?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: format FREE_SPACE_TREE{,_VALID} nicely in dump-super
[btrfs-progs-unstable/devel.git] / volumes.c
blobe39f21ea1ea5491d741918a1a997a49ab0e673c4
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 * this uses a pretty simple search, the expectation is that it is
280 * called very infrequently and that a given device has a small number
281 * of extents
282 */
283 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
284 struct btrfs_device *device,
285 struct btrfs_path *path,
286 u64 num_bytes, u64 *start)
287 {
288 struct btrfs_key key;
289 struct btrfs_root *root = device->dev_root;
290 struct btrfs_dev_extent *dev_extent = NULL;
291 u64 hole_size = 0;
292 u64 last_byte = 0;
293 u64 search_start = root->fs_info->alloc_start;
294 u64 search_end = device->total_bytes;
295 int ret;
296 int slot = 0;
297 int start_found;
298 struct extent_buffer *l;
299
300 start_found = 0;
301 path->reada = 2;
302
303 /* FIXME use last free of some kind */
304
305 /* we don't want to overwrite the superblock on the drive,
306 * so we make sure to start at an offset of at least 1MB
307 */
308 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
309
310 if (search_start >= search_end) {
311 ret = -ENOSPC;
312 goto error;
313 }
314
315 key.objectid = device->devid;
316 key.offset = search_start;
317 key.type = BTRFS_DEV_EXTENT_KEY;
318 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
319 if (ret < 0)
320 goto error;
321 ret = btrfs_previous_item(root, path, 0, key.type);
322 if (ret < 0)
323 goto error;
324 l = path->nodes[0];
325 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
326 while (1) {
327 l = path->nodes[0];
328 slot = path->slots[0];
329 if (slot >= btrfs_header_nritems(l)) {
330 ret = btrfs_next_leaf(root, path);
331 if (ret == 0)
332 continue;
333 if (ret < 0)
334 goto error;
335 no_more_items:
336 if (!start_found) {
337 if (search_start >= search_end) {
338 ret = -ENOSPC;
339 goto error;
340 }
341 *start = search_start;
342 start_found = 1;
343 goto check_pending;
344 }
345 *start = last_byte > search_start ?
346 last_byte : search_start;
347 if (search_end <= *start) {
348 ret = -ENOSPC;
349 goto error;
350 }
351 goto check_pending;
352 }
353 btrfs_item_key_to_cpu(l, &key, slot);
354
355 if (key.objectid < device->devid)
356 goto next;
357
358 if (key.objectid > device->devid)
359 goto no_more_items;
360
361 if (key.offset >= search_start && key.offset > last_byte &&
362 start_found) {
363 if (last_byte < search_start)
364 last_byte = search_start;
365 hole_size = key.offset - last_byte;
366 if (key.offset > last_byte &&
367 hole_size >= num_bytes) {
368 *start = last_byte;
369 goto check_pending;
370 }
371 }
372 if (key.type != BTRFS_DEV_EXTENT_KEY) {
373 goto next;
374 }
375
376 start_found = 1;
377 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
378 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
379 next:
380 path->slots[0]++;
381 cond_resched();
382 }
383 check_pending:
384 /* we have to make sure we didn't find an extent that has already
385 * been allocated by the map tree or the original allocation
386 */
387 btrfs_release_path(path);
388 BUG_ON(*start < search_start);
389
390 if (*start + num_bytes > search_end) {
391 ret = -ENOSPC;
392 goto error;
393 }
394 /* check for pending inserts here */
395 return 0;
396
397 error:
398 btrfs_release_path(path);
399 return ret;
400 }
401
402 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
403 struct btrfs_device *device,
404 u64 chunk_tree, u64 chunk_objectid,
405 u64 chunk_offset,
406 u64 num_bytes, u64 *start, int convert)
407 {
408 int ret;
409 struct btrfs_path *path;
410 struct btrfs_root *root = device->dev_root;
411 struct btrfs_dev_extent *extent;
412 struct extent_buffer *leaf;
413 struct btrfs_key key;
414
415 path = btrfs_alloc_path();
416 if (!path)
417 return -ENOMEM;
418
419 /*
420 * For convert case, just skip search free dev_extent, as caller
421 * is responsible to make sure it's free.
422 */
423 if (!convert) {
424 ret = find_free_dev_extent(trans, device, path, num_bytes,
425 start);
426 if (ret)
427 goto err;
428 }
429
430 key.objectid = device->devid;
431 key.offset = *start;
432 key.type = BTRFS_DEV_EXTENT_KEY;
433 ret = btrfs_insert_empty_item(trans, root, path, &key,
434 sizeof(*extent));
435 BUG_ON(ret);
436
437 leaf = path->nodes[0];
438 extent = btrfs_item_ptr(leaf, path->slots[0],
439 struct btrfs_dev_extent);
440 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
441 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
442 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
443
444 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
445 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
446 BTRFS_UUID_SIZE);
447
448 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
449 btrfs_mark_buffer_dirty(leaf);
450 err:
451 btrfs_free_path(path);
452 return ret;
453 }
454
455 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
456 {
457 struct btrfs_path *path;
458 int ret;
459 struct btrfs_key key;
460 struct btrfs_chunk *chunk;
461 struct btrfs_key found_key;
462
463 path = btrfs_alloc_path();
464 if (!path)
465 return -ENOMEM;
466
467 key.objectid = objectid;
468 key.offset = (u64)-1;
469 key.type = BTRFS_CHUNK_ITEM_KEY;
470
471 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
472 if (ret < 0)
473 goto error;
474
475 BUG_ON(ret == 0);
476
477 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
478 if (ret) {
479 *offset = 0;
480 } else {
481 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
482 path->slots[0]);
483 if (found_key.objectid != objectid)
484 *offset = 0;
485 else {
486 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
487 struct btrfs_chunk);
488 *offset = found_key.offset +
489 btrfs_chunk_length(path->nodes[0], chunk);
490 }
491 }
492 ret = 0;
493 error:
494 btrfs_free_path(path);
495 return ret;
496 }
497
498 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
499 u64 *objectid)
500 {
501 int ret;
502 struct btrfs_key key;
503 struct btrfs_key found_key;
504
505 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
506 key.type = BTRFS_DEV_ITEM_KEY;
507 key.offset = (u64)-1;
508
509 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
510 if (ret < 0)
511 goto error;
512
513 BUG_ON(ret == 0);
514
515 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
516 BTRFS_DEV_ITEM_KEY);
517 if (ret) {
518 *objectid = 1;
519 } else {
520 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
521 path->slots[0]);
522 *objectid = found_key.offset + 1;
523 }
524 ret = 0;
525 error:
526 btrfs_release_path(path);
527 return ret;
528 }
529
530 /*
531 * the device information is stored in the chunk root
532 * the btrfs_device struct should be fully filled in
533 */
534 int btrfs_add_device(struct btrfs_trans_handle *trans,
535 struct btrfs_root *root,
536 struct btrfs_device *device)
537 {
538 int ret;
539 struct btrfs_path *path;
540 struct btrfs_dev_item *dev_item;
541 struct extent_buffer *leaf;
542 struct btrfs_key key;
543 unsigned long ptr;
544 u64 free_devid = 0;
545
546 root = root->fs_info->chunk_root;
547
548 path = btrfs_alloc_path();
549 if (!path)
550 return -ENOMEM;
551
552 ret = find_next_devid(root, path, &free_devid);
553 if (ret)
554 goto out;
555
556 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
557 key.type = BTRFS_DEV_ITEM_KEY;
558 key.offset = free_devid;
559
560 ret = btrfs_insert_empty_item(trans, root, path, &key,
561 sizeof(*dev_item));
562 if (ret)
563 goto out;
564
565 leaf = path->nodes[0];
566 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
567
568 device->devid = free_devid;
569 btrfs_set_device_id(leaf, dev_item, device->devid);
570 btrfs_set_device_generation(leaf, dev_item, 0);
571 btrfs_set_device_type(leaf, dev_item, device->type);
572 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
573 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
574 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
575 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
576 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
577 btrfs_set_device_group(leaf, dev_item, 0);
578 btrfs_set_device_seek_speed(leaf, dev_item, 0);
579 btrfs_set_device_bandwidth(leaf, dev_item, 0);
580 btrfs_set_device_start_offset(leaf, dev_item, 0);
581
582 ptr = (unsigned long)btrfs_device_uuid(dev_item);
583 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
584 ptr = (unsigned long)btrfs_device_fsid(dev_item);
585 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
586 btrfs_mark_buffer_dirty(leaf);
587 ret = 0;
588
589 out:
590 btrfs_free_path(path);
591 return ret;
592 }
593
594 int btrfs_update_device(struct btrfs_trans_handle *trans,
595 struct btrfs_device *device)
596 {
597 int ret;
598 struct btrfs_path *path;
599 struct btrfs_root *root;
600 struct btrfs_dev_item *dev_item;
601 struct extent_buffer *leaf;
602 struct btrfs_key key;
603
604 root = device->dev_root->fs_info->chunk_root;
605
606 path = btrfs_alloc_path();
607 if (!path)
608 return -ENOMEM;
609
610 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
611 key.type = BTRFS_DEV_ITEM_KEY;
612 key.offset = device->devid;
613
614 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
615 if (ret < 0)
616 goto out;
617
618 if (ret > 0) {
619 ret = -ENOENT;
620 goto out;
621 }
622
623 leaf = path->nodes[0];
624 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
625
626 btrfs_set_device_id(leaf, dev_item, device->devid);
627 btrfs_set_device_type(leaf, dev_item, device->type);
628 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
629 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
630 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
631 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
632 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
633 btrfs_mark_buffer_dirty(leaf);
634
635 out:
636 btrfs_free_path(path);
637 return ret;
638 }
639
640 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
641 struct btrfs_root *root,
642 struct btrfs_key *key,
643 struct btrfs_chunk *chunk, int item_size)
644 {
645 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
646 struct btrfs_disk_key disk_key;
647 u32 array_size;
648 u8 *ptr;
649
650 array_size = btrfs_super_sys_array_size(super_copy);
651 if (array_size + item_size + sizeof(disk_key)
652 > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
653 return -EFBIG;
654
655 ptr = super_copy->sys_chunk_array + array_size;
656 btrfs_cpu_key_to_disk(&disk_key, key);
657 memcpy(ptr, &disk_key, sizeof(disk_key));
658 ptr += sizeof(disk_key);
659 memcpy(ptr, chunk, item_size);
660 item_size += sizeof(disk_key);
661 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
662 return 0;
663 }
664
665 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
666 int sub_stripes)
667 {
668 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
669 return calc_size;
670 else if (type & BTRFS_BLOCK_GROUP_RAID10)
671 return calc_size * (num_stripes / sub_stripes);
672 else if (type & BTRFS_BLOCK_GROUP_RAID5)
673 return calc_size * (num_stripes - 1);
674 else if (type & BTRFS_BLOCK_GROUP_RAID6)
675 return calc_size * (num_stripes - 2);
676 else
677 return calc_size * num_stripes;
678 }
679
680
681 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
682 {
683 /* TODO, add a way to store the preferred stripe size */
684 return BTRFS_STRIPE_LEN;
685 }
686
687 /*
688 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
689 *
690 * It is not equal to "device->total_bytes - device->bytes_used".
691 * We do not allocate any chunk in 1M at beginning of device, and not
692 * allowed to allocate any chunk before alloc_start if it is specified.
693 * So search holes from max(1M, alloc_start) to device->total_bytes.
694 */
695 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
696 struct btrfs_device *device,
697 u64 *avail_bytes)
698 {
699 struct btrfs_path *path;
700 struct btrfs_root *root = device->dev_root;
701 struct btrfs_key key;
702 struct btrfs_dev_extent *dev_extent = NULL;
703 struct extent_buffer *l;
704 u64 search_start = root->fs_info->alloc_start;
705 u64 search_end = device->total_bytes;
706 u64 extent_end = 0;
707 u64 free_bytes = 0;
708 int ret;
709 int slot = 0;
710
711 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
712
713 path = btrfs_alloc_path();
714 if (!path)
715 return -ENOMEM;
716
717 key.objectid = device->devid;
718 key.offset = root->fs_info->alloc_start;
719 key.type = BTRFS_DEV_EXTENT_KEY;
720
721 path->reada = 2;
722 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
723 if (ret < 0)
724 goto error;
725 ret = btrfs_previous_item(root, path, 0, key.type);
726 if (ret < 0)
727 goto error;
728
729 while (1) {
730 l = path->nodes[0];
731 slot = path->slots[0];
732 if (slot >= btrfs_header_nritems(l)) {
733 ret = btrfs_next_leaf(root, path);
734 if (ret == 0)
735 continue;
736 if (ret < 0)
737 goto error;
738 break;
739 }
740 btrfs_item_key_to_cpu(l, &key, slot);
741
742 if (key.objectid < device->devid)
743 goto next;
744 if (key.objectid > device->devid)
745 break;
746 if (key.type != BTRFS_DEV_EXTENT_KEY)
747 goto next;
748 if (key.offset > search_end)
749 break;
750 if (key.offset > search_start)
751 free_bytes += key.offset - search_start;
752
753 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
754 extent_end = key.offset + btrfs_dev_extent_length(l,
755 dev_extent);
756 if (extent_end > search_start)
757 search_start = extent_end;
758 if (search_start > search_end)
759 break;
760 next:
761 path->slots[0]++;
762 cond_resched();
763 }
764
765 if (search_start < search_end)
766 free_bytes += search_end - search_start;
767
768 *avail_bytes = free_bytes;
769 ret = 0;
770 error:
771 btrfs_free_path(path);
772 return ret;
773 }
774
775 #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \
776 - sizeof(struct btrfs_item) \
777 - sizeof(struct btrfs_chunk)) \
778 / sizeof(struct btrfs_stripe) + 1)
779
780 #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \
781 - 2 * sizeof(struct btrfs_disk_key) \
782 - 2 * sizeof(struct btrfs_chunk)) \
783 / sizeof(struct btrfs_stripe) + 1)
784
785 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
786 struct btrfs_root *extent_root, u64 *start,
787 u64 *num_bytes, u64 type)
788 {
789 u64 dev_offset;
790 struct btrfs_fs_info *info = extent_root->fs_info;
791 struct btrfs_root *chunk_root = info->chunk_root;
792 struct btrfs_stripe *stripes;
793 struct btrfs_device *device = NULL;
794 struct btrfs_chunk *chunk;
795 struct list_head private_devs;
796 struct list_head *dev_list = &info->fs_devices->devices;
797 struct list_head *cur;
798 struct map_lookup *map;
799 int min_stripe_size = 1 * 1024 * 1024;
800 u64 calc_size = 8 * 1024 * 1024;
801 u64 min_free;
802 u64 max_chunk_size = 4 * calc_size;
803 u64 avail = 0;
804 u64 max_avail = 0;
805 u64 percent_max;
806 int num_stripes = 1;
807 int max_stripes = 0;
808 int min_stripes = 1;
809 int sub_stripes = 0;
810 int looped = 0;
811 int ret;
812 int index;
813 int stripe_len = BTRFS_STRIPE_LEN;
814 struct btrfs_key key;
815 u64 offset;
816
817 if (list_empty(dev_list)) {
818 return -ENOSPC;
819 }
820
821 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
822 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
823 BTRFS_BLOCK_GROUP_RAID10 |
824 BTRFS_BLOCK_GROUP_DUP)) {
825 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
826 calc_size = 8 * 1024 * 1024;
827 max_chunk_size = calc_size * 2;
828 min_stripe_size = 1 * 1024 * 1024;
829 max_stripes = BTRFS_MAX_DEVS_SYS_CHUNK;
830 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
831 calc_size = 1024 * 1024 * 1024;
832 max_chunk_size = 10 * calc_size;
833 min_stripe_size = 64 * 1024 * 1024;
834 max_stripes = BTRFS_MAX_DEVS(chunk_root);
835 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
836 calc_size = 1024 * 1024 * 1024;
837 max_chunk_size = 4 * calc_size;
838 min_stripe_size = 32 * 1024 * 1024;
839 max_stripes = BTRFS_MAX_DEVS(chunk_root);
840 }
841 }
842 if (type & BTRFS_BLOCK_GROUP_RAID1) {
843 num_stripes = min_t(u64, 2,
844 btrfs_super_num_devices(info->super_copy));
845 if (num_stripes < 2)
846 return -ENOSPC;
847 min_stripes = 2;
848 }
849 if (type & BTRFS_BLOCK_GROUP_DUP) {
850 num_stripes = 2;
851 min_stripes = 2;
852 }
853 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
854 num_stripes = btrfs_super_num_devices(info->super_copy);
855 if (num_stripes > max_stripes)
856 num_stripes = max_stripes;
857 min_stripes = 2;
858 }
859 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
860 num_stripes = btrfs_super_num_devices(info->super_copy);
861 if (num_stripes > max_stripes)
862 num_stripes = max_stripes;
863 if (num_stripes < 4)
864 return -ENOSPC;
865 num_stripes &= ~(u32)1;
866 sub_stripes = 2;
867 min_stripes = 4;
868 }
869 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
870 num_stripes = btrfs_super_num_devices(info->super_copy);
871 if (num_stripes > max_stripes)
872 num_stripes = max_stripes;
873 if (num_stripes < 2)
874 return -ENOSPC;
875 min_stripes = 2;
876 stripe_len = find_raid56_stripe_len(num_stripes - 1,
877 btrfs_super_stripesize(info->super_copy));
878 }
879 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
880 num_stripes = btrfs_super_num_devices(info->super_copy);
881 if (num_stripes > max_stripes)
882 num_stripes = max_stripes;
883 if (num_stripes < 3)
884 return -ENOSPC;
885 min_stripes = 3;
886 stripe_len = find_raid56_stripe_len(num_stripes - 2,
887 btrfs_super_stripesize(info->super_copy));
888 }
889
890 /* we don't want a chunk larger than 10% of the FS */
891 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
892 max_chunk_size = min(percent_max, max_chunk_size);
893
894 again:
895 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
896 max_chunk_size) {
897 calc_size = max_chunk_size;
898 calc_size /= num_stripes;
899 calc_size /= stripe_len;
900 calc_size *= stripe_len;
901 }
902 /* we don't want tiny stripes */
903 calc_size = max_t(u64, calc_size, min_stripe_size);
904
905 calc_size /= stripe_len;
906 calc_size *= stripe_len;
907 INIT_LIST_HEAD(&private_devs);
908 cur = dev_list->next;
909 index = 0;
910
911 if (type & BTRFS_BLOCK_GROUP_DUP)
912 min_free = calc_size * 2;
913 else
914 min_free = calc_size;
915
916 /* build a private list of devices we will allocate from */
917 while(index < num_stripes) {
918 device = list_entry(cur, struct btrfs_device, dev_list);
919 ret = btrfs_device_avail_bytes(trans, device, &avail);
920 if (ret)
921 return ret;
922 cur = cur->next;
923 if (avail >= min_free) {
924 list_move_tail(&device->dev_list, &private_devs);
925 index++;
926 if (type & BTRFS_BLOCK_GROUP_DUP)
927 index++;
928 } else if (avail > max_avail)
929 max_avail = avail;
930 if (cur == dev_list)
931 break;
932 }
933 if (index < num_stripes) {
934 list_splice(&private_devs, dev_list);
935 if (index >= min_stripes) {
936 num_stripes = index;
937 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
938 num_stripes /= sub_stripes;
939 num_stripes *= sub_stripes;
940 }
941 looped = 1;
942 goto again;
943 }
944 if (!looped && max_avail > 0) {
945 looped = 1;
946 calc_size = max_avail;
947 goto again;
948 }
949 return -ENOSPC;
950 }
951 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
952 &offset);
953 if (ret)
954 return ret;
955 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
956 key.type = BTRFS_CHUNK_ITEM_KEY;
957 key.offset = offset;
958
959 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
960 if (!chunk)
961 return -ENOMEM;
962
963 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
964 if (!map) {
965 kfree(chunk);
966 return -ENOMEM;
967 }
968
969 stripes = &chunk->stripe;
970 *num_bytes = chunk_bytes_by_type(type, calc_size,
971 num_stripes, sub_stripes);
972 index = 0;
973 while(index < num_stripes) {
974 struct btrfs_stripe *stripe;
975 BUG_ON(list_empty(&private_devs));
976 cur = private_devs.next;
977 device = list_entry(cur, struct btrfs_device, dev_list);
978
979 /* loop over this device again if we're doing a dup group */
980 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
981 (index == num_stripes - 1))
982 list_move_tail(&device->dev_list, dev_list);
983
984 ret = btrfs_alloc_dev_extent(trans, device,
985 info->chunk_root->root_key.objectid,
986 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
987 calc_size, &dev_offset, 0);
988 BUG_ON(ret);
989
990 device->bytes_used += calc_size;
991 ret = btrfs_update_device(trans, device);
992 BUG_ON(ret);
993
994 map->stripes[index].dev = device;
995 map->stripes[index].physical = dev_offset;
996 stripe = stripes + index;
997 btrfs_set_stack_stripe_devid(stripe, device->devid);
998 btrfs_set_stack_stripe_offset(stripe, dev_offset);
999 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1000 index++;
1001 }
1002 BUG_ON(!list_empty(&private_devs));
1003
1004 /* key was set above */
1005 btrfs_set_stack_chunk_length(chunk, *num_bytes);
1006 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1007 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1008 btrfs_set_stack_chunk_type(chunk, type);
1009 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1010 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1011 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1012 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1013 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1014 map->sector_size = extent_root->sectorsize;
1015 map->stripe_len = stripe_len;
1016 map->io_align = stripe_len;
1017 map->io_width = stripe_len;
1018 map->type = type;
1019 map->num_stripes = num_stripes;
1020 map->sub_stripes = sub_stripes;
1021
1022 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1023 btrfs_chunk_item_size(num_stripes));
1024 BUG_ON(ret);
1025 *start = key.offset;;
1026
1027 map->ce.start = key.offset;
1028 map->ce.size = *num_bytes;
1029
1030 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1031 BUG_ON(ret);
1032
1033 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
1034 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
1035 chunk, btrfs_chunk_item_size(num_stripes));
1036 BUG_ON(ret);
1037 }
1038
1039 kfree(chunk);
1040 return ret;
1041 }
1042
1043 /*
1044 * Alloc a DATA chunk with SINGLE profile.
1045 *
1046 * If 'convert' is set, it will alloc a chunk with 1:1 mapping
1047 * (btrfs logical bytenr == on-disk bytenr)
1048 * For that case, caller must make sure the chunk and dev_extent are not
1049 * occupied.
1050 */
1051 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1052 struct btrfs_root *extent_root, u64 *start,
1053 u64 num_bytes, u64 type, int convert)
1054 {
1055 u64 dev_offset;
1056 struct btrfs_fs_info *info = extent_root->fs_info;
1057 struct btrfs_root *chunk_root = info->chunk_root;
1058 struct btrfs_stripe *stripes;
1059 struct btrfs_device *device = NULL;
1060 struct btrfs_chunk *chunk;
1061 struct list_head *dev_list = &info->fs_devices->devices;
1062 struct list_head *cur;
1063 struct map_lookup *map;
1064 u64 calc_size = 8 * 1024 * 1024;
1065 int num_stripes = 1;
1066 int sub_stripes = 0;
1067 int ret;
1068 int index;
1069 int stripe_len = BTRFS_STRIPE_LEN;
1070 struct btrfs_key key;
1071
1072 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1073 key.type = BTRFS_CHUNK_ITEM_KEY;
1074 if (convert) {
1075 if (*start != round_down(*start, extent_root->sectorsize)) {
1076 error("DATA chunk start not sectorsize aligned: %llu",
1077 (unsigned long long)*start);
1078 return -EINVAL;
1079 }
1080 key.offset = *start;
1081 dev_offset = *start;
1082 } else {
1083 u64 tmp;
1084
1085 ret = find_next_chunk(chunk_root,
1086 BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1087 &tmp);
1088 key.offset = tmp;
1089 if (ret)
1090 return ret;
1091 }
1092
1093 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1094 if (!chunk)
1095 return -ENOMEM;
1096
1097 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1098 if (!map) {
1099 kfree(chunk);
1100 return -ENOMEM;
1101 }
1102
1103 stripes = &chunk->stripe;
1104 calc_size = num_bytes;
1105
1106 index = 0;
1107 cur = dev_list->next;
1108 device = list_entry(cur, struct btrfs_device, dev_list);
1109
1110 while (index < num_stripes) {
1111 struct btrfs_stripe *stripe;
1112
1113 ret = btrfs_alloc_dev_extent(trans, device,
1114 info->chunk_root->root_key.objectid,
1115 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1116 calc_size, &dev_offset, convert);
1117 BUG_ON(ret);
1118
1119 device->bytes_used += calc_size;
1120 ret = btrfs_update_device(trans, device);
1121 BUG_ON(ret);
1122
1123 map->stripes[index].dev = device;
1124 map->stripes[index].physical = dev_offset;
1125 stripe = stripes + index;
1126 btrfs_set_stack_stripe_devid(stripe, device->devid);
1127 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1128 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1129 index++;
1130 }
1131
1132 /* key was set above */
1133 btrfs_set_stack_chunk_length(chunk, num_bytes);
1134 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1135 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1136 btrfs_set_stack_chunk_type(chunk, type);
1137 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1138 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1139 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1140 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1141 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1142 map->sector_size = extent_root->sectorsize;
1143 map->stripe_len = stripe_len;
1144 map->io_align = stripe_len;
1145 map->io_width = stripe_len;
1146 map->type = type;
1147 map->num_stripes = num_stripes;
1148 map->sub_stripes = sub_stripes;
1149
1150 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1151 btrfs_chunk_item_size(num_stripes));
1152 BUG_ON(ret);
1153 if (!convert)
1154 *start = key.offset;
1155
1156 map->ce.start = key.offset;
1157 map->ce.size = num_bytes;
1158
1159 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1160 BUG_ON(ret);
1161
1162 kfree(chunk);
1163 return ret;
1164 }
1165
1166 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1167 {
1168 struct cache_extent *ce;
1169 struct map_lookup *map;
1170 int ret;
1171
1172 ce = search_cache_extent(&map_tree->cache_tree, logical);
1173 if (!ce) {
1174 fprintf(stderr, "No mapping for %llu-%llu\n",
1175 (unsigned long long)logical,
1176 (unsigned long long)logical+len);
1177 return 1;
1178 }
1179 if (ce->start > logical || ce->start + ce->size < logical) {
1180 fprintf(stderr, "Invalid mapping for %llu-%llu, got "
1181 "%llu-%llu\n", (unsigned long long)logical,
1182 (unsigned long long)logical+len,
1183 (unsigned long long)ce->start,
1184 (unsigned long long)ce->start + ce->size);
1185 return 1;
1186 }
1187 map = container_of(ce, struct map_lookup, ce);
1188
1189 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1190 ret = map->num_stripes;
1191 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1192 ret = map->sub_stripes;
1193 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1194 ret = 2;
1195 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1196 ret = 3;
1197 else
1198 ret = 1;
1199 return ret;
1200 }
1201
1202 int btrfs_next_bg(struct btrfs_mapping_tree *map_tree, u64 *logical,
1203 u64 *size, u64 type)
1204 {
1205 struct cache_extent *ce;
1206 struct map_lookup *map;
1207 u64 cur = *logical;
1208
1209 ce = search_cache_extent(&map_tree->cache_tree, cur);
1210
1211 while (ce) {
1212 /*
1213 * only jump to next bg if our cur is not 0
1214 * As the initial logical for btrfs_next_bg() is 0, and
1215 * if we jump to next bg, we skipped a valid bg.
1216 */
1217 if (cur) {
1218 ce = next_cache_extent(ce);
1219 if (!ce)
1220 return -ENOENT;
1221 }
1222
1223 cur = ce->start;
1224 map = container_of(ce, struct map_lookup, ce);
1225 if (map->type & type) {
1226 *logical = ce->start;
1227 *size = ce->size;
1228 return 0;
1229 }
1230 }
1231
1232 return -ENOENT;
1233 }
1234
1235 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1236 u64 chunk_start, u64 physical, u64 devid,
1237 u64 **logical, int *naddrs, int *stripe_len)
1238 {
1239 struct cache_extent *ce;
1240 struct map_lookup *map;
1241 u64 *buf;
1242 u64 bytenr;
1243 u64 length;
1244 u64 stripe_nr;
1245 u64 rmap_len;
1246 int i, j, nr = 0;
1247
1248 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1249 BUG_ON(!ce);
1250 map = container_of(ce, struct map_lookup, ce);
1251
1252 length = ce->size;
1253 rmap_len = map->stripe_len;
1254 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1255 length = ce->size / (map->num_stripes / map->sub_stripes);
1256 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1257 length = ce->size / map->num_stripes;
1258 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1259 BTRFS_BLOCK_GROUP_RAID6)) {
1260 length = ce->size / nr_data_stripes(map);
1261 rmap_len = map->stripe_len * nr_data_stripes(map);
1262 }
1263
1264 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1265
1266 for (i = 0; i < map->num_stripes; i++) {
1267 if (devid && map->stripes[i].dev->devid != devid)
1268 continue;
1269 if (map->stripes[i].physical > physical ||
1270 map->stripes[i].physical + length <= physical)
1271 continue;
1272
1273 stripe_nr = (physical - map->stripes[i].physical) /
1274 map->stripe_len;
1275
1276 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1277 stripe_nr = (stripe_nr * map->num_stripes + i) /
1278 map->sub_stripes;
1279 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1280 stripe_nr = stripe_nr * map->num_stripes + i;
1281 } /* else if RAID[56], multiply by nr_data_stripes().
1282 * Alternatively, just use rmap_len below instead of
1283 * map->stripe_len */
1284
1285 bytenr = ce->start + stripe_nr * rmap_len;
1286 for (j = 0; j < nr; j++) {
1287 if (buf[j] == bytenr)
1288 break;
1289 }
1290 if (j == nr)
1291 buf[nr++] = bytenr;
1292 }
1293
1294 *logical = buf;
1295 *naddrs = nr;
1296 *stripe_len = rmap_len;
1297
1298 return 0;
1299 }
1300
1301 static inline int parity_smaller(u64 a, u64 b)
1302 {
1303 return a > b;
1304 }
1305
1306 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1307 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1308 {
1309 struct btrfs_bio_stripe s;
1310 int i;
1311 u64 l;
1312 int again = 1;
1313
1314 while (again) {
1315 again = 0;
1316 for (i = 0; i < bbio->num_stripes - 1; i++) {
1317 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1318 s = bbio->stripes[i];
1319 l = raid_map[i];
1320 bbio->stripes[i] = bbio->stripes[i+1];
1321 raid_map[i] = raid_map[i+1];
1322 bbio->stripes[i+1] = s;
1323 raid_map[i+1] = l;
1324 again = 1;
1325 }
1326 }
1327 }
1328 }
1329
1330 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1331 u64 logical, u64 *length,
1332 struct btrfs_multi_bio **multi_ret, int mirror_num,
1333 u64 **raid_map_ret)
1334 {
1335 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1336 multi_ret, mirror_num, raid_map_ret);
1337 }
1338
1339 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1340 u64 logical, u64 *length, u64 *type,
1341 struct btrfs_multi_bio **multi_ret, int mirror_num,
1342 u64 **raid_map_ret)
1343 {
1344 struct cache_extent *ce;
1345 struct map_lookup *map;
1346 u64 offset;
1347 u64 stripe_offset;
1348 u64 stripe_nr;
1349 u64 *raid_map = NULL;
1350 int stripes_allocated = 8;
1351 int stripes_required = 1;
1352 int stripe_index;
1353 int i;
1354 struct btrfs_multi_bio *multi = NULL;
1355
1356 if (multi_ret && rw == READ) {
1357 stripes_allocated = 1;
1358 }
1359 again:
1360 ce = search_cache_extent(&map_tree->cache_tree, logical);
1361 if (!ce) {
1362 kfree(multi);
1363 *length = (u64)-1;
1364 return -ENOENT;
1365 }
1366 if (ce->start > logical) {
1367 kfree(multi);
1368 *length = ce->start - logical;
1369 return -ENOENT;
1370 }
1371
1372 if (multi_ret) {
1373 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1374 GFP_NOFS);
1375 if (!multi)
1376 return -ENOMEM;
1377 }
1378 map = container_of(ce, struct map_lookup, ce);
1379 offset = logical - ce->start;
1380
1381 if (rw == WRITE) {
1382 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1383 BTRFS_BLOCK_GROUP_DUP)) {
1384 stripes_required = map->num_stripes;
1385 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1386 stripes_required = map->sub_stripes;
1387 }
1388 }
1389 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1390 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1391 /* RAID[56] write or recovery. Return all stripes */
1392 stripes_required = map->num_stripes;
1393
1394 /* Only allocate the map if we've already got a large enough multi_ret */
1395 if (stripes_allocated >= stripes_required) {
1396 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1397 if (!raid_map) {
1398 kfree(multi);
1399 return -ENOMEM;
1400 }
1401 }
1402 }
1403
1404 /* if our multi bio struct is too small, back off and try again */
1405 if (multi_ret && stripes_allocated < stripes_required) {
1406 stripes_allocated = stripes_required;
1407 kfree(multi);
1408 multi = NULL;
1409 goto again;
1410 }
1411 stripe_nr = offset;
1412 /*
1413 * stripe_nr counts the total number of stripes we have to stride
1414 * to get to this block
1415 */
1416 stripe_nr = stripe_nr / map->stripe_len;
1417
1418 stripe_offset = stripe_nr * map->stripe_len;
1419 BUG_ON(offset < stripe_offset);
1420
1421 /* stripe_offset is the offset of this block in its stripe*/
1422 stripe_offset = offset - stripe_offset;
1423
1424 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1425 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1426 BTRFS_BLOCK_GROUP_RAID10 |
1427 BTRFS_BLOCK_GROUP_DUP)) {
1428 /* we limit the length of each bio to what fits in a stripe */
1429 *length = min_t(u64, ce->size - offset,
1430 map->stripe_len - stripe_offset);
1431 } else {
1432 *length = ce->size - offset;
1433 }
1434
1435 if (!multi_ret)
1436 goto out;
1437
1438 multi->num_stripes = 1;
1439 stripe_index = 0;
1440 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1441 if (rw == WRITE)
1442 multi->num_stripes = map->num_stripes;
1443 else if (mirror_num)
1444 stripe_index = mirror_num - 1;
1445 else
1446 stripe_index = stripe_nr % map->num_stripes;
1447 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1448 int factor = map->num_stripes / map->sub_stripes;
1449
1450 stripe_index = stripe_nr % factor;
1451 stripe_index *= map->sub_stripes;
1452
1453 if (rw == WRITE)
1454 multi->num_stripes = map->sub_stripes;
1455 else if (mirror_num)
1456 stripe_index += mirror_num - 1;
1457
1458 stripe_nr = stripe_nr / factor;
1459 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1460 if (rw == WRITE)
1461 multi->num_stripes = map->num_stripes;
1462 else if (mirror_num)
1463 stripe_index = mirror_num - 1;
1464 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1465 BTRFS_BLOCK_GROUP_RAID6)) {
1466
1467 if (raid_map) {
1468 int rot;
1469 u64 tmp;
1470 u64 raid56_full_stripe_start;
1471 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1472
1473 /*
1474 * align the start of our data stripe in the logical
1475 * address space
1476 */
1477 raid56_full_stripe_start = offset / full_stripe_len;
1478 raid56_full_stripe_start *= full_stripe_len;
1479
1480 /* get the data stripe number */
1481 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1482 stripe_nr = stripe_nr / nr_data_stripes(map);
1483
1484 /* Work out the disk rotation on this stripe-set */
1485 rot = stripe_nr % map->num_stripes;
1486
1487 /* Fill in the logical address of each stripe */
1488 tmp = stripe_nr * nr_data_stripes(map);
1489
1490 for (i = 0; i < nr_data_stripes(map); i++)
1491 raid_map[(i+rot) % map->num_stripes] =
1492 ce->start + (tmp + i) * map->stripe_len;
1493
1494 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1495 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1496 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1497
1498 *length = map->stripe_len;
1499 stripe_index = 0;
1500 stripe_offset = 0;
1501 multi->num_stripes = map->num_stripes;
1502 } else {
1503 stripe_index = stripe_nr % nr_data_stripes(map);
1504 stripe_nr = stripe_nr / nr_data_stripes(map);
1505
1506 /*
1507 * Mirror #0 or #1 means the original data block.
1508 * Mirror #2 is RAID5 parity block.
1509 * Mirror #3 is RAID6 Q block.
1510 */
1511 if (mirror_num > 1)
1512 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1513
1514 /* We distribute the parity blocks across stripes */
1515 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1516 }
1517 } else {
1518 /*
1519 * after this do_div call, stripe_nr is the number of stripes
1520 * on this device we have to walk to find the data, and
1521 * stripe_index is the number of our device in the stripe array
1522 */
1523 stripe_index = stripe_nr % map->num_stripes;
1524 stripe_nr = stripe_nr / map->num_stripes;
1525 }
1526 BUG_ON(stripe_index >= map->num_stripes);
1527
1528 for (i = 0; i < multi->num_stripes; i++) {
1529 multi->stripes[i].physical =
1530 map->stripes[stripe_index].physical + stripe_offset +
1531 stripe_nr * map->stripe_len;
1532 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1533 stripe_index++;
1534 }
1535 *multi_ret = multi;
1536
1537 if (type)
1538 *type = map->type;
1539
1540 if (raid_map) {
1541 sort_parity_stripes(multi, raid_map);
1542 *raid_map_ret = raid_map;
1543 }
1544 out:
1545 return 0;
1546 }
1547
1548 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1549 u8 *uuid, u8 *fsid)
1550 {
1551 struct btrfs_device *device;
1552 struct btrfs_fs_devices *cur_devices;
1553
1554 cur_devices = root->fs_info->fs_devices;
1555 while (cur_devices) {
1556 if (!fsid ||
1557 (!memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE) ||
1558 root->fs_info->ignore_fsid_mismatch)) {
1559 device = __find_device(&cur_devices->devices,
1560 devid, uuid);
1561 if (device)
1562 return device;
1563 }
1564 cur_devices = cur_devices->seed;
1565 }
1566 return NULL;
1567 }
1568
1569 struct btrfs_device *
1570 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1571 u64 devid, int instance)
1572 {
1573 struct list_head *head = &fs_devices->devices;
1574 struct btrfs_device *dev;
1575 int num_found = 0;
1576
1577 list_for_each_entry(dev, head, dev_list) {
1578 if (dev->devid == devid && num_found++ == instance)
1579 return dev;
1580 }
1581 return NULL;
1582 }
1583
1584 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1585 {
1586 struct cache_extent *ce;
1587 struct map_lookup *map;
1588 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1589 int readonly = 0;
1590 int i;
1591
1592 /*
1593 * During chunk recovering, we may fail to find block group's
1594 * corresponding chunk, we will rebuild it later
1595 */
1596 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1597 if (!root->fs_info->is_chunk_recover)
1598 BUG_ON(!ce);
1599 else
1600 return 0;
1601
1602 map = container_of(ce, struct map_lookup, ce);
1603 for (i = 0; i < map->num_stripes; i++) {
1604 if (!map->stripes[i].dev->writeable) {
1605 readonly = 1;
1606 break;
1607 }
1608 }
1609
1610 return readonly;
1611 }
1612
1613 static struct btrfs_device *fill_missing_device(u64 devid)
1614 {
1615 struct btrfs_device *device;
1616
1617 device = kzalloc(sizeof(*device), GFP_NOFS);
1618 device->devid = devid;
1619 device->fd = -1;
1620 return device;
1621 }
1622
1623 /*
1624 * slot == -1: SYSTEM chunk
1625 * return -EIO on error, otherwise return 0
1626 */
1627 int btrfs_check_chunk_valid(struct btrfs_root *root,
1628 struct extent_buffer *leaf,
1629 struct btrfs_chunk *chunk,
1630 int slot, u64 logical)
1631 {
1632 u64 length;
1633 u64 stripe_len;
1634 u16 num_stripes;
1635 u16 sub_stripes;
1636 u64 type;
1637
1638 length = btrfs_chunk_length(leaf, chunk);
1639 stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1640 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1641 sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1642 type = btrfs_chunk_type(leaf, chunk);
1643
1644 /*
1645 * These valid checks may be insufficient to cover every corner cases.
1646 */
1647 if (!IS_ALIGNED(logical, root->sectorsize)) {
1648 error("invalid chunk logical %llu", logical);
1649 return -EIO;
1650 }
1651 if (btrfs_chunk_sector_size(leaf, chunk) != root->sectorsize) {
1652 error("invalid chunk sectorsize %llu",
1653 (unsigned long long)btrfs_chunk_sector_size(leaf, chunk));
1654 return -EIO;
1655 }
1656 if (!length || !IS_ALIGNED(length, root->sectorsize)) {
1657 error("invalid chunk length %llu", length);
1658 return -EIO;
1659 }
1660 if (stripe_len != BTRFS_STRIPE_LEN) {
1661 error("invalid chunk stripe length: %llu", stripe_len);
1662 return -EIO;
1663 }
1664 /* Check on chunk item type */
1665 if (slot == -1 && (type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) {
1666 error("invalid chunk type %llu", type);
1667 return -EIO;
1668 }
1669 if (type & ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1670 BTRFS_BLOCK_GROUP_PROFILE_MASK)) {
1671 error("unrecognized chunk type: %llu",
1672 ~(BTRFS_BLOCK_GROUP_TYPE_MASK |
1673 BTRFS_BLOCK_GROUP_PROFILE_MASK) & type);
1674 return -EIO;
1675 }
1676 /*
1677 * Btrfs_chunk contains at least one stripe, and for sys_chunk
1678 * it can't exceed the system chunk array size
1679 * For normal chunk, it should match its chunk item size.
1680 */
1681 if (num_stripes < 1 ||
1682 (slot == -1 && sizeof(struct btrfs_stripe) * num_stripes >
1683 BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) ||
1684 (slot >= 0 && sizeof(struct btrfs_stripe) * (num_stripes - 1) >
1685 btrfs_item_size_nr(leaf, slot))) {
1686 error("invalid num_stripes: %u", num_stripes);
1687 return -EIO;
1688 }
1689 /*
1690 * Device number check against profile
1691 */
1692 if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes == 0) ||
1693 (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) ||
1694 (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) ||
1695 (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) ||
1696 (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) ||
1697 ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 &&
1698 num_stripes != 1)) {
1699 error("Invalid num_stripes:sub_stripes %u:%u for profile %llu",
1700 num_stripes, sub_stripes,
1701 type & BTRFS_BLOCK_GROUP_PROFILE_MASK);
1702 return -EIO;
1703 }
1704
1705 return 0;
1706 }
1707
1708 /*
1709 * Slot is used to verify the chunk item is valid
1710 *
1711 * For sys chunk in superblock, pass -1 to indicate sys chunk.
1712 */
1713 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1714 struct extent_buffer *leaf,
1715 struct btrfs_chunk *chunk, int slot)
1716 {
1717 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1718 struct map_lookup *map;
1719 struct cache_extent *ce;
1720 u64 logical;
1721 u64 length;
1722 u64 devid;
1723 u8 uuid[BTRFS_UUID_SIZE];
1724 int num_stripes;
1725 int ret;
1726 int i;
1727
1728 logical = key->offset;
1729 length = btrfs_chunk_length(leaf, chunk);
1730 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1731 /* Validation check */
1732 ret = btrfs_check_chunk_valid(root, leaf, chunk, slot, logical);
1733 if (ret) {
1734 error("%s checksums match, but it has an invalid chunk, %s",
1735 (slot == -1) ? "Superblock" : "Metadata",
1736 (slot == -1) ? "try btrfsck --repair -s <superblock> ie, 0,1,2" : "");
1737 return ret;
1738 }
1739
1740 ce = search_cache_extent(&map_tree->cache_tree, logical);
1741
1742 /* already mapped? */
1743 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1744 return 0;
1745 }
1746
1747 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1748 if (!map)
1749 return -ENOMEM;
1750
1751 map->ce.start = logical;
1752 map->ce.size = length;
1753 map->num_stripes = num_stripes;
1754 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1755 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1756 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1757 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1758 map->type = btrfs_chunk_type(leaf, chunk);
1759 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1760
1761 for (i = 0; i < num_stripes; i++) {
1762 map->stripes[i].physical =
1763 btrfs_stripe_offset_nr(leaf, chunk, i);
1764 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1765 read_extent_buffer(leaf, uuid, (unsigned long)
1766 btrfs_stripe_dev_uuid_nr(chunk, i),
1767 BTRFS_UUID_SIZE);
1768 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1769 NULL);
1770 if (!map->stripes[i].dev) {
1771 map->stripes[i].dev = fill_missing_device(devid);
1772 printf("warning, device %llu is missing\n",
1773 (unsigned long long)devid);
1774 list_add(&map->stripes[i].dev->dev_list,
1775 &root->fs_info->fs_devices->devices);
1776 }
1777
1778 }
1779 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1780 BUG_ON(ret);
1781
1782 return 0;
1783 }
1784
1785 static int fill_device_from_item(struct extent_buffer *leaf,
1786 struct btrfs_dev_item *dev_item,
1787 struct btrfs_device *device)
1788 {
1789 unsigned long ptr;
1790
1791 device->devid = btrfs_device_id(leaf, dev_item);
1792 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1793 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1794 device->type = btrfs_device_type(leaf, dev_item);
1795 device->io_align = btrfs_device_io_align(leaf, dev_item);
1796 device->io_width = btrfs_device_io_width(leaf, dev_item);
1797 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1798
1799 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1800 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1801
1802 return 0;
1803 }
1804
1805 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1806 {
1807 struct btrfs_fs_devices *fs_devices;
1808 int ret;
1809
1810 fs_devices = root->fs_info->fs_devices->seed;
1811 while (fs_devices) {
1812 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1813 ret = 0;
1814 goto out;
1815 }
1816 fs_devices = fs_devices->seed;
1817 }
1818
1819 fs_devices = find_fsid(fsid);
1820 if (!fs_devices) {
1821 /* missing all seed devices */
1822 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
1823 if (!fs_devices) {
1824 ret = -ENOMEM;
1825 goto out;
1826 }
1827 INIT_LIST_HEAD(&fs_devices->devices);
1828 list_add(&fs_devices->list, &fs_uuids);
1829 memcpy(fs_devices->fsid, fsid, BTRFS_FSID_SIZE);
1830 }
1831
1832 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1833 if (ret)
1834 goto out;
1835
1836 fs_devices->seed = root->fs_info->fs_devices->seed;
1837 root->fs_info->fs_devices->seed = fs_devices;
1838 out:
1839 return ret;
1840 }
1841
1842 static int read_one_dev(struct btrfs_root *root,
1843 struct extent_buffer *leaf,
1844 struct btrfs_dev_item *dev_item)
1845 {
1846 struct btrfs_device *device;
1847 u64 devid;
1848 int ret = 0;
1849 u8 fs_uuid[BTRFS_UUID_SIZE];
1850 u8 dev_uuid[BTRFS_UUID_SIZE];
1851
1852 devid = btrfs_device_id(leaf, dev_item);
1853 read_extent_buffer(leaf, dev_uuid,
1854 (unsigned long)btrfs_device_uuid(dev_item),
1855 BTRFS_UUID_SIZE);
1856 read_extent_buffer(leaf, fs_uuid,
1857 (unsigned long)btrfs_device_fsid(dev_item),
1858 BTRFS_UUID_SIZE);
1859
1860 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1861 ret = open_seed_devices(root, fs_uuid);
1862 if (ret)
1863 return ret;
1864 }
1865
1866 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1867 if (!device) {
1868 device = kzalloc(sizeof(*device), GFP_NOFS);
1869 if (!device)
1870 return -ENOMEM;
1871 device->fd = -1;
1872 list_add(&device->dev_list,
1873 &root->fs_info->fs_devices->devices);
1874 }
1875
1876 fill_device_from_item(leaf, dev_item, device);
1877 device->dev_root = root->fs_info->dev_root;
1878 return ret;
1879 }
1880
1881 int btrfs_read_sys_array(struct btrfs_root *root)
1882 {
1883 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1884 struct extent_buffer *sb;
1885 struct btrfs_disk_key *disk_key;
1886 struct btrfs_chunk *chunk;
1887 u8 *array_ptr;
1888 unsigned long sb_array_offset;
1889 int ret = 0;
1890 u32 num_stripes;
1891 u32 array_size;
1892 u32 len = 0;
1893 u32 cur_offset;
1894 struct btrfs_key key;
1895
1896 sb = btrfs_find_create_tree_block(root->fs_info,
1897 BTRFS_SUPER_INFO_OFFSET,
1898 BTRFS_SUPER_INFO_SIZE);
1899 if (!sb)
1900 return -ENOMEM;
1901 btrfs_set_buffer_uptodate(sb);
1902 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1903 array_size = btrfs_super_sys_array_size(super_copy);
1904
1905 array_ptr = super_copy->sys_chunk_array;
1906 sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array);
1907 cur_offset = 0;
1908
1909 while (cur_offset < array_size) {
1910 disk_key = (struct btrfs_disk_key *)array_ptr;
1911 len = sizeof(*disk_key);
1912 if (cur_offset + len > array_size)
1913 goto out_short_read;
1914
1915 btrfs_disk_key_to_cpu(&key, disk_key);
1916
1917 array_ptr += len;
1918 sb_array_offset += len;
1919 cur_offset += len;
1920
1921 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1922 chunk = (struct btrfs_chunk *)sb_array_offset;
1923 /*
1924 * At least one btrfs_chunk with one stripe must be
1925 * present, exact stripe count check comes afterwards
1926 */
1927 len = btrfs_chunk_item_size(1);
1928 if (cur_offset + len > array_size)
1929 goto out_short_read;
1930
1931 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1932 if (!num_stripes) {
1933 printk(
1934 "ERROR: invalid number of stripes %u in sys_array at offset %u\n",
1935 num_stripes, cur_offset);
1936 ret = -EIO;
1937 break;
1938 }
1939
1940 len = btrfs_chunk_item_size(num_stripes);
1941 if (cur_offset + len > array_size)
1942 goto out_short_read;
1943
1944 ret = read_one_chunk(root, &key, sb, chunk, -1);
1945 if (ret)
1946 break;
1947 } else {
1948 printk(
1949 "ERROR: unexpected item type %u in sys_array at offset %u\n",
1950 (u32)key.type, cur_offset);
1951 ret = -EIO;
1952 break;
1953 }
1954 array_ptr += len;
1955 sb_array_offset += len;
1956 cur_offset += len;
1957 }
1958 free_extent_buffer(sb);
1959 return ret;
1960
1961 out_short_read:
1962 printk("ERROR: sys_array too short to read %u bytes at offset %u\n",
1963 len, cur_offset);
1964 free_extent_buffer(sb);
1965 return -EIO;
1966 }
1967
1968 int btrfs_read_chunk_tree(struct btrfs_root *root)
1969 {
1970 struct btrfs_path *path;
1971 struct extent_buffer *leaf;
1972 struct btrfs_key key;
1973 struct btrfs_key found_key;
1974 int ret;
1975 int slot;
1976
1977 root = root->fs_info->chunk_root;
1978
1979 path = btrfs_alloc_path();
1980 if (!path)
1981 return -ENOMEM;
1982
1983 /*
1984 * Read all device items, and then all the chunk items. All
1985 * device items are found before any chunk item (their object id
1986 * is smaller than the lowest possible object id for a chunk
1987 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1988 */
1989 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1990 key.offset = 0;
1991 key.type = 0;
1992 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1993 if (ret < 0)
1994 goto error;
1995 while(1) {
1996 leaf = path->nodes[0];
1997 slot = path->slots[0];
1998 if (slot >= btrfs_header_nritems(leaf)) {
1999 ret = btrfs_next_leaf(root, path);
2000 if (ret == 0)
2001 continue;
2002 if (ret < 0)
2003 goto error;
2004 break;
2005 }
2006 btrfs_item_key_to_cpu(leaf, &found_key, slot);
2007 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
2008 struct btrfs_dev_item *dev_item;
2009 dev_item = btrfs_item_ptr(leaf, slot,
2010 struct btrfs_dev_item);
2011 ret = read_one_dev(root, leaf, dev_item);
2012 BUG_ON(ret);
2013 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
2014 struct btrfs_chunk *chunk;
2015 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
2016 ret = read_one_chunk(root, &found_key, leaf, chunk,
2017 slot);
2018 BUG_ON(ret);
2019 }
2020 path->slots[0]++;
2021 }
2022
2023 ret = 0;
2024 error:
2025 btrfs_free_path(path);
2026 return ret;
2027 }
2028
2029 struct list_head *btrfs_scanned_uuids(void)
2030 {
2031 return &fs_uuids;
2032 }
2033
2034 static int rmw_eb(struct btrfs_fs_info *info,
2035 struct extent_buffer *eb, struct extent_buffer *orig_eb)
2036 {
2037 int ret;
2038 unsigned long orig_off = 0;
2039 unsigned long dest_off = 0;
2040 unsigned long copy_len = eb->len;
2041
2042 ret = read_whole_eb(info, eb, 0);
2043 if (ret)
2044 return ret;
2045
2046 if (eb->start + eb->len <= orig_eb->start ||
2047 eb->start >= orig_eb->start + orig_eb->len)
2048 return 0;
2049 /*
2050 * | ----- orig_eb ------- |
2051 * | ----- stripe ------- |
2052 * | ----- orig_eb ------- |
2053 * | ----- orig_eb ------- |
2054 */
2055 if (eb->start > orig_eb->start)
2056 orig_off = eb->start - orig_eb->start;
2057 if (orig_eb->start > eb->start)
2058 dest_off = orig_eb->start - eb->start;
2059
2060 if (copy_len > orig_eb->len - orig_off)
2061 copy_len = orig_eb->len - orig_off;
2062 if (copy_len > eb->len - dest_off)
2063 copy_len = eb->len - dest_off;
2064
2065 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
2066 return 0;
2067 }
2068
2069 static int split_eb_for_raid56(struct btrfs_fs_info *info,
2070 struct extent_buffer *orig_eb,
2071 struct extent_buffer **ebs,
2072 u64 stripe_len, u64 *raid_map,
2073 int num_stripes)
2074 {
2075 struct extent_buffer **tmp_ebs;
2076 u64 start = orig_eb->start;
2077 u64 this_eb_start;
2078 int i;
2079 int ret = 0;
2080
2081 tmp_ebs = calloc(num_stripes, sizeof(*tmp_ebs));
2082 if (!tmp_ebs)
2083 return -ENOMEM;
2084
2085 /* Alloc memory in a row for data stripes */
2086 for (i = 0; i < num_stripes; i++) {
2087 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2088 break;
2089
2090 tmp_ebs[i] = calloc(1, sizeof(**tmp_ebs) + stripe_len);
2091 if (!tmp_ebs[i]) {
2092 ret = -ENOMEM;
2093 goto clean_up;
2094 }
2095 }
2096
2097 for (i = 0; i < num_stripes; i++) {
2098 struct extent_buffer *eb = tmp_ebs[i];
2099
2100 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
2101 break;
2102
2103 eb->start = raid_map[i];
2104 eb->len = stripe_len;
2105 eb->refs = 1;
2106 eb->flags = 0;
2107 eb->fd = -1;
2108 eb->dev_bytenr = (u64)-1;
2109
2110 this_eb_start = raid_map[i];
2111
2112 if (start > this_eb_start ||
2113 start + orig_eb->len < this_eb_start + stripe_len) {
2114 ret = rmw_eb(info, eb, orig_eb);
2115 if (ret)
2116 goto clean_up;
2117 } else {
2118 memcpy(eb->data, orig_eb->data + eb->start - start,
2119 stripe_len);
2120 }
2121 ebs[i] = eb;
2122 }
2123 free(tmp_ebs);
2124 return ret;
2125 clean_up:
2126 for (i = 0; i < num_stripes; i++)
2127 free(tmp_ebs[i]);
2128 free(tmp_ebs);
2129 return ret;
2130 }
2131
2132 int write_raid56_with_parity(struct btrfs_fs_info *info,
2133 struct extent_buffer *eb,
2134 struct btrfs_multi_bio *multi,
2135 u64 stripe_len, u64 *raid_map)
2136 {
2137 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
2138 int i;
2139 int j;
2140 int ret;
2141 int alloc_size = eb->len;
2142 void **pointers;
2143
2144 ebs = malloc(sizeof(*ebs) * multi->num_stripes);
2145 pointers = malloc(sizeof(*pointers) * multi->num_stripes);
2146 if (!ebs || !pointers) {
2147 free(ebs);
2148 free(pointers);
2149 return -ENOMEM;
2150 }
2151
2152 if (stripe_len > alloc_size)
2153 alloc_size = stripe_len;
2154
2155 ret = split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
2156 multi->num_stripes);
2157 if (ret)
2158 goto out;
2159
2160 for (i = 0; i < multi->num_stripes; i++) {
2161 struct extent_buffer *new_eb;
2162 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
2163 ebs[i]->dev_bytenr = multi->stripes[i].physical;
2164 ebs[i]->fd = multi->stripes[i].dev->fd;
2165 multi->stripes[i].dev->total_ios++;
2166 if (ebs[i]->start != raid_map[i]) {
2167 ret = -EINVAL;
2168 goto out_free_split;
2169 }
2170 continue;
2171 }
2172 new_eb = malloc(sizeof(*eb) + alloc_size);
2173 if (!new_eb) {
2174 ret = -ENOMEM;
2175 goto out_free_split;
2176 }
2177 new_eb->dev_bytenr = multi->stripes[i].physical;
2178 new_eb->fd = multi->stripes[i].dev->fd;
2179 multi->stripes[i].dev->total_ios++;
2180 new_eb->len = stripe_len;
2181
2182 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
2183 p_eb = new_eb;
2184 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
2185 q_eb = new_eb;
2186 }
2187 if (q_eb) {
2188 ebs[multi->num_stripes - 2] = p_eb;
2189 ebs[multi->num_stripes - 1] = q_eb;
2190
2191 for (i = 0; i < multi->num_stripes; i++)
2192 pointers[i] = ebs[i]->data;
2193
2194 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
2195 } else {
2196 ebs[multi->num_stripes - 1] = p_eb;
2197 memcpy(p_eb->data, ebs[0]->data, stripe_len);
2198 for (j = 1; j < multi->num_stripes - 1; j++) {
2199 for (i = 0; i < stripe_len; i += sizeof(u64)) {
2200 u64 p_eb_data;
2201 u64 ebs_data;
2202
2203 p_eb_data = get_unaligned_64(p_eb->data + i);
2204 ebs_data = get_unaligned_64(ebs[j]->data + i);
2205 p_eb_data ^= ebs_data;
2206 put_unaligned_64(p_eb_data, p_eb->data + i);
2207 }
2208 }
2209 }
2210
2211 for (i = 0; i < multi->num_stripes; i++) {
2212 ret = write_extent_to_disk(ebs[i]);
2213 if (ret < 0)
2214 goto out_free_split;
2215 }
2216
2217 out_free_split:
2218 for (i = 0; i < multi->num_stripes; i++) {
2219 if (ebs[i] != eb)
2220 free(ebs[i]);
2221 }
2222 out:
2223 free(ebs);
2224 free(pointers);
2225
2226 return ret;
2227 }
(deprecated) Btrfs progs developments and patch integration