Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
8a4b83cc | 20 | #include <linux/buffer_head.h> |
593060d7 | 21 | #include <asm/div64.h> |
0b86a832 CM |
22 | #include "ctree.h" |
23 | #include "extent_map.h" | |
24 | #include "disk-io.h" | |
25 | #include "transaction.h" | |
26 | #include "print-tree.h" | |
27 | #include "volumes.h" | |
28 | ||
593060d7 CM |
29 | struct map_lookup { |
30 | u64 type; | |
31 | int io_align; | |
32 | int io_width; | |
33 | int stripe_len; | |
34 | int sector_size; | |
35 | int num_stripes; | |
321aecc6 | 36 | int sub_stripes; |
cea9e445 | 37 | struct btrfs_bio_stripe stripes[]; |
593060d7 CM |
38 | }; |
39 | ||
40 | #define map_lookup_size(n) (sizeof(struct map_lookup) + \ | |
cea9e445 | 41 | (sizeof(struct btrfs_bio_stripe) * (n))) |
593060d7 | 42 | |
8a4b83cc CM |
43 | static DEFINE_MUTEX(uuid_mutex); |
44 | static LIST_HEAD(fs_uuids); | |
45 | ||
46 | int btrfs_cleanup_fs_uuids(void) | |
47 | { | |
48 | struct btrfs_fs_devices *fs_devices; | |
49 | struct list_head *uuid_cur; | |
50 | struct list_head *devices_cur; | |
51 | struct btrfs_device *dev; | |
52 | ||
53 | list_for_each(uuid_cur, &fs_uuids) { | |
54 | fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices, | |
55 | list); | |
56 | while(!list_empty(&fs_devices->devices)) { | |
57 | devices_cur = fs_devices->devices.next; | |
58 | dev = list_entry(devices_cur, struct btrfs_device, | |
59 | dev_list); | |
8a4b83cc | 60 | if (dev->bdev) { |
8a4b83cc CM |
61 | close_bdev_excl(dev->bdev); |
62 | } | |
63 | list_del(&dev->dev_list); | |
64 | kfree(dev); | |
65 | } | |
66 | } | |
67 | return 0; | |
68 | } | |
69 | ||
a443755f CM |
70 | static struct btrfs_device *__find_device(struct list_head *head, u64 devid, |
71 | u8 *uuid) | |
8a4b83cc CM |
72 | { |
73 | struct btrfs_device *dev; | |
74 | struct list_head *cur; | |
75 | ||
76 | list_for_each(cur, head) { | |
77 | dev = list_entry(cur, struct btrfs_device, dev_list); | |
a443755f CM |
78 | if (dev->devid == devid && |
79 | !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE)) { | |
8a4b83cc | 80 | return dev; |
a443755f | 81 | } |
8a4b83cc CM |
82 | } |
83 | return NULL; | |
84 | } | |
85 | ||
86 | static struct btrfs_fs_devices *find_fsid(u8 *fsid) | |
87 | { | |
88 | struct list_head *cur; | |
89 | struct btrfs_fs_devices *fs_devices; | |
90 | ||
91 | list_for_each(cur, &fs_uuids) { | |
92 | fs_devices = list_entry(cur, struct btrfs_fs_devices, list); | |
93 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
94 | return fs_devices; | |
95 | } | |
96 | return NULL; | |
97 | } | |
98 | ||
99 | static int device_list_add(const char *path, | |
100 | struct btrfs_super_block *disk_super, | |
101 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
102 | { | |
103 | struct btrfs_device *device; | |
104 | struct btrfs_fs_devices *fs_devices; | |
105 | u64 found_transid = btrfs_super_generation(disk_super); | |
106 | ||
107 | fs_devices = find_fsid(disk_super->fsid); | |
108 | if (!fs_devices) { | |
109 | fs_devices = kmalloc(sizeof(*fs_devices), GFP_NOFS); | |
110 | if (!fs_devices) | |
111 | return -ENOMEM; | |
112 | INIT_LIST_HEAD(&fs_devices->devices); | |
113 | list_add(&fs_devices->list, &fs_uuids); | |
114 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
115 | fs_devices->latest_devid = devid; | |
116 | fs_devices->latest_trans = found_transid; | |
117 | fs_devices->lowest_devid = (u64)-1; | |
118 | fs_devices->num_devices = 0; | |
119 | device = NULL; | |
120 | } else { | |
a443755f CM |
121 | device = __find_device(&fs_devices->devices, devid, |
122 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
123 | } |
124 | if (!device) { | |
125 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
126 | if (!device) { | |
127 | /* we can safely leave the fs_devices entry around */ | |
128 | return -ENOMEM; | |
129 | } | |
130 | device->devid = devid; | |
a443755f CM |
131 | memcpy(device->uuid, disk_super->dev_item.uuid, |
132 | BTRFS_UUID_SIZE); | |
f2984462 | 133 | device->barriers = 1; |
b248a415 | 134 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
135 | device->name = kstrdup(path, GFP_NOFS); |
136 | if (!device->name) { | |
137 | kfree(device); | |
138 | return -ENOMEM; | |
139 | } | |
140 | list_add(&device->dev_list, &fs_devices->devices); | |
141 | fs_devices->num_devices++; | |
142 | } | |
143 | ||
144 | if (found_transid > fs_devices->latest_trans) { | |
145 | fs_devices->latest_devid = devid; | |
146 | fs_devices->latest_trans = found_transid; | |
147 | } | |
148 | if (fs_devices->lowest_devid > devid) { | |
149 | fs_devices->lowest_devid = devid; | |
8a4b83cc CM |
150 | } |
151 | *fs_devices_ret = fs_devices; | |
152 | return 0; | |
153 | } | |
154 | ||
155 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) | |
156 | { | |
157 | struct list_head *head = &fs_devices->devices; | |
158 | struct list_head *cur; | |
159 | struct btrfs_device *device; | |
160 | ||
161 | mutex_lock(&uuid_mutex); | |
162 | list_for_each(cur, head) { | |
163 | device = list_entry(cur, struct btrfs_device, dev_list); | |
164 | if (device->bdev) { | |
165 | close_bdev_excl(device->bdev); | |
8a4b83cc CM |
166 | } |
167 | device->bdev = NULL; | |
168 | } | |
169 | mutex_unlock(&uuid_mutex); | |
170 | return 0; | |
171 | } | |
172 | ||
173 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
174 | int flags, void *holder) | |
175 | { | |
176 | struct block_device *bdev; | |
177 | struct list_head *head = &fs_devices->devices; | |
178 | struct list_head *cur; | |
179 | struct btrfs_device *device; | |
180 | int ret; | |
181 | ||
182 | mutex_lock(&uuid_mutex); | |
183 | list_for_each(cur, head) { | |
184 | device = list_entry(cur, struct btrfs_device, dev_list); | |
185 | bdev = open_bdev_excl(device->name, flags, holder); | |
e17cade2 | 186 | |
8a4b83cc CM |
187 | if (IS_ERR(bdev)) { |
188 | printk("open %s failed\n", device->name); | |
189 | ret = PTR_ERR(bdev); | |
190 | goto fail; | |
191 | } | |
192 | if (device->devid == fs_devices->latest_devid) | |
193 | fs_devices->latest_bdev = bdev; | |
194 | if (device->devid == fs_devices->lowest_devid) { | |
195 | fs_devices->lowest_bdev = bdev; | |
8a4b83cc CM |
196 | } |
197 | device->bdev = bdev; | |
198 | } | |
199 | mutex_unlock(&uuid_mutex); | |
200 | return 0; | |
201 | fail: | |
202 | mutex_unlock(&uuid_mutex); | |
203 | btrfs_close_devices(fs_devices); | |
204 | return ret; | |
205 | } | |
206 | ||
207 | int btrfs_scan_one_device(const char *path, int flags, void *holder, | |
208 | struct btrfs_fs_devices **fs_devices_ret) | |
209 | { | |
210 | struct btrfs_super_block *disk_super; | |
211 | struct block_device *bdev; | |
212 | struct buffer_head *bh; | |
213 | int ret; | |
214 | u64 devid; | |
f2984462 | 215 | u64 transid; |
8a4b83cc CM |
216 | |
217 | mutex_lock(&uuid_mutex); | |
218 | ||
8a4b83cc CM |
219 | bdev = open_bdev_excl(path, flags, holder); |
220 | ||
221 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
222 | ret = PTR_ERR(bdev); |
223 | goto error; | |
224 | } | |
225 | ||
226 | ret = set_blocksize(bdev, 4096); | |
227 | if (ret) | |
228 | goto error_close; | |
229 | bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096); | |
230 | if (!bh) { | |
231 | ret = -EIO; | |
232 | goto error_close; | |
233 | } | |
234 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
235 | if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, | |
236 | sizeof(disk_super->magic))) { | |
e58ca020 | 237 | ret = -EINVAL; |
8a4b83cc CM |
238 | goto error_brelse; |
239 | } | |
240 | devid = le64_to_cpu(disk_super->dev_item.devid); | |
f2984462 | 241 | transid = btrfs_super_generation(disk_super); |
7ae9c09d CM |
242 | if (disk_super->label[0]) |
243 | printk("device label %s ", disk_super->label); | |
244 | else { | |
245 | /* FIXME, make a readl uuid parser */ | |
246 | printk("device fsid %llx-%llx ", | |
247 | *(unsigned long long *)disk_super->fsid, | |
248 | *(unsigned long long *)(disk_super->fsid + 8)); | |
249 | } | |
250 | printk("devid %Lu transid %Lu %s\n", devid, transid, path); | |
8a4b83cc CM |
251 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
252 | ||
253 | error_brelse: | |
254 | brelse(bh); | |
255 | error_close: | |
256 | close_bdev_excl(bdev); | |
8a4b83cc CM |
257 | error: |
258 | mutex_unlock(&uuid_mutex); | |
259 | return ret; | |
260 | } | |
0b86a832 CM |
261 | |
262 | /* | |
263 | * this uses a pretty simple search, the expectation is that it is | |
264 | * called very infrequently and that a given device has a small number | |
265 | * of extents | |
266 | */ | |
267 | static int find_free_dev_extent(struct btrfs_trans_handle *trans, | |
268 | struct btrfs_device *device, | |
269 | struct btrfs_path *path, | |
270 | u64 num_bytes, u64 *start) | |
271 | { | |
272 | struct btrfs_key key; | |
273 | struct btrfs_root *root = device->dev_root; | |
274 | struct btrfs_dev_extent *dev_extent = NULL; | |
275 | u64 hole_size = 0; | |
276 | u64 last_byte = 0; | |
277 | u64 search_start = 0; | |
278 | u64 search_end = device->total_bytes; | |
279 | int ret; | |
280 | int slot = 0; | |
281 | int start_found; | |
282 | struct extent_buffer *l; | |
283 | ||
284 | start_found = 0; | |
285 | path->reada = 2; | |
286 | ||
287 | /* FIXME use last free of some kind */ | |
288 | ||
8a4b83cc CM |
289 | /* we don't want to overwrite the superblock on the drive, |
290 | * so we make sure to start at an offset of at least 1MB | |
291 | */ | |
292 | search_start = max((u64)1024 * 1024, search_start); | |
0b86a832 CM |
293 | key.objectid = device->devid; |
294 | key.offset = search_start; | |
295 | key.type = BTRFS_DEV_EXTENT_KEY; | |
296 | ret = btrfs_search_slot(trans, root, &key, path, 0, 0); | |
297 | if (ret < 0) | |
298 | goto error; | |
299 | ret = btrfs_previous_item(root, path, 0, key.type); | |
300 | if (ret < 0) | |
301 | goto error; | |
302 | l = path->nodes[0]; | |
303 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
304 | while (1) { | |
305 | l = path->nodes[0]; | |
306 | slot = path->slots[0]; | |
307 | if (slot >= btrfs_header_nritems(l)) { | |
308 | ret = btrfs_next_leaf(root, path); | |
309 | if (ret == 0) | |
310 | continue; | |
311 | if (ret < 0) | |
312 | goto error; | |
313 | no_more_items: | |
314 | if (!start_found) { | |
315 | if (search_start >= search_end) { | |
316 | ret = -ENOSPC; | |
317 | goto error; | |
318 | } | |
319 | *start = search_start; | |
320 | start_found = 1; | |
321 | goto check_pending; | |
322 | } | |
323 | *start = last_byte > search_start ? | |
324 | last_byte : search_start; | |
325 | if (search_end <= *start) { | |
326 | ret = -ENOSPC; | |
327 | goto error; | |
328 | } | |
329 | goto check_pending; | |
330 | } | |
331 | btrfs_item_key_to_cpu(l, &key, slot); | |
332 | ||
333 | if (key.objectid < device->devid) | |
334 | goto next; | |
335 | ||
336 | if (key.objectid > device->devid) | |
337 | goto no_more_items; | |
338 | ||
339 | if (key.offset >= search_start && key.offset > last_byte && | |
340 | start_found) { | |
341 | if (last_byte < search_start) | |
342 | last_byte = search_start; | |
343 | hole_size = key.offset - last_byte; | |
344 | if (key.offset > last_byte && | |
345 | hole_size >= num_bytes) { | |
346 | *start = last_byte; | |
347 | goto check_pending; | |
348 | } | |
349 | } | |
350 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) { | |
351 | goto next; | |
352 | } | |
353 | ||
354 | start_found = 1; | |
355 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
356 | last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent); | |
357 | next: | |
358 | path->slots[0]++; | |
359 | cond_resched(); | |
360 | } | |
361 | check_pending: | |
362 | /* we have to make sure we didn't find an extent that has already | |
363 | * been allocated by the map tree or the original allocation | |
364 | */ | |
365 | btrfs_release_path(root, path); | |
366 | BUG_ON(*start < search_start); | |
367 | ||
6324fbf3 | 368 | if (*start + num_bytes > search_end) { |
0b86a832 CM |
369 | ret = -ENOSPC; |
370 | goto error; | |
371 | } | |
372 | /* check for pending inserts here */ | |
373 | return 0; | |
374 | ||
375 | error: | |
376 | btrfs_release_path(root, path); | |
377 | return ret; | |
378 | } | |
379 | ||
380 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, | |
381 | struct btrfs_device *device, | |
e17cade2 CM |
382 | u64 chunk_tree, u64 chunk_objectid, |
383 | u64 chunk_offset, | |
384 | u64 num_bytes, u64 *start) | |
0b86a832 CM |
385 | { |
386 | int ret; | |
387 | struct btrfs_path *path; | |
388 | struct btrfs_root *root = device->dev_root; | |
389 | struct btrfs_dev_extent *extent; | |
390 | struct extent_buffer *leaf; | |
391 | struct btrfs_key key; | |
392 | ||
393 | path = btrfs_alloc_path(); | |
394 | if (!path) | |
395 | return -ENOMEM; | |
396 | ||
397 | ret = find_free_dev_extent(trans, device, path, num_bytes, start); | |
6324fbf3 | 398 | if (ret) { |
0b86a832 | 399 | goto err; |
6324fbf3 | 400 | } |
0b86a832 CM |
401 | |
402 | key.objectid = device->devid; | |
403 | key.offset = *start; | |
404 | key.type = BTRFS_DEV_EXTENT_KEY; | |
405 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
406 | sizeof(*extent)); | |
407 | BUG_ON(ret); | |
408 | ||
409 | leaf = path->nodes[0]; | |
410 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
411 | struct btrfs_dev_extent); | |
e17cade2 CM |
412 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
413 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
414 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
415 | ||
416 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
417 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
418 | BTRFS_UUID_SIZE); | |
419 | ||
0b86a832 CM |
420 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
421 | btrfs_mark_buffer_dirty(leaf); | |
422 | err: | |
423 | btrfs_free_path(path); | |
424 | return ret; | |
425 | } | |
426 | ||
e17cade2 | 427 | static int find_next_chunk(struct btrfs_root *root, u64 objectid, u64 *offset) |
0b86a832 CM |
428 | { |
429 | struct btrfs_path *path; | |
430 | int ret; | |
431 | struct btrfs_key key; | |
e17cade2 | 432 | struct btrfs_chunk *chunk; |
0b86a832 CM |
433 | struct btrfs_key found_key; |
434 | ||
435 | path = btrfs_alloc_path(); | |
436 | BUG_ON(!path); | |
437 | ||
e17cade2 | 438 | key.objectid = objectid; |
0b86a832 CM |
439 | key.offset = (u64)-1; |
440 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
441 | ||
442 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
443 | if (ret < 0) | |
444 | goto error; | |
445 | ||
446 | BUG_ON(ret == 0); | |
447 | ||
448 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
449 | if (ret) { | |
e17cade2 | 450 | *offset = 0; |
0b86a832 CM |
451 | } else { |
452 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
453 | path->slots[0]); | |
e17cade2 CM |
454 | if (found_key.objectid != objectid) |
455 | *offset = 0; | |
456 | else { | |
457 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
458 | struct btrfs_chunk); | |
459 | *offset = found_key.offset + | |
460 | btrfs_chunk_length(path->nodes[0], chunk); | |
461 | } | |
0b86a832 CM |
462 | } |
463 | ret = 0; | |
464 | error: | |
465 | btrfs_free_path(path); | |
466 | return ret; | |
467 | } | |
468 | ||
0b86a832 CM |
469 | static int find_next_devid(struct btrfs_root *root, struct btrfs_path *path, |
470 | u64 *objectid) | |
471 | { | |
472 | int ret; | |
473 | struct btrfs_key key; | |
474 | struct btrfs_key found_key; | |
475 | ||
476 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
477 | key.type = BTRFS_DEV_ITEM_KEY; | |
478 | key.offset = (u64)-1; | |
479 | ||
480 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
481 | if (ret < 0) | |
482 | goto error; | |
483 | ||
484 | BUG_ON(ret == 0); | |
485 | ||
486 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
487 | BTRFS_DEV_ITEM_KEY); | |
488 | if (ret) { | |
489 | *objectid = 1; | |
490 | } else { | |
491 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
492 | path->slots[0]); | |
493 | *objectid = found_key.offset + 1; | |
494 | } | |
495 | ret = 0; | |
496 | error: | |
497 | btrfs_release_path(root, path); | |
498 | return ret; | |
499 | } | |
500 | ||
501 | /* | |
502 | * the device information is stored in the chunk root | |
503 | * the btrfs_device struct should be fully filled in | |
504 | */ | |
505 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
506 | struct btrfs_root *root, | |
507 | struct btrfs_device *device) | |
508 | { | |
509 | int ret; | |
510 | struct btrfs_path *path; | |
511 | struct btrfs_dev_item *dev_item; | |
512 | struct extent_buffer *leaf; | |
513 | struct btrfs_key key; | |
514 | unsigned long ptr; | |
515 | u64 free_devid; | |
516 | ||
517 | root = root->fs_info->chunk_root; | |
518 | ||
519 | path = btrfs_alloc_path(); | |
520 | if (!path) | |
521 | return -ENOMEM; | |
522 | ||
523 | ret = find_next_devid(root, path, &free_devid); | |
524 | if (ret) | |
525 | goto out; | |
526 | ||
527 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
528 | key.type = BTRFS_DEV_ITEM_KEY; | |
529 | key.offset = free_devid; | |
530 | ||
531 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 532 | sizeof(*dev_item)); |
0b86a832 CM |
533 | if (ret) |
534 | goto out; | |
535 | ||
536 | leaf = path->nodes[0]; | |
537 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
538 | ||
8a4b83cc | 539 | device->devid = free_devid; |
0b86a832 CM |
540 | btrfs_set_device_id(leaf, dev_item, device->devid); |
541 | btrfs_set_device_type(leaf, dev_item, device->type); | |
542 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
543 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
544 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
545 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
546 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
547 | btrfs_set_device_group(leaf, dev_item, 0); |
548 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
549 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
0b86a832 | 550 | |
0b86a832 | 551 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 552 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
553 | btrfs_mark_buffer_dirty(leaf); |
554 | ret = 0; | |
555 | ||
556 | out: | |
557 | btrfs_free_path(path); | |
558 | return ret; | |
559 | } | |
560 | int btrfs_update_device(struct btrfs_trans_handle *trans, | |
561 | struct btrfs_device *device) | |
562 | { | |
563 | int ret; | |
564 | struct btrfs_path *path; | |
565 | struct btrfs_root *root; | |
566 | struct btrfs_dev_item *dev_item; | |
567 | struct extent_buffer *leaf; | |
568 | struct btrfs_key key; | |
569 | ||
570 | root = device->dev_root->fs_info->chunk_root; | |
571 | ||
572 | path = btrfs_alloc_path(); | |
573 | if (!path) | |
574 | return -ENOMEM; | |
575 | ||
576 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
577 | key.type = BTRFS_DEV_ITEM_KEY; | |
578 | key.offset = device->devid; | |
579 | ||
580 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
581 | if (ret < 0) | |
582 | goto out; | |
583 | ||
584 | if (ret > 0) { | |
585 | ret = -ENOENT; | |
586 | goto out; | |
587 | } | |
588 | ||
589 | leaf = path->nodes[0]; | |
590 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
591 | ||
592 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
593 | btrfs_set_device_type(leaf, dev_item, device->type); | |
594 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
595 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
596 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
597 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
598 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
599 | btrfs_mark_buffer_dirty(leaf); | |
600 | ||
601 | out: | |
602 | btrfs_free_path(path); | |
603 | return ret; | |
604 | } | |
605 | ||
606 | int btrfs_add_system_chunk(struct btrfs_trans_handle *trans, | |
607 | struct btrfs_root *root, | |
608 | struct btrfs_key *key, | |
609 | struct btrfs_chunk *chunk, int item_size) | |
610 | { | |
611 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
612 | struct btrfs_disk_key disk_key; | |
613 | u32 array_size; | |
614 | u8 *ptr; | |
615 | ||
616 | array_size = btrfs_super_sys_array_size(super_copy); | |
617 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
618 | return -EFBIG; | |
619 | ||
620 | ptr = super_copy->sys_chunk_array + array_size; | |
621 | btrfs_cpu_key_to_disk(&disk_key, key); | |
622 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
623 | ptr += sizeof(disk_key); | |
624 | memcpy(ptr, chunk, item_size); | |
625 | item_size += sizeof(disk_key); | |
626 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
627 | return 0; | |
628 | } | |
629 | ||
9b3f68b9 CM |
630 | static u64 div_factor(u64 num, int factor) |
631 | { | |
632 | if (factor == 10) | |
633 | return num; | |
634 | num *= factor; | |
635 | do_div(num, 10); | |
636 | return num; | |
637 | } | |
638 | ||
639 | static u64 chunk_bytes_by_type(u64 type, u64 calc_size, int num_stripes, | |
640 | int sub_stripes) | |
641 | { | |
642 | if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP)) | |
643 | return calc_size; | |
644 | else if (type & BTRFS_BLOCK_GROUP_RAID10) | |
645 | return calc_size * (num_stripes / sub_stripes); | |
646 | else | |
647 | return calc_size * num_stripes; | |
648 | } | |
649 | ||
650 | ||
0b86a832 CM |
651 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
652 | struct btrfs_root *extent_root, u64 *start, | |
6324fbf3 | 653 | u64 *num_bytes, u64 type) |
0b86a832 CM |
654 | { |
655 | u64 dev_offset; | |
593060d7 | 656 | struct btrfs_fs_info *info = extent_root->fs_info; |
0b86a832 CM |
657 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; |
658 | struct btrfs_stripe *stripes; | |
659 | struct btrfs_device *device = NULL; | |
660 | struct btrfs_chunk *chunk; | |
6324fbf3 | 661 | struct list_head private_devs; |
8a4b83cc | 662 | struct list_head *dev_list = &extent_root->fs_info->fs_devices->devices; |
6324fbf3 | 663 | struct list_head *cur; |
0b86a832 CM |
664 | struct extent_map_tree *em_tree; |
665 | struct map_lookup *map; | |
666 | struct extent_map *em; | |
a40a90a0 | 667 | int min_stripe_size = 1 * 1024 * 1024; |
0b86a832 CM |
668 | u64 physical; |
669 | u64 calc_size = 1024 * 1024 * 1024; | |
9b3f68b9 CM |
670 | u64 max_chunk_size = calc_size; |
671 | u64 min_free; | |
6324fbf3 CM |
672 | u64 avail; |
673 | u64 max_avail = 0; | |
9b3f68b9 | 674 | u64 percent_max; |
6324fbf3 | 675 | int num_stripes = 1; |
a40a90a0 | 676 | int min_stripes = 1; |
321aecc6 | 677 | int sub_stripes = 0; |
6324fbf3 | 678 | int looped = 0; |
0b86a832 | 679 | int ret; |
6324fbf3 | 680 | int index; |
593060d7 | 681 | int stripe_len = 64 * 1024; |
0b86a832 CM |
682 | struct btrfs_key key; |
683 | ||
6324fbf3 CM |
684 | if (list_empty(dev_list)) |
685 | return -ENOSPC; | |
593060d7 | 686 | |
a40a90a0 | 687 | if (type & (BTRFS_BLOCK_GROUP_RAID0)) { |
593060d7 | 688 | num_stripes = btrfs_super_num_devices(&info->super_copy); |
a40a90a0 CM |
689 | min_stripes = 2; |
690 | } | |
691 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
611f0e00 | 692 | num_stripes = 2; |
a40a90a0 CM |
693 | min_stripes = 2; |
694 | } | |
8790d502 CM |
695 | if (type & (BTRFS_BLOCK_GROUP_RAID1)) { |
696 | num_stripes = min_t(u64, 2, | |
697 | btrfs_super_num_devices(&info->super_copy)); | |
9b3f68b9 CM |
698 | if (num_stripes < 2) |
699 | return -ENOSPC; | |
a40a90a0 | 700 | min_stripes = 2; |
8790d502 | 701 | } |
321aecc6 CM |
702 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { |
703 | num_stripes = btrfs_super_num_devices(&info->super_copy); | |
704 | if (num_stripes < 4) | |
705 | return -ENOSPC; | |
706 | num_stripes &= ~(u32)1; | |
707 | sub_stripes = 2; | |
a40a90a0 | 708 | min_stripes = 4; |
321aecc6 | 709 | } |
9b3f68b9 CM |
710 | |
711 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
712 | max_chunk_size = 10 * calc_size; | |
a40a90a0 | 713 | min_stripe_size = 64 * 1024 * 1024; |
9b3f68b9 CM |
714 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
715 | max_chunk_size = 4 * calc_size; | |
a40a90a0 CM |
716 | min_stripe_size = 32 * 1024 * 1024; |
717 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
718 | calc_size = 8 * 1024 * 1024; | |
719 | max_chunk_size = calc_size * 2; | |
720 | min_stripe_size = 1 * 1024 * 1024; | |
9b3f68b9 CM |
721 | } |
722 | ||
723 | /* we don't want a chunk larger than 10% of the FS */ | |
724 | percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1); | |
725 | max_chunk_size = min(percent_max, max_chunk_size); | |
726 | ||
a40a90a0 | 727 | again: |
9b3f68b9 CM |
728 | if (calc_size * num_stripes > max_chunk_size) { |
729 | calc_size = max_chunk_size; | |
730 | do_div(calc_size, num_stripes); | |
731 | do_div(calc_size, stripe_len); | |
732 | calc_size *= stripe_len; | |
733 | } | |
734 | /* we don't want tiny stripes */ | |
a40a90a0 | 735 | calc_size = max_t(u64, min_stripe_size, calc_size); |
9b3f68b9 | 736 | |
9b3f68b9 CM |
737 | do_div(calc_size, stripe_len); |
738 | calc_size *= stripe_len; | |
739 | ||
6324fbf3 CM |
740 | INIT_LIST_HEAD(&private_devs); |
741 | cur = dev_list->next; | |
742 | index = 0; | |
611f0e00 CM |
743 | |
744 | if (type & BTRFS_BLOCK_GROUP_DUP) | |
745 | min_free = calc_size * 2; | |
9b3f68b9 CM |
746 | else |
747 | min_free = calc_size; | |
611f0e00 | 748 | |
ad5bd91e CM |
749 | /* we add 1MB because we never use the first 1MB of the device */ |
750 | min_free += 1024 * 1024; | |
751 | ||
6324fbf3 CM |
752 | /* build a private list of devices we will allocate from */ |
753 | while(index < num_stripes) { | |
754 | device = list_entry(cur, struct btrfs_device, dev_list); | |
611f0e00 | 755 | |
6324fbf3 CM |
756 | avail = device->total_bytes - device->bytes_used; |
757 | cur = cur->next; | |
611f0e00 | 758 | if (avail >= min_free) { |
6324fbf3 CM |
759 | list_move_tail(&device->dev_list, &private_devs); |
760 | index++; | |
611f0e00 CM |
761 | if (type & BTRFS_BLOCK_GROUP_DUP) |
762 | index++; | |
a40a90a0 CM |
763 | } else if (avail > max_avail) |
764 | max_avail = avail; | |
6324fbf3 CM |
765 | if (cur == dev_list) |
766 | break; | |
767 | } | |
768 | if (index < num_stripes) { | |
769 | list_splice(&private_devs, dev_list); | |
a40a90a0 CM |
770 | if (index >= min_stripes) { |
771 | num_stripes = index; | |
772 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
773 | num_stripes /= sub_stripes; | |
774 | num_stripes *= sub_stripes; | |
775 | } | |
776 | looped = 1; | |
777 | goto again; | |
778 | } | |
6324fbf3 CM |
779 | if (!looped && max_avail > 0) { |
780 | looped = 1; | |
781 | calc_size = max_avail; | |
782 | goto again; | |
783 | } | |
784 | return -ENOSPC; | |
785 | } | |
e17cade2 CM |
786 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
787 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
788 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
789 | &key.offset); | |
0b86a832 CM |
790 | if (ret) |
791 | return ret; | |
792 | ||
0b86a832 CM |
793 | chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS); |
794 | if (!chunk) | |
795 | return -ENOMEM; | |
796 | ||
593060d7 CM |
797 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
798 | if (!map) { | |
799 | kfree(chunk); | |
800 | return -ENOMEM; | |
801 | } | |
802 | ||
0b86a832 | 803 | stripes = &chunk->stripe; |
9b3f68b9 CM |
804 | *num_bytes = chunk_bytes_by_type(type, calc_size, |
805 | num_stripes, sub_stripes); | |
0b86a832 | 806 | |
8790d502 | 807 | |
6324fbf3 | 808 | index = 0; |
e17cade2 | 809 | printk("new chunk type %Lu start %Lu size %Lu\n", type, key.offset, *num_bytes); |
0b86a832 | 810 | while(index < num_stripes) { |
e17cade2 | 811 | struct btrfs_stripe *stripe; |
6324fbf3 CM |
812 | BUG_ON(list_empty(&private_devs)); |
813 | cur = private_devs.next; | |
814 | device = list_entry(cur, struct btrfs_device, dev_list); | |
611f0e00 CM |
815 | |
816 | /* loop over this device again if we're doing a dup group */ | |
817 | if (!(type & BTRFS_BLOCK_GROUP_DUP) || | |
818 | (index == num_stripes - 1)) | |
819 | list_move_tail(&device->dev_list, dev_list); | |
0b86a832 CM |
820 | |
821 | ret = btrfs_alloc_dev_extent(trans, device, | |
e17cade2 CM |
822 | info->chunk_root->root_key.objectid, |
823 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset, | |
824 | calc_size, &dev_offset); | |
0b86a832 | 825 | BUG_ON(ret); |
e17cade2 | 826 | printk("alloc chunk start %Lu size %Lu from dev %Lu type %Lu\n", key.offset, calc_size, device->devid, type); |
0b86a832 CM |
827 | device->bytes_used += calc_size; |
828 | ret = btrfs_update_device(trans, device); | |
829 | BUG_ON(ret); | |
830 | ||
593060d7 CM |
831 | map->stripes[index].dev = device; |
832 | map->stripes[index].physical = dev_offset; | |
e17cade2 CM |
833 | stripe = stripes + index; |
834 | btrfs_set_stack_stripe_devid(stripe, device->devid); | |
835 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
836 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
0b86a832 CM |
837 | physical = dev_offset; |
838 | index++; | |
839 | } | |
6324fbf3 | 840 | BUG_ON(!list_empty(&private_devs)); |
0b86a832 | 841 | |
e17cade2 CM |
842 | /* key was set above */ |
843 | btrfs_set_stack_chunk_length(chunk, *num_bytes); | |
0b86a832 | 844 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
593060d7 | 845 | btrfs_set_stack_chunk_stripe_len(chunk, stripe_len); |
0b86a832 CM |
846 | btrfs_set_stack_chunk_type(chunk, type); |
847 | btrfs_set_stack_chunk_num_stripes(chunk, num_stripes); | |
593060d7 CM |
848 | btrfs_set_stack_chunk_io_align(chunk, stripe_len); |
849 | btrfs_set_stack_chunk_io_width(chunk, stripe_len); | |
0b86a832 | 850 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
321aecc6 | 851 | btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes); |
593060d7 CM |
852 | map->sector_size = extent_root->sectorsize; |
853 | map->stripe_len = stripe_len; | |
854 | map->io_align = stripe_len; | |
855 | map->io_width = stripe_len; | |
856 | map->type = type; | |
857 | map->num_stripes = num_stripes; | |
321aecc6 | 858 | map->sub_stripes = sub_stripes; |
0b86a832 CM |
859 | |
860 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, | |
861 | btrfs_chunk_item_size(num_stripes)); | |
862 | BUG_ON(ret); | |
e17cade2 | 863 | *start = key.offset;; |
0b86a832 CM |
864 | |
865 | em = alloc_extent_map(GFP_NOFS); | |
866 | if (!em) | |
867 | return -ENOMEM; | |
0b86a832 | 868 | em->bdev = (struct block_device *)map; |
e17cade2 CM |
869 | em->start = key.offset; |
870 | em->len = *num_bytes; | |
0b86a832 CM |
871 | em->block_start = 0; |
872 | ||
0b86a832 CM |
873 | kfree(chunk); |
874 | ||
875 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; | |
876 | spin_lock(&em_tree->lock); | |
877 | ret = add_extent_mapping(em_tree, em); | |
0b86a832 | 878 | spin_unlock(&em_tree->lock); |
b248a415 | 879 | BUG_ON(ret); |
0b86a832 CM |
880 | free_extent_map(em); |
881 | return ret; | |
882 | } | |
883 | ||
884 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
885 | { | |
886 | extent_map_tree_init(&tree->map_tree, GFP_NOFS); | |
887 | } | |
888 | ||
889 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
890 | { | |
891 | struct extent_map *em; | |
892 | ||
893 | while(1) { | |
894 | spin_lock(&tree->map_tree.lock); | |
895 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); | |
896 | if (em) | |
897 | remove_extent_mapping(&tree->map_tree, em); | |
898 | spin_unlock(&tree->map_tree.lock); | |
899 | if (!em) | |
900 | break; | |
901 | kfree(em->bdev); | |
902 | /* once for us */ | |
903 | free_extent_map(em); | |
904 | /* once for the tree */ | |
905 | free_extent_map(em); | |
906 | } | |
907 | } | |
908 | ||
f188591e CM |
909 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
910 | { | |
911 | struct extent_map *em; | |
912 | struct map_lookup *map; | |
913 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
914 | int ret; | |
915 | ||
916 | spin_lock(&em_tree->lock); | |
917 | em = lookup_extent_mapping(em_tree, logical, len); | |
b248a415 | 918 | spin_unlock(&em_tree->lock); |
f188591e CM |
919 | BUG_ON(!em); |
920 | ||
921 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
922 | map = (struct map_lookup *)em->bdev; | |
923 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
924 | ret = map->num_stripes; | |
321aecc6 CM |
925 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
926 | ret = map->sub_stripes; | |
f188591e CM |
927 | else |
928 | ret = 1; | |
929 | free_extent_map(em); | |
f188591e CM |
930 | return ret; |
931 | } | |
932 | ||
8790d502 | 933 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
cea9e445 | 934 | u64 logical, u64 *length, |
f188591e | 935 | struct btrfs_multi_bio **multi_ret, int mirror_num) |
0b86a832 CM |
936 | { |
937 | struct extent_map *em; | |
938 | struct map_lookup *map; | |
939 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
940 | u64 offset; | |
593060d7 CM |
941 | u64 stripe_offset; |
942 | u64 stripe_nr; | |
cea9e445 | 943 | int stripes_allocated = 8; |
321aecc6 | 944 | int stripes_required = 1; |
593060d7 | 945 | int stripe_index; |
cea9e445 CM |
946 | int i; |
947 | struct btrfs_multi_bio *multi = NULL; | |
0b86a832 | 948 | |
cea9e445 CM |
949 | if (multi_ret && !(rw & (1 << BIO_RW))) { |
950 | stripes_allocated = 1; | |
951 | } | |
952 | again: | |
953 | if (multi_ret) { | |
954 | multi = kzalloc(btrfs_multi_bio_size(stripes_allocated), | |
955 | GFP_NOFS); | |
956 | if (!multi) | |
957 | return -ENOMEM; | |
958 | } | |
0b86a832 CM |
959 | |
960 | spin_lock(&em_tree->lock); | |
961 | em = lookup_extent_mapping(em_tree, logical, *length); | |
b248a415 | 962 | spin_unlock(&em_tree->lock); |
3b951516 CM |
963 | if (!em) { |
964 | printk("unable to find logical %Lu\n", logical); | |
965 | } | |
0b86a832 CM |
966 | BUG_ON(!em); |
967 | ||
968 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
969 | map = (struct map_lookup *)em->bdev; | |
970 | offset = logical - em->start; | |
593060d7 | 971 | |
f188591e CM |
972 | if (mirror_num > map->num_stripes) |
973 | mirror_num = 0; | |
974 | ||
cea9e445 | 975 | /* if our multi bio struct is too small, back off and try again */ |
321aecc6 CM |
976 | if (rw & (1 << BIO_RW)) { |
977 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
978 | BTRFS_BLOCK_GROUP_DUP)) { | |
979 | stripes_required = map->num_stripes; | |
980 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
981 | stripes_required = map->sub_stripes; | |
982 | } | |
983 | } | |
984 | if (multi_ret && rw == WRITE && | |
985 | stripes_allocated < stripes_required) { | |
cea9e445 | 986 | stripes_allocated = map->num_stripes; |
cea9e445 CM |
987 | free_extent_map(em); |
988 | kfree(multi); | |
989 | goto again; | |
990 | } | |
593060d7 CM |
991 | stripe_nr = offset; |
992 | /* | |
993 | * stripe_nr counts the total number of stripes we have to stride | |
994 | * to get to this block | |
995 | */ | |
996 | do_div(stripe_nr, map->stripe_len); | |
997 | ||
998 | stripe_offset = stripe_nr * map->stripe_len; | |
999 | BUG_ON(offset < stripe_offset); | |
1000 | ||
1001 | /* stripe_offset is the offset of this block in its stripe*/ | |
1002 | stripe_offset = offset - stripe_offset; | |
1003 | ||
cea9e445 | 1004 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1005 | BTRFS_BLOCK_GROUP_RAID10 | |
cea9e445 CM |
1006 | BTRFS_BLOCK_GROUP_DUP)) { |
1007 | /* we limit the length of each bio to what fits in a stripe */ | |
1008 | *length = min_t(u64, em->len - offset, | |
1009 | map->stripe_len - stripe_offset); | |
1010 | } else { | |
1011 | *length = em->len - offset; | |
1012 | } | |
1013 | if (!multi_ret) | |
1014 | goto out; | |
1015 | ||
1016 | multi->num_stripes = 1; | |
1017 | stripe_index = 0; | |
8790d502 | 1018 | if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
8790d502 | 1019 | if (rw & (1 << BIO_RW)) |
cea9e445 | 1020 | multi->num_stripes = map->num_stripes; |
f188591e CM |
1021 | else if (mirror_num) { |
1022 | stripe_index = mirror_num - 1; | |
1023 | } else { | |
8790d502 CM |
1024 | int i; |
1025 | u64 least = (u64)-1; | |
1026 | struct btrfs_device *cur; | |
1027 | ||
1028 | for (i = 0; i < map->num_stripes; i++) { | |
1029 | cur = map->stripes[i].dev; | |
1030 | spin_lock(&cur->io_lock); | |
1031 | if (cur->total_ios < least) { | |
1032 | least = cur->total_ios; | |
1033 | stripe_index = i; | |
1034 | } | |
1035 | spin_unlock(&cur->io_lock); | |
1036 | } | |
8790d502 | 1037 | } |
611f0e00 | 1038 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
cea9e445 CM |
1039 | if (rw & (1 << BIO_RW)) |
1040 | multi->num_stripes = map->num_stripes; | |
f188591e CM |
1041 | else if (mirror_num) |
1042 | stripe_index = mirror_num - 1; | |
321aecc6 CM |
1043 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
1044 | int factor = map->num_stripes / map->sub_stripes; | |
1045 | int orig_stripe_nr = stripe_nr; | |
1046 | ||
1047 | stripe_index = do_div(stripe_nr, factor); | |
1048 | stripe_index *= map->sub_stripes; | |
1049 | ||
1050 | if (rw & (1 << BIO_RW)) | |
1051 | multi->num_stripes = map->sub_stripes; | |
1052 | else if (mirror_num) | |
1053 | stripe_index += mirror_num - 1; | |
1054 | else | |
1055 | stripe_index += orig_stripe_nr % map->sub_stripes; | |
8790d502 CM |
1056 | } else { |
1057 | /* | |
1058 | * after this do_div call, stripe_nr is the number of stripes | |
1059 | * on this device we have to walk to find the data, and | |
1060 | * stripe_index is the number of our device in the stripe array | |
1061 | */ | |
1062 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
1063 | } | |
593060d7 | 1064 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 CM |
1065 | |
1066 | for (i = 0; i < multi->num_stripes; i++) { | |
1067 | multi->stripes[i].physical = | |
1068 | map->stripes[stripe_index].physical + stripe_offset + | |
1069 | stripe_nr * map->stripe_len; | |
1070 | multi->stripes[i].dev = map->stripes[stripe_index].dev; | |
1071 | stripe_index++; | |
593060d7 | 1072 | } |
cea9e445 CM |
1073 | *multi_ret = multi; |
1074 | out: | |
0b86a832 | 1075 | free_extent_map(em); |
0b86a832 CM |
1076 | return 0; |
1077 | } | |
1078 | ||
8790d502 CM |
1079 | #if LINUX_VERSION_CODE > KERNEL_VERSION(2,6,23) |
1080 | static void end_bio_multi_stripe(struct bio *bio, int err) | |
1081 | #else | |
1082 | static int end_bio_multi_stripe(struct bio *bio, | |
1083 | unsigned int bytes_done, int err) | |
1084 | #endif | |
1085 | { | |
cea9e445 | 1086 | struct btrfs_multi_bio *multi = bio->bi_private; |
8790d502 CM |
1087 | |
1088 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1089 | if (bio->bi_size) | |
1090 | return 1; | |
1091 | #endif | |
1092 | if (err) | |
1093 | multi->error = err; | |
1094 | ||
cea9e445 | 1095 | if (atomic_dec_and_test(&multi->stripes_pending)) { |
8790d502 CM |
1096 | bio->bi_private = multi->private; |
1097 | bio->bi_end_io = multi->end_io; | |
1098 | ||
1099 | if (!err && multi->error) | |
1100 | err = multi->error; | |
1101 | kfree(multi); | |
1102 | ||
73f61b2a M |
1103 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) |
1104 | bio_endio(bio, bio->bi_size, err); | |
1105 | #else | |
8790d502 | 1106 | bio_endio(bio, err); |
73f61b2a | 1107 | #endif |
8790d502 CM |
1108 | } else { |
1109 | bio_put(bio); | |
1110 | } | |
1111 | #if LINUX_VERSION_CODE <= KERNEL_VERSION(2,6,23) | |
1112 | return 0; | |
1113 | #endif | |
1114 | } | |
1115 | ||
f188591e CM |
1116 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
1117 | int mirror_num) | |
0b86a832 CM |
1118 | { |
1119 | struct btrfs_mapping_tree *map_tree; | |
1120 | struct btrfs_device *dev; | |
8790d502 | 1121 | struct bio *first_bio = bio; |
0b86a832 | 1122 | u64 logical = bio->bi_sector << 9; |
0b86a832 CM |
1123 | u64 length = 0; |
1124 | u64 map_length; | |
1125 | struct bio_vec *bvec; | |
cea9e445 | 1126 | struct btrfs_multi_bio *multi = NULL; |
0b86a832 CM |
1127 | int i; |
1128 | int ret; | |
8790d502 CM |
1129 | int dev_nr = 0; |
1130 | int total_devs = 1; | |
0b86a832 CM |
1131 | |
1132 | bio_for_each_segment(bvec, bio, i) { | |
1133 | length += bvec->bv_len; | |
1134 | } | |
8790d502 | 1135 | |
0b86a832 CM |
1136 | map_tree = &root->fs_info->mapping_tree; |
1137 | map_length = length; | |
cea9e445 | 1138 | |
f188591e CM |
1139 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi, |
1140 | mirror_num); | |
cea9e445 CM |
1141 | BUG_ON(ret); |
1142 | ||
1143 | total_devs = multi->num_stripes; | |
1144 | if (map_length < length) { | |
1145 | printk("mapping failed logical %Lu bio len %Lu " | |
1146 | "len %Lu\n", logical, length, map_length); | |
1147 | BUG(); | |
1148 | } | |
1149 | multi->end_io = first_bio->bi_end_io; | |
1150 | multi->private = first_bio->bi_private; | |
1151 | atomic_set(&multi->stripes_pending, multi->num_stripes); | |
1152 | ||
8790d502 | 1153 | while(dev_nr < total_devs) { |
8790d502 | 1154 | if (total_devs > 1) { |
8790d502 CM |
1155 | if (dev_nr < total_devs - 1) { |
1156 | bio = bio_clone(first_bio, GFP_NOFS); | |
1157 | BUG_ON(!bio); | |
1158 | } else { | |
1159 | bio = first_bio; | |
1160 | } | |
1161 | bio->bi_private = multi; | |
1162 | bio->bi_end_io = end_bio_multi_stripe; | |
1163 | } | |
cea9e445 CM |
1164 | bio->bi_sector = multi->stripes[dev_nr].physical >> 9; |
1165 | dev = multi->stripes[dev_nr].dev; | |
8790d502 CM |
1166 | bio->bi_bdev = dev->bdev; |
1167 | spin_lock(&dev->io_lock); | |
1168 | dev->total_ios++; | |
1169 | spin_unlock(&dev->io_lock); | |
1170 | submit_bio(rw, bio); | |
1171 | dev_nr++; | |
1172 | } | |
cea9e445 CM |
1173 | if (total_devs == 1) |
1174 | kfree(multi); | |
0b86a832 CM |
1175 | return 0; |
1176 | } | |
1177 | ||
a443755f CM |
1178 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
1179 | u8 *uuid) | |
0b86a832 | 1180 | { |
8a4b83cc | 1181 | struct list_head *head = &root->fs_info->fs_devices->devices; |
0b86a832 | 1182 | |
a443755f | 1183 | return __find_device(head, devid, uuid); |
0b86a832 CM |
1184 | } |
1185 | ||
1186 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, | |
1187 | struct extent_buffer *leaf, | |
1188 | struct btrfs_chunk *chunk) | |
1189 | { | |
1190 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
1191 | struct map_lookup *map; | |
1192 | struct extent_map *em; | |
1193 | u64 logical; | |
1194 | u64 length; | |
1195 | u64 devid; | |
a443755f | 1196 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 1197 | int num_stripes; |
0b86a832 | 1198 | int ret; |
593060d7 | 1199 | int i; |
0b86a832 | 1200 | |
e17cade2 CM |
1201 | logical = key->offset; |
1202 | length = btrfs_chunk_length(leaf, chunk); | |
0b86a832 CM |
1203 | spin_lock(&map_tree->map_tree.lock); |
1204 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); | |
b248a415 | 1205 | spin_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
1206 | |
1207 | /* already mapped? */ | |
1208 | if (em && em->start <= logical && em->start + em->len > logical) { | |
1209 | free_extent_map(em); | |
0b86a832 CM |
1210 | return 0; |
1211 | } else if (em) { | |
1212 | free_extent_map(em); | |
1213 | } | |
0b86a832 CM |
1214 | |
1215 | map = kzalloc(sizeof(*map), GFP_NOFS); | |
1216 | if (!map) | |
1217 | return -ENOMEM; | |
1218 | ||
1219 | em = alloc_extent_map(GFP_NOFS); | |
1220 | if (!em) | |
1221 | return -ENOMEM; | |
593060d7 CM |
1222 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
1223 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
1224 | if (!map) { |
1225 | free_extent_map(em); | |
1226 | return -ENOMEM; | |
1227 | } | |
1228 | ||
1229 | em->bdev = (struct block_device *)map; | |
1230 | em->start = logical; | |
1231 | em->len = length; | |
1232 | em->block_start = 0; | |
1233 | ||
593060d7 CM |
1234 | map->num_stripes = num_stripes; |
1235 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
1236 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
1237 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
1238 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
1239 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 1240 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
1241 | for (i = 0; i < num_stripes; i++) { |
1242 | map->stripes[i].physical = | |
1243 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
1244 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
1245 | read_extent_buffer(leaf, uuid, (unsigned long) |
1246 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
1247 | BTRFS_UUID_SIZE); | |
1248 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid); | |
593060d7 CM |
1249 | if (!map->stripes[i].dev) { |
1250 | kfree(map); | |
1251 | free_extent_map(em); | |
1252 | return -EIO; | |
1253 | } | |
0b86a832 CM |
1254 | } |
1255 | ||
1256 | spin_lock(&map_tree->map_tree.lock); | |
1257 | ret = add_extent_mapping(&map_tree->map_tree, em); | |
0b86a832 | 1258 | spin_unlock(&map_tree->map_tree.lock); |
b248a415 | 1259 | BUG_ON(ret); |
0b86a832 CM |
1260 | free_extent_map(em); |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | ||
1265 | static int fill_device_from_item(struct extent_buffer *leaf, | |
1266 | struct btrfs_dev_item *dev_item, | |
1267 | struct btrfs_device *device) | |
1268 | { | |
1269 | unsigned long ptr; | |
0b86a832 CM |
1270 | |
1271 | device->devid = btrfs_device_id(leaf, dev_item); | |
1272 | device->total_bytes = btrfs_device_total_bytes(leaf, dev_item); | |
1273 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); | |
1274 | device->type = btrfs_device_type(leaf, dev_item); | |
1275 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
1276 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
1277 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
1278 | |
1279 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 1280 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1281 | |
0b86a832 CM |
1282 | return 0; |
1283 | } | |
1284 | ||
0d81ba5d | 1285 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
1286 | struct extent_buffer *leaf, |
1287 | struct btrfs_dev_item *dev_item) | |
1288 | { | |
1289 | struct btrfs_device *device; | |
1290 | u64 devid; | |
1291 | int ret; | |
a443755f CM |
1292 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
1293 | ||
0b86a832 | 1294 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
1295 | read_extent_buffer(leaf, dev_uuid, |
1296 | (unsigned long)btrfs_device_uuid(dev_item), | |
1297 | BTRFS_UUID_SIZE); | |
1298 | device = btrfs_find_device(root, devid, dev_uuid); | |
6324fbf3 | 1299 | if (!device) { |
8a4b83cc | 1300 | printk("warning devid %Lu not found already\n", devid); |
f2984462 | 1301 | device = kzalloc(sizeof(*device), GFP_NOFS); |
6324fbf3 CM |
1302 | if (!device) |
1303 | return -ENOMEM; | |
8a4b83cc CM |
1304 | list_add(&device->dev_list, |
1305 | &root->fs_info->fs_devices->devices); | |
b248a415 | 1306 | device->barriers = 1; |
8790d502 | 1307 | spin_lock_init(&device->io_lock); |
6324fbf3 | 1308 | } |
0b86a832 CM |
1309 | |
1310 | fill_device_from_item(leaf, dev_item, device); | |
1311 | device->dev_root = root->fs_info->dev_root; | |
0b86a832 CM |
1312 | ret = 0; |
1313 | #if 0 | |
1314 | ret = btrfs_open_device(device); | |
1315 | if (ret) { | |
1316 | kfree(device); | |
1317 | } | |
1318 | #endif | |
1319 | return ret; | |
1320 | } | |
1321 | ||
0d81ba5d CM |
1322 | int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf) |
1323 | { | |
1324 | struct btrfs_dev_item *dev_item; | |
1325 | ||
1326 | dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block, | |
1327 | dev_item); | |
1328 | return read_one_dev(root, buf, dev_item); | |
1329 | } | |
1330 | ||
0b86a832 CM |
1331 | int btrfs_read_sys_array(struct btrfs_root *root) |
1332 | { | |
1333 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
1334 | struct extent_buffer *sb = root->fs_info->sb_buffer; | |
1335 | struct btrfs_disk_key *disk_key; | |
0b86a832 CM |
1336 | struct btrfs_chunk *chunk; |
1337 | struct btrfs_key key; | |
1338 | u32 num_stripes; | |
1339 | u32 array_size; | |
1340 | u32 len = 0; | |
1341 | u8 *ptr; | |
1342 | unsigned long sb_ptr; | |
1343 | u32 cur; | |
1344 | int ret; | |
0b86a832 CM |
1345 | |
1346 | array_size = btrfs_super_sys_array_size(super_copy); | |
1347 | ||
1348 | /* | |
1349 | * we do this loop twice, once for the device items and | |
1350 | * once for all of the chunks. This way there are device | |
1351 | * structs filled in for every chunk | |
1352 | */ | |
0b86a832 CM |
1353 | ptr = super_copy->sys_chunk_array; |
1354 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
1355 | cur = 0; | |
1356 | ||
1357 | while (cur < array_size) { | |
1358 | disk_key = (struct btrfs_disk_key *)ptr; | |
1359 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1360 | ||
1361 | len = sizeof(*disk_key); | |
1362 | ptr += len; | |
1363 | sb_ptr += len; | |
1364 | cur += len; | |
1365 | ||
0d81ba5d | 1366 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 1367 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d CM |
1368 | ret = read_one_chunk(root, &key, sb, chunk); |
1369 | BUG_ON(ret); | |
0b86a832 CM |
1370 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
1371 | len = btrfs_chunk_item_size(num_stripes); | |
1372 | } else { | |
1373 | BUG(); | |
1374 | } | |
1375 | ptr += len; | |
1376 | sb_ptr += len; | |
1377 | cur += len; | |
1378 | } | |
0b86a832 CM |
1379 | return 0; |
1380 | } | |
1381 | ||
1382 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
1383 | { | |
1384 | struct btrfs_path *path; | |
1385 | struct extent_buffer *leaf; | |
1386 | struct btrfs_key key; | |
1387 | struct btrfs_key found_key; | |
1388 | int ret; | |
1389 | int slot; | |
1390 | ||
1391 | root = root->fs_info->chunk_root; | |
1392 | ||
1393 | path = btrfs_alloc_path(); | |
1394 | if (!path) | |
1395 | return -ENOMEM; | |
1396 | ||
1397 | /* first we search for all of the device items, and then we | |
1398 | * read in all of the chunk items. This way we can create chunk | |
1399 | * mappings that reference all of the devices that are afound | |
1400 | */ | |
1401 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1402 | key.offset = 0; | |
1403 | key.type = 0; | |
1404 | again: | |
1405 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1406 | while(1) { | |
1407 | leaf = path->nodes[0]; | |
1408 | slot = path->slots[0]; | |
1409 | if (slot >= btrfs_header_nritems(leaf)) { | |
1410 | ret = btrfs_next_leaf(root, path); | |
1411 | if (ret == 0) | |
1412 | continue; | |
1413 | if (ret < 0) | |
1414 | goto error; | |
1415 | break; | |
1416 | } | |
1417 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
1418 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
1419 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
1420 | break; | |
1421 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
1422 | struct btrfs_dev_item *dev_item; | |
1423 | dev_item = btrfs_item_ptr(leaf, slot, | |
1424 | struct btrfs_dev_item); | |
0d81ba5d | 1425 | ret = read_one_dev(root, leaf, dev_item); |
0b86a832 CM |
1426 | BUG_ON(ret); |
1427 | } | |
1428 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1429 | struct btrfs_chunk *chunk; | |
1430 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
1431 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
1432 | } | |
1433 | path->slots[0]++; | |
1434 | } | |
1435 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
1436 | key.objectid = 0; | |
1437 | btrfs_release_path(root, path); | |
1438 | goto again; | |
1439 | } | |
1440 | ||
1441 | btrfs_free_path(path); | |
1442 | ret = 0; | |
1443 | error: | |
1444 | return ret; | |
1445 | } | |
1446 |