?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Add root tree pointer transaction ids
[btrfs-progs-unstable.git] / volumes.c
blobc94f73c1d5426fb728c28aaf3a734f48bfabd28d
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
33 struct stripe {
34 struct btrfs_device *dev;
35 u64 physical;
36 };
37
38 struct map_lookup {
39 struct cache_extent ce;
40 u64 type;
41 int io_align;
42 int io_width;
43 int stripe_len;
44 int sector_size;
45 int num_stripes;
46 int sub_stripes;
47 struct btrfs_bio_stripe stripes[];
48 };
49
50 #define map_lookup_size(n) (sizeof(struct map_lookup) + \
51 (sizeof(struct btrfs_bio_stripe) * (n)))
52
53 static LIST_HEAD(fs_uuids);
54
55 static struct btrfs_device *__find_device(struct list_head *head, u64 devid,
56 u8 *uuid)
57 {
58 struct btrfs_device *dev;
59 struct list_head *cur;
60
61 list_for_each(cur, head) {
62 dev = list_entry(cur, struct btrfs_device, dev_list);
63 if (dev->devid == devid &&
64 !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) {
65 return dev;
66 }
67 }
68 return NULL;
69 }
70
71 static struct btrfs_fs_devices *find_fsid(u8 *fsid)
72 {
73 struct list_head *cur;
74 struct btrfs_fs_devices *fs_devices;
75
76 list_for_each(cur, &fs_uuids) {
77 fs_devices = list_entry(cur, struct btrfs_fs_devices, list);
78 if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0)
79 return fs_devices;
80 }
81 return NULL;
82 }
83
84 static int device_list_add(const char *path,
85 struct btrfs_super_block *disk_super,
86 u64 devid, struct btrfs_fs_devices **fs_devices_ret)
87 {
88 struct btrfs_device *device;
89 struct btrfs_fs_devices *fs_devices;
90 u64 found_transid = btrfs_super_generation(disk_super);
91
92 fs_devices = find_fsid(disk_super->fsid);
93 if (!fs_devices) {
94 fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS);
95 if (!fs_devices)
96 return -ENOMEM;
97 INIT_LIST_HEAD(&fs_devices->devices);
98 list_add(&fs_devices->list, &fs_uuids);
99 memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE);
100 fs_devices->latest_devid = devid;
101 fs_devices->latest_trans = found_transid;
102 fs_devices->lowest_devid = (u64)-1;
103 device = NULL;
104 } else {
105 device = __find_device(&fs_devices->devices, devid,
106 disk_super->dev_item.uuid);
107 }
108 if (!device) {
109 device = kzalloc(sizeof(*device), GFP_NOFS);
110 if (!device) {
111 /* we can safely leave the fs_devices entry around */
112 return -ENOMEM;
113 }
114 device->devid = devid;
115 memcpy(device->uuid, disk_super->dev_item.uuid,
116 BTRFS_UUID_SIZE);
117 device->name = kstrdup(path, GFP_NOFS);
118 if (!device->name) {
119 kfree(device);
120 return -ENOMEM;
121 }
122 device->label = kstrdup(disk_super->label, GFP_NOFS);
123 device->total_devs = btrfs_super_num_devices(disk_super);
124 device->super_bytes_used = btrfs_super_bytes_used(disk_super);
125 device->total_bytes =
126 btrfs_stack_device_total_bytes(&disk_super->dev_item);
127 device->bytes_used =
128 btrfs_stack_device_bytes_used(&disk_super->dev_item);
129 list_add(&device->dev_list, &fs_devices->devices);
130 }
131
132 if (found_transid > fs_devices->latest_trans) {
133 fs_devices->latest_devid = devid;
134 fs_devices->latest_trans = found_transid;
135 }
136 if (fs_devices->lowest_devid > devid) {
137 fs_devices->lowest_devid = devid;
138 }
139 *fs_devices_ret = fs_devices;
140 return 0;
141 }
142
143 int btrfs_close_devices(struct btrfs_fs_devices *fs_devices)
144 {
145 struct list_head *head = &fs_devices->devices;
146 struct list_head *cur;
147 struct btrfs_device *device;
148
149 list_for_each(cur, head) {
150 device = list_entry(cur, struct btrfs_device, dev_list);
151 close(device->fd);
152 device->fd = -1;
153 }
154 return 0;
155 }
156
157 int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, int flags)
158 {
159 int fd;
160 struct list_head *head = &fs_devices->devices;
161 struct list_head *cur;
162 struct btrfs_device *device;
163 int ret;
164
165 list_for_each(cur, head) {
166 device = list_entry(cur, struct btrfs_device, dev_list);
167
168 fd = open(device->name, flags);
169 if (fd < 0) {
170 ret = -errno;
171 goto fail;
172 }
173
174 if (device->devid == fs_devices->latest_devid)
175 fs_devices->latest_bdev = fd;
176 if (device->devid == fs_devices->lowest_devid)
177 fs_devices->lowest_bdev = fd;
178 device->fd = fd;
179 }
180 return 0;
181 fail:
182 btrfs_close_devices(fs_devices);
183 return ret;
184 }
185
186 int btrfs_scan_one_device(int fd, const char *path,
187 struct btrfs_fs_devices **fs_devices_ret,
188 u64 *total_devs, u64 super_offset)
189 {
190 struct btrfs_super_block *disk_super;
191 char *buf;
192 int ret;
193 u64 devid;
194 char uuidbuf[37];
195
196 buf = malloc(4096);
197 if (!buf) {
198 ret = -ENOMEM;
199 goto error;
200 }
201 ret = pread(fd, buf, 4096, super_offset);
202 if (ret != 4096) {
203 ret = -EIO;
204 goto error;
205 }
206 disk_super = (struct btrfs_super_block *)buf;
207 if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC,
208 sizeof(disk_super->magic))) {
209 ret = -ENOENT;
210 goto error_brelse;
211 }
212 devid = le64_to_cpu(disk_super->dev_item.devid);
213 *total_devs = btrfs_super_num_devices(disk_super);
214 uuid_unparse(disk_super->fsid, uuidbuf);
215
216 ret = device_list_add(path, disk_super, devid, fs_devices_ret);
217
218 error_brelse:
219 free(buf);
220 error:
221 return ret;
222 }
223
224 /*
225 * this uses a pretty simple search, the expectation is that it is
226 * called very infrequently and that a given device has a small number
227 * of extents
228 */
229 static int find_free_dev_extent(struct btrfs_trans_handle *trans,
230 struct btrfs_device *device,
231 struct btrfs_path *path,
232 u64 num_bytes, u64 *start)
233 {
234 struct btrfs_key key;
235 struct btrfs_root *root = device->dev_root;
236 struct btrfs_dev_extent *dev_extent = NULL;
237 u64 hole_size = 0;
238 u64 last_byte = 0;
239 u64 search_start = 0;
240 u64 search_end = device->total_bytes;
241 int ret;
242 int slot = 0;
243 int start_found;
244 struct extent_buffer *l;
245
246 start_found = 0;
247 path->reada = 2;
248
249 /* FIXME use last free of some kind */
250
251 /* we don't want to overwrite the superblock on the drive,
252 * so we make sure to start at an offset of at least 1MB
253 */
254 search_start = max((u64)1024 * 1024, search_start);
255
256 if (root->fs_info->alloc_start + num_bytes <= device->total_bytes)
257 search_start = max(root->fs_info->alloc_start, search_start);
258
259 key.objectid = device->devid;
260 key.offset = search_start;
261 key.type = BTRFS_DEV_EXTENT_KEY;
262 ret = btrfs_search_slot(trans, root, &key, path, 0, 0);
263 if (ret < 0)
264 goto error;
265 ret = btrfs_previous_item(root, path, 0, key.type);
266 if (ret < 0)
267 goto error;
268 l = path->nodes[0];
269 btrfs_item_key_to_cpu(l, &key, path->slots[0]);
270 while (1) {
271 l = path->nodes[0];
272 slot = path->slots[0];
273 if (slot >= btrfs_header_nritems(l)) {
274 ret = btrfs_next_leaf(root, path);
275 if (ret == 0)
276 continue;
277 if (ret < 0)
278 goto error;
279 no_more_items:
280 if (!start_found) {
281 if (search_start >= search_end) {
282 ret = -ENOSPC;
283 goto error;
284 }
285 *start = search_start;
286 start_found = 1;
287 goto check_pending;
288 }
289 *start = last_byte > search_start ?
290 last_byte : search_start;
291 if (search_end <= *start) {
292 ret = -ENOSPC;
293 goto error;
294 }
295 goto check_pending;
296 }
297 btrfs_item_key_to_cpu(l, &key, slot);
298
299 if (key.objectid < device->devid)
300 goto next;
301
302 if (key.objectid > device->devid)
303 goto no_more_items;
304
305 if (key.offset >= search_start && key.offset > last_byte &&
306 start_found) {
307 if (last_byte < search_start)
308 last_byte = search_start;
309 hole_size = key.offset - last_byte;
310 if (key.offset > last_byte &&
311 hole_size >= num_bytes) {
312 *start = last_byte;
313 goto check_pending;
314 }
315 }
316 if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) {
317 goto next;
318 }
319
320 start_found = 1;
321 dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent);
322 last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent);
323 next:
324 path->slots[0]++;
325 cond_resched();
326 }
327 check_pending:
328 /* we have to make sure we didn't find an extent that has already
329 * been allocated by the map tree or the original allocation
330 */
331 btrfs_release_path(root, path);
332 BUG_ON(*start < search_start);
333
334 if (*start + num_bytes > search_end) {
335 ret = -ENOSPC;
336 goto error;
337 }
338 /* check for pending inserts here */
339 return 0;
340
341 error:
342 btrfs_release_path(root, path);
343 return ret;
344 }
345
346 int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans,
347 struct btrfs_device *device,
348 u64 chunk_tree, u64 chunk_objectid,
349 u64 chunk_offset,
350 u64 num_bytes, u64 *start)
351 {
352 int ret;
353 struct btrfs_path *path;
354 struct btrfs_root *root = device->dev_root;
355 struct btrfs_dev_extent *extent;
356 struct extent_buffer *leaf;
357 struct btrfs_key key;
358
359 path = btrfs_alloc_path();
360 if (!path)
361 return -ENOMEM;
362
363 ret = find_free_dev_extent(trans, device, path, num_bytes, start);
364 if (ret) {
365 goto err;
366 }
367
368 key.objectid = device->devid;
369 key.offset = *start;
370 key.type = BTRFS_DEV_EXTENT_KEY;
371 ret = btrfs_insert_empty_item(trans, root, path, &key,
372 sizeof(*extent));
373 BUG_ON(ret);
374
375 leaf = path->nodes[0];
376 extent = btrfs_item_ptr(leaf, path->slots[0],
377 struct btrfs_dev_extent);
378 btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree);
379 btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid);
380 btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset);
381
382 write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid,
383 (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent),
384 BTRFS_UUID_SIZE);
385
386 btrfs_set_dev_extent_length(leaf, extent, num_bytes);
387 btrfs_mark_buffer_dirty(leaf);
388 err:
389 btrfs_free_path(path);
390 return ret;
391 }
392
393 static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset)
394 {
395 struct btrfs_path *path;
396 int ret;
397 struct btrfs_key key;
398 struct btrfs_chunk *chunk;
399 struct btrfs_key found_key;
400
401 path = btrfs_alloc_path();
402 BUG_ON(!path);
403
404 key.objectid = objectid;
405 key.offset = (u64)-1;
406 key.type = BTRFS_CHUNK_ITEM_KEY;
407
408 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
409 if (ret < 0)
410 goto error;
411
412 BUG_ON(ret == 0);
413
414 ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY);
415 if (ret) {
416 *offset = 0;
417 } else {
418 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
419 path->slots[0]);
420 if (found_key.objectid != objectid)
421 *offset = 0;
422 else {
423 chunk = btrfs_item_ptr(path->nodes[0], path->slots[0],
424 struct btrfs_chunk);
425 *offset = found_key.offset +
426 btrfs_chunk_length(path->nodes[0], chunk);
427 }
428 }
429 ret = 0;
430 error:
431 btrfs_free_path(path);
432 return ret;
433 }
434
435 static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path,
436 u64 *objectid)
437 {
438 int ret;
439 struct btrfs_key key;
440 struct btrfs_key found_key;
441
442 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
443 key.type = BTRFS_DEV_ITEM_KEY;
444 key.offset = (u64)-1;
445
446 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
447 if (ret < 0)
448 goto error;
449
450 BUG_ON(ret == 0);
451
452 ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID,
453 BTRFS_DEV_ITEM_KEY);
454 if (ret) {
455 *objectid = 1;
456 } else {
457 btrfs_item_key_to_cpu(path->nodes[0], &found_key,
458 path->slots[0]);
459 *objectid = found_key.offset + 1;
460 }
461 ret = 0;
462 error:
463 btrfs_release_path(root, path);
464 return ret;
465 }
466
467 /*
468 * the device information is stored in the chunk root
469 * the btrfs_device struct should be fully filled in
470 */
471 int btrfs_add_device(struct btrfs_trans_handle *trans,
472 struct btrfs_root *root,
473 struct btrfs_device *device)
474 {
475 int ret;
476 struct btrfs_path *path;
477 struct btrfs_dev_item *dev_item;
478 struct extent_buffer *leaf;
479 struct btrfs_key key;
480 unsigned long ptr;
481 u64 free_devid = 0;
482
483 root = root->fs_info->chunk_root;
484
485 path = btrfs_alloc_path();
486 if (!path)
487 return -ENOMEM;
488
489 ret = find_next_devid(root, path, &free_devid);
490 if (ret)
491 goto out;
492
493 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
494 key.type = BTRFS_DEV_ITEM_KEY;
495 key.offset = free_devid;
496
497 ret = btrfs_insert_empty_item(trans, root, path, &key,
498 sizeof(*dev_item));
499 if (ret)
500 goto out;
501
502 leaf = path->nodes[0];
503 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
504
505 device->devid = free_devid;
506 btrfs_set_device_id(leaf, dev_item, device->devid);
507 btrfs_set_device_type(leaf, dev_item, device->type);
508 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
509 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
510 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
511 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
512 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
513 btrfs_set_device_group(leaf, dev_item, 0);
514 btrfs_set_device_seek_speed(leaf, dev_item, 0);
515 btrfs_set_device_bandwidth(leaf, dev_item, 0);
516
517 ptr = (unsigned long)btrfs_device_uuid(dev_item);
518 write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
519 btrfs_mark_buffer_dirty(leaf);
520 ret = 0;
521
522 out:
523 btrfs_free_path(path);
524 return ret;
525 }
526
527 int btrfs_update_device(struct btrfs_trans_handle *trans,
528 struct btrfs_device *device)
529 {
530 int ret;
531 struct btrfs_path *path;
532 struct btrfs_root *root;
533 struct btrfs_dev_item *dev_item;
534 struct extent_buffer *leaf;
535 struct btrfs_key key;
536
537 root = device->dev_root->fs_info->chunk_root;
538
539 path = btrfs_alloc_path();
540 if (!path)
541 return -ENOMEM;
542
543 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
544 key.type = BTRFS_DEV_ITEM_KEY;
545 key.offset = device->devid;
546
547 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
548 if (ret < 0)
549 goto out;
550
551 if (ret > 0) {
552 ret = -ENOENT;
553 goto out;
554 }
555
556 leaf = path->nodes[0];
557 dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item);
558
559 btrfs_set_device_id(leaf, dev_item, device->devid);
560 btrfs_set_device_type(leaf, dev_item, device->type);
561 btrfs_set_device_io_align(leaf, dev_item, device->io_align);
562 btrfs_set_device_io_width(leaf, dev_item, device->io_width);
563 btrfs_set_device_sector_size(leaf, dev_item, device->sector_size);
564 btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes);
565 btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used);
566 btrfs_mark_buffer_dirty(leaf);
567
568 out:
569 btrfs_free_path(path);
570 return ret;
571 }
572
573 int btrfs_add_system_chunk(struct btrfs_trans_handle *trans,
574 struct btrfs_root *root,
575 struct btrfs_key *key,
576 struct btrfs_chunk *chunk, int item_size)
577 {
578 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
579 struct btrfs_disk_key disk_key;
580 u32 array_size;
581 u8 *ptr;
582
583 array_size = btrfs_super_sys_array_size(super_copy);
584 if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE)
585 return -EFBIG;
586
587 ptr = super_copy->sys_chunk_array + array_size;
588 btrfs_cpu_key_to_disk(&disk_key, key);
589 memcpy(ptr, &disk_key, sizeof(disk_key));
590 ptr += sizeof(disk_key);
591 memcpy(ptr, chunk, item_size);
592 item_size += sizeof(disk_key);
593 btrfs_set_super_sys_array_size(super_copy, array_size + item_size);
594 return 0;
595 }
596
597 static u64 div_factor(u64 num, int factor)
598 {
599 if (factor == 10)
600 return num;
601 num *= factor;
602 return num / 10;
603 }
604
605 static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes,
606 int sub_stripes)
607 {
608 if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP))
609 return calc_size;
610 else if (type & BTRFS_BLOCK_GROUP_RAID10)
611 return calc_size * (num_stripes / sub_stripes);
612 else
613 return calc_size * num_stripes;
614 }
615
616
617 int btrfs_alloc_chunk(struct btrfs_trans_handle *trans,
618 struct btrfs_root *extent_root, u64 *start,
619 u64 *num_bytes, u64 type)
620 {
621 u64 dev_offset;
622 struct btrfs_fs_info *info = extent_root->fs_info;
623 struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root;
624 struct btrfs_stripe *stripes;
625 struct btrfs_device *device = NULL;
626 struct btrfs_chunk *chunk;
627 struct list_head private_devs;
628 struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices;
629 struct list_head *cur;
630 struct map_lookup *map;
631 int min_stripe_size = 1 * 1024 * 1024;
632 u64 physical;
633 u64 calc_size = 8 * 1024 * 1024;
634 u64 min_free;
635 u64 max_chunk_size = 4 * calc_size;
636 u64 avail;
637 u64 max_avail = 0;
638 u64 percent_max;
639 int num_stripes = 1;
640 int min_stripes = 1;
641 int sub_stripes = 0;
642 int looped = 0;
643 int ret;
644 int index;
645 int stripe_len = 64 * 1024;
646 struct btrfs_key key;
647
648 if (list_empty(dev_list)) {
649 return -ENOSPC;
650 }
651
652 if (type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
653 BTRFS_BLOCK_GROUP_RAID10 |
654 BTRFS_BLOCK_GROUP_DUP)) {
655 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
656 calc_size = 8 * 1024 * 1024;
657 max_chunk_size = calc_size * 2;
658 min_stripe_size = 1 * 1024 * 1024;
659 } else if (type & BTRFS_BLOCK_GROUP_DATA) {
660 calc_size = 1024 * 1024 * 1024;
661 max_chunk_size = 10 * calc_size;
662 min_stripe_size = 64 * 1024 * 1024;
663 } else if (type & BTRFS_BLOCK_GROUP_METADATA) {
664 calc_size = 1024 * 1024 * 1024;
665 max_chunk_size = 4 * calc_size;
666 min_stripe_size = 32 * 1024 * 1024;
667 }
668 }
669 if (type & BTRFS_BLOCK_GROUP_RAID1) {
670 num_stripes = min_t(u64, 2,
671 btrfs_super_num_devices(&info->super_copy));
672 if (num_stripes < 2)
673 return -ENOSPC;
674 min_stripes = 2;
675 }
676 if (type & BTRFS_BLOCK_GROUP_DUP) {
677 num_stripes = 2;
678 min_stripes = 2;
679 }
680 if (type & (BTRFS_BLOCK_GROUP_RAID0)) {
681 num_stripes = btrfs_super_num_devices(&info->super_copy);
682 min_stripes = 2;
683 }
684 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
685 num_stripes = btrfs_super_num_devices(&info->super_copy);
686 if (num_stripes < 4)
687 return -ENOSPC;
688 num_stripes &= ~(u32)1;
689 sub_stripes = 2;
690 min_stripes = 4;
691 }
692
693 /* we don't want a chunk larger than 10% of the FS */
694 percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1);
695 max_chunk_size = min(percent_max, max_chunk_size);
696
697 again:
698 if (chunk_bytes_by_type(type, calc_size, num_stripes, sub_stripes) >
699 max_chunk_size) {
700 calc_size = max_chunk_size;
701 calc_size /= num_stripes;
702 calc_size /= stripe_len;
703 calc_size *= stripe_len;
704 }
705 /* we don't want tiny stripes */
706 calc_size = max_t(u64, calc_size, min_stripe_size);
707
708 calc_size /= stripe_len;
709 calc_size *= stripe_len;
710 INIT_LIST_HEAD(&private_devs);
711 cur = dev_list->next;
712 index = 0;
713
714 if (type & BTRFS_BLOCK_GROUP_DUP)
715 min_free = calc_size * 2;
716 else
717 min_free = calc_size;
718
719 /* build a private list of devices we will allocate from */
720 while(index < num_stripes) {
721 device = list_entry(cur, struct btrfs_device, dev_list);
722 avail = device->total_bytes - device->bytes_used;
723 cur = cur->next;
724 if (avail >= min_free) {
725 list_move_tail(&device->dev_list, &private_devs);
726 index++;
727 if (type & BTRFS_BLOCK_GROUP_DUP)
728 index++;
729 } else if (avail > max_avail)
730 max_avail = avail;
731 if (cur == dev_list)
732 break;
733 }
734 if (index < num_stripes) {
735 list_splice(&private_devs, dev_list);
736 if (index >= min_stripes) {
737 num_stripes = index;
738 if (type & (BTRFS_BLOCK_GROUP_RAID10)) {
739 num_stripes /= sub_stripes;
740 num_stripes *= sub_stripes;
741 }
742 looped = 1;
743 goto again;
744 }
745 if (!looped && max_avail > 0) {
746 looped = 1;
747 calc_size = max_avail;
748 goto again;
749 }
750 return -ENOSPC;
751 }
752 key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID;
753 key.type = BTRFS_CHUNK_ITEM_KEY;
754 ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID,
755 &key.offset);
756 if (ret)
757 return ret;
758
759 chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS);
760 if (!chunk)
761 return -ENOMEM;
762
763 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
764 if (!map) {
765 kfree(chunk);
766 return -ENOMEM;
767 }
768
769 stripes = &chunk->stripe;
770 *num_bytes = chunk_bytes_by_type(type, calc_size,
771 num_stripes, sub_stripes);
772 index = 0;
773 while(index < num_stripes) {
774 struct btrfs_stripe *stripe;
775 BUG_ON(list_empty(&private_devs));
776 cur = private_devs.next;
777 device = list_entry(cur, struct btrfs_device, dev_list);
778
779 /* loop over this device again if we're doing a dup group */
780 if (!(type & BTRFS_BLOCK_GROUP_DUP) ||
781 (index == num_stripes - 1))
782 list_move_tail(&device->dev_list, dev_list);
783
784 ret = btrfs_alloc_dev_extent(trans, device,
785 info->chunk_root->root_key.objectid,
786 BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset,
787 calc_size, &dev_offset);
788 BUG_ON(ret);
789
790 device->bytes_used += calc_size;
791 ret = btrfs_update_device(trans, device);
792 BUG_ON(ret);
793
794 map->stripes[index].dev = device;
795 map->stripes[index].physical = dev_offset;
796 stripe = stripes + index;
797 btrfs_set_stack_stripe_devid(stripe, device->devid);
798 btrfs_set_stack_stripe_offset(stripe, dev_offset);
799 memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE);
800 physical = dev_offset;
801 index++;
802 }
803 BUG_ON(!list_empty(&private_devs));
804
805 /* key was set above */
806 btrfs_set_stack_chunk_length(chunk, *num_bytes);
807 btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid);
808 btrfs_set_stack_chunk_stripe_len(chunk, stripe_len);
809 btrfs_set_stack_chunk_type(chunk, type);
810 btrfs_set_stack_chunk_num_stripes(chunk, num_stripes);
811 btrfs_set_stack_chunk_io_align(chunk, stripe_len);
812 btrfs_set_stack_chunk_io_width(chunk, stripe_len);
813 btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize);
814 btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes);
815 map->sector_size = extent_root->sectorsize;
816 map->stripe_len = stripe_len;
817 map->io_align = stripe_len;
818 map->io_width = stripe_len;
819 map->type = type;
820 map->num_stripes = num_stripes;
821 map->sub_stripes = sub_stripes;
822
823 ret = btrfs_insert_item(trans, chunk_root, &key, chunk,
824 btrfs_chunk_item_size(num_stripes));
825 BUG_ON(ret);
826 *start = key.offset;;
827
828 map->ce.start = key.offset;
829 map->ce.size = *num_bytes;
830
831 ret = insert_existing_cache_extent(
832 &extent_root->fs_info->mapping_tree.cache_tree,
833 &map->ce);
834 BUG_ON(ret);
835
836 if (type & BTRFS_BLOCK_GROUP_SYSTEM) {
837 ret = btrfs_add_system_chunk(trans, chunk_root, &key,
838 chunk, btrfs_chunk_item_size(num_stripes));
839 BUG_ON(ret);
840 }
841
842 kfree(chunk);
843 return ret;
844 }
845
846 void btrfs_mapping_init(struct btrfs_mapping_tree *tree)
847 {
848 cache_tree_init(&tree->cache_tree);
849 }
850
851 int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len)
852 {
853 struct cache_extent *ce;
854 struct map_lookup *map;
855 int ret;
856 u64 offset;
857
858 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
859 BUG_ON(!ce);
860 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
861 map = container_of(ce, struct map_lookup, ce);
862
863 offset = logical - ce->start;
864 if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1))
865 ret = map->num_stripes;
866 else if (map->type & BTRFS_BLOCK_GROUP_RAID10)
867 ret = map->sub_stripes;
868 else
869 ret = 1;
870 return ret;
871 }
872
873 int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw,
874 u64 logical, u64 *length,
875 struct btrfs_multi_bio **multi_ret, int mirror_num)
876 {
877 struct cache_extent *ce;
878 struct map_lookup *map;
879 u64 offset;
880 u64 stripe_offset;
881 u64 stripe_nr;
882 int stripes_allocated = 8;
883 int stripes_required = 1;
884 int stripe_index;
885 int i;
886 struct btrfs_multi_bio *multi = NULL;
887
888 if (multi_ret && rw == READ) {
889 stripes_allocated = 1;
890 }
891 again:
892 if (multi_ret) {
893 multi = kzalloc(btrfs_multi_bio_size(stripes_allocated),
894 GFP_NOFS);
895 if (!multi)
896 return -ENOMEM;
897 }
898
899 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
900 BUG_ON(!ce);
901 BUG_ON(ce->start > logical || ce->start + ce->size < logical);
902 map = container_of(ce, struct map_lookup, ce);
903 offset = logical - ce->start;
904
905 if (rw == WRITE) {
906 if (map->type & (BTRFS_BLOCK_GROUP_RAID1 |
907 BTRFS_BLOCK_GROUP_DUP)) {
908 stripes_required = map->num_stripes;
909 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
910 stripes_required = map->sub_stripes;
911 }
912 }
913 /* if our multi bio struct is too small, back off and try again */
914 if (multi_ret && rw == WRITE &&
915 stripes_allocated < stripes_required) {
916 stripes_allocated = map->num_stripes;
917 kfree(multi);
918 goto again;
919 }
920 stripe_nr = offset;
921 /*
922 * stripe_nr counts the total number of stripes we have to stride
923 * to get to this block
924 */
925 stripe_nr = stripe_nr / map->stripe_len;
926
927 stripe_offset = stripe_nr * map->stripe_len;
928 BUG_ON(offset < stripe_offset);
929
930 /* stripe_offset is the offset of this block in its stripe*/
931 stripe_offset = offset - stripe_offset;
932
933 if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 |
934 BTRFS_BLOCK_GROUP_RAID10 |
935 BTRFS_BLOCK_GROUP_DUP)) {
936 /* we limit the length of each bio to what fits in a stripe */
937 *length = min_t(u64, ce->size - offset,
938 map->stripe_len - stripe_offset);
939 } else {
940 *length = ce->size - offset;
941 }
942
943 if (!multi_ret)
944 goto out;
945
946 multi->num_stripes = 1;
947 stripe_index = 0;
948 if (map->type & BTRFS_BLOCK_GROUP_RAID1) {
949 if (rw == WRITE)
950 multi->num_stripes = map->num_stripes;
951 else if (mirror_num)
952 stripe_index = mirror_num - 1;
953 else
954 stripe_index = stripe_nr % map->num_stripes;
955 } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) {
956 int factor = map->num_stripes / map->sub_stripes;
957
958 stripe_index = stripe_nr % factor;
959 stripe_index *= map->sub_stripes;
960
961 if (rw == WRITE)
962 multi->num_stripes = map->sub_stripes;
963 else if (mirror_num)
964 stripe_index += mirror_num - 1;
965 else
966 stripe_index = stripe_nr % map->sub_stripes;
967
968 stripe_nr = stripe_nr / factor;
969 } else if (map->type & BTRFS_BLOCK_GROUP_DUP) {
970 if (rw == WRITE)
971 multi->num_stripes = map->num_stripes;
972 else if (mirror_num)
973 stripe_index = mirror_num - 1;
974 } else {
975 /*
976 * after this do_div call, stripe_nr is the number of stripes
977 * on this device we have to walk to find the data, and
978 * stripe_index is the number of our device in the stripe array
979 */
980 stripe_index = stripe_nr % map->num_stripes;
981 stripe_nr = stripe_nr / map->num_stripes;
982 }
983 BUG_ON(stripe_index >= map->num_stripes);
984
985 BUG_ON(stripe_index != 0 && multi->num_stripes > 1);
986 for (i = 0; i < multi->num_stripes; i++) {
987 multi->stripes[i].physical =
988 map->stripes[stripe_index].physical + stripe_offset +
989 stripe_nr * map->stripe_len;
990 multi->stripes[i].dev = map->stripes[stripe_index].dev;
991 stripe_index++;
992 }
993 *multi_ret = multi;
994 out:
995 return 0;
996 }
997
998 struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid,
999 u8 *uuid)
1000 {
1001 struct list_head *head = &root->fs_info->fs_devices->devices;
1002
1003 return __find_device(head, devid, uuid);
1004 }
1005
1006 int btrfs_bootstrap_super_map(struct btrfs_mapping_tree *map_tree,
1007 struct btrfs_fs_devices *fs_devices)
1008 {
1009 struct map_lookup *map;
1010 u64 logical = BTRFS_SUPER_INFO_OFFSET;
1011 u64 length = BTRFS_SUPER_INFO_SIZE;
1012 int num_stripes = 0;
1013 int sub_stripes = 0;
1014 int ret;
1015 int i;
1016 struct list_head *cur;
1017
1018 list_for_each(cur, &fs_devices->devices) {
1019 num_stripes++;
1020 }
1021 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1022 if (!map)
1023 return -ENOMEM;
1024
1025 map->ce.start = logical;
1026 map->ce.size = length;
1027 map->num_stripes = num_stripes;
1028 map->sub_stripes = sub_stripes;
1029 map->io_width = length;
1030 map->io_align = length;
1031 map->sector_size = length;
1032 map->stripe_len = length;
1033 map->type = BTRFS_BLOCK_GROUP_RAID1;
1034
1035 i = 0;
1036 list_for_each(cur, &fs_devices->devices) {
1037 struct btrfs_device *device = list_entry(cur,
1038 struct btrfs_device,
1039 dev_list);
1040 map->stripes[i].physical = logical;
1041 map->stripes[i].dev = device;
1042 i++;
1043 }
1044 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1045 if (ret == -EEXIST) {
1046 struct cache_extent *old;
1047 struct map_lookup *old_map;
1048 old = find_cache_extent(&map_tree->cache_tree, logical, length);
1049 old_map = container_of(old, struct map_lookup, ce);
1050 remove_cache_extent(&map_tree->cache_tree, old);
1051 kfree(old_map);
1052 ret = insert_existing_cache_extent(&map_tree->cache_tree,
1053 &map->ce);
1054 }
1055 BUG_ON(ret);
1056 return 0;
1057 }
1058
1059 static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key,
1060 struct extent_buffer *leaf,
1061 struct btrfs_chunk *chunk)
1062 {
1063 struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree;
1064 struct map_lookup *map;
1065 struct cache_extent *ce;
1066 u64 logical;
1067 u64 length;
1068 u64 devid;
1069 u64 super_offset_diff = 0;
1070 u8 uuid[BTRFS_UUID_SIZE];
1071 int num_stripes;
1072 int ret;
1073 int i;
1074
1075 logical = key->offset;
1076 length = btrfs_chunk_length(leaf, chunk);
1077
1078 if (logical < BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE) {
1079 super_offset_diff = BTRFS_SUPER_INFO_OFFSET +
1080 BTRFS_SUPER_INFO_SIZE - logical;
1081 logical = BTRFS_SUPER_INFO_OFFSET + BTRFS_SUPER_INFO_SIZE;
1082 }
1083
1084 ce = find_first_cache_extent(&map_tree->cache_tree, logical);
1085
1086 /* already mapped? */
1087 if (ce && ce->start <= logical && ce->start + ce->size > logical) {
1088 return 0;
1089 }
1090
1091 num_stripes = btrfs_chunk_num_stripes(leaf, chunk);
1092 map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS);
1093 if (!map)
1094 return -ENOMEM;
1095
1096 map->ce.start = logical;
1097 map->ce.size = length - super_offset_diff;
1098 map->num_stripes = num_stripes;
1099 map->io_width = btrfs_chunk_io_width(leaf, chunk);
1100 map->io_align = btrfs_chunk_io_align(leaf, chunk);
1101 map->sector_size = btrfs_chunk_sector_size(leaf, chunk);
1102 map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk);
1103 map->type = btrfs_chunk_type(leaf, chunk);
1104 map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk);
1105
1106 for (i = 0; i < num_stripes; i++) {
1107 map->stripes[i].physical =
1108 btrfs_stripe_offset_nr(leaf, chunk, i) +
1109 super_offset_diff;
1110 devid = btrfs_stripe_devid_nr(leaf, chunk, i);
1111 read_extent_buffer(leaf, uuid, (unsigned long)
1112 btrfs_stripe_dev_uuid_nr(chunk, i),
1113 BTRFS_UUID_SIZE);
1114 map->stripes[i].dev = btrfs_find_device(root, devid, uuid);
1115 if (!map->stripes[i].dev) {
1116 kfree(map);
1117 return -EIO;
1118 }
1119
1120 }
1121 ret = insert_existing_cache_extent(&map_tree->cache_tree, &map->ce);
1122 BUG_ON(ret);
1123
1124 return 0;
1125 }
1126
1127 static int fill_device_from_item(struct extent_buffer *leaf,
1128 struct btrfs_dev_item *dev_item,
1129 struct btrfs_device *device)
1130 {
1131 unsigned long ptr;
1132
1133 device->devid = btrfs_device_id(leaf, dev_item);
1134 device->total_bytes = btrfs_device_total_bytes(leaf, dev_item);
1135 device->bytes_used = btrfs_device_bytes_used(leaf, dev_item);
1136 device->type = btrfs_device_type(leaf, dev_item);
1137 device->io_align = btrfs_device_io_align(leaf, dev_item);
1138 device->io_width = btrfs_device_io_width(leaf, dev_item);
1139 device->sector_size = btrfs_device_sector_size(leaf, dev_item);
1140
1141 ptr = (unsigned long)btrfs_device_uuid(dev_item);
1142 read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE);
1143
1144 return 0;
1145 }
1146
1147 static int read_one_dev(struct btrfs_root *root,
1148 struct extent_buffer *leaf,
1149 struct btrfs_dev_item *dev_item)
1150 {
1151 struct btrfs_device *device;
1152 u64 devid;
1153 int ret = 0;
1154 u8 dev_uuid[BTRFS_UUID_SIZE];
1155
1156 devid = btrfs_device_id(leaf, dev_item);
1157 read_extent_buffer(leaf, dev_uuid,
1158 (unsigned long)btrfs_device_uuid(dev_item),
1159 BTRFS_UUID_SIZE);
1160 device = btrfs_find_device(root, devid, dev_uuid);
1161 if (!device) {
1162 printk("warning devid %llu not found already\n",
1163 (unsigned long long)devid);
1164 device = kmalloc(sizeof(*device), GFP_NOFS);
1165 if (!device)
1166 return -ENOMEM;
1167 device->total_ios = 0;
1168 list_add(&device->dev_list,
1169 &root->fs_info->fs_devices->devices);
1170 }
1171
1172 fill_device_from_item(leaf, dev_item, device);
1173 device->dev_root = root->fs_info->dev_root;
1174 return ret;
1175 }
1176
1177 int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf)
1178 {
1179 struct btrfs_dev_item *dev_item;
1180
1181 dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block,
1182 dev_item);
1183 return read_one_dev(root, buf, dev_item);
1184 }
1185
1186 int btrfs_read_sys_array(struct btrfs_root *root)
1187 {
1188 struct btrfs_super_block *super_copy = &root->fs_info->super_copy;
1189 struct extent_buffer *sb = root->fs_info->sb_buffer;
1190 struct btrfs_disk_key *disk_key;
1191 struct btrfs_chunk *chunk;
1192 struct btrfs_key key;
1193 u32 num_stripes;
1194 u32 array_size;
1195 u32 len = 0;
1196 u8 *ptr;
1197 unsigned long sb_ptr;
1198 u32 cur;
1199 int ret;
1200
1201 array_size = btrfs_super_sys_array_size(super_copy);
1202
1203 /*
1204 * we do this loop twice, once for the device items and
1205 * once for all of the chunks. This way there are device
1206 * structs filled in for every chunk
1207 */
1208 ptr = super_copy->sys_chunk_array;
1209 sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array);
1210 cur = 0;
1211
1212 while (cur < array_size) {
1213 disk_key = (struct btrfs_disk_key *)ptr;
1214 btrfs_disk_key_to_cpu(&key, disk_key);
1215
1216 len = sizeof(*disk_key);
1217 ptr += len;
1218 sb_ptr += len;
1219 cur += len;
1220
1221 if (key.type == BTRFS_CHUNK_ITEM_KEY) {
1222 chunk = (struct btrfs_chunk *)sb_ptr;
1223 ret = read_one_chunk(root, &key, sb, chunk);
1224 BUG_ON(ret);
1225 num_stripes = btrfs_chunk_num_stripes(sb, chunk);
1226 len = btrfs_chunk_item_size(num_stripes);
1227 } else {
1228 BUG();
1229 }
1230 ptr += len;
1231 sb_ptr += len;
1232 cur += len;
1233 }
1234 return 0;
1235 }
1236
1237 int btrfs_read_chunk_tree(struct btrfs_root *root)
1238 {
1239 struct btrfs_path *path;
1240 struct extent_buffer *leaf;
1241 struct btrfs_key key;
1242 struct btrfs_key found_key;
1243 int ret;
1244 int slot;
1245
1246 root = root->fs_info->chunk_root;
1247
1248 path = btrfs_alloc_path();
1249 if (!path)
1250 return -ENOMEM;
1251
1252 /* first we search for all of the device items, and then we
1253 * read in all of the chunk items. This way we can create chunk
1254 * mappings that reference all of the devices that are afound
1255 */
1256 key.objectid = BTRFS_DEV_ITEMS_OBJECTID;
1257 key.offset = 0;
1258 key.type = 0;
1259 again:
1260 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
1261 while(1) {
1262 leaf = path->nodes[0];
1263 slot = path->slots[0];
1264 if (slot >= btrfs_header_nritems(leaf)) {
1265 ret = btrfs_next_leaf(root, path);
1266 if (ret == 0)
1267 continue;
1268 if (ret < 0)
1269 goto error;
1270 break;
1271 }
1272 btrfs_item_key_to_cpu(leaf, &found_key, slot);
1273 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1274 if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID)
1275 break;
1276 if (found_key.type == BTRFS_DEV_ITEM_KEY) {
1277 struct btrfs_dev_item *dev_item;
1278 dev_item = btrfs_item_ptr(leaf, slot,
1279 struct btrfs_dev_item);
1280 ret = read_one_dev(root, leaf, dev_item);
1281 BUG_ON(ret);
1282 }
1283 } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) {
1284 struct btrfs_chunk *chunk;
1285 chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk);
1286 ret = read_one_chunk(root, &found_key, leaf, chunk);
1287 }
1288 path->slots[0]++;
1289 }
1290 if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) {
1291 key.objectid = 0;
1292 btrfs_release_path(root, path);
1293 goto again;
1294 }
1295
1296 btrfs_free_path(path);
1297 ret = 0;
1298 error:
1299 return ret;
1300 }
1301
1302 struct list_head *btrfs_scanned_uuids(void)
1303 {
1304 return &fs_uuids;
1305 }
Btrfs-progs-unstable mirror