
664 key.offset = device->devid;
665
666 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
667 if (ret < 0)
668 goto out;
669
670 if (ret > 0) {
671 ret = -ENOENT;
672 goto out;
673 }
674
675 leaf = path->nodes[0];
676 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
677
678 btrfs_set_device_id(leaf, dev_item, device->devid);
679 btrfs_set_device_type(leaf, dev_item, device->type);
680 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
681 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
682 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
683 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
684 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
685 btrfs_mark_buffer_dirty(leaf);
686
687 out:
688 btrfs_free_path(path);
689 return ret;
690 }
691
692 int btrfs_add_system_chunk(struct btrfs_root *root,
693 struct btrfs_key *key,
694 struct btrfs_chunk *chunk, int item_size)
695 {
696 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
697 struct btrfs_disk_key disk_key;
698 u32 array_size;
699 u8 *ptr;
700
701 array_size = btrfs_super_sys_array_size(super_copy);
702 if (array_size + item_size + sizeof(disk_key)
703 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
704 return -EFBIG;
705
706 ptr = super_copy->sys_chunk_array + array_size;
707 btrfs_cpu_key_to_disk(&disk_key, key);
708 memcpy(ptr, &disk_key, sizeof(disk_key));
709 ptr += sizeof(disk_key);
710 memcpy(ptr, chunk, item_size);
711 item_size += sizeof(disk_key);
712 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
713 return 0;
714 }
715
716 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
717 int sub_stripes)
718 {
719 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
720 return calc_size;
721 else if (type & BTRFS_BLOCK_GROUP_RAID10)
722 return calc_size * (num_stripes / sub_stripes);
723 else if (type & BTRFS_BLOCK_GROUP_RAID5)
724 return calc_size * (num_stripes - 1);
725 else if (type & BTRFS_BLOCK_GROUP_RAID6)
726 return calc_size * (num_stripes - 2);
727 else
728 return calc_size * num_stripes;
729 }
730
731
732 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
733 {
734 /* TODO, add a way to store the preferred stripe size */
735 return BTRFS_STRIPE_LEN;
736 }
737
738 /*
739 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
740 *
741 * It is not equal to "device->total_bytes - device->bytes_used".
742 * We do not allocate any chunk in 1M at beginning of device, and not
743 * allowed to allocate any chunk before alloc_start if it is specified.
744 * So search holes from max(1M, alloc_start) to device->total_bytes.
745 */
746 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
747 struct btrfs_device *device,
748 u64 *avail_bytes)
749 {
750 struct btrfs_path *path;
751 struct btrfs_root *root = device->dev_root;
752 struct btrfs_key key;
753 struct btrfs_dev_extent *dev_extent = NULL;
754 struct extent_buffer *l;
755 u64 search_start = root->fs_info->alloc_start;
756 u64 search_end = device->total_bytes;
757 u64 extent_end = 0;
758 u64 free_bytes = 0;
759 int ret;
760 int slot = 0;
761
762 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
763
764 path = btrfs_alloc_path();
765 if (!path)
766 return -ENOMEM;
767
768 key.objectid = device->devid;
769 key.offset = root->fs_info->alloc_start;
770 key.type = BTRFS_DEV_EXTENT_KEY;
771
772 path->reada = 2;
773 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
774 if (ret < 0)
775 goto error;
776 ret = btrfs_previous_item(root, path, 0, key.type);
777 if (ret < 0)
778 goto error;
779
780 while (1) {
781 l = path->nodes[0];
782 slot = path->slots[0];
783 if (slot >= btrfs_header_nritems(l)) {
784 ret = btrfs_next_leaf(root, path);
785 if (ret == 0)
786 continue;
787 if (ret < 0)
788 goto error;
789 break;
790 }
791 btrfs_item_key_to_cpu(l, &key, slot);
792
793 if (key.objectid < device->devid)
794 goto next;
795 if (key.objectid > device->devid)
796 break;
797 if (key.type != BTRFS_DEV_EXTENT_KEY)
798 goto next;
799 if (key.offset > search_end)
800 break;
801 if (key.offset > search_start)
802 free_bytes += key.offset - search_start;
803
804 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
805 extent_end = key.offset + btrfs_dev_extent_length(l,
806 dev_extent);
807 if (extent_end > search_start)
808 search_start = extent_end;
809 if (search_start > search_end)
810 break;
811 next:
812 path->slots[0]++;
813 cond_resched();
814 }
815
816 if (search_start < search_end)
817 free_bytes += search_end - search_start;
818
819 *avail_bytes = free_bytes;
820 ret = 0;
821 error:
822 btrfs_free_path(path);
823 return ret;
824 }
825
826 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
827 - sizeof(struct btrfs_item) \
828 - sizeof(struct btrfs_chunk)) \
829 / sizeof(struct btrfs_stripe) + 1)
830
831 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
832 - 2 * sizeof(struct btrfs_disk_key) \
833 - 2 * sizeof(struct btrfs_chunk)) \
834 / sizeof(struct btrfs_stripe) + 1)
835
836 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
837 struct btrfs_root *extent_root, u64 *start,
838 u64 *num_bytes, u64 type)
839 {
840 u64 dev_offset;
841 struct btrfs_fs_info *info = extent_root->fs_info;
842 struct btrfs_root *chunk_root = info->chunk_root;
843 struct btrfs_stripe *stripes;
844 struct btrfs_device *device = NULL;
845 struct btrfs_chunk *chunk;
846 struct list_head private_devs;
847 struct list_head *dev_list = &info->fs_devices->devices;
848 struct list_head *cur;
849 struct map_lookup *map;
850 int min_stripe_size = SZ_1M;
851 u64 calc_size = SZ_8M;
852 u64 min_free;
853 u64 max_chunk_size = 4 * calc_size;
854 u64 avail = 0;
855 u64 max_avail = 0;
856 u64 percent_max;
857 int num_stripes = 1;
858 int max_stripes = 0;
859 int min_stripes = 1;
860 int sub_stripes = 0;
861 int looped = 0;
862 int ret;
863 int index;
864 int stripe_len = BTRFS_STRIPE_LEN;
865 struct btrfs_key key;
866 u64 offset;
867
868 if (list_empty(dev_list)) {
869 return -ENOSPC;
870 }
871
872 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
873 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
874 BTRFS_BLOCK_GROUP_RAID10 |
875 BTRFS_BLOCK_GROUP_DUP)) {
876 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
877 calc_size = SZ_8M;
878 max_chunk_size = calc_size * 2;
879 min_stripe_size = SZ_1M;
880 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
881 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
882 calc_size = SZ_1G;
883 max_chunk_size = 10 * calc_size;
884 min_stripe_size = SZ_64M;
885 max_stripes = BTRFS_MAX_DEVS(chunk_root);
886 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
887 calc_size = SZ_1G;
888 max_chunk_size = 4 * calc_size;
889 min_stripe_size = SZ_32M;
890 max_stripes = BTRFS_MAX_DEVS(chunk_root);
891 }
892 }
893 if (type & BTRFS_BLOCK_GROUP_RAID1) {
894 num_stripes = min_t(u64, 2,
895 btrfs_super_num_devices(info->super_copy));
896 if (num_stripes < 2)
897 return -ENOSPC;
898 min_stripes = 2;
899 }
900 if (type & BTRFS_BLOCK_GROUP_DUP) {
901 num_stripes = 2;
902 min_stripes = 2;
903 }
904 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
905 num_stripes = btrfs_super_num_devices(info->super_copy);
906 if (num_stripes > max_stripes)
907 num_stripes = max_stripes;
908 min_stripes = 2;
909 }
910 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
911 num_stripes = btrfs_super_num_devices(info->super_copy);
912 if (num_stripes > max_stripes)
913 num_stripes = max_stripes;
914 if (num_stripes < 4)
915 return -ENOSPC;
916 num_stripes &= ~(u32)1;
917 sub_stripes = 2;
918 min_stripes = 4;
919 }
920 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
921 num_stripes = btrfs_super_num_devices(info->super_copy);
922 if (num_stripes > max_stripes)
923 num_stripes = max_stripes;
924 if (num_stripes < 2)
925 return -ENOSPC;
926 min_stripes = 2;
927 stripe_len = find_raid56_stripe_len(num_stripes - 1,
928 btrfs_super_stripesize(info->super_copy));
929 }
930 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
931 num_stripes = btrfs_super_num_devices(info->super_copy);
932 if (num_stripes > max_stripes)
933 num_stripes = max_stripes;
934 if (num_stripes < 3)
935 return -ENOSPC;
936 min_stripes = 3;
937 stripe_len = find_raid56_stripe_len(num_stripes - 2,
938 btrfs_super_stripesize(info->super_copy));
939 }
940
941 /* we don't want a chunk larger than 10% of the FS */
942 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
943 max_chunk_size = min(percent_max, max_chunk_size);
944
945 again:
946 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
947 max_chunk_size) {
948 calc_size = max_chunk_size;
949 calc_size /= num_stripes;
950 calc_size /= stripe_len;
951 calc_size *= stripe_len;
952 }
953 /* we don't want tiny stripes */
954 calc_size = max_t(u64, calc_size, min_stripe_size);
955
956 calc_size /= stripe_len;
957 calc_size *= stripe_len;
958 INIT_LIST_HEAD(&private_devs);
959 cur = dev_list->next;
960 index = 0;
961
962 if (type & BTRFS_BLOCK_GROUP_DUP)
963 min_free = calc_size * 2;
964 else
965 min_free = calc_size;
966
967 /* build a private list of devices we will allocate from */
968 while(index < num_stripes) {
969 device = list_entry(cur, struct btrfs_device, dev_list);
970 ret = btrfs_device_avail_bytes(trans, device, &avail);
971 if (ret)
972 return ret;
973 cur = cur->next;
974 if (avail >= min_free) {
975 list_move_tail(&device->dev_list, &private_devs);
976 index++;
977 if (type & BTRFS_BLOCK_GROUP_DUP)
978 index++;
979 } else if (avail > max_avail)
980 max_avail = avail;
981 if (cur == dev_list)
982 break;
983 }
984 if (index < num_stripes) {
985 list_splice(&private_devs, dev_list);
986 if (index >= min_stripes) {
987 num_stripes = index;
988 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
989 num_stripes /= sub_stripes;
990 num_stripes *= sub_stripes;
991 }
992 looped = 1;
993 goto again;
994 }
995 if (!looped && max_avail > 0) {
996 looped = 1;
997 calc_size = max_avail;
998 goto again;
999 }
1000 return -ENOSPC;
1001 }
1002 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1003 &offset);
1004 if (ret)
1005 return ret;
1006 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1007 key.type = BTRFS_CHUNK_ITEM_KEY;
1008 key.offset = offset;
1009
1010 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1011 if (!chunk)
1012 return -ENOMEM;
1013
1014 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1015 if (!map) {
1016 kfree(chunk);
1017 return -ENOMEM;
1018 }
1019
1020 stripes = &chunk->stripe;
1021 *num_bytes = chunk_bytes_by_type(type, calc_size,
1022 num_stripes, sub_stripes);
1023 index = 0;
1024 while(index < num_stripes) {
1025 struct btrfs_stripe *stripe;
1026 BUG_ON(list_empty(&private_devs));
1027 cur = private_devs.next;
1028 device = list_entry(cur, struct btrfs_device, dev_list);
1029
1030 /* loop over this device again if we're doing a dup group */
1031 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
1032 (index == num_stripes - 1))
1033 list_move_tail(&device->dev_list, dev_list);
1034
1035 ret = btrfs_alloc_dev_extent(trans, device,
1036 info->chunk_root->root_key.objectid,
1037 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1038 calc_size, &dev_offset, 0);
1039 BUG_ON(ret);
1040
1041 device->bytes_used += calc_size;
1042 ret = btrfs_update_device(trans, device);
1043 BUG_ON(ret);
1044
1045 map->stripes[index].dev = device;
1046 map->stripes[index].physical = dev_offset;
1047 stripe = stripes + index;
1048 btrfs_set_stack_stripe_devid(stripe, device->devid);
1049 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1050 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1051 index++;
1052 }
1053 BUG_ON(!list_empty(&private_devs));
1054
1055 /* key was set above */
1056 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1057 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1058 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1059 btrfs_set_stack_chunk_type(chunk, type);
1060 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1061 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1062 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1063 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1064 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1065 map->sector_size = extent_root->sectorsize;
1066 map->stripe_len = stripe_len;
1067 map->io_align = stripe_len;
1068 map->io_width = stripe_len;
1069 map->type = type;
1070 map->num_stripes = num_stripes;
1071 map->sub_stripes = sub_stripes;
1072
1073 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1074 btrfs_chunk_item_size(num_stripes));
1075 BUG_ON(ret);
1076 *start = key.offset;;
1077
1078 map->ce.start = key.offset;
1079 map->ce.size = *num_bytes;
1080
1081 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1082 BUG_ON(ret);
1083
1084 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1085 ret = btrfs_add_system_chunk(chunk_root, &key,
1086 chunk, btrfs_chunk_item_size(num_stripes));
1087 BUG_ON(ret);
1088 }
1089
1090 kfree(chunk);
1091 return ret;
1092 }
1093
1094 /*
1095 * Alloc a DATA chunk with SINGLE profile.
1096 *
1097 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1098 * (btrfs logical bytenr == on-disk bytenr)
1099 * For that case, caller must make sure the chunk and dev_extent are not
1100 * occupied.
1101 */
1102 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1103 struct btrfs_root *extent_root, u64 *start,
1104 u64 num_bytes, u64 type, int convert)
1105 {
1106 u64 dev_offset;
1107 struct btrfs_fs_info *info = extent_root->fs_info;
1108 struct btrfs_root *chunk_root = info->chunk_root;
1109 struct btrfs_stripe *stripes;
1110 struct btrfs_device *device = NULL;
1111 struct btrfs_chunk *chunk;
1112 struct list_head *dev_list = &info->fs_devices->devices;
1113 struct list_head *cur;
1114 struct map_lookup *map;
1115 u64 calc_size = SZ_8M;
1116 int num_stripes = 1;
1117 int sub_stripes = 0;
1118 int ret;
1119 int index;
1120 int stripe_len = BTRFS_STRIPE_LEN;
1121 struct btrfs_key key;
1122
1123 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1124 key.type = BTRFS_CHUNK_ITEM_KEY;
1125 if (convert) {
1126 if (*start != round_down(*start, extent_root->sectorsize)) {
1127 error("DATA chunk start not sectorsize aligned: %llu",
1128 (unsigned long long)*start);
1129 return -EINVAL;
1130 }
1131 key.offset = *start;
1132 dev_offset = *start;
1133 } else {
1134 u64 tmp;
1135
1136 ret = find_next_chunk(chunk_root,
1137 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1138 &tmp);
1139 key.offset = tmp;
1140 if (ret)
1141 return ret;
1142 }
1143
1144 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1145 if (!chunk)
1146 return -ENOMEM;
1147
1148 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1149 if (!map) {
1150 kfree(chunk);
1151 return -ENOMEM;
1152 }
1153
1154 stripes = &chunk->stripe;
1155 calc_size = num_bytes;
1156
1157 index = 0;
1158 cur = dev_list->next;
1159 device = list_entry(cur, struct btrfs_device, dev_list);
1160
1161 while (index < num_stripes) {
1162 struct btrfs_stripe *stripe;
1163
1164 ret = btrfs_alloc_dev_extent(trans, device,
1165 info->chunk_root->root_key.objectid,
1166 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1167 calc_size, &dev_offset, convert);
1168 BUG_ON(ret);
1169
1170 device->bytes_used += calc_size;
1171 ret = btrfs_update_device(trans, device);
1172 BUG_ON(ret);
1173
1174 map->stripes[index].dev = device;
1175 map->stripes[index].physical = dev_offset;
1176 stripe = stripes + index;
1177 btrfs_set_stack_stripe_devid(stripe, device->devid);
1178 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1179 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1180 index++;
1181 }
1182
1183 /* key was set above */
1184 btrfs_set_stack_chunk_length(chunk, num_bytes);
1185 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1186 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1187 btrfs_set_stack_chunk_type(chunk, type);
1188 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1189 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1190 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1191 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1192 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1193 map->sector_size = extent_root->sectorsize;
1194 map->stripe_len = stripe_len;
1195 map->io_align = stripe_len;
1196 map->io_width = stripe_len;
1197 map->type = type;
1198 map->num_stripes = num_stripes;
1199 map->sub_stripes = sub_stripes;
1200
1201 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1202 btrfs_chunk_item_size(num_stripes));
1203 BUG_ON(ret);
1204 if (!convert)
1205 *start = key.offset;
1206
1207 map->ce.start = key.offset;
1208 map->ce.size = num_bytes;
1209
1210 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1211 BUG_ON(ret);
1212
1213 kfree(chunk);
1214 return ret;
1215 }
1216
1217 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1218 {
1219 struct cache_extent *ce;
1220 struct map_lookup *map;
1221 int ret;
1222
1223 ce = search_cache_extent(&map_tree->cache_tree, logical);
1224 if (!ce) {
1225 fprintf(stderr, "No mapping for %llu-%llu\n",
1226 (unsigned long long)logical,
1227 (unsigned long long)logical+len);
1228 return 1;
1229 }
1230 if (ce->start > logical || ce->start + ce->size < logical) {
1231 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1232 "%llu-%llu\n", (unsigned long long)logical,
1233 (unsigned long long)logical+len,
1234 (unsigned long long)ce->start,
1235 (unsigned long long)ce->start + ce->size);
1236 return 1;
1237 }
1238 map = container_of(ce, struct map_lookup, ce);
1239
1240 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1241 ret = map->num_stripes;
1242 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1243 ret = map->sub_stripes;
1244 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1245 ret = 2;
1246 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1247 ret = 3;
1248 else
1249 ret = 1;
1250 return ret;
1251 }
1252
1253 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1254 u64 *size, u64 type)
1255 {
1256 struct cache_extent *ce;
1257 struct map_lookup *map;
1258 u64 cur = *logical;
1259
1260 ce = search_cache_extent(&map_tree->cache_tree, cur);
1261
1262 while (ce) {
1263 /*
1264 * only jump to next bg if our cur is not 0
1265 * As the initial logical for btrfs_next_bg() is 0, and
1266 * if we jump to next bg, we skipped a valid bg.
1267 */
1268 if (cur) {
1269 ce = next_cache_extent(ce);
1270 if (!ce)
1271 return -ENOENT;
1272 }
1273
1274 cur = ce->start;
1275 map = container_of(ce, struct map_lookup, ce);
1276 if (map->type & type) {
1277 *logical = ce->start;
1278 *size = ce->size;
1279 return 0;
1280 }
1281 }
1282
1283 return -ENOENT;
1284 }
1285
1286 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1287 u64 chunk_start, u64 physical, u64 devid,
1288 u64 **logical, int *naddrs, int *stripe_len)
1289 {
1290 struct cache_extent *ce;
1291 struct map_lookup *map;
1292 u64 *buf;
1293 u64 bytenr;
1294 u64 length;
1295 u64 stripe_nr;
1296 u64 rmap_len;
1297 int i, j, nr = 0;
1298
1299 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1300 BUG_ON(!ce);
1301 map = container_of(ce, struct map_lookup, ce);
1302
1303 length = ce->size;
1304 rmap_len = map->stripe_len;
1305 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1306 length = ce->size / (map->num_stripes / map->sub_stripes);
1307 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1308 length = ce->size / map->num_stripes;
1309 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1310 BTRFS_BLOCK_GROUP_RAID6)) {
1311 length = ce->size / nr_data_stripes(map);
1312 rmap_len = map->stripe_len * nr_data_stripes(map);
1313 }
1314
1315 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1316
1317 for (i = 0; i < map->num_stripes; i++) {
1318 if (devid && map->stripes[i].dev->devid != devid)
1319 continue;
1320 if (map->stripes[i].physical > physical ||
1321 map->stripes[i].physical + length <= physical)
1322 continue;
1323
1324 stripe_nr = (physical - map->stripes[i].physical) /
1325 map->stripe_len;
1326
1327 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1328 stripe_nr = (stripe_nr * map->num_stripes + i) /
1329 map->sub_stripes;
1330 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1331 stripe_nr = stripe_nr * map->num_stripes + i;
1332 } /* else if RAID[56], multiply by nr_data_stripes().
1333 * Alternatively, just use rmap_len below instead of
1334 * map->stripe_len */
1335
1336 bytenr = ce->start + stripe_nr * rmap_len;
1337 for (j = 0; j < nr; j++) {
1338 if (buf[j] == bytenr)
1339 break;
1340 }
1341 if (j == nr)
1342 buf[nr++] = bytenr;
1343 }
1344
1345 *logical = buf;
1346 *naddrs = nr;
1347 *stripe_len = rmap_len;
1348
1349 return 0;
1350 }
1351
1352 static inline int parity_smaller(u64 a, u64 b)
1353 {
1354 return a > b;
1355 }
1356
1357 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1358 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1359 {
1360 struct btrfs_bio_stripe s;
1361 int i;
1362 u64 l;
1363 int again = 1;
1364
1365 while (again) {
1366 again = 0;
1367 for (i = 0; i < bbio->num_stripes - 1; i++) {
1368 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1369 s = bbio->stripes[i];
1370 l = raid_map[i];
1371 bbio->stripes[i] = bbio->stripes[i+1];
1372 raid_map[i] = raid_map[i+1];
1373 bbio->stripes[i+1] = s;
1374 raid_map[i+1] = l;
1375 again = 1;
1376 }
1377 }
1378 }
1379 }
1380
1381 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1382 u64 logical, u64 *length,
1383 struct btrfs_multi_bio **multi_ret, int mirror_num,
1384 u64 **raid_map_ret)
1385 {
1386 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1387 multi_ret, mirror_num, raid_map_ret);
1388 }
1389
1390 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1391 u64 logical, u64 *length, u64 *type,
1392 struct btrfs_multi_bio **multi_ret, int mirror_num,
1393 u64 **raid_map_ret)
1394 {
1395 struct cache_extent *ce;
1396 struct map_lookup *map;
1397 u64 offset;
1398 u64 stripe_offset;
1399 u64 stripe_nr;
1400 u64 *raid_map = NULL;
1401 int stripes_allocated = 8;
1402 int stripes_required = 1;
1403 int stripe_index;
1404 int i;
1405 struct btrfs_multi_bio *multi = NULL;
1406
1407 if (multi_ret && rw == READ) {
1408 stripes_allocated = 1;
1409 }
1410 again:
1411 ce = search_cache_extent(&map_tree->cache_tree, logical);
1412 if (!ce) {
1413 kfree(multi);
1414 *length = (u64)-1;
1415 return -ENOENT;
1416 }
1417 if (ce->start > logical) {
1418 kfree(multi);
1419 *length = ce->start - logical;
1420 return -ENOENT;
1421 }
1422
1423 if (multi_ret) {
1424 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1425 GFP_NOFS);
1426 if (!multi)
1427 return -ENOMEM;
1428 }
1429 map = container_of(ce, struct map_lookup, ce);
1430 offset = logical - ce->start;
1431
1432 if (rw == WRITE) {
1433 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1434 BTRFS_BLOCK_GROUP_DUP)) {
1435 stripes_required = map->num_stripes;
1436 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1437 stripes_required = map->sub_stripes;
1438 }
1439 }
1440 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1441 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1442 /* RAID[56] write or recovery. Return all stripes */
1443 stripes_required = map->num_stripes;
1444
1445 /* Only allocate the map if we've already got a large enough multi_ret */
1446 if (stripes_allocated >= stripes_required) {
1447 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1448 if (!raid_map) {
1449 kfree(multi);
1450 return -ENOMEM;
1451 }
1452 }
1453 }
1454
1455 /* if our multi bio struct is too small, back off and try again */
1456 if (multi_ret && stripes_allocated < stripes_required) {
1457 stripes_allocated = stripes_required;
1458 kfree(multi);
1459 multi = NULL;
1460 goto again;
1461 }
1462 stripe_nr = offset;
1463 /*
1464 * stripe_nr counts the total number of stripes we have to stride
1465 * to get to this block
1466 */
1467 stripe_nr = stripe_nr / map->stripe_len;
1468
1469 stripe_offset = stripe_nr * map->stripe_len;
1470 BUG_ON(offset < stripe_offset);
1471
1472 /* stripe_offset is the offset of this block in its stripe*/
1473 stripe_offset = offset - stripe_offset;
1474
1475 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1476 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1477 BTRFS_BLOCK_GROUP_RAID10 |
1478 BTRFS_BLOCK_GROUP_DUP)) {
1479 /* we limit the length of each bio to what fits in a stripe */
1480 *length = min_t(u64, ce->size - offset,
1481 map->stripe_len - stripe_offset);
1482 } else {
1483 *length = ce->size - offset;
1484 }
1485
1486 if (!multi_ret)
1487 goto out;
1488
1489 multi->num_stripes = 1;
1490 stripe_index = 0;
1491 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1492 if (rw == WRITE)
1493 multi->num_stripes = map->num_stripes;
1494 else if (mirror_num)
1495 stripe_index = mirror_num - 1;
1496 else
1497 stripe_index = stripe_nr % map->num_stripes;
1498 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1499 int factor = map->num_stripes / map->sub_stripes;
1500
1501 stripe_index = stripe_nr % factor;
1502 stripe_index *= map->sub_stripes;
1503
1504 if (rw == WRITE)
1505 multi->num_stripes = map->sub_stripes;
1506 else if (mirror_num)
1507 stripe_index += mirror_num - 1;
1508
1509 stripe_nr = stripe_nr / factor;
1510 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1511 if (rw == WRITE)
1512 multi->num_stripes = map->num_stripes;
1513 else if (mirror_num)
1514 stripe_index = mirror_num - 1;
1515 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1516 BTRFS_BLOCK_GROUP_RAID6)) {
1517
1518 if (raid_map) {
1519 int rot;
1520 u64 tmp;
1521 u64 raid56_full_stripe_start;
1522 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1523
1524 /*
1525 * align the start of our data stripe in the logical
1526 * address space
1527 */
1528 raid56_full_stripe_start = offset / full_stripe_len;
1529 raid56_full_stripe_start *= full_stripe_len;
1530
1531 /* get the data stripe number */
1532 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1533 stripe_nr = stripe_nr / nr_data_stripes(map);
1534
1535 /* Work out the disk rotation on this stripe-set */
1536 rot = stripe_nr % map->num_stripes;
1537
1538 /* Fill in the logical address of each stripe */
1539 tmp = stripe_nr * nr_data_stripes(map);
1540
1541 for (i = 0; i < nr_data_stripes(map); i++)
1542 raid_map[(i+rot) % map->num_stripes] =
1543 ce->start + (tmp + i) * map->stripe_len;
1544
1545 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1546 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1547 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1548
1549 *length = map->stripe_len;
1550 stripe_index = 0;
1551 stripe_offset = 0;
1552 multi->num_stripes = map->num_stripes;
1553 } else {
1554 stripe_index = stripe_nr % nr_data_stripes(map);
1555 stripe_nr = stripe_nr / nr_data_stripes(map);
1556
1557 /*
1558 * Mirror #0 or #1 means the original data block.
1559 * Mirror #2 is RAID5 parity block.
1560 * Mirror #3 is RAID6 Q block.
1561 */
1562 if (mirror_num > 1)
1563 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1564
1565 /* We distribute the parity blocks across stripes */
1566 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1567 }
1568 } else {
1569 /*
1570 * after this do_div call, stripe_nr is the number of stripes
1571 * on this device we have to walk to find the data, and
1572 * stripe_index is the number of our device in the stripe array
1573 */
1574 stripe_index = stripe_nr % map->num_stripes;
1575 stripe_nr = stripe_nr / map->num_stripes;
1576 }
1577 BUG_ON(stripe_index >= map->num_stripes);
1578
1579 for (i = 0; i < multi->num_stripes; i++) {
1580 multi->stripes[i].physical =
1581 map->stripes[stripe_index].physical + stripe_offset +
1582 stripe_nr * map->stripe_len;
1583 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1584 stripe_index++;
1585 }
1586 *multi_ret = multi;
1587
1588 if (type)
1589 *type = map->type;
1590
1591 if (raid_map) {
1592 sort_parity_stripes(multi, raid_map);
1593 *raid_map_ret = raid_map;
1594 }
1595 out:
1596 return 0;
1597 }
1598
1599 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1600 u8 *uuid, u8 *fsid)
1601 {
1602 struct btrfs_device *device;
1603 struct btrfs_fs_devices *cur_devices;
1604
1605 cur_devices = root->fs_info->fs_devices;
1606 while (cur_devices) {
1607 if (!fsid ||
1608 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1609 root->fs_info->ignore_fsid_mismatch)) {
1610 device = __find_device(&cur_devices->devices,
1611 devid, uuid);
1612 if (device)
1613 return device;
1614 }
1615 cur_devices = cur_devices->seed;
1616 }
1617 return NULL;
1618 }
1619
1620 struct btrfs_device *
1621 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1622 u64 devid, int instance)
1623 {
1624 struct list_head *head = &fs_devices->devices;
1625 struct btrfs_device *dev;
1626 int num_found = 0;
1627
1628 list_for_each_entry(dev, head, dev_list) {
1629 if (dev->devid == devid && num_found++ == instance)
1630 return dev;
1631 }
1632 return NULL;
1633 }
1634
1635 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1636 {
1637 struct cache_extent *ce;
1638 struct map_lookup *map;
1639 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1640 int readonly = 0;
1641 int i;
1642
1643 /*
1644 * During chunk recovering, we may fail to find block group's
1645 * corresponding chunk, we will rebuild it later
1646 */
1647 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1648 if (!root->fs_info->is_chunk_recover)
1649 BUG_ON(!ce);
1650 else
1651 return 0;
1652
1653 map = container_of(ce, struct map_lookup, ce);
1654 for (i = 0; i < map->num_stripes; i++) {
1655 if (!map->stripes[i].dev->writeable) {
1656 readonly = 1;
1657 break;
1658 }
1659 }
1660
1661 return readonly;
1662 }
1663
1664 static struct btrfs_device *fill_missing_device(u64 devid)
1665 {
1666 struct btrfs_device *device;
1667
1668 device = kzalloc(sizeof(*device), GFP_NOFS);
1669 device->devid = devid;
1670 device->fd = -1;
1671 return device;
1672 }
1673
1674 /*
1675 * slot == -1: SYSTEM chunk
1676 * return -EIO on error, otherwise return 0
1677 */
1678 int btrfs_check_chunk_valid(struct btrfs_root *root,
1679 struct extent_buffer *leaf,
1680 struct btrfs_chunk *chunk,
1681 int slot, u64 logical)
1682 {
1683 u64 length;
1684 u64 stripe_len;
1685 u16 num_stripes;
1686 u16 sub_stripes;
1687 u64 type;
1688
1689 length = btrfs_chunk_length(leaf, chunk);
1690 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1691 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1692 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1693 type = btrfs_chunk_type(leaf, chunk);
1694
1695 /*
1696 * These valid checks may be insufficient to cover every corner cases.
1697 */
1698 if (!IS_ALIGNED(logical, root->sectorsize)) {
1699 error("invalid chunk logical %llu", logical);
1700 return -EIO;
1701 }
1702 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1703 error("invalid chunk sectorsize %llu",
1704 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1705 return -EIO;
1706 }
1707 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1708 error("invalid chunk length %llu", length);
1709 return -EIO;
1710 }
1711 if (stripe_len != BTRFS_STRIPE_LEN) {
1712 error("invalid chunk stripe length: %llu", stripe_len);
1713 return -EIO;
1714 }
1715 /* Check on chunk item type */
1716 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1717 error("invalid chunk type %llu", type);
1718 return -EIO;
1719 }
1720 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1721 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1722 error("unrecognized chunk type: %llu",
1723 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1724 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1725 return -EIO;
1726 }
1727 /*
1728 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1729 * it can't exceed the system chunk array size
1730 * For normal chunk, it should match its chunk item size.
1731 */
1732 if (num_stripes < 1 ||
1733 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1734 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1735 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1736 btrfs_item_size_nr(leaf, slot))) {
1737 error("invalid num_stripes: %u", num_stripes);
1738 return -EIO;
1739 }
1740 /*
1741 * Device number check against profile
1742 */
1743 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1744 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1745 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1746 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1747 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1748 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1749 num_stripes != 1)) {
1750 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1751 num_stripes, sub_stripes,
1752 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1753 return -EIO;
1754 }
1755
1756 return 0;
1757 }
1758
1759 /*
1760 * Slot is used to verify the chunk item is valid
1761 *
1762 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1763 */
1764 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1765 struct extent_buffer *leaf,
1766 struct btrfs_chunk *chunk, int slot)
1767 {
1768 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1769 struct map_lookup *map;
1770 struct cache_extent *ce;
1771 u64 logical;
1772 u64 length;
1773 u64 devid;
1774 u8 uuid[BTRFS_UUID_SIZE];
1775 int num_stripes;
1776 int ret;
1777 int i;
1778
1779 logical = key->offset;
1780 length = btrfs_chunk_length(leaf, chunk);
1781 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1782 /* Validation check */
1783 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1784 if (ret) {
1785 error("%s checksums match, but it has an invalid chunk, %s",
1786 (slot == -1) ? "Superblock" : "Metadata",
1787 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1788 return ret;
1789 }
1790
1791 ce = search_cache_extent(&map_tree->cache_tree, logical);
1792
1793 /* already mapped? */
1794 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1795 return 0;
1796 }
1797
1798 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1799 if (!map)
1800 return -ENOMEM;
1801
1802 map->ce.start = logical;
1803 map->ce.size = length;
1804 map->num_stripes = num_stripes;
1805 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1806 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1807 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1808 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1809 map->type = btrfs_chunk_type(leaf, chunk);
1810 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1811
1812 for (i = 0; i < num_stripes; i++) {
1813 map->stripes[i].physical =
1814 btrfs_stripe_offset_nr(leaf, chunk, i);
1815 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1816 read_extent_buffer(leaf, uuid, (unsigned long)
1817 btrfs_stripe_dev_uuid_nr(chunk, i),
1818 BTRFS_UUID_SIZE);
1819 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1820 NULL);
1821 if (!map->stripes[i].dev) {
1822 map->stripes[i].dev = fill_missing_device(devid);
1823 printf("warning, device %llu is missing\n",
1824 (unsigned long long)devid);
1825 list_add(&map->stripes[i].dev->dev_list,
1826 &root->fs_info->fs_devices->devices);
1827 }
1828
1829 }
1830 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1831 BUG_ON(ret);
1832
1833 return 0;
1834 }
1835
1836 static int fill_device_from_item(struct extent_buffer *leaf,
1837 struct btrfs_dev_item *dev_item,
1838 struct btrfs_device *device)
1839 {
1840 unsigned long ptr;
1841
1842 device->devid = btrfs_device_id(leaf, dev_item);
1843 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1844 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1845 device->type = btrfs_device_type(leaf, dev_item);
1846 device->io_align = btrfs_device_io_align(leaf, dev_item);
1847 device->io_width = btrfs_device_io_width(leaf, dev_item);
1848 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1849
1850 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1851 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1852
1853 return 0;
1854 }
1855
1856 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1857 {
1858 struct btrfs_fs_devices *fs_devices;
1859 int ret;
1860
1861 fs_devices = root->fs_info->fs_devices->seed;
1862 while (fs_devices) {
1863 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1864 ret = 0;
1865 goto out;
1866 }
1867 fs_devices = fs_devices->seed;
1868 }
1869
1870 fs_devices = find_fsid(fsid);
1871 if (!fs_devices) {
1872 /* missing all seed devices */
1873 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1874 if (!fs_devices) {
1875 ret = -ENOMEM;
1876 goto out;
1877 }
1878 INIT_LIST_HEAD(&fs_devices->devices);
1879 list_add(&fs_devices->list, &fs_uuids);
1880 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1881 }
1882
1883 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1884 if (ret)
1885 goto out;
1886
1887 fs_devices->seed = root->fs_info->fs_devices->seed;
1888 root->fs_info->fs_devices->seed = fs_devices;
1889 out:
1890 return ret;
1891 }
1892
1893 static int read_one_dev(struct btrfs_root *root,
1894 struct extent_buffer *leaf,
1895 struct btrfs_dev_item *dev_item)
1896 {
1897 struct btrfs_device *device;
1898 u64 devid;
1899 int ret = 0;
1900 u8 fs_uuid[BTRFS_UUID_SIZE];
1901 u8 dev_uuid[BTRFS_UUID_SIZE];
1902
1903 devid = btrfs_device_id(leaf, dev_item);
1904 read_extent_buffer(leaf, dev_uuid,
1905 (unsigned long)btrfs_device_uuid(dev_item),
1906 BTRFS_UUID_SIZE);
1907 read_extent_buffer(leaf, fs_uuid,
1908 (unsigned long)btrfs_device_fsid(dev_item),
1909 BTRFS_UUID_SIZE);
1910
1911 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1912 ret = open_seed_devices(root, fs_uuid);
1913 if (ret)
1914 return ret;
1915 }
1916
1917 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1918 if (!device) {
1919 device = kzalloc(sizeof(*device), GFP_NOFS);
1920 if (!device)
1921 return -ENOMEM;
1922 device->fd = -1;
1923 list_add(&device->dev_list,
1924 &root->fs_info->fs_devices->devices);
1925 }
1926
1927 fill_device_from_item(leaf, dev_item, device);
1928 device->dev_root = root->fs_info->dev_root;
1929 return ret;
1930 }
1931
1932 int btrfs_read_sys_array(struct btrfs_root *root)
1933 {
1934 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1935 struct extent_buffer *sb;
1936 struct btrfs_disk_key *disk_key;
1937 struct btrfs_chunk *chunk;
1938 u8 *array_ptr;
1939 unsigned long sb_array_offset;
1940 int ret = 0;
1941 u32 num_stripes;
1942 u32 array_size;
1943 u32 len = 0;
1944 u32 cur_offset;
1945 struct btrfs_key key;
1946
1947 sb = btrfs_find_create_tree_block(root->fs_info,
1948 BTRFS_SUPER_INFO_OFFSET,
1949 BTRFS_SUPER_INFO_SIZE);
1950 if (!sb)
1951 return -ENOMEM;
1952 btrfs_set_buffer_uptodate(sb);
1953 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1954 array_size = btrfs_super_sys_array_size(super_copy);
1955
1956 array_ptr = super_copy->sys_chunk_array;
1957 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1958 cur_offset = 0;
1959
1960 while (cur_offset < array_size) {
1961 disk_key = (struct btrfs_disk_key *)array_ptr;
1962 len = sizeof(*disk_key);
1963 if (cur_offset + len > array_size)
1964 goto out_short_read;
1965
1966 btrfs_disk_key_to_cpu(&key, disk_key);
1967
1968 array_ptr += len;
1969 sb_array_offset += len;
1970 cur_offset += len;
1971
1972 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1973 chunk = (struct btrfs_chunk *)sb_array_offset;
1974 /*
1975 * At least one btrfs_chunk with one stripe must be
1976 * present, exact stripe count check comes afterwards
1977 */
1978 len = btrfs_chunk_item_size(1);
1979 if (cur_offset + len > array_size)
1980 goto out_short_read;
1981
1982 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1983 if (!num_stripes) {
1984 printk(
1985 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1986 num_stripes, cur_offset);
1987 ret = -EIO;
1988 break;
1989 }
1990
1991 len = btrfs_chunk_item_size(num_stripes);
1992 if (cur_offset + len > array_size)
1993 goto out_short_read;
1994
1995 ret = read_one_chunk(root, &key, sb, chunk, -1);
1996 if (ret)
1997 break;
1998 } else {
1999 printk(
2000 "ERROR: unexpected item type %u in sys_array at offset %u\n",
2001 (u32)key.type, cur_offset);
2002 ret = -EIO;
2003 break;
2004 }
2005 array_ptr += len;
2006 sb_array_offset += len;
2007 cur_offset += len;
2008 }
2009 free_extent_buffer(sb);
2010 return ret;
2011
2012 out_short_read:
2013 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
2014 len, cur_offset);
2015 free_extent_buffer(sb);
2016 return -EIO;
2017 }
2018
2019 int btrfs_read_chunk_tree(struct btrfs_root *root)
2020 {
2021 struct btrfs_path *path;
2022 struct extent_buffer *leaf;
2023 struct btrfs_key key;
2024 struct btrfs_key found_key;
2025 int ret;
2026 int slot;
2027
2028 root = root->fs_info->chunk_root;
2029
2030 path = btrfs_alloc_path();
2031 if (!path)
2032 return -ENOMEM;
2033
2034 /*
2035 * Read all device items, and then all the chunk items. All
2036 * device items are found before any chunk item (their object id
2037 * is smaller than the lowest possible object id for a chunk
2038 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
2039 */
2040 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
2041 key.offset = 0;
2042 key.type = 0;
2043 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
2044 if (ret < 0)
2045 goto error;
2046 while(1) {
2047 leaf = path->nodes[0];
2048 slot = path->slots[0];
2049 if (slot >= btrfs_header_nritems(leaf)) {
2050 ret = btrfs_next_leaf(root, path);
2051 if (ret == 0)
2052 continue;
2053 if (ret < 0)
2054 goto error;
2055 break;
2056 }
2057 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2058 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2059 struct btrfs_dev_item *dev_item;
2060 dev_item = btrfs_item_ptr(leaf, slot,
2061 struct btrfs_dev_item);
2062 ret = read_one_dev(root, leaf, dev_item);
2063 BUG_ON(ret);
2064 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2065 struct btrfs_chunk *chunk;
2066 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2067 ret = read_one_chunk(root, &found_key, leaf, chunk,
2068 slot);
2069 BUG_ON(ret);
2070 }
2071 path->slots[0]++;
2072 }
2073
2074 ret = 0;
2075 error:
2076 btrfs_free_path(path);
2077 return ret;
2078 }
2079
2080 struct list_head *btrfs_scanned_uuids(void)
2081 {
2082 return &fs_uuids;
2083 }
2084
2085 static int rmw_eb(struct btrfs_fs_info *info,
2086 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2087 {
2088 int ret;
2089 unsigned long orig_off = 0;
2090 unsigned long dest_off = 0;
2091 unsigned long copy_len = eb->len;
2092
2093 ret = read_whole_eb(info, eb, 0);
2094 if (ret)
2095 return ret;
2096
2097 if (eb->start + eb->len <= orig_eb->start ||
2098 eb->start >= orig_eb->start + orig_eb->len)
2099 return 0;
2100 /*
2101 * | ----- orig_eb ------- |
2102 * | ----- stripe ------- |
2103 * | ----- orig_eb ------- |
2104 * | ----- orig_eb ------- |
2105 */
2106 if (eb->start > orig_eb->start)
2107 orig_off = eb->start - orig_eb->start;
2108 if (orig_eb->start > eb->start)
2109 dest_off = orig_eb->start - eb->start;
2110
2111 if (copy_len > orig_eb->len - orig_off)
2112 copy_len = orig_eb->len - orig_off;
2113 if (copy_len > eb->len - dest_off)
2114 copy_len = eb->len - dest_off;
2115
2116 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2117 return 0;
2118 }
2119
2120 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2121 struct extent_buffer *orig_eb,
2122 struct extent_buffer **ebs,
2123 u64 stripe_len, u64 *raid_map,
2124 int num_stripes)
2125 {
2126 struct extent_buffer **tmp_ebs;
2127 u64 start = orig_eb->start;
2128 u64 this_eb_start;
2129 int i;
2130 int ret = 0;
2131
2132 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2133 if (!tmp_ebs)
2134 return -ENOMEM;
2135
2136 /* Alloc memory in a row for data stripes */
2137 for (i = 0; i < num_stripes; i++) {
2138 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2139 break;
2140
2141 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2142 if (!tmp_ebs[i]) {
2143 ret = -ENOMEM;
2144 goto clean_up;
2145 }
2146 }
2147
2148 for (i = 0; i < num_stripes; i++) {
2149 struct extent_buffer *eb = tmp_ebs[i];
2150
2151 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2152 break;
2153
2154 eb->start = raid_map[i];
2155 eb->len = stripe_len;
2156 eb->refs = 1;
2157 eb->flags = 0;
2158 eb->fd = -1;
2159 eb->dev_bytenr = (u64)-1;
2160
2161 this_eb_start = raid_map[i];
2162
2163 if (start > this_eb_start ||
2164 start + orig_eb->len < this_eb_start + stripe_len) {
2165 ret = rmw_eb(info, eb, orig_eb);
2166 if (ret)
2167 goto clean_up;
2168 } else {
2169 memcpy(eb->data, orig_eb->data + eb->start - start,
2170 stripe_len);
2171 }
2172 ebs[i] = eb;
2173 }
2174 free(tmp_ebs);
2175 return ret;
2176 clean_up:
2177 for (i = 0; i < num_stripes; i++)
2178 free(tmp_ebs[i]);
2179 free(tmp_ebs);
2180 return ret;
2181 }
2182
2183 int write_raid56_with_parity(struct btrfs_fs_info *info,
2184 struct extent_buffer *eb,
2185 struct btrfs_multi_bio *multi,
2186 u64 stripe_len, u64 *raid_map)
2187 {
2188 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2189 int i;
2190 int ret;
2191 int alloc_size = eb->len;
2192 void **pointers;
2193
2194 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2195 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2196 if (!ebs || !pointers) {
2197 free(ebs);
2198 free(pointers);
2199 return -ENOMEM;
2200 }
2201
2202 if (stripe_len > alloc_size)
2203 alloc_size = stripe_len;
2204
2205 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2206 multi->num_stripes);
2207 if (ret)
2208 goto out;
2209
2210 for (i = 0; i < multi->num_stripes; i++) {
2211 struct extent_buffer *new_eb;
2212 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2213 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2214 ebs[i]->fd = multi->stripes[i].dev->fd;
2215 multi->stripes[i].dev->total_ios++;
2216 if (ebs[i]->start != raid_map[i]) {
2217 ret = -EINVAL;
2218 goto out_free_split;
2219 }
2220 continue;
2221 }
2222 new_eb = malloc(sizeof(*eb) + alloc_size);
2223 if (!new_eb) {
2224 ret = -ENOMEM;
2225 goto out_free_split;
2226 }
2227 new_eb->dev_bytenr = multi->stripes[i].physical;
2228 new_eb->fd = multi->stripes[i].dev->fd;
2229 multi->stripes[i].dev->total_ios++;
2230 new_eb->len = stripe_len;
2231
2232 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2233 p_eb = new_eb;
2234 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2235 q_eb = new_eb;
2236 }
2237 if (q_eb) {
2238 ebs[multi->num_stripes - 2] = p_eb;
2239 ebs[multi->num_stripes - 1] = q_eb;
2240
2241 for (i = 0; i < multi->num_stripes; i++)
2242 pointers[i] = ebs[i]->data;
2243
2244 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2245 } else {
2246 ebs[multi->num_stripes - 1] = p_eb;
2247 for (i = 0; i < multi->num_stripes; i++)
2248 pointers[i] = ebs[i]->data;
2249 ret = raid5_gen_result(multi->num_stripes, stripe_len,
2250 multi->num_stripes - 1, pointers);
2251 if (ret < 0)
2252 goto out_free_split;
2253 }
2254
2255 for (i = 0; i < multi->num_stripes; i++) {
2256 ret = write_extent_to_disk(ebs[i]);
2257 if (ret < 0)
2258 goto out_free_split;
2259 }
2260
2261 out_free_split:
2262 for (i = 0; i < multi->num_stripes; i++) {
2263 if (ebs[i] != eb)
2264 free(ebs[i]);
2265 }
2266 out:
2267 free(ebs);
2268 free(pointers);
2269
2270 return ret;
2271 }
(deprecated) Btrfs progs developments and patch integration