?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
btrfs-progs: Fix a memleak in btrfs_scan_one_device.
[btrfs-progs-unstable/devel.git] / volumes.c
blob77ffd3252c38a2fed20cc28cabc44a5443f7e92b
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 #define _XOPEN_SOURCE 600
19 #define __USE_XOPEN2K
20 #include <stdio.h>
21 #include <stdlib.h>
22 #include <sys/types.h>
23 #include <sys/stat.h>
24 #include <uuid/uuid.h>
25 #include <fcntl.h>
26 #include <unistd.h>
27 #include "ctree.h"
28 #include "disk-io.h"
29 #include "transaction.h"
30 #include "print-tree.h"
31 #include "volumes.h"
32 #include "math.h"
33
34 struct stripe {
35 struct btrfs_device *dev;
36 u64 physical;
37 };
38
39 static inline int nr_parity_stripes(struct map_lookup *map)
40 {
41 if (map->type & BTRFS_BLOCK_GROUP_RAID5)
42 return 1;
43 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
44 return 2;
45 else
46 return 0;
47 }
48
49 static inline int nr_data_stripes(struct map_lookup *map)
50 {
51 return map->num_stripes - nr_parity_stripes(map);
52 }
53
54 #define is_parity_stripe(x) ( ((x) == BTRFS_RAID5_P_STRIPE) || ((x) == BTRFS_RAID6_Q_STRIPE) )
55
56 static LIST_HEAD(fs_uuids);
57
58 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
59 u8 *uuid)
60 {
61 struct btrfs_device *dev;
62 struct list_head *cur;
63
64 list_for_each(cur, head) {
65 dev = list_entry(cur, struct btrfs_device, dev_list);
66 if (dev->devid == devid &&
67 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
68 return dev;
69 }
70 }
71 return NULL;
72 }
73
74 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
75 {
76 struct list_head *cur;
77 struct btrfs_fs_devices *fs_devices;
78
79 list_for_each(cur, &fs_uuids) {
80 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
81 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
82 return fs_devices;
83 }
84 return NULL;
85 }
86
87 static int device_list_add(const char *path,
88 struct btrfs_super_block *disk_super,
89 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
90 {
91 struct btrfs_device *device;
92 struct btrfs_fs_devices *fs_devices;
93 u64 found_transid = btrfs_super_generation(disk_super);
94
95 fs_devices = find_fsid(disk_super->fsid);
96 if (!fs_devices) {
97 fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS);
98 if (!fs_devices)
99 return -ENOMEM;
100 INIT_LIST_HEAD(&fs_devices->devices);
101 list_add(&fs_devices->list, &fs_uuids);
102 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
103 fs_devices->latest_devid = devid;
104 fs_devices->latest_trans = found_transid;
105 fs_devices->lowest_devid = (u64)-1;
106 device = NULL;
107 } else {
108 device = __find_device(&fs_devices->devices, devid,
109 disk_super->dev_item.uuid);
110 }
111 if (!device) {
112 device = kzalloc(sizeof(*device), GFP_NOFS);
113 if (!device) {
114 /* we can safely leave the fs_devices entry around */
115 return -ENOMEM;
116 }
117 device->fd = -1;
118 device->devid = devid;
119 memcpy(device->uuid, disk_super->dev_item.uuid,
120 BTRFS_UUID_SIZE);
121 device->name = kstrdup(path, GFP_NOFS);
122 if (!device->name) {
123 kfree(device);
124 return -ENOMEM;
125 }
126 device->label = kstrdup(disk_super->label, GFP_NOFS);
127 if (!device->label) {
128 kfree(device->name);
129 kfree(device);
130 return -ENOMEM;
131 }
132 device->total_devs = btrfs_super_num_devices(disk_super);
133 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
134 device->total_bytes =
135 btrfs_stack_device_total_bytes(&disk_super->dev_item);
136 device->bytes_used =
137 btrfs_stack_device_bytes_used(&disk_super->dev_item);
138 list_add(&device->dev_list, &fs_devices->devices);
139 device->fs_devices = fs_devices;
140 } else if (!device->name || strcmp(device->name, path)) {
141 char *name = strdup(path);
142 if (!name)
143 return -ENOMEM;
144 kfree(device->name);
145 device->name = name;
146 }
147
148
149 if (found_transid > fs_devices->latest_trans) {
150 fs_devices->latest_devid = devid;
151 fs_devices->latest_trans = found_transid;
152 }
153 if (fs_devices->lowest_devid > devid) {
154 fs_devices->lowest_devid = devid;
155 }
156 *fs_devices_ret = fs_devices;
157 return 0;
158 }
159
160 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
161 {
162 struct btrfs_fs_devices *seed_devices;
163 struct btrfs_device *device;
164
165 again:
166 while (!list_empty(&fs_devices->devices)) {
167 device = list_entry(fs_devices->devices.next,
168 struct btrfs_device, dev_list);
169 if (device->fd != -1) {
170 fsync(device->fd);
171 if (posix_fadvise(device->fd, 0, 0, POSIX_FADV_DONTNEED))
172 fprintf(stderr, "Warning, could not drop caches\n");
173 close(device->fd);
174 device->fd = -1;
175 }
176 device->writeable = 0;
177 list_del(&device->dev_list);
178 /* free the memory */
179 free(device->name);
180 free(device->label);
181 free(device);
182 }
183
184 seed_devices = fs_devices->seed;
185 fs_devices->seed = NULL;
186 if (seed_devices) {
187 fs_devices = seed_devices;
188 goto again;
189 }
190
191 free(fs_devices);
192 return 0;
193 }
194
195 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
196 {
197 int fd;
198 struct list_head *head = &fs_devices->devices;
199 struct list_head *cur;
200 struct btrfs_device *device;
201 int ret;
202
203 list_for_each(cur, head) {
204 device = list_entry(cur, struct btrfs_device, dev_list);
205 if (!device->name) {
206 printk("no name for device %llu, skip it now\n", device->devid);
207 continue;
208 }
209
210 fd = open(device->name, flags);
211 if (fd < 0) {
212 ret = -errno;
213 goto fail;
214 }
215
216 if (posix_fadvise(fd, 0, 0, POSIX_FADV_DONTNEED))
217 fprintf(stderr, "Warning, could not drop caches\n");
218
219 if (device->devid == fs_devices->latest_devid)
220 fs_devices->latest_bdev = fd;
221 if (device->devid == fs_devices->lowest_devid)
222 fs_devices->lowest_bdev = fd;
223 device->fd = fd;
224 if (flags & O_RDWR)
225 device->writeable = 1;
226 }
227 return 0;
228 fail:
229 btrfs_close_devices(fs_devices);
230 return ret;
231 }
232
233 int btrfs_scan_one_device(int fd, const char *path,
234 struct btrfs_fs_devices **fs_devices_ret,
235 u64 *total_devs, u64 super_offset)
236 {
237 struct btrfs_super_block *disk_super;
238 char *buf;
239 int ret;
240 u64 devid;
241
242 buf = malloc(4096);
243 if (!buf) {
244 ret = -ENOMEM;
245 goto error;
246 }
247 disk_super = (struct btrfs_super_block *)buf;
248 ret = btrfs_read_dev_super(fd, disk_super, super_offset);
249 if (ret < 0) {
250 ret = -EIO;
251 goto error_brelse;
252 }
253 devid = btrfs_stack_device_id(&disk_super->dev_item);
254 if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_METADUMP)
255 *total_devs = 1;
256 else
257 *total_devs = btrfs_super_num_devices(disk_super);
258
259 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
260
261 error_brelse:
262 free(buf);
263 error:
264 return ret;
265 }
266
267 /*
268 * this uses a pretty simple search, the expectation is that it is
269 * called very infrequently and that a given device has a small number
270 * of extents
271 */
272 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
273 struct btrfs_device *device,
274 struct btrfs_path *path,
275 u64 num_bytes, u64 *start)
276 {
277 struct btrfs_key key;
278 struct btrfs_root *root = device->dev_root;
279 struct btrfs_dev_extent *dev_extent = NULL;
280 u64 hole_size = 0;
281 u64 last_byte = 0;
282 u64 search_start = root->fs_info->alloc_start;
283 u64 search_end = device->total_bytes;
284 int ret;
285 int slot = 0;
286 int start_found;
287 struct extent_buffer *l;
288
289 start_found = 0;
290 path->reada = 2;
291
292 /* FIXME use last free of some kind */
293
294 /* we don't want to overwrite the superblock on the drive,
295 * so we make sure to start at an offset of at least 1MB
296 */
297 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
298
299 if (search_start >= search_end) {
300 ret = -ENOSPC;
301 goto error;
302 }
303
304 key.objectid = device->devid;
305 key.offset = search_start;
306 key.type = BTRFS_DEV_EXTENT_KEY;
307 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
308 if (ret < 0)
309 goto error;
310 ret = btrfs_previous_item(root, path, 0, key.type);
311 if (ret < 0)
312 goto error;
313 l = path->nodes[0];
314 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
315 while (1) {
316 l = path->nodes[0];
317 slot = path->slots[0];
318 if (slot >= btrfs_header_nritems(l)) {
319 ret = btrfs_next_leaf(root, path);
320 if (ret == 0)
321 continue;
322 if (ret < 0)
323 goto error;
324 no_more_items:
325 if (!start_found) {
326 if (search_start >= search_end) {
327 ret = -ENOSPC;
328 goto error;
329 }
330 *start = search_start;
331 start_found = 1;
332 goto check_pending;
333 }
334 *start = last_byte > search_start ?
335 last_byte : search_start;
336 if (search_end <= *start) {
337 ret = -ENOSPC;
338 goto error;
339 }
340 goto check_pending;
341 }
342 btrfs_item_key_to_cpu(l, &key, slot);
343
344 if (key.objectid < device->devid)
345 goto next;
346
347 if (key.objectid > device->devid)
348 goto no_more_items;
349
350 if (key.offset >= search_start && key.offset > last_byte &&
351 start_found) {
352 if (last_byte < search_start)
353 last_byte = search_start;
354 hole_size = key.offset - last_byte;
355 if (key.offset > last_byte &&
356 hole_size >= num_bytes) {
357 *start = last_byte;
358 goto check_pending;
359 }
360 }
361 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
362 goto next;
363 }
364
365 start_found = 1;
366 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
367 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
368 next:
369 path->slots[0]++;
370 cond_resched();
371 }
372 check_pending:
373 /* we have to make sure we didn't find an extent that has already
374 * been allocated by the map tree or the original allocation
375 */
376 btrfs_release_path(path);
377 BUG_ON(*start < search_start);
378
379 if (*start + num_bytes > search_end) {
380 ret = -ENOSPC;
381 goto error;
382 }
383 /* check for pending inserts here */
384 return 0;
385
386 error:
387 btrfs_release_path(path);
388 return ret;
389 }
390
391 static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
392 struct btrfs_device *device,
393 u64 chunk_tree, u64 chunk_objectid,
394 u64 chunk_offset,
395 u64 num_bytes, u64 *start)
396 {
397 int ret;
398 struct btrfs_path *path;
399 struct btrfs_root *root = device->dev_root;
400 struct btrfs_dev_extent *extent;
401 struct extent_buffer *leaf;
402 struct btrfs_key key;
403
404 path = btrfs_alloc_path();
405 if (!path)
406 return -ENOMEM;
407
408 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
409 if (ret) {
410 goto err;
411 }
412
413 key.objectid = device->devid;
414 key.offset = *start;
415 key.type = BTRFS_DEV_EXTENT_KEY;
416 ret = btrfs_insert_empty_item(trans, root, path, &key,
417 sizeof(*extent));
418 BUG_ON(ret);
419
420 leaf = path->nodes[0];
421 extent = btrfs_item_ptr(leaf, path->slots[0],
422 struct btrfs_dev_extent);
423 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
424 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
425 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
426
427 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
428 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
429 BTRFS_UUID_SIZE);
430
431 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
432 btrfs_mark_buffer_dirty(leaf);
433 err:
434 btrfs_free_path(path);
435 return ret;
436 }
437
438 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
439 {
440 struct btrfs_path *path;
441 int ret;
442 struct btrfs_key key;
443 struct btrfs_chunk *chunk;
444 struct btrfs_key found_key;
445
446 path = btrfs_alloc_path();
447 BUG_ON(!path);
448
449 key.objectid = objectid;
450 key.offset = (u64)-1;
451 key.type = BTRFS_CHUNK_ITEM_KEY;
452
453 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
454 if (ret < 0)
455 goto error;
456
457 BUG_ON(ret == 0);
458
459 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
460 if (ret) {
461 *offset = 0;
462 } else {
463 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
464 path->slots[0]);
465 if (found_key.objectid != objectid)
466 *offset = 0;
467 else {
468 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
469 struct btrfs_chunk);
470 *offset = found_key.offset +
471 btrfs_chunk_length(path->nodes[0], chunk);
472 }
473 }
474 ret = 0;
475 error:
476 btrfs_free_path(path);
477 return ret;
478 }
479
480 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
481 u64 *objectid)
482 {
483 int ret;
484 struct btrfs_key key;
485 struct btrfs_key found_key;
486
487 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
488 key.type = BTRFS_DEV_ITEM_KEY;
489 key.offset = (u64)-1;
490
491 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
492 if (ret < 0)
493 goto error;
494
495 BUG_ON(ret == 0);
496
497 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
498 BTRFS_DEV_ITEM_KEY);
499 if (ret) {
500 *objectid = 1;
501 } else {
502 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
503 path->slots[0]);
504 *objectid = found_key.offset + 1;
505 }
506 ret = 0;
507 error:
508 btrfs_release_path(path);
509 return ret;
510 }
511
512 /*
513 * the device information is stored in the chunk root
514 * the btrfs_device struct should be fully filled in
515 */
516 int btrfs_add_device(struct btrfs_trans_handle *trans,
517 struct btrfs_root *root,
518 struct btrfs_device *device)
519 {
520 int ret;
521 struct btrfs_path *path;
522 struct btrfs_dev_item *dev_item;
523 struct extent_buffer *leaf;
524 struct btrfs_key key;
525 unsigned long ptr;
526 u64 free_devid = 0;
527
528 root = root->fs_info->chunk_root;
529
530 path = btrfs_alloc_path();
531 if (!path)
532 return -ENOMEM;
533
534 ret = find_next_devid(root, path, &free_devid);
535 if (ret)
536 goto out;
537
538 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
539 key.type = BTRFS_DEV_ITEM_KEY;
540 key.offset = free_devid;
541
542 ret = btrfs_insert_empty_item(trans, root, path, &key,
543 sizeof(*dev_item));
544 if (ret)
545 goto out;
546
547 leaf = path->nodes[0];
548 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
549
550 device->devid = free_devid;
551 btrfs_set_device_id(leaf, dev_item, device->devid);
552 btrfs_set_device_generation(leaf, dev_item, 0);
553 btrfs_set_device_type(leaf, dev_item, device->type);
554 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
555 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
556 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
557 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
558 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
559 btrfs_set_device_group(leaf, dev_item, 0);
560 btrfs_set_device_seek_speed(leaf, dev_item, 0);
561 btrfs_set_device_bandwidth(leaf, dev_item, 0);
562 btrfs_set_device_start_offset(leaf, dev_item, 0);
563
564 ptr = (unsigned long)btrfs_device_uuid(dev_item);
565 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
566 ptr = (unsigned long)btrfs_device_fsid(dev_item);
567 write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE);
568 btrfs_mark_buffer_dirty(leaf);
569 ret = 0;
570
571 out:
572 btrfs_free_path(path);
573 return ret;
574 }
575
576 int btrfs_update_device(struct btrfs_trans_handle *trans,
577 struct btrfs_device *device)
578 {
579 int ret;
580 struct btrfs_path *path;
581 struct btrfs_root *root;
582 struct btrfs_dev_item *dev_item;
583 struct extent_buffer *leaf;
584 struct btrfs_key key;
585
586 root = device->dev_root->fs_info->chunk_root;
587
588 path = btrfs_alloc_path();
589 if (!path)
590 return -ENOMEM;
591
592 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
593 key.type = BTRFS_DEV_ITEM_KEY;
594 key.offset = device->devid;
595
596 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
597 if (ret < 0)
598 goto out;
599
600 if (ret > 0) {
601 ret = -ENOENT;
602 goto out;
603 }
604
605 leaf = path->nodes[0];
606 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
607
608 btrfs_set_device_id(leaf, dev_item, device->devid);
609 btrfs_set_device_type(leaf, dev_item, device->type);
610 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
611 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
612 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
613 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
614 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
615 btrfs_mark_buffer_dirty(leaf);
616
617 out:
618 btrfs_free_path(path);
619 return ret;
620 }
621
622 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
623 struct btrfs_root *root,
624 struct btrfs_key *key,
625 struct btrfs_chunk *chunk, int item_size)
626 {
627 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
628 struct btrfs_disk_key disk_key;
629 u32 array_size;
630 u8 *ptr;
631
632 array_size = btrfs_super_sys_array_size(super_copy);
633 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
634 return -EFBIG;
635
636 ptr = super_copy->sys_chunk_array + array_size;
637 btrfs_cpu_key_to_disk(&disk_key, key);
638 memcpy(ptr, &disk_key, sizeof(disk_key));
639 ptr += sizeof(disk_key);
640 memcpy(ptr, chunk, item_size);
641 item_size += sizeof(disk_key);
642 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
643 return 0;
644 }
645
646 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
647 int sub_stripes)
648 {
649 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
650 return calc_size;
651 else if (type & BTRFS_BLOCK_GROUP_RAID10)
652 return calc_size * (num_stripes / sub_stripes);
653 else if (type & BTRFS_BLOCK_GROUP_RAID5)
654 return calc_size * (num_stripes - 1);
655 else if (type & BTRFS_BLOCK_GROUP_RAID6)
656 return calc_size * (num_stripes - 2);
657 else
658 return calc_size * num_stripes;
659 }
660
661
662 static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target)
663 {
664 /* TODO, add a way to store the preferred stripe size */
665 return BTRFS_STRIPE_LEN;
666 }
667
668 /*
669 * btrfs_device_avail_bytes - count bytes available for alloc_chunk
670 *
671 * It is not equal to "device->total_bytes - device->bytes_used".
672 * We do not allocate any chunk in 1M at beginning of device, and not
673 * allowed to allocate any chunk before alloc_start if it is specified.
674 * So search holes from max(1M, alloc_start) to device->total_bytes.
675 */
676 static int btrfs_device_avail_bytes(struct btrfs_trans_handle *trans,
677 struct btrfs_device *device,
678 u64 *avail_bytes)
679 {
680 struct btrfs_path *path;
681 struct btrfs_root *root = device->dev_root;
682 struct btrfs_key key;
683 struct btrfs_dev_extent *dev_extent = NULL;
684 struct extent_buffer *l;
685 u64 search_start = root->fs_info->alloc_start;
686 u64 search_end = device->total_bytes;
687 u64 extent_end = 0;
688 u64 free_bytes = 0;
689 int ret;
690 int slot = 0;
691
692 search_start = max(BTRFS_BLOCK_RESERVED_1M_FOR_SUPER, search_start);
693
694 path = btrfs_alloc_path();
695 if (!path)
696 return -ENOMEM;
697
698 key.objectid = device->devid;
699 key.offset = root->fs_info->alloc_start;
700 key.type = BTRFS_DEV_EXTENT_KEY;
701
702 path->reada = 2;
703 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
704 if (ret < 0)
705 goto error;
706 ret = btrfs_previous_item(root, path, 0, key.type);
707 if (ret < 0)
708 goto error;
709
710 while (1) {
711 l = path->nodes[0];
712 slot = path->slots[0];
713 if (slot >= btrfs_header_nritems(l)) {
714 ret = btrfs_next_leaf(root, path);
715 if (ret == 0)
716 continue;
717 if (ret < 0)
718 goto error;
719 break;
720 }
721 btrfs_item_key_to_cpu(l, &key, slot);
722
723 if (key.objectid < device->devid)
724 goto next;
725 if (key.objectid > device->devid)
726 break;
727 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY)
728 goto next;
729 if (key.offset > search_end)
730 break;
731 if (key.offset > search_start)
732 free_bytes += key.offset - search_start;
733
734 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
735 extent_end = key.offset + btrfs_dev_extent_length(l,
736 dev_extent);
737 if (extent_end > search_start)
738 search_start = extent_end;
739 if (search_start > search_end)
740 break;
741 next:
742 path->slots[0]++;
743 cond_resched();
744 }
745
746 if (search_start < search_end)
747 free_bytes += search_end - search_start;
748
749 *avail_bytes = free_bytes;
750 ret = 0;
751 error:
752 btrfs_free_path(path);
753 return ret;
754 }
755
756 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
757 struct btrfs_root *extent_root, u64 *start,
758 u64 *num_bytes, u64 type)
759 {
760 u64 dev_offset;
761 struct btrfs_fs_info *info = extent_root->fs_info;
762 struct btrfs_root *chunk_root = info->chunk_root;
763 struct btrfs_stripe *stripes;
764 struct btrfs_device *device = NULL;
765 struct btrfs_chunk *chunk;
766 struct list_head private_devs;
767 struct list_head *dev_list = &info->fs_devices->devices;
768 struct list_head *cur;
769 struct map_lookup *map;
770 int min_stripe_size = 1 * 1024 * 1024;
771 u64 calc_size = 8 * 1024 * 1024;
772 u64 min_free;
773 u64 max_chunk_size = 4 * calc_size;
774 u64 avail = 0;
775 u64 max_avail = 0;
776 u64 percent_max;
777 int num_stripes = 1;
778 int min_stripes = 1;
779 int sub_stripes = 0;
780 int looped = 0;
781 int ret;
782 int index;
783 int stripe_len = BTRFS_STRIPE_LEN;
784 struct btrfs_key key;
785 u64 offset;
786
787 if (list_empty(dev_list)) {
788 return -ENOSPC;
789 }
790
791 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
792 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
793 BTRFS_BLOCK_GROUP_RAID10 |
794 BTRFS_BLOCK_GROUP_DUP)) {
795 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
796 calc_size = 8 * 1024 * 1024;
797 max_chunk_size = calc_size * 2;
798 min_stripe_size = 1 * 1024 * 1024;
799 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
800 calc_size = 1024 * 1024 * 1024;
801 max_chunk_size = 10 * calc_size;
802 min_stripe_size = 64 * 1024 * 1024;
803 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
804 calc_size = 1024 * 1024 * 1024;
805 max_chunk_size = 4 * calc_size;
806 min_stripe_size = 32 * 1024 * 1024;
807 }
808 }
809 if (type & BTRFS_BLOCK_GROUP_RAID1) {
810 num_stripes = min_t(u64, 2,
811 btrfs_super_num_devices(info->super_copy));
812 if (num_stripes < 2)
813 return -ENOSPC;
814 min_stripes = 2;
815 }
816 if (type & BTRFS_BLOCK_GROUP_DUP) {
817 num_stripes = 2;
818 min_stripes = 2;
819 }
820 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
821 num_stripes = btrfs_super_num_devices(info->super_copy);
822 min_stripes = 2;
823 }
824 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
825 num_stripes = btrfs_super_num_devices(info->super_copy);
826 if (num_stripes < 4)
827 return -ENOSPC;
828 num_stripes &= ~(u32)1;
829 sub_stripes = 2;
830 min_stripes = 4;
831 }
832 if (type & (BTRFS_BLOCK_GROUP_RAID5)) {
833 num_stripes = btrfs_super_num_devices(info->super_copy);
834 if (num_stripes < 2)
835 return -ENOSPC;
836 min_stripes = 2;
837 stripe_len = find_raid56_stripe_len(num_stripes - 1,
838 btrfs_super_stripesize(info->super_copy));
839 }
840 if (type & (BTRFS_BLOCK_GROUP_RAID6)) {
841 num_stripes = btrfs_super_num_devices(info->super_copy);
842 if (num_stripes < 3)
843 return -ENOSPC;
844 min_stripes = 3;
845 stripe_len = find_raid56_stripe_len(num_stripes - 2,
846 btrfs_super_stripesize(info->super_copy));
847 }
848
849 /* we don't want a chunk larger than 10% of the FS */
850 percent_max = div_factor(btrfs_super_total_bytes(info->super_copy), 1);
851 max_chunk_size = min(percent_max, max_chunk_size);
852
853 again:
854 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
855 max_chunk_size) {
856 calc_size = max_chunk_size;
857 calc_size /= num_stripes;
858 calc_size /= stripe_len;
859 calc_size *= stripe_len;
860 }
861 /* we don't want tiny stripes */
862 calc_size = max_t(u64, calc_size, min_stripe_size);
863
864 calc_size /= stripe_len;
865 calc_size *= stripe_len;
866 INIT_LIST_HEAD(&private_devs);
867 cur = dev_list->next;
868 index = 0;
869
870 if (type & BTRFS_BLOCK_GROUP_DUP)
871 min_free = calc_size * 2;
872 else
873 min_free = calc_size;
874
875 /* build a private list of devices we will allocate from */
876 while(index < num_stripes) {
877 device = list_entry(cur, struct btrfs_device, dev_list);
878 ret = btrfs_device_avail_bytes(trans, device, &avail);
879 if (ret)
880 return ret;
881 cur = cur->next;
882 if (avail >= min_free) {
883 list_move_tail(&device->dev_list, &private_devs);
884 index++;
885 if (type & BTRFS_BLOCK_GROUP_DUP)
886 index++;
887 } else if (avail > max_avail)
888 max_avail = avail;
889 if (cur == dev_list)
890 break;
891 }
892 if (index < num_stripes) {
893 list_splice(&private_devs, dev_list);
894 if (index >= min_stripes) {
895 num_stripes = index;
896 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
897 num_stripes /= sub_stripes;
898 num_stripes *= sub_stripes;
899 }
900 looped = 1;
901 goto again;
902 }
903 if (!looped && max_avail > 0) {
904 looped = 1;
905 calc_size = max_avail;
906 goto again;
907 }
908 return -ENOSPC;
909 }
910 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
911 &offset);
912 if (ret)
913 return ret;
914 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
915 key.type = BTRFS_CHUNK_ITEM_KEY;
916 key.offset = offset;
917
918 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
919 if (!chunk)
920 return -ENOMEM;
921
922 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
923 if (!map) {
924 kfree(chunk);
925 return -ENOMEM;
926 }
927
928 stripes = &chunk->stripe;
929 *num_bytes = chunk_bytes_by_type(type, calc_size,
930 num_stripes, sub_stripes);
931 index = 0;
932 while(index < num_stripes) {
933 struct btrfs_stripe *stripe;
934 BUG_ON(list_empty(&private_devs));
935 cur = private_devs.next;
936 device = list_entry(cur, struct btrfs_device, dev_list);
937
938 /* loop over this device again if we're doing a dup group */
939 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
940 (index == num_stripes - 1))
941 list_move_tail(&device->dev_list, dev_list);
942
943 ret = btrfs_alloc_dev_extent(trans, device,
944 info->chunk_root->root_key.objectid,
945 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
946 calc_size, &dev_offset);
947 BUG_ON(ret);
948
949 device->bytes_used += calc_size;
950 ret = btrfs_update_device(trans, device);
951 BUG_ON(ret);
952
953 map->stripes[index].dev = device;
954 map->stripes[index].physical = dev_offset;
955 stripe = stripes + index;
956 btrfs_set_stack_stripe_devid(stripe, device->devid);
957 btrfs_set_stack_stripe_offset(stripe, dev_offset);
958 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
959 index++;
960 }
961 BUG_ON(!list_empty(&private_devs));
962
963 /* key was set above */
964 btrfs_set_stack_chunk_length(chunk, *num_bytes);
965 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
966 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
967 btrfs_set_stack_chunk_type(chunk, type);
968 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
969 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
970 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
971 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
972 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
973 map->sector_size = extent_root->sectorsize;
974 map->stripe_len = stripe_len;
975 map->io_align = stripe_len;
976 map->io_width = stripe_len;
977 map->type = type;
978 map->num_stripes = num_stripes;
979 map->sub_stripes = sub_stripes;
980
981 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
982 btrfs_chunk_item_size(num_stripes));
983 BUG_ON(ret);
984 *start = key.offset;;
985
986 map->ce.start = key.offset;
987 map->ce.size = *num_bytes;
988
989 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
990 BUG_ON(ret);
991
992 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
993 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
994 chunk, btrfs_chunk_item_size(num_stripes));
995 BUG_ON(ret);
996 }
997
998 kfree(chunk);
999 return ret;
1000 }
1001
1002 int btrfs_alloc_data_chunk(struct btrfs_trans_handle *trans,
1003 struct btrfs_root *extent_root, u64 *start,
1004 u64 num_bytes, u64 type)
1005 {
1006 u64 dev_offset;
1007 struct btrfs_fs_info *info = extent_root->fs_info;
1008 struct btrfs_root *chunk_root = info->chunk_root;
1009 struct btrfs_stripe *stripes;
1010 struct btrfs_device *device = NULL;
1011 struct btrfs_chunk *chunk;
1012 struct list_head *dev_list = &info->fs_devices->devices;
1013 struct list_head *cur;
1014 struct map_lookup *map;
1015 u64 calc_size = 8 * 1024 * 1024;
1016 int num_stripes = 1;
1017 int sub_stripes = 0;
1018 int ret;
1019 int index;
1020 int stripe_len = BTRFS_STRIPE_LEN;
1021 struct btrfs_key key;
1022
1023 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
1024 key.type = BTRFS_CHUNK_ITEM_KEY;
1025 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
1026 &key.offset);
1027 if (ret)
1028 return ret;
1029
1030 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
1031 if (!chunk)
1032 return -ENOMEM;
1033
1034 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1035 if (!map) {
1036 kfree(chunk);
1037 return -ENOMEM;
1038 }
1039
1040 stripes = &chunk->stripe;
1041 calc_size = num_bytes;
1042
1043 index = 0;
1044 cur = dev_list->next;
1045 device = list_entry(cur, struct btrfs_device, dev_list);
1046
1047 while (index < num_stripes) {
1048 struct btrfs_stripe *stripe;
1049
1050 ret = btrfs_alloc_dev_extent(trans, device,
1051 info->chunk_root->root_key.objectid,
1052 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
1053 calc_size, &dev_offset);
1054 BUG_ON(ret);
1055
1056 device->bytes_used += calc_size;
1057 ret = btrfs_update_device(trans, device);
1058 BUG_ON(ret);
1059
1060 map->stripes[index].dev = device;
1061 map->stripes[index].physical = dev_offset;
1062 stripe = stripes + index;
1063 btrfs_set_stack_stripe_devid(stripe, device->devid);
1064 btrfs_set_stack_stripe_offset(stripe, dev_offset);
1065 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
1066 index++;
1067 }
1068
1069 /* key was set above */
1070 btrfs_set_stack_chunk_length(chunk, num_bytes);
1071 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
1072 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
1073 btrfs_set_stack_chunk_type(chunk, type);
1074 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
1075 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
1076 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
1077 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
1078 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
1079 map->sector_size = extent_root->sectorsize;
1080 map->stripe_len = stripe_len;
1081 map->io_align = stripe_len;
1082 map->io_width = stripe_len;
1083 map->type = type;
1084 map->num_stripes = num_stripes;
1085 map->sub_stripes = sub_stripes;
1086
1087 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
1088 btrfs_chunk_item_size(num_stripes));
1089 BUG_ON(ret);
1090 *start = key.offset;
1091
1092 map->ce.start = key.offset;
1093 map->ce.size = num_bytes;
1094
1095 ret = insert_cache_extent(&info->mapping_tree.cache_tree, &map->ce);
1096 BUG_ON(ret);
1097
1098 kfree(chunk);
1099 return ret;
1100 }
1101
1102 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
1103 {
1104 struct cache_extent *ce;
1105 struct map_lookup *map;
1106 int ret;
1107
1108 ce = search_cache_extent(&map_tree->cache_tree, logical);
1109 BUG_ON(!ce);
1110 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
1111 map = container_of(ce, struct map_lookup, ce);
1112
1113 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
1114 ret = map->num_stripes;
1115 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1116 ret = map->sub_stripes;
1117 else if (map->type & BTRFS_BLOCK_GROUP_RAID5)
1118 ret = 2;
1119 else if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1120 ret = 3;
1121 else
1122 ret = 1;
1123 return ret;
1124 }
1125
1126 int btrfs_next_metadata(struct btrfs_mapping_tree *map_tree, u64 *logical,
1127 u64 *size)
1128 {
1129 struct cache_extent *ce;
1130 struct map_lookup *map;
1131
1132 ce = search_cache_extent(&map_tree->cache_tree, *logical);
1133
1134 while (ce) {
1135 ce = next_cache_extent(ce);
1136 if (!ce)
1137 return -ENOENT;
1138
1139 map = container_of(ce, struct map_lookup, ce);
1140 if (map->type & BTRFS_BLOCK_GROUP_METADATA) {
1141 *logical = ce->start;
1142 *size = ce->size;
1143 return 0;
1144 }
1145 }
1146
1147 return -ENOENT;
1148 }
1149
1150 int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree,
1151 u64 chunk_start, u64 physical, u64 devid,
1152 u64 **logical, int *naddrs, int *stripe_len)
1153 {
1154 struct cache_extent *ce;
1155 struct map_lookup *map;
1156 u64 *buf;
1157 u64 bytenr;
1158 u64 length;
1159 u64 stripe_nr;
1160 u64 rmap_len;
1161 int i, j, nr = 0;
1162
1163 ce = search_cache_extent(&map_tree->cache_tree, chunk_start);
1164 BUG_ON(!ce);
1165 map = container_of(ce, struct map_lookup, ce);
1166
1167 length = ce->size;
1168 rmap_len = map->stripe_len;
1169 if (map->type & BTRFS_BLOCK_GROUP_RAID10)
1170 length = ce->size / (map->num_stripes / map->sub_stripes);
1171 else if (map->type & BTRFS_BLOCK_GROUP_RAID0)
1172 length = ce->size / map->num_stripes;
1173 else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1174 BTRFS_BLOCK_GROUP_RAID6)) {
1175 length = ce->size / nr_data_stripes(map);
1176 rmap_len = map->stripe_len * nr_data_stripes(map);
1177 }
1178
1179 buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1180
1181 for (i = 0; i < map->num_stripes; i++) {
1182 if (devid && map->stripes[i].dev->devid != devid)
1183 continue;
1184 if (map->stripes[i].physical > physical ||
1185 map->stripes[i].physical + length <= physical)
1186 continue;
1187
1188 stripe_nr = (physical - map->stripes[i].physical) /
1189 map->stripe_len;
1190
1191 if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1192 stripe_nr = (stripe_nr * map->num_stripes + i) /
1193 map->sub_stripes;
1194 } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) {
1195 stripe_nr = stripe_nr * map->num_stripes + i;
1196 } /* else if RAID[56], multiply by nr_data_stripes().
1197 * Alternatively, just use rmap_len below instead of
1198 * map->stripe_len */
1199
1200 bytenr = ce->start + stripe_nr * rmap_len;
1201 for (j = 0; j < nr; j++) {
1202 if (buf[j] == bytenr)
1203 break;
1204 }
1205 if (j == nr)
1206 buf[nr++] = bytenr;
1207 }
1208
1209 *logical = buf;
1210 *naddrs = nr;
1211 *stripe_len = rmap_len;
1212
1213 return 0;
1214 }
1215
1216 static inline int parity_smaller(u64 a, u64 b)
1217 {
1218 return a > b;
1219 }
1220
1221 /* Bubble-sort the stripe set to put the parity/syndrome stripes last */
1222 static void sort_parity_stripes(struct btrfs_multi_bio *bbio, u64 *raid_map)
1223 {
1224 struct btrfs_bio_stripe s;
1225 int i;
1226 u64 l;
1227 int again = 1;
1228
1229 while (again) {
1230 again = 0;
1231 for (i = 0; i < bbio->num_stripes - 1; i++) {
1232 if (parity_smaller(raid_map[i], raid_map[i+1])) {
1233 s = bbio->stripes[i];
1234 l = raid_map[i];
1235 bbio->stripes[i] = bbio->stripes[i+1];
1236 raid_map[i] = raid_map[i+1];
1237 bbio->stripes[i+1] = s;
1238 raid_map[i+1] = l;
1239 again = 1;
1240 }
1241 }
1242 }
1243 }
1244
1245 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1246 u64 logical, u64 *length,
1247 struct btrfs_multi_bio **multi_ret, int mirror_num,
1248 u64 **raid_map_ret)
1249 {
1250 return __btrfs_map_block(map_tree, rw, logical, length, NULL,
1251 multi_ret, mirror_num, raid_map_ret);
1252 }
1253
1254 int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
1255 u64 logical, u64 *length, u64 *type,
1256 struct btrfs_multi_bio **multi_ret, int mirror_num,
1257 u64 **raid_map_ret)
1258 {
1259 struct cache_extent *ce;
1260 struct map_lookup *map;
1261 u64 offset;
1262 u64 stripe_offset;
1263 u64 stripe_nr;
1264 u64 *raid_map = NULL;
1265 int stripes_allocated = 8;
1266 int stripes_required = 1;
1267 int stripe_index;
1268 int i;
1269 struct btrfs_multi_bio *multi = NULL;
1270
1271 if (multi_ret && rw == READ) {
1272 stripes_allocated = 1;
1273 }
1274 again:
1275 ce = search_cache_extent(&map_tree->cache_tree, logical);
1276 if (!ce) {
1277 kfree(multi);
1278 return -ENOENT;
1279 }
1280 if (ce->start > logical || ce->start + ce->size < logical) {
1281 kfree(multi);
1282 return -ENOENT;
1283 }
1284
1285 if (multi_ret) {
1286 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
1287 GFP_NOFS);
1288 if (!multi)
1289 return -ENOMEM;
1290 }
1291 map = container_of(ce, struct map_lookup, ce);
1292 offset = logical - ce->start;
1293
1294 if (rw == WRITE) {
1295 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
1296 BTRFS_BLOCK_GROUP_DUP)) {
1297 stripes_required = map->num_stripes;
1298 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1299 stripes_required = map->sub_stripes;
1300 }
1301 }
1302 if (map->type & (BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6)
1303 && multi_ret && ((rw & WRITE) || mirror_num > 1) && raid_map_ret) {
1304 /* RAID[56] write or recovery. Return all stripes */
1305 stripes_required = map->num_stripes;
1306
1307 /* Only allocate the map if we've already got a large enough multi_ret */
1308 if (stripes_allocated >= stripes_required) {
1309 raid_map = kmalloc(sizeof(u64) * map->num_stripes, GFP_NOFS);
1310 if (!raid_map) {
1311 kfree(multi);
1312 return -ENOMEM;
1313 }
1314 }
1315 }
1316
1317 /* if our multi bio struct is too small, back off and try again */
1318 if (multi_ret && stripes_allocated < stripes_required) {
1319 stripes_allocated = stripes_required;
1320 kfree(multi);
1321 multi = NULL;
1322 goto again;
1323 }
1324 stripe_nr = offset;
1325 /*
1326 * stripe_nr counts the total number of stripes we have to stride
1327 * to get to this block
1328 */
1329 stripe_nr = stripe_nr / map->stripe_len;
1330
1331 stripe_offset = stripe_nr * map->stripe_len;
1332 BUG_ON(offset < stripe_offset);
1333
1334 /* stripe_offset is the offset of this block in its stripe*/
1335 stripe_offset = offset - stripe_offset;
1336
1337 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
1338 BTRFS_BLOCK_GROUP_RAID5 | BTRFS_BLOCK_GROUP_RAID6 |
1339 BTRFS_BLOCK_GROUP_RAID10 |
1340 BTRFS_BLOCK_GROUP_DUP)) {
1341 /* we limit the length of each bio to what fits in a stripe */
1342 *length = min_t(u64, ce->size - offset,
1343 map->stripe_len - stripe_offset);
1344 } else {
1345 *length = ce->size - offset;
1346 }
1347
1348 if (!multi_ret)
1349 goto out;
1350
1351 multi->num_stripes = 1;
1352 stripe_index = 0;
1353 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
1354 if (rw == WRITE)
1355 multi->num_stripes = map->num_stripes;
1356 else if (mirror_num)
1357 stripe_index = mirror_num - 1;
1358 else
1359 stripe_index = stripe_nr % map->num_stripes;
1360 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
1361 int factor = map->num_stripes / map->sub_stripes;
1362
1363 stripe_index = stripe_nr % factor;
1364 stripe_index *= map->sub_stripes;
1365
1366 if (rw == WRITE)
1367 multi->num_stripes = map->sub_stripes;
1368 else if (mirror_num)
1369 stripe_index += mirror_num - 1;
1370
1371 stripe_nr = stripe_nr / factor;
1372 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
1373 if (rw == WRITE)
1374 multi->num_stripes = map->num_stripes;
1375 else if (mirror_num)
1376 stripe_index = mirror_num - 1;
1377 } else if (map->type & (BTRFS_BLOCK_GROUP_RAID5 |
1378 BTRFS_BLOCK_GROUP_RAID6)) {
1379
1380 if (raid_map) {
1381 int rot;
1382 u64 tmp;
1383 u64 raid56_full_stripe_start;
1384 u64 full_stripe_len = nr_data_stripes(map) * map->stripe_len;
1385
1386 /*
1387 * align the start of our data stripe in the logical
1388 * address space
1389 */
1390 raid56_full_stripe_start = offset / full_stripe_len;
1391 raid56_full_stripe_start *= full_stripe_len;
1392
1393 /* get the data stripe number */
1394 stripe_nr = raid56_full_stripe_start / map->stripe_len;
1395 stripe_nr = stripe_nr / nr_data_stripes(map);
1396
1397 /* Work out the disk rotation on this stripe-set */
1398 rot = stripe_nr % map->num_stripes;
1399
1400 /* Fill in the logical address of each stripe */
1401 tmp = stripe_nr * nr_data_stripes(map);
1402
1403 for (i = 0; i < nr_data_stripes(map); i++)
1404 raid_map[(i+rot) % map->num_stripes] =
1405 ce->start + (tmp + i) * map->stripe_len;
1406
1407 raid_map[(i+rot) % map->num_stripes] = BTRFS_RAID5_P_STRIPE;
1408 if (map->type & BTRFS_BLOCK_GROUP_RAID6)
1409 raid_map[(i+rot+1) % map->num_stripes] = BTRFS_RAID6_Q_STRIPE;
1410
1411 *length = map->stripe_len;
1412 stripe_index = 0;
1413 stripe_offset = 0;
1414 multi->num_stripes = map->num_stripes;
1415 } else {
1416 stripe_index = stripe_nr % nr_data_stripes(map);
1417 stripe_nr = stripe_nr / nr_data_stripes(map);
1418
1419 /*
1420 * Mirror #0 or #1 means the original data block.
1421 * Mirror #2 is RAID5 parity block.
1422 * Mirror #3 is RAID6 Q block.
1423 */
1424 if (mirror_num > 1)
1425 stripe_index = nr_data_stripes(map) + mirror_num - 2;
1426
1427 /* We distribute the parity blocks across stripes */
1428 stripe_index = (stripe_nr + stripe_index) % map->num_stripes;
1429 }
1430 } else {
1431 /*
1432 * after this do_div call, stripe_nr is the number of stripes
1433 * on this device we have to walk to find the data, and
1434 * stripe_index is the number of our device in the stripe array
1435 */
1436 stripe_index = stripe_nr % map->num_stripes;
1437 stripe_nr = stripe_nr / map->num_stripes;
1438 }
1439 BUG_ON(stripe_index >= map->num_stripes);
1440
1441 for (i = 0; i < multi->num_stripes; i++) {
1442 multi->stripes[i].physical =
1443 map->stripes[stripe_index].physical + stripe_offset +
1444 stripe_nr * map->stripe_len;
1445 multi->stripes[i].dev = map->stripes[stripe_index].dev;
1446 stripe_index++;
1447 }
1448 *multi_ret = multi;
1449
1450 if (type)
1451 *type = map->type;
1452
1453 if (raid_map) {
1454 sort_parity_stripes(multi, raid_map);
1455 *raid_map_ret = raid_map;
1456 }
1457 out:
1458 return 0;
1459 }
1460
1461 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
1462 u8 *uuid, u8 *fsid)
1463 {
1464 struct btrfs_device *device;
1465 struct btrfs_fs_devices *cur_devices;
1466
1467 cur_devices = root->fs_info->fs_devices;
1468 while (cur_devices) {
1469 if (!fsid ||
1470 !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1471 device = __find_device(&cur_devices->devices,
1472 devid, uuid);
1473 if (device)
1474 return device;
1475 }
1476 cur_devices = cur_devices->seed;
1477 }
1478 return NULL;
1479 }
1480
1481 struct btrfs_device *
1482 btrfs_find_device_by_devid(struct btrfs_fs_devices *fs_devices,
1483 u64 devid, int instance)
1484 {
1485 struct list_head *head = &fs_devices->devices;
1486 struct btrfs_device *dev;
1487 int num_found = 0;
1488
1489 list_for_each_entry(dev, head, dev_list) {
1490 if (dev->devid == devid && num_found++ == instance)
1491 return dev;
1492 }
1493 return NULL;
1494 }
1495
1496 int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset)
1497 {
1498 struct cache_extent *ce;
1499 struct map_lookup *map;
1500 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1501 int readonly = 0;
1502 int i;
1503
1504 /*
1505 * During chunk recovering, we may fail to find block group's
1506 * corresponding chunk, we will rebuild it later
1507 */
1508 ce = search_cache_extent(&map_tree->cache_tree, chunk_offset);
1509 if (!root->fs_info->is_chunk_recover)
1510 BUG_ON(!ce);
1511 else
1512 return 0;
1513
1514 map = container_of(ce, struct map_lookup, ce);
1515 for (i = 0; i < map->num_stripes; i++) {
1516 if (!map->stripes[i].dev->writeable) {
1517 readonly = 1;
1518 break;
1519 }
1520 }
1521
1522 return readonly;
1523 }
1524
1525 static struct btrfs_device *fill_missing_device(u64 devid)
1526 {
1527 struct btrfs_device *device;
1528
1529 device = kzalloc(sizeof(*device), GFP_NOFS);
1530 device->devid = devid;
1531 device->fd = -1;
1532 return device;
1533 }
1534
1535 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1536 struct extent_buffer *leaf,
1537 struct btrfs_chunk *chunk)
1538 {
1539 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1540 struct map_lookup *map;
1541 struct cache_extent *ce;
1542 u64 logical;
1543 u64 length;
1544 u64 devid;
1545 u8 uuid[BTRFS_UUID_SIZE];
1546 int num_stripes;
1547 int ret;
1548 int i;
1549
1550 logical = key->offset;
1551 length = btrfs_chunk_length(leaf, chunk);
1552
1553 ce = search_cache_extent(&map_tree->cache_tree, logical);
1554
1555 /* already mapped? */
1556 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1557 return 0;
1558 }
1559
1560 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1561 map = kmalloc(btrfs_map_lookup_size(num_stripes), GFP_NOFS);
1562 if (!map)
1563 return -ENOMEM;
1564
1565 map->ce.start = logical;
1566 map->ce.size = length;
1567 map->num_stripes = num_stripes;
1568 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1569 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1570 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1571 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1572 map->type = btrfs_chunk_type(leaf, chunk);
1573 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1574
1575 for (i = 0; i < num_stripes; i++) {
1576 map->stripes[i].physical =
1577 btrfs_stripe_offset_nr(leaf, chunk, i);
1578 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1579 read_extent_buffer(leaf, uuid, (unsigned long)
1580 btrfs_stripe_dev_uuid_nr(chunk, i),
1581 BTRFS_UUID_SIZE);
1582 map->stripes[i].dev = btrfs_find_device(root, devid, uuid,
1583 NULL);
1584 if (!map->stripes[i].dev) {
1585 map->stripes[i].dev = fill_missing_device(devid);
1586 printf("warning, device %llu is missing\n",
1587 (unsigned long long)devid);
1588 }
1589
1590 }
1591 ret = insert_cache_extent(&map_tree->cache_tree, &map->ce);
1592 BUG_ON(ret);
1593
1594 return 0;
1595 }
1596
1597 static int fill_device_from_item(struct extent_buffer *leaf,
1598 struct btrfs_dev_item *dev_item,
1599 struct btrfs_device *device)
1600 {
1601 unsigned long ptr;
1602
1603 device->devid = btrfs_device_id(leaf, dev_item);
1604 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1605 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1606 device->type = btrfs_device_type(leaf, dev_item);
1607 device->io_align = btrfs_device_io_align(leaf, dev_item);
1608 device->io_width = btrfs_device_io_width(leaf, dev_item);
1609 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1610
1611 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1612 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1613
1614 return 0;
1615 }
1616
1617 static int open_seed_devices(struct btrfs_root *root, u8 *fsid)
1618 {
1619 struct btrfs_fs_devices *fs_devices;
1620 int ret;
1621
1622 fs_devices = root->fs_info->fs_devices->seed;
1623 while (fs_devices) {
1624 if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) {
1625 ret = 0;
1626 goto out;
1627 }
1628 fs_devices = fs_devices->seed;
1629 }
1630
1631 fs_devices = find_fsid(fsid);
1632 if (!fs_devices) {
1633 ret = -ENOENT;
1634 goto out;
1635 }
1636
1637 ret = btrfs_open_devices(fs_devices, O_RDONLY);
1638 if (ret)
1639 goto out;
1640
1641 fs_devices->seed = root->fs_info->fs_devices->seed;
1642 root->fs_info->fs_devices->seed = fs_devices;
1643 out:
1644 return ret;
1645 }
1646
1647 static int read_one_dev(struct btrfs_root *root,
1648 struct extent_buffer *leaf,
1649 struct btrfs_dev_item *dev_item)
1650 {
1651 struct btrfs_device *device;
1652 u64 devid;
1653 int ret = 0;
1654 u8 fs_uuid[BTRFS_UUID_SIZE];
1655 u8 dev_uuid[BTRFS_UUID_SIZE];
1656
1657 devid = btrfs_device_id(leaf, dev_item);
1658 read_extent_buffer(leaf, dev_uuid,
1659 (unsigned long)btrfs_device_uuid(dev_item),
1660 BTRFS_UUID_SIZE);
1661 read_extent_buffer(leaf, fs_uuid,
1662 (unsigned long)btrfs_device_fsid(dev_item),
1663 BTRFS_UUID_SIZE);
1664
1665 if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) {
1666 ret = open_seed_devices(root, fs_uuid);
1667 if (ret)
1668 return ret;
1669 }
1670
1671 device = btrfs_find_device(root, devid, dev_uuid, fs_uuid);
1672 if (!device) {
1673 printk("warning devid %llu not found already\n",
1674 (unsigned long long)devid);
1675 device = kzalloc(sizeof(*device), GFP_NOFS);
1676 if (!device)
1677 return -ENOMEM;
1678 device->fd = -1;
1679 list_add(&device->dev_list,
1680 &root->fs_info->fs_devices->devices);
1681 }
1682
1683 fill_device_from_item(leaf, dev_item, device);
1684 device->dev_root = root->fs_info->dev_root;
1685 return ret;
1686 }
1687
1688 int btrfs_read_sys_array(struct btrfs_root *root)
1689 {
1690 struct btrfs_super_block *super_copy = root->fs_info->super_copy;
1691 struct extent_buffer *sb;
1692 struct btrfs_disk_key *disk_key;
1693 struct btrfs_chunk *chunk;
1694 struct btrfs_key key;
1695 u32 num_stripes;
1696 u32 len = 0;
1697 u8 *ptr;
1698 u8 *array_end;
1699 int ret = 0;
1700
1701 sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET,
1702 BTRFS_SUPER_INFO_SIZE);
1703 if (!sb)
1704 return -ENOMEM;
1705 btrfs_set_buffer_uptodate(sb);
1706 write_extent_buffer(sb, super_copy, 0, sizeof(*super_copy));
1707 array_end = ((u8 *)super_copy->sys_chunk_array) +
1708 btrfs_super_sys_array_size(super_copy);
1709
1710 /*
1711 * we do this loop twice, once for the device items and
1712 * once for all of the chunks. This way there are device
1713 * structs filled in for every chunk
1714 */
1715 ptr = super_copy->sys_chunk_array;
1716
1717 while (ptr < array_end) {
1718 disk_key = (struct btrfs_disk_key *)ptr;
1719 btrfs_disk_key_to_cpu(&key, disk_key);
1720
1721 len = sizeof(*disk_key);
1722 ptr += len;
1723
1724 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1725 chunk = (struct btrfs_chunk *)(ptr - (u8 *)super_copy);
1726 ret = read_one_chunk(root, &key, sb, chunk);
1727 if (ret)
1728 break;
1729 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1730 len = btrfs_chunk_item_size(num_stripes);
1731 } else {
1732 BUG();
1733 }
1734 ptr += len;
1735 }
1736 free_extent_buffer(sb);
1737 return ret;
1738 }
1739
1740 int btrfs_read_chunk_tree(struct btrfs_root *root)
1741 {
1742 struct btrfs_path *path;
1743 struct extent_buffer *leaf;
1744 struct btrfs_key key;
1745 struct btrfs_key found_key;
1746 int ret;
1747 int slot;
1748
1749 root = root->fs_info->chunk_root;
1750
1751 path = btrfs_alloc_path();
1752 if (!path)
1753 return -ENOMEM;
1754
1755 /*
1756 * Read all device items, and then all the chunk items. All
1757 * device items are found before any chunk item (their object id
1758 * is smaller than the lowest possible object id for a chunk
1759 * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID).
1760 */
1761 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1762 key.offset = 0;
1763 key.type = 0;
1764 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1765 if (ret < 0)
1766 goto error;
1767 while(1) {
1768 leaf = path->nodes[0];
1769 slot = path->slots[0];
1770 if (slot >= btrfs_header_nritems(leaf)) {
1771 ret = btrfs_next_leaf(root, path);
1772 if (ret == 0)
1773 continue;
1774 if (ret < 0)
1775 goto error;
1776 break;
1777 }
1778 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1779 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1780 struct btrfs_dev_item *dev_item;
1781 dev_item = btrfs_item_ptr(leaf, slot,
1782 struct btrfs_dev_item);
1783 ret = read_one_dev(root, leaf, dev_item);
1784 BUG_ON(ret);
1785 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1786 struct btrfs_chunk *chunk;
1787 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1788 ret = read_one_chunk(root, &found_key, leaf, chunk);
1789 BUG_ON(ret);
1790 }
1791 path->slots[0]++;
1792 }
1793
1794 ret = 0;
1795 error:
1796 btrfs_free_path(path);
1797 return ret;
1798 }
1799
1800 struct list_head *btrfs_scanned_uuids(void)
1801 {
1802 return &fs_uuids;
1803 }
1804
1805 static int rmw_eb(struct btrfs_fs_info *info,
1806 struct extent_buffer *eb, struct extent_buffer *orig_eb)
1807 {
1808 int ret;
1809 unsigned long orig_off = 0;
1810 unsigned long dest_off = 0;
1811 unsigned long copy_len = eb->len;
1812
1813 ret = read_whole_eb(info, eb, 0);
1814 if (ret)
1815 return ret;
1816
1817 if (eb->start + eb->len <= orig_eb->start ||
1818 eb->start >= orig_eb->start + orig_eb->len)
1819 return 0;
1820 /*
1821 * | ----- orig_eb ------- |
1822 * | ----- stripe ------- |
1823 * | ----- orig_eb ------- |
1824 * | ----- orig_eb ------- |
1825 */
1826 if (eb->start > orig_eb->start)
1827 orig_off = eb->start - orig_eb->start;
1828 if (orig_eb->start > eb->start)
1829 dest_off = orig_eb->start - eb->start;
1830
1831 if (copy_len > orig_eb->len - orig_off)
1832 copy_len = orig_eb->len - orig_off;
1833 if (copy_len > eb->len - dest_off)
1834 copy_len = eb->len - dest_off;
1835
1836 memcpy(eb->data + dest_off, orig_eb->data + orig_off, copy_len);
1837 return 0;
1838 }
1839
1840 static void split_eb_for_raid56(struct btrfs_fs_info *info,
1841 struct extent_buffer *orig_eb,
1842 struct extent_buffer **ebs,
1843 u64 stripe_len, u64 *raid_map,
1844 int num_stripes)
1845 {
1846 struct extent_buffer *eb;
1847 u64 start = orig_eb->start;
1848 u64 this_eb_start;
1849 int i;
1850 int ret;
1851
1852 for (i = 0; i < num_stripes; i++) {
1853 if (raid_map[i] >= BTRFS_RAID5_P_STRIPE)
1854 break;
1855
1856 eb = malloc(sizeof(struct extent_buffer) + stripe_len);
1857 if (!eb)
1858 BUG();
1859 memset(eb, 0, sizeof(struct extent_buffer) + stripe_len);
1860
1861 eb->start = raid_map[i];
1862 eb->len = stripe_len;
1863 eb->refs = 1;
1864 eb->flags = 0;
1865 eb->fd = -1;
1866 eb->dev_bytenr = (u64)-1;
1867
1868 this_eb_start = raid_map[i];
1869
1870 if (start > this_eb_start ||
1871 start + orig_eb->len < this_eb_start + stripe_len) {
1872 ret = rmw_eb(info, eb, orig_eb);
1873 BUG_ON(ret);
1874 } else {
1875 memcpy(eb->data, orig_eb->data + eb->start - start, stripe_len);
1876 }
1877 ebs[i] = eb;
1878 }
1879 }
1880
1881 int write_raid56_with_parity(struct btrfs_fs_info *info,
1882 struct extent_buffer *eb,
1883 struct btrfs_multi_bio *multi,
1884 u64 stripe_len, u64 *raid_map)
1885 {
1886 struct extent_buffer **ebs, *p_eb = NULL, *q_eb = NULL;
1887 int i;
1888 int j;
1889 int ret;
1890 int alloc_size = eb->len;
1891
1892 ebs = kmalloc(sizeof(*ebs) * multi->num_stripes, GFP_NOFS);
1893 BUG_ON(!ebs);
1894
1895 if (stripe_len > alloc_size)
1896 alloc_size = stripe_len;
1897
1898 split_eb_for_raid56(info, eb, ebs, stripe_len, raid_map,
1899 multi->num_stripes);
1900
1901 for (i = 0; i < multi->num_stripes; i++) {
1902 struct extent_buffer *new_eb;
1903 if (raid_map[i] < BTRFS_RAID5_P_STRIPE) {
1904 ebs[i]->dev_bytenr = multi->stripes[i].physical;
1905 ebs[i]->fd = multi->stripes[i].dev->fd;
1906 multi->stripes[i].dev->total_ios++;
1907 BUG_ON(ebs[i]->start != raid_map[i]);
1908 continue;
1909 }
1910 new_eb = kmalloc(sizeof(*eb) + alloc_size, GFP_NOFS);
1911 BUG_ON(!new_eb);
1912 new_eb->dev_bytenr = multi->stripes[i].physical;
1913 new_eb->fd = multi->stripes[i].dev->fd;
1914 multi->stripes[i].dev->total_ios++;
1915 new_eb->len = stripe_len;
1916
1917 if (raid_map[i] == BTRFS_RAID5_P_STRIPE)
1918 p_eb = new_eb;
1919 else if (raid_map[i] == BTRFS_RAID6_Q_STRIPE)
1920 q_eb = new_eb;
1921 }
1922 if (q_eb) {
1923 void **pointers;
1924
1925 pointers = kmalloc(sizeof(*pointers) * multi->num_stripes,
1926 GFP_NOFS);
1927 BUG_ON(!pointers);
1928
1929 ebs[multi->num_stripes - 2] = p_eb;
1930 ebs[multi->num_stripes - 1] = q_eb;
1931
1932 for (i = 0; i < multi->num_stripes; i++)
1933 pointers[i] = ebs[i]->data;
1934
1935 raid6_gen_syndrome(multi->num_stripes, stripe_len, pointers);
1936 kfree(pointers);
1937 } else {
1938 ebs[multi->num_stripes - 1] = p_eb;
1939 memcpy(p_eb->data, ebs[0]->data, stripe_len);
1940 for (j = 1; j < multi->num_stripes - 1; j++) {
1941 for (i = 0; i < stripe_len; i += sizeof(unsigned long)) {
1942 *(unsigned long *)(p_eb->data + i) ^=
1943 *(unsigned long *)(ebs[j]->data + i);
1944 }
1945 }
1946 }
1947
1948 for (i = 0; i < multi->num_stripes; i++) {
1949 ret = write_extent_to_disk(ebs[i]);
1950 BUG_ON(ret);
1951 if (ebs[i] != eb)
1952 kfree(ebs[i]);
1953 }
1954
1955 kfree(ebs);
1956
1957 return 0;
1958 }
(deprecated) Btrfs progs developments and patch integration