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> |
f2d8d74d | 21 | #include <linux/blkdev.h> |
788f20eb | 22 | #include <linux/random.h> |
593060d7 | 23 | #include <asm/div64.h> |
0b86a832 CM |
24 | #include "ctree.h" |
25 | #include "extent_map.h" | |
26 | #include "disk-io.h" | |
27 | #include "transaction.h" | |
28 | #include "print-tree.h" | |
29 | #include "volumes.h" | |
8b712842 | 30 | #include "async-thread.h" |
0b86a832 | 31 | |
593060d7 CM |
32 | struct map_lookup { |
33 | u64 type; | |
34 | int io_align; | |
35 | int io_width; | |
36 | int stripe_len; | |
37 | int sector_size; | |
38 | int num_stripes; | |
321aecc6 | 39 | int sub_stripes; |
cea9e445 | 40 | struct btrfs_bio_stripe stripes[]; |
593060d7 CM |
41 | }; |
42 | ||
43 | #define map_lookup_size(n) (sizeof(struct map_lookup) + \ | |
cea9e445 | 44 | (sizeof(struct btrfs_bio_stripe) * (n))) |
593060d7 | 45 | |
8a4b83cc CM |
46 | static DEFINE_MUTEX(uuid_mutex); |
47 | static LIST_HEAD(fs_uuids); | |
48 | ||
a061fc8d CM |
49 | void btrfs_lock_volumes(void) |
50 | { | |
51 | mutex_lock(&uuid_mutex); | |
52 | } | |
53 | ||
54 | void btrfs_unlock_volumes(void) | |
55 | { | |
56 | mutex_unlock(&uuid_mutex); | |
57 | } | |
58 | ||
7d9eb12c CM |
59 | static void lock_chunks(struct btrfs_root *root) |
60 | { | |
7d9eb12c CM |
61 | mutex_lock(&root->fs_info->chunk_mutex); |
62 | } | |
63 | ||
64 | static void unlock_chunks(struct btrfs_root *root) | |
65 | { | |
7d9eb12c CM |
66 | mutex_unlock(&root->fs_info->chunk_mutex); |
67 | } | |
68 | ||
8a4b83cc CM |
69 | int btrfs_cleanup_fs_uuids(void) |
70 | { | |
71 | struct btrfs_fs_devices *fs_devices; | |
72 | struct list_head *uuid_cur; | |
73 | struct list_head *devices_cur; | |
74 | struct btrfs_device *dev; | |
75 | ||
76 | list_for_each(uuid_cur, &fs_uuids) { | |
77 | fs_devices = list_entry(uuid_cur, struct btrfs_fs_devices, | |
78 | list); | |
79 | while(!list_empty(&fs_devices->devices)) { | |
80 | devices_cur = fs_devices->devices.next; | |
81 | dev = list_entry(devices_cur, struct btrfs_device, | |
82 | dev_list); | |
8a4b83cc | 83 | if (dev->bdev) { |
8a4b83cc | 84 | close_bdev_excl(dev->bdev); |
a0af469b | 85 | fs_devices->open_devices--; |
8a4b83cc CM |
86 | } |
87 | list_del(&dev->dev_list); | |
dfe25020 | 88 | kfree(dev->name); |
8a4b83cc CM |
89 | kfree(dev); |
90 | } | |
91 | } | |
92 | return 0; | |
93 | } | |
94 | ||
a1b32a59 CM |
95 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
96 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
97 | { |
98 | struct btrfs_device *dev; | |
99 | struct list_head *cur; | |
100 | ||
101 | list_for_each(cur, head) { | |
102 | dev = list_entry(cur, struct btrfs_device, dev_list); | |
a443755f | 103 | if (dev->devid == devid && |
8f18cf13 | 104 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 105 | return dev; |
a443755f | 106 | } |
8a4b83cc CM |
107 | } |
108 | return NULL; | |
109 | } | |
110 | ||
a1b32a59 | 111 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc CM |
112 | { |
113 | struct list_head *cur; | |
114 | struct btrfs_fs_devices *fs_devices; | |
115 | ||
116 | list_for_each(cur, &fs_uuids) { | |
117 | fs_devices = list_entry(cur, struct btrfs_fs_devices, list); | |
118 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
119 | return fs_devices; | |
120 | } | |
121 | return NULL; | |
122 | } | |
123 | ||
8b712842 CM |
124 | /* |
125 | * we try to collect pending bios for a device so we don't get a large | |
126 | * number of procs sending bios down to the same device. This greatly | |
127 | * improves the schedulers ability to collect and merge the bios. | |
128 | * | |
129 | * But, it also turns into a long list of bios to process and that is sure | |
130 | * to eventually make the worker thread block. The solution here is to | |
131 | * make some progress and then put this work struct back at the end of | |
132 | * the list if the block device is congested. This way, multiple devices | |
133 | * can make progress from a single worker thread. | |
134 | */ | |
a1b32a59 | 135 | static int noinline run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
136 | { |
137 | struct bio *pending; | |
138 | struct backing_dev_info *bdi; | |
b64a2851 | 139 | struct btrfs_fs_info *fs_info; |
8b712842 CM |
140 | struct bio *tail; |
141 | struct bio *cur; | |
142 | int again = 0; | |
143 | unsigned long num_run = 0; | |
b64a2851 | 144 | unsigned long limit; |
8b712842 CM |
145 | |
146 | bdi = device->bdev->bd_inode->i_mapping->backing_dev_info; | |
b64a2851 CM |
147 | fs_info = device->dev_root->fs_info; |
148 | limit = btrfs_async_submit_limit(fs_info); | |
149 | limit = limit * 2 / 3; | |
150 | ||
8b712842 CM |
151 | loop: |
152 | spin_lock(&device->io_lock); | |
153 | ||
154 | /* take all the bios off the list at once and process them | |
155 | * later on (without the lock held). But, remember the | |
156 | * tail and other pointers so the bios can be properly reinserted | |
157 | * into the list if we hit congestion | |
158 | */ | |
159 | pending = device->pending_bios; | |
160 | tail = device->pending_bio_tail; | |
161 | WARN_ON(pending && !tail); | |
162 | device->pending_bios = NULL; | |
163 | device->pending_bio_tail = NULL; | |
164 | ||
165 | /* | |
166 | * if pending was null this time around, no bios need processing | |
167 | * at all and we can stop. Otherwise it'll loop back up again | |
168 | * and do an additional check so no bios are missed. | |
169 | * | |
170 | * device->running_pending is used to synchronize with the | |
171 | * schedule_bio code. | |
172 | */ | |
173 | if (pending) { | |
174 | again = 1; | |
175 | device->running_pending = 1; | |
176 | } else { | |
177 | again = 0; | |
178 | device->running_pending = 0; | |
179 | } | |
180 | spin_unlock(&device->io_lock); | |
181 | ||
182 | while(pending) { | |
183 | cur = pending; | |
184 | pending = pending->bi_next; | |
185 | cur->bi_next = NULL; | |
b64a2851 CM |
186 | atomic_dec(&fs_info->nr_async_bios); |
187 | ||
188 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
189 | waitqueue_active(&fs_info->async_submit_wait)) | |
190 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
191 | |
192 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
193 | bio_get(cur); | |
8b712842 | 194 | submit_bio(cur->bi_rw, cur); |
492bb6de | 195 | bio_put(cur); |
8b712842 CM |
196 | num_run++; |
197 | ||
198 | /* | |
199 | * we made progress, there is more work to do and the bdi | |
200 | * is now congested. Back off and let other work structs | |
201 | * run instead | |
202 | */ | |
5f2cc086 CM |
203 | if (pending && bdi_write_congested(bdi) && |
204 | fs_info->fs_devices->open_devices > 1) { | |
8b712842 CM |
205 | struct bio *old_head; |
206 | ||
207 | spin_lock(&device->io_lock); | |
492bb6de | 208 | |
8b712842 CM |
209 | old_head = device->pending_bios; |
210 | device->pending_bios = pending; | |
211 | if (device->pending_bio_tail) | |
212 | tail->bi_next = old_head; | |
213 | else | |
214 | device->pending_bio_tail = tail; | |
215 | ||
216 | spin_unlock(&device->io_lock); | |
217 | btrfs_requeue_work(&device->work); | |
218 | goto done; | |
219 | } | |
220 | } | |
221 | if (again) | |
222 | goto loop; | |
223 | done: | |
224 | return 0; | |
225 | } | |
226 | ||
227 | void pending_bios_fn(struct btrfs_work *work) | |
228 | { | |
229 | struct btrfs_device *device; | |
230 | ||
231 | device = container_of(work, struct btrfs_device, work); | |
232 | run_scheduled_bios(device); | |
233 | } | |
234 | ||
a1b32a59 | 235 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
236 | struct btrfs_super_block *disk_super, |
237 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
238 | { | |
239 | struct btrfs_device *device; | |
240 | struct btrfs_fs_devices *fs_devices; | |
241 | u64 found_transid = btrfs_super_generation(disk_super); | |
242 | ||
243 | fs_devices = find_fsid(disk_super->fsid); | |
244 | if (!fs_devices) { | |
515dc322 | 245 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
246 | if (!fs_devices) |
247 | return -ENOMEM; | |
248 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 249 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
250 | list_add(&fs_devices->list, &fs_uuids); |
251 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
252 | fs_devices->latest_devid = devid; | |
253 | fs_devices->latest_trans = found_transid; | |
8a4b83cc CM |
254 | device = NULL; |
255 | } else { | |
a443755f CM |
256 | device = __find_device(&fs_devices->devices, devid, |
257 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
258 | } |
259 | if (!device) { | |
260 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
261 | if (!device) { | |
262 | /* we can safely leave the fs_devices entry around */ | |
263 | return -ENOMEM; | |
264 | } | |
265 | device->devid = devid; | |
8b712842 | 266 | device->work.func = pending_bios_fn; |
a443755f CM |
267 | memcpy(device->uuid, disk_super->dev_item.uuid, |
268 | BTRFS_UUID_SIZE); | |
f2984462 | 269 | device->barriers = 1; |
b248a415 | 270 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
271 | device->name = kstrdup(path, GFP_NOFS); |
272 | if (!device->name) { | |
273 | kfree(device); | |
274 | return -ENOMEM; | |
275 | } | |
276 | list_add(&device->dev_list, &fs_devices->devices); | |
b3075717 | 277 | list_add(&device->dev_alloc_list, &fs_devices->alloc_list); |
8a4b83cc CM |
278 | fs_devices->num_devices++; |
279 | } | |
280 | ||
281 | if (found_transid > fs_devices->latest_trans) { | |
282 | fs_devices->latest_devid = devid; | |
283 | fs_devices->latest_trans = found_transid; | |
284 | } | |
8a4b83cc CM |
285 | *fs_devices_ret = fs_devices; |
286 | return 0; | |
287 | } | |
288 | ||
dfe25020 CM |
289 | int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
290 | { | |
291 | struct list_head *head = &fs_devices->devices; | |
292 | struct list_head *cur; | |
293 | struct btrfs_device *device; | |
294 | ||
295 | mutex_lock(&uuid_mutex); | |
296 | again: | |
297 | list_for_each(cur, head) { | |
298 | device = list_entry(cur, struct btrfs_device, dev_list); | |
299 | if (!device->in_fs_metadata) { | |
a74a4b97 | 300 | struct block_device *bdev; |
dfe25020 CM |
301 | list_del(&device->dev_list); |
302 | list_del(&device->dev_alloc_list); | |
303 | fs_devices->num_devices--; | |
a74a4b97 CM |
304 | if (device->bdev) { |
305 | bdev = device->bdev; | |
306 | fs_devices->open_devices--; | |
307 | mutex_unlock(&uuid_mutex); | |
308 | close_bdev_excl(bdev); | |
309 | mutex_lock(&uuid_mutex); | |
310 | } | |
dfe25020 CM |
311 | kfree(device->name); |
312 | kfree(device); | |
313 | goto again; | |
314 | } | |
315 | } | |
316 | mutex_unlock(&uuid_mutex); | |
317 | return 0; | |
318 | } | |
a0af469b | 319 | |
8a4b83cc CM |
320 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
321 | { | |
322 | struct list_head *head = &fs_devices->devices; | |
323 | struct list_head *cur; | |
324 | struct btrfs_device *device; | |
325 | ||
326 | mutex_lock(&uuid_mutex); | |
327 | list_for_each(cur, head) { | |
328 | device = list_entry(cur, struct btrfs_device, dev_list); | |
329 | if (device->bdev) { | |
330 | close_bdev_excl(device->bdev); | |
a0af469b | 331 | fs_devices->open_devices--; |
8a4b83cc CM |
332 | } |
333 | device->bdev = NULL; | |
dfe25020 | 334 | device->in_fs_metadata = 0; |
8a4b83cc | 335 | } |
a0af469b | 336 | fs_devices->mounted = 0; |
8a4b83cc CM |
337 | mutex_unlock(&uuid_mutex); |
338 | return 0; | |
339 | } | |
340 | ||
341 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
342 | int flags, void *holder) | |
343 | { | |
344 | struct block_device *bdev; | |
345 | struct list_head *head = &fs_devices->devices; | |
346 | struct list_head *cur; | |
347 | struct btrfs_device *device; | |
a0af469b CM |
348 | struct block_device *latest_bdev = NULL; |
349 | struct buffer_head *bh; | |
350 | struct btrfs_super_block *disk_super; | |
351 | u64 latest_devid = 0; | |
352 | u64 latest_transid = 0; | |
353 | u64 transid; | |
354 | u64 devid; | |
355 | int ret = 0; | |
8a4b83cc CM |
356 | |
357 | mutex_lock(&uuid_mutex); | |
a0af469b CM |
358 | if (fs_devices->mounted) |
359 | goto out; | |
360 | ||
8a4b83cc CM |
361 | list_for_each(cur, head) { |
362 | device = list_entry(cur, struct btrfs_device, dev_list); | |
c1c4d91c CM |
363 | if (device->bdev) |
364 | continue; | |
365 | ||
dfe25020 CM |
366 | if (!device->name) |
367 | continue; | |
368 | ||
8a4b83cc | 369 | bdev = open_bdev_excl(device->name, flags, holder); |
e17cade2 | 370 | |
8a4b83cc CM |
371 | if (IS_ERR(bdev)) { |
372 | printk("open %s failed\n", device->name); | |
a0af469b | 373 | goto error; |
8a4b83cc | 374 | } |
a061fc8d | 375 | set_blocksize(bdev, 4096); |
a0af469b CM |
376 | |
377 | bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096); | |
378 | if (!bh) | |
379 | goto error_close; | |
380 | ||
381 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
382 | if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, | |
383 | sizeof(disk_super->magic))) | |
384 | goto error_brelse; | |
385 | ||
386 | devid = le64_to_cpu(disk_super->dev_item.devid); | |
387 | if (devid != device->devid) | |
388 | goto error_brelse; | |
389 | ||
390 | transid = btrfs_super_generation(disk_super); | |
6af5ac3c | 391 | if (!latest_transid || transid > latest_transid) { |
a0af469b CM |
392 | latest_devid = devid; |
393 | latest_transid = transid; | |
394 | latest_bdev = bdev; | |
395 | } | |
396 | ||
8a4b83cc | 397 | device->bdev = bdev; |
dfe25020 | 398 | device->in_fs_metadata = 0; |
a0af469b CM |
399 | fs_devices->open_devices++; |
400 | continue; | |
a061fc8d | 401 | |
a0af469b CM |
402 | error_brelse: |
403 | brelse(bh); | |
404 | error_close: | |
405 | close_bdev_excl(bdev); | |
406 | error: | |
407 | continue; | |
8a4b83cc | 408 | } |
a0af469b CM |
409 | if (fs_devices->open_devices == 0) { |
410 | ret = -EIO; | |
411 | goto out; | |
412 | } | |
413 | fs_devices->mounted = 1; | |
414 | fs_devices->latest_bdev = latest_bdev; | |
415 | fs_devices->latest_devid = latest_devid; | |
416 | fs_devices->latest_trans = latest_transid; | |
417 | out: | |
8a4b83cc | 418 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
419 | return ret; |
420 | } | |
421 | ||
422 | int btrfs_scan_one_device(const char *path, int flags, void *holder, | |
423 | struct btrfs_fs_devices **fs_devices_ret) | |
424 | { | |
425 | struct btrfs_super_block *disk_super; | |
426 | struct block_device *bdev; | |
427 | struct buffer_head *bh; | |
428 | int ret; | |
429 | u64 devid; | |
f2984462 | 430 | u64 transid; |
8a4b83cc CM |
431 | |
432 | mutex_lock(&uuid_mutex); | |
433 | ||
8a4b83cc CM |
434 | bdev = open_bdev_excl(path, flags, holder); |
435 | ||
436 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
437 | ret = PTR_ERR(bdev); |
438 | goto error; | |
439 | } | |
440 | ||
441 | ret = set_blocksize(bdev, 4096); | |
442 | if (ret) | |
443 | goto error_close; | |
444 | bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096); | |
445 | if (!bh) { | |
446 | ret = -EIO; | |
447 | goto error_close; | |
448 | } | |
449 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
450 | if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, | |
451 | sizeof(disk_super->magic))) { | |
e58ca020 | 452 | ret = -EINVAL; |
8a4b83cc CM |
453 | goto error_brelse; |
454 | } | |
455 | devid = le64_to_cpu(disk_super->dev_item.devid); | |
f2984462 | 456 | transid = btrfs_super_generation(disk_super); |
7ae9c09d CM |
457 | if (disk_super->label[0]) |
458 | printk("device label %s ", disk_super->label); | |
459 | else { | |
460 | /* FIXME, make a readl uuid parser */ | |
461 | printk("device fsid %llx-%llx ", | |
462 | *(unsigned long long *)disk_super->fsid, | |
463 | *(unsigned long long *)(disk_super->fsid + 8)); | |
464 | } | |
465 | printk("devid %Lu transid %Lu %s\n", devid, transid, path); | |
8a4b83cc CM |
466 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
467 | ||
468 | error_brelse: | |
469 | brelse(bh); | |
470 | error_close: | |
471 | close_bdev_excl(bdev); | |
8a4b83cc CM |
472 | error: |
473 | mutex_unlock(&uuid_mutex); | |
474 | return ret; | |
475 | } | |
0b86a832 CM |
476 | |
477 | /* | |
478 | * this uses a pretty simple search, the expectation is that it is | |
479 | * called very infrequently and that a given device has a small number | |
480 | * of extents | |
481 | */ | |
a1b32a59 CM |
482 | static noinline int find_free_dev_extent(struct btrfs_trans_handle *trans, |
483 | struct btrfs_device *device, | |
484 | struct btrfs_path *path, | |
485 | u64 num_bytes, u64 *start) | |
0b86a832 CM |
486 | { |
487 | struct btrfs_key key; | |
488 | struct btrfs_root *root = device->dev_root; | |
489 | struct btrfs_dev_extent *dev_extent = NULL; | |
490 | u64 hole_size = 0; | |
491 | u64 last_byte = 0; | |
492 | u64 search_start = 0; | |
493 | u64 search_end = device->total_bytes; | |
494 | int ret; | |
495 | int slot = 0; | |
496 | int start_found; | |
497 | struct extent_buffer *l; | |
498 | ||
499 | start_found = 0; | |
500 | path->reada = 2; | |
501 | ||
502 | /* FIXME use last free of some kind */ | |
503 | ||
8a4b83cc CM |
504 | /* we don't want to overwrite the superblock on the drive, |
505 | * so we make sure to start at an offset of at least 1MB | |
506 | */ | |
507 | search_start = max((u64)1024 * 1024, search_start); | |
8f18cf13 CM |
508 | |
509 | if (root->fs_info->alloc_start + num_bytes <= device->total_bytes) | |
510 | search_start = max(root->fs_info->alloc_start, search_start); | |
511 | ||
0b86a832 CM |
512 | key.objectid = device->devid; |
513 | key.offset = search_start; | |
514 | key.type = BTRFS_DEV_EXTENT_KEY; | |
515 | ret = btrfs_search_slot(trans, root, &key, path, 0, 0); | |
516 | if (ret < 0) | |
517 | goto error; | |
518 | ret = btrfs_previous_item(root, path, 0, key.type); | |
519 | if (ret < 0) | |
520 | goto error; | |
521 | l = path->nodes[0]; | |
522 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
523 | while (1) { | |
524 | l = path->nodes[0]; | |
525 | slot = path->slots[0]; | |
526 | if (slot >= btrfs_header_nritems(l)) { | |
527 | ret = btrfs_next_leaf(root, path); | |
528 | if (ret == 0) | |
529 | continue; | |
530 | if (ret < 0) | |
531 | goto error; | |
532 | no_more_items: | |
533 | if (!start_found) { | |
534 | if (search_start >= search_end) { | |
535 | ret = -ENOSPC; | |
536 | goto error; | |
537 | } | |
538 | *start = search_start; | |
539 | start_found = 1; | |
540 | goto check_pending; | |
541 | } | |
542 | *start = last_byte > search_start ? | |
543 | last_byte : search_start; | |
544 | if (search_end <= *start) { | |
545 | ret = -ENOSPC; | |
546 | goto error; | |
547 | } | |
548 | goto check_pending; | |
549 | } | |
550 | btrfs_item_key_to_cpu(l, &key, slot); | |
551 | ||
552 | if (key.objectid < device->devid) | |
553 | goto next; | |
554 | ||
555 | if (key.objectid > device->devid) | |
556 | goto no_more_items; | |
557 | ||
558 | if (key.offset >= search_start && key.offset > last_byte && | |
559 | start_found) { | |
560 | if (last_byte < search_start) | |
561 | last_byte = search_start; | |
562 | hole_size = key.offset - last_byte; | |
563 | if (key.offset > last_byte && | |
564 | hole_size >= num_bytes) { | |
565 | *start = last_byte; | |
566 | goto check_pending; | |
567 | } | |
568 | } | |
569 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) { | |
570 | goto next; | |
571 | } | |
572 | ||
573 | start_found = 1; | |
574 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
575 | last_byte = key.offset + btrfs_dev_extent_length(l, dev_extent); | |
576 | next: | |
577 | path->slots[0]++; | |
578 | cond_resched(); | |
579 | } | |
580 | check_pending: | |
581 | /* we have to make sure we didn't find an extent that has already | |
582 | * been allocated by the map tree or the original allocation | |
583 | */ | |
584 | btrfs_release_path(root, path); | |
585 | BUG_ON(*start < search_start); | |
586 | ||
6324fbf3 | 587 | if (*start + num_bytes > search_end) { |
0b86a832 CM |
588 | ret = -ENOSPC; |
589 | goto error; | |
590 | } | |
591 | /* check for pending inserts here */ | |
592 | return 0; | |
593 | ||
594 | error: | |
595 | btrfs_release_path(root, path); | |
596 | return ret; | |
597 | } | |
598 | ||
8f18cf13 CM |
599 | int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
600 | struct btrfs_device *device, | |
601 | u64 start) | |
602 | { | |
603 | int ret; | |
604 | struct btrfs_path *path; | |
605 | struct btrfs_root *root = device->dev_root; | |
606 | struct btrfs_key key; | |
a061fc8d CM |
607 | struct btrfs_key found_key; |
608 | struct extent_buffer *leaf = NULL; | |
609 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
610 | |
611 | path = btrfs_alloc_path(); | |
612 | if (!path) | |
613 | return -ENOMEM; | |
614 | ||
615 | key.objectid = device->devid; | |
616 | key.offset = start; | |
617 | key.type = BTRFS_DEV_EXTENT_KEY; | |
618 | ||
619 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
a061fc8d CM |
620 | if (ret > 0) { |
621 | ret = btrfs_previous_item(root, path, key.objectid, | |
622 | BTRFS_DEV_EXTENT_KEY); | |
623 | BUG_ON(ret); | |
624 | leaf = path->nodes[0]; | |
625 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
626 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
627 | struct btrfs_dev_extent); | |
628 | BUG_ON(found_key.offset > start || found_key.offset + | |
629 | btrfs_dev_extent_length(leaf, extent) < start); | |
630 | ret = 0; | |
631 | } else if (ret == 0) { | |
632 | leaf = path->nodes[0]; | |
633 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
634 | struct btrfs_dev_extent); | |
635 | } | |
8f18cf13 CM |
636 | BUG_ON(ret); |
637 | ||
dfe25020 CM |
638 | if (device->bytes_used > 0) |
639 | device->bytes_used -= btrfs_dev_extent_length(leaf, extent); | |
8f18cf13 CM |
640 | ret = btrfs_del_item(trans, root, path); |
641 | BUG_ON(ret); | |
642 | ||
643 | btrfs_free_path(path); | |
644 | return ret; | |
645 | } | |
646 | ||
a1b32a59 | 647 | int noinline btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 648 | struct btrfs_device *device, |
e17cade2 CM |
649 | u64 chunk_tree, u64 chunk_objectid, |
650 | u64 chunk_offset, | |
651 | u64 num_bytes, u64 *start) | |
0b86a832 CM |
652 | { |
653 | int ret; | |
654 | struct btrfs_path *path; | |
655 | struct btrfs_root *root = device->dev_root; | |
656 | struct btrfs_dev_extent *extent; | |
657 | struct extent_buffer *leaf; | |
658 | struct btrfs_key key; | |
659 | ||
dfe25020 | 660 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
661 | path = btrfs_alloc_path(); |
662 | if (!path) | |
663 | return -ENOMEM; | |
664 | ||
665 | ret = find_free_dev_extent(trans, device, path, num_bytes, start); | |
6324fbf3 | 666 | if (ret) { |
0b86a832 | 667 | goto err; |
6324fbf3 | 668 | } |
0b86a832 CM |
669 | |
670 | key.objectid = device->devid; | |
671 | key.offset = *start; | |
672 | key.type = BTRFS_DEV_EXTENT_KEY; | |
673 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
674 | sizeof(*extent)); | |
675 | BUG_ON(ret); | |
676 | ||
677 | leaf = path->nodes[0]; | |
678 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
679 | struct btrfs_dev_extent); | |
e17cade2 CM |
680 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
681 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
682 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
683 | ||
684 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
685 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
686 | BTRFS_UUID_SIZE); | |
687 | ||
0b86a832 CM |
688 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
689 | btrfs_mark_buffer_dirty(leaf); | |
690 | err: | |
691 | btrfs_free_path(path); | |
692 | return ret; | |
693 | } | |
694 | ||
a1b32a59 CM |
695 | static noinline int find_next_chunk(struct btrfs_root *root, |
696 | u64 objectid, u64 *offset) | |
0b86a832 CM |
697 | { |
698 | struct btrfs_path *path; | |
699 | int ret; | |
700 | struct btrfs_key key; | |
e17cade2 | 701 | struct btrfs_chunk *chunk; |
0b86a832 CM |
702 | struct btrfs_key found_key; |
703 | ||
704 | path = btrfs_alloc_path(); | |
705 | BUG_ON(!path); | |
706 | ||
e17cade2 | 707 | key.objectid = objectid; |
0b86a832 CM |
708 | key.offset = (u64)-1; |
709 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
710 | ||
711 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
712 | if (ret < 0) | |
713 | goto error; | |
714 | ||
715 | BUG_ON(ret == 0); | |
716 | ||
717 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
718 | if (ret) { | |
e17cade2 | 719 | *offset = 0; |
0b86a832 CM |
720 | } else { |
721 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
722 | path->slots[0]); | |
e17cade2 CM |
723 | if (found_key.objectid != objectid) |
724 | *offset = 0; | |
725 | else { | |
726 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
727 | struct btrfs_chunk); | |
728 | *offset = found_key.offset + | |
729 | btrfs_chunk_length(path->nodes[0], chunk); | |
730 | } | |
0b86a832 CM |
731 | } |
732 | ret = 0; | |
733 | error: | |
734 | btrfs_free_path(path); | |
735 | return ret; | |
736 | } | |
737 | ||
a1b32a59 CM |
738 | static noinline int find_next_devid(struct btrfs_root *root, |
739 | struct btrfs_path *path, u64 *objectid) | |
0b86a832 CM |
740 | { |
741 | int ret; | |
742 | struct btrfs_key key; | |
743 | struct btrfs_key found_key; | |
744 | ||
745 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
746 | key.type = BTRFS_DEV_ITEM_KEY; | |
747 | key.offset = (u64)-1; | |
748 | ||
749 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
750 | if (ret < 0) | |
751 | goto error; | |
752 | ||
753 | BUG_ON(ret == 0); | |
754 | ||
755 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
756 | BTRFS_DEV_ITEM_KEY); | |
757 | if (ret) { | |
758 | *objectid = 1; | |
759 | } else { | |
760 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
761 | path->slots[0]); | |
762 | *objectid = found_key.offset + 1; | |
763 | } | |
764 | ret = 0; | |
765 | error: | |
766 | btrfs_release_path(root, path); | |
767 | return ret; | |
768 | } | |
769 | ||
770 | /* | |
771 | * the device information is stored in the chunk root | |
772 | * the btrfs_device struct should be fully filled in | |
773 | */ | |
774 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
775 | struct btrfs_root *root, | |
776 | struct btrfs_device *device) | |
777 | { | |
778 | int ret; | |
779 | struct btrfs_path *path; | |
780 | struct btrfs_dev_item *dev_item; | |
781 | struct extent_buffer *leaf; | |
782 | struct btrfs_key key; | |
783 | unsigned long ptr; | |
006a58a2 | 784 | u64 free_devid = 0; |
0b86a832 CM |
785 | |
786 | root = root->fs_info->chunk_root; | |
787 | ||
788 | path = btrfs_alloc_path(); | |
789 | if (!path) | |
790 | return -ENOMEM; | |
791 | ||
792 | ret = find_next_devid(root, path, &free_devid); | |
793 | if (ret) | |
794 | goto out; | |
795 | ||
796 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
797 | key.type = BTRFS_DEV_ITEM_KEY; | |
798 | key.offset = free_devid; | |
799 | ||
800 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 801 | sizeof(*dev_item)); |
0b86a832 CM |
802 | if (ret) |
803 | goto out; | |
804 | ||
805 | leaf = path->nodes[0]; | |
806 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
807 | ||
8a4b83cc | 808 | device->devid = free_devid; |
0b86a832 CM |
809 | btrfs_set_device_id(leaf, dev_item, device->devid); |
810 | btrfs_set_device_type(leaf, dev_item, device->type); | |
811 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
812 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
813 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
814 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
815 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
816 | btrfs_set_device_group(leaf, dev_item, 0); |
817 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
818 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
0b86a832 | 819 | |
0b86a832 | 820 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 821 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
822 | btrfs_mark_buffer_dirty(leaf); |
823 | ret = 0; | |
824 | ||
825 | out: | |
826 | btrfs_free_path(path); | |
827 | return ret; | |
828 | } | |
8f18cf13 | 829 | |
a061fc8d CM |
830 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
831 | struct btrfs_device *device) | |
832 | { | |
833 | int ret; | |
834 | struct btrfs_path *path; | |
835 | struct block_device *bdev = device->bdev; | |
836 | struct btrfs_device *next_dev; | |
837 | struct btrfs_key key; | |
838 | u64 total_bytes; | |
839 | struct btrfs_fs_devices *fs_devices; | |
840 | struct btrfs_trans_handle *trans; | |
841 | ||
842 | root = root->fs_info->chunk_root; | |
843 | ||
844 | path = btrfs_alloc_path(); | |
845 | if (!path) | |
846 | return -ENOMEM; | |
847 | ||
848 | trans = btrfs_start_transaction(root, 1); | |
849 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
850 | key.type = BTRFS_DEV_ITEM_KEY; | |
851 | key.offset = device->devid; | |
7d9eb12c | 852 | lock_chunks(root); |
a061fc8d CM |
853 | |
854 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
855 | if (ret < 0) | |
856 | goto out; | |
857 | ||
858 | if (ret > 0) { | |
859 | ret = -ENOENT; | |
860 | goto out; | |
861 | } | |
862 | ||
863 | ret = btrfs_del_item(trans, root, path); | |
864 | if (ret) | |
865 | goto out; | |
866 | ||
867 | /* | |
868 | * at this point, the device is zero sized. We want to | |
869 | * remove it from the devices list and zero out the old super | |
870 | */ | |
871 | list_del_init(&device->dev_list); | |
872 | list_del_init(&device->dev_alloc_list); | |
873 | fs_devices = root->fs_info->fs_devices; | |
874 | ||
875 | next_dev = list_entry(fs_devices->devices.next, struct btrfs_device, | |
876 | dev_list); | |
a061fc8d CM |
877 | if (bdev == root->fs_info->sb->s_bdev) |
878 | root->fs_info->sb->s_bdev = next_dev->bdev; | |
879 | if (bdev == fs_devices->latest_bdev) | |
880 | fs_devices->latest_bdev = next_dev->bdev; | |
881 | ||
a061fc8d CM |
882 | total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy); |
883 | btrfs_set_super_num_devices(&root->fs_info->super_copy, | |
884 | total_bytes - 1); | |
885 | out: | |
886 | btrfs_free_path(path); | |
7d9eb12c | 887 | unlock_chunks(root); |
a061fc8d CM |
888 | btrfs_commit_transaction(trans, root); |
889 | return ret; | |
890 | } | |
891 | ||
892 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
893 | { | |
894 | struct btrfs_device *device; | |
895 | struct block_device *bdev; | |
dfe25020 | 896 | struct buffer_head *bh = NULL; |
a061fc8d CM |
897 | struct btrfs_super_block *disk_super; |
898 | u64 all_avail; | |
899 | u64 devid; | |
900 | int ret = 0; | |
901 | ||
a061fc8d | 902 | mutex_lock(&uuid_mutex); |
7d9eb12c | 903 | mutex_lock(&root->fs_info->volume_mutex); |
a061fc8d CM |
904 | |
905 | all_avail = root->fs_info->avail_data_alloc_bits | | |
906 | root->fs_info->avail_system_alloc_bits | | |
907 | root->fs_info->avail_metadata_alloc_bits; | |
908 | ||
909 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
dfe25020 | 910 | btrfs_super_num_devices(&root->fs_info->super_copy) <= 4) { |
a061fc8d CM |
911 | printk("btrfs: unable to go below four devices on raid10\n"); |
912 | ret = -EINVAL; | |
913 | goto out; | |
914 | } | |
915 | ||
916 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
dfe25020 | 917 | btrfs_super_num_devices(&root->fs_info->super_copy) <= 2) { |
a061fc8d CM |
918 | printk("btrfs: unable to go below two devices on raid1\n"); |
919 | ret = -EINVAL; | |
920 | goto out; | |
921 | } | |
922 | ||
dfe25020 CM |
923 | if (strcmp(device_path, "missing") == 0) { |
924 | struct list_head *cur; | |
925 | struct list_head *devices; | |
926 | struct btrfs_device *tmp; | |
a061fc8d | 927 | |
dfe25020 CM |
928 | device = NULL; |
929 | devices = &root->fs_info->fs_devices->devices; | |
930 | list_for_each(cur, devices) { | |
931 | tmp = list_entry(cur, struct btrfs_device, dev_list); | |
932 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
933 | device = tmp; | |
934 | break; | |
935 | } | |
936 | } | |
937 | bdev = NULL; | |
938 | bh = NULL; | |
939 | disk_super = NULL; | |
940 | if (!device) { | |
941 | printk("btrfs: no missing devices found to remove\n"); | |
942 | goto out; | |
943 | } | |
944 | ||
945 | } else { | |
946 | bdev = open_bdev_excl(device_path, 0, | |
947 | root->fs_info->bdev_holder); | |
948 | if (IS_ERR(bdev)) { | |
949 | ret = PTR_ERR(bdev); | |
950 | goto out; | |
951 | } | |
a061fc8d | 952 | |
dfe25020 CM |
953 | bh = __bread(bdev, BTRFS_SUPER_INFO_OFFSET / 4096, 4096); |
954 | if (!bh) { | |
955 | ret = -EIO; | |
956 | goto error_close; | |
957 | } | |
958 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
959 | if (strncmp((char *)(&disk_super->magic), BTRFS_MAGIC, | |
960 | sizeof(disk_super->magic))) { | |
961 | ret = -ENOENT; | |
962 | goto error_brelse; | |
963 | } | |
964 | if (memcmp(disk_super->fsid, root->fs_info->fsid, | |
965 | BTRFS_FSID_SIZE)) { | |
966 | ret = -ENOENT; | |
967 | goto error_brelse; | |
968 | } | |
969 | devid = le64_to_cpu(disk_super->dev_item.devid); | |
970 | device = btrfs_find_device(root, devid, NULL); | |
971 | if (!device) { | |
972 | ret = -ENOENT; | |
973 | goto error_brelse; | |
974 | } | |
975 | ||
976 | } | |
a061fc8d | 977 | root->fs_info->fs_devices->num_devices--; |
0ef3e66b | 978 | root->fs_info->fs_devices->open_devices--; |
a061fc8d CM |
979 | |
980 | ret = btrfs_shrink_device(device, 0); | |
981 | if (ret) | |
982 | goto error_brelse; | |
983 | ||
984 | ||
985 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); | |
986 | if (ret) | |
987 | goto error_brelse; | |
988 | ||
dfe25020 CM |
989 | if (bh) { |
990 | /* make sure this device isn't detected as part of | |
991 | * the FS anymore | |
992 | */ | |
993 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
994 | set_buffer_dirty(bh); | |
995 | sync_dirty_buffer(bh); | |
a061fc8d | 996 | |
dfe25020 CM |
997 | brelse(bh); |
998 | } | |
a061fc8d | 999 | |
dfe25020 CM |
1000 | if (device->bdev) { |
1001 | /* one close for the device struct or super_block */ | |
1002 | close_bdev_excl(device->bdev); | |
1003 | } | |
1004 | if (bdev) { | |
1005 | /* one close for us */ | |
1006 | close_bdev_excl(bdev); | |
1007 | } | |
a061fc8d CM |
1008 | kfree(device->name); |
1009 | kfree(device); | |
1010 | ret = 0; | |
1011 | goto out; | |
1012 | ||
1013 | error_brelse: | |
1014 | brelse(bh); | |
1015 | error_close: | |
dfe25020 CM |
1016 | if (bdev) |
1017 | close_bdev_excl(bdev); | |
a061fc8d | 1018 | out: |
7d9eb12c | 1019 | mutex_unlock(&root->fs_info->volume_mutex); |
a061fc8d | 1020 | mutex_unlock(&uuid_mutex); |
a061fc8d CM |
1021 | return ret; |
1022 | } | |
1023 | ||
788f20eb CM |
1024 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1025 | { | |
1026 | struct btrfs_trans_handle *trans; | |
1027 | struct btrfs_device *device; | |
1028 | struct block_device *bdev; | |
1029 | struct list_head *cur; | |
1030 | struct list_head *devices; | |
1031 | u64 total_bytes; | |
1032 | int ret = 0; | |
1033 | ||
1034 | ||
1035 | bdev = open_bdev_excl(device_path, 0, root->fs_info->bdev_holder); | |
1036 | if (!bdev) { | |
1037 | return -EIO; | |
1038 | } | |
a2135011 | 1039 | |
8c8bee1d | 1040 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
7d9eb12c | 1041 | mutex_lock(&root->fs_info->volume_mutex); |
a2135011 | 1042 | |
788f20eb | 1043 | trans = btrfs_start_transaction(root, 1); |
7d9eb12c | 1044 | lock_chunks(root); |
788f20eb CM |
1045 | devices = &root->fs_info->fs_devices->devices; |
1046 | list_for_each(cur, devices) { | |
1047 | device = list_entry(cur, struct btrfs_device, dev_list); | |
1048 | if (device->bdev == bdev) { | |
1049 | ret = -EEXIST; | |
1050 | goto out; | |
1051 | } | |
1052 | } | |
1053 | ||
1054 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1055 | if (!device) { | |
1056 | /* we can safely leave the fs_devices entry around */ | |
1057 | ret = -ENOMEM; | |
1058 | goto out_close_bdev; | |
1059 | } | |
1060 | ||
1061 | device->barriers = 1; | |
8b712842 | 1062 | device->work.func = pending_bios_fn; |
788f20eb CM |
1063 | generate_random_uuid(device->uuid); |
1064 | spin_lock_init(&device->io_lock); | |
1065 | device->name = kstrdup(device_path, GFP_NOFS); | |
1066 | if (!device->name) { | |
1067 | kfree(device); | |
1068 | goto out_close_bdev; | |
1069 | } | |
1070 | device->io_width = root->sectorsize; | |
1071 | device->io_align = root->sectorsize; | |
1072 | device->sector_size = root->sectorsize; | |
1073 | device->total_bytes = i_size_read(bdev->bd_inode); | |
1074 | device->dev_root = root->fs_info->dev_root; | |
1075 | device->bdev = bdev; | |
dfe25020 | 1076 | device->in_fs_metadata = 1; |
788f20eb CM |
1077 | |
1078 | ret = btrfs_add_device(trans, root, device); | |
1079 | if (ret) | |
1080 | goto out_close_bdev; | |
1081 | ||
325cd4ba ZY |
1082 | set_blocksize(device->bdev, 4096); |
1083 | ||
788f20eb CM |
1084 | total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
1085 | btrfs_set_super_total_bytes(&root->fs_info->super_copy, | |
1086 | total_bytes + device->total_bytes); | |
1087 | ||
1088 | total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy); | |
1089 | btrfs_set_super_num_devices(&root->fs_info->super_copy, | |
1090 | total_bytes + 1); | |
1091 | ||
1092 | list_add(&device->dev_list, &root->fs_info->fs_devices->devices); | |
1093 | list_add(&device->dev_alloc_list, | |
1094 | &root->fs_info->fs_devices->alloc_list); | |
1095 | root->fs_info->fs_devices->num_devices++; | |
a0af469b | 1096 | root->fs_info->fs_devices->open_devices++; |
788f20eb | 1097 | out: |
7d9eb12c | 1098 | unlock_chunks(root); |
788f20eb | 1099 | btrfs_end_transaction(trans, root); |
7d9eb12c | 1100 | mutex_unlock(&root->fs_info->volume_mutex); |
a2135011 | 1101 | |
788f20eb CM |
1102 | return ret; |
1103 | ||
1104 | out_close_bdev: | |
1105 | close_bdev_excl(bdev); | |
1106 | goto out; | |
1107 | } | |
1108 | ||
a1b32a59 CM |
1109 | int noinline btrfs_update_device(struct btrfs_trans_handle *trans, |
1110 | struct btrfs_device *device) | |
0b86a832 CM |
1111 | { |
1112 | int ret; | |
1113 | struct btrfs_path *path; | |
1114 | struct btrfs_root *root; | |
1115 | struct btrfs_dev_item *dev_item; | |
1116 | struct extent_buffer *leaf; | |
1117 | struct btrfs_key key; | |
1118 | ||
1119 | root = device->dev_root->fs_info->chunk_root; | |
1120 | ||
1121 | path = btrfs_alloc_path(); | |
1122 | if (!path) | |
1123 | return -ENOMEM; | |
1124 | ||
1125 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1126 | key.type = BTRFS_DEV_ITEM_KEY; | |
1127 | key.offset = device->devid; | |
1128 | ||
1129 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1130 | if (ret < 0) | |
1131 | goto out; | |
1132 | ||
1133 | if (ret > 0) { | |
1134 | ret = -ENOENT; | |
1135 | goto out; | |
1136 | } | |
1137 | ||
1138 | leaf = path->nodes[0]; | |
1139 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1140 | ||
1141 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1142 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1143 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1144 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1145 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1146 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1147 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
1148 | btrfs_mark_buffer_dirty(leaf); | |
1149 | ||
1150 | out: | |
1151 | btrfs_free_path(path); | |
1152 | return ret; | |
1153 | } | |
1154 | ||
7d9eb12c | 1155 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1156 | struct btrfs_device *device, u64 new_size) |
1157 | { | |
1158 | struct btrfs_super_block *super_copy = | |
1159 | &device->dev_root->fs_info->super_copy; | |
1160 | u64 old_total = btrfs_super_total_bytes(super_copy); | |
1161 | u64 diff = new_size - device->total_bytes; | |
1162 | ||
1163 | btrfs_set_super_total_bytes(super_copy, old_total + diff); | |
1164 | return btrfs_update_device(trans, device); | |
1165 | } | |
1166 | ||
7d9eb12c CM |
1167 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1168 | struct btrfs_device *device, u64 new_size) | |
1169 | { | |
1170 | int ret; | |
1171 | lock_chunks(device->dev_root); | |
1172 | ret = __btrfs_grow_device(trans, device, new_size); | |
1173 | unlock_chunks(device->dev_root); | |
1174 | return ret; | |
1175 | } | |
1176 | ||
8f18cf13 CM |
1177 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1178 | struct btrfs_root *root, | |
1179 | u64 chunk_tree, u64 chunk_objectid, | |
1180 | u64 chunk_offset) | |
1181 | { | |
1182 | int ret; | |
1183 | struct btrfs_path *path; | |
1184 | struct btrfs_key key; | |
1185 | ||
1186 | root = root->fs_info->chunk_root; | |
1187 | path = btrfs_alloc_path(); | |
1188 | if (!path) | |
1189 | return -ENOMEM; | |
1190 | ||
1191 | key.objectid = chunk_objectid; | |
1192 | key.offset = chunk_offset; | |
1193 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1194 | ||
1195 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1196 | BUG_ON(ret); | |
1197 | ||
1198 | ret = btrfs_del_item(trans, root, path); | |
1199 | BUG_ON(ret); | |
1200 | ||
1201 | btrfs_free_path(path); | |
1202 | return 0; | |
1203 | } | |
1204 | ||
1205 | int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 | |
1206 | chunk_offset) | |
1207 | { | |
1208 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
1209 | struct btrfs_disk_key *disk_key; | |
1210 | struct btrfs_chunk *chunk; | |
1211 | u8 *ptr; | |
1212 | int ret = 0; | |
1213 | u32 num_stripes; | |
1214 | u32 array_size; | |
1215 | u32 len = 0; | |
1216 | u32 cur; | |
1217 | struct btrfs_key key; | |
1218 | ||
1219 | array_size = btrfs_super_sys_array_size(super_copy); | |
1220 | ||
1221 | ptr = super_copy->sys_chunk_array; | |
1222 | cur = 0; | |
1223 | ||
1224 | while (cur < array_size) { | |
1225 | disk_key = (struct btrfs_disk_key *)ptr; | |
1226 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1227 | ||
1228 | len = sizeof(*disk_key); | |
1229 | ||
1230 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1231 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1232 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1233 | len += btrfs_chunk_item_size(num_stripes); | |
1234 | } else { | |
1235 | ret = -EIO; | |
1236 | break; | |
1237 | } | |
1238 | if (key.objectid == chunk_objectid && | |
1239 | key.offset == chunk_offset) { | |
1240 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1241 | array_size -= len; | |
1242 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1243 | } else { | |
1244 | ptr += len; | |
1245 | cur += len; | |
1246 | } | |
1247 | } | |
1248 | return ret; | |
1249 | } | |
1250 | ||
1251 | ||
1252 | int btrfs_relocate_chunk(struct btrfs_root *root, | |
1253 | u64 chunk_tree, u64 chunk_objectid, | |
1254 | u64 chunk_offset) | |
1255 | { | |
1256 | struct extent_map_tree *em_tree; | |
1257 | struct btrfs_root *extent_root; | |
1258 | struct btrfs_trans_handle *trans; | |
1259 | struct extent_map *em; | |
1260 | struct map_lookup *map; | |
1261 | int ret; | |
1262 | int i; | |
1263 | ||
323da79c CM |
1264 | printk("btrfs relocating chunk %llu\n", |
1265 | (unsigned long long)chunk_offset); | |
8f18cf13 CM |
1266 | root = root->fs_info->chunk_root; |
1267 | extent_root = root->fs_info->extent_root; | |
1268 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
1269 | ||
1270 | /* step one, relocate all the extents inside this chunk */ | |
1a40e23b | 1271 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
8f18cf13 CM |
1272 | BUG_ON(ret); |
1273 | ||
1274 | trans = btrfs_start_transaction(root, 1); | |
1275 | BUG_ON(!trans); | |
1276 | ||
7d9eb12c CM |
1277 | lock_chunks(root); |
1278 | ||
8f18cf13 CM |
1279 | /* |
1280 | * step two, delete the device extents and the | |
1281 | * chunk tree entries | |
1282 | */ | |
1283 | spin_lock(&em_tree->lock); | |
1284 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
1285 | spin_unlock(&em_tree->lock); | |
1286 | ||
a061fc8d CM |
1287 | BUG_ON(em->start > chunk_offset || |
1288 | em->start + em->len < chunk_offset); | |
8f18cf13 CM |
1289 | map = (struct map_lookup *)em->bdev; |
1290 | ||
1291 | for (i = 0; i < map->num_stripes; i++) { | |
1292 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
1293 | map->stripes[i].physical); | |
1294 | BUG_ON(ret); | |
a061fc8d | 1295 | |
dfe25020 CM |
1296 | if (map->stripes[i].dev) { |
1297 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
1298 | BUG_ON(ret); | |
1299 | } | |
8f18cf13 CM |
1300 | } |
1301 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
1302 | chunk_offset); | |
1303 | ||
1304 | BUG_ON(ret); | |
1305 | ||
1306 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
1307 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
1308 | BUG_ON(ret); | |
8f18cf13 CM |
1309 | } |
1310 | ||
1a40e23b ZY |
1311 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
1312 | BUG_ON(ret); | |
1313 | ||
8f18cf13 CM |
1314 | spin_lock(&em_tree->lock); |
1315 | remove_extent_mapping(em_tree, em); | |
1a40e23b ZY |
1316 | spin_unlock(&em_tree->lock); |
1317 | ||
8f18cf13 CM |
1318 | kfree(map); |
1319 | em->bdev = NULL; | |
1320 | ||
1321 | /* once for the tree */ | |
1322 | free_extent_map(em); | |
8f18cf13 CM |
1323 | /* once for us */ |
1324 | free_extent_map(em); | |
1325 | ||
7d9eb12c | 1326 | unlock_chunks(root); |
8f18cf13 CM |
1327 | btrfs_end_transaction(trans, root); |
1328 | return 0; | |
1329 | } | |
1330 | ||
ec44a35c CM |
1331 | static u64 div_factor(u64 num, int factor) |
1332 | { | |
1333 | if (factor == 10) | |
1334 | return num; | |
1335 | num *= factor; | |
1336 | do_div(num, 10); | |
1337 | return num; | |
1338 | } | |
1339 | ||
1340 | ||
1341 | int btrfs_balance(struct btrfs_root *dev_root) | |
1342 | { | |
1343 | int ret; | |
1344 | struct list_head *cur; | |
1345 | struct list_head *devices = &dev_root->fs_info->fs_devices->devices; | |
1346 | struct btrfs_device *device; | |
1347 | u64 old_size; | |
1348 | u64 size_to_free; | |
1349 | struct btrfs_path *path; | |
1350 | struct btrfs_key key; | |
1351 | struct btrfs_chunk *chunk; | |
1352 | struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root; | |
1353 | struct btrfs_trans_handle *trans; | |
1354 | struct btrfs_key found_key; | |
1355 | ||
1356 | ||
7d9eb12c | 1357 | mutex_lock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
1358 | dev_root = dev_root->fs_info->dev_root; |
1359 | ||
ec44a35c CM |
1360 | /* step one make some room on all the devices */ |
1361 | list_for_each(cur, devices) { | |
1362 | device = list_entry(cur, struct btrfs_device, dev_list); | |
1363 | old_size = device->total_bytes; | |
1364 | size_to_free = div_factor(old_size, 1); | |
1365 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
1366 | if (device->total_bytes - device->bytes_used > size_to_free) | |
1367 | continue; | |
1368 | ||
1369 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
1370 | BUG_ON(ret); | |
1371 | ||
1372 | trans = btrfs_start_transaction(dev_root, 1); | |
1373 | BUG_ON(!trans); | |
1374 | ||
1375 | ret = btrfs_grow_device(trans, device, old_size); | |
1376 | BUG_ON(ret); | |
1377 | ||
1378 | btrfs_end_transaction(trans, dev_root); | |
1379 | } | |
1380 | ||
1381 | /* step two, relocate all the chunks */ | |
1382 | path = btrfs_alloc_path(); | |
1383 | BUG_ON(!path); | |
1384 | ||
1385 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
1386 | key.offset = (u64)-1; | |
1387 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1388 | ||
1389 | while(1) { | |
1390 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
1391 | if (ret < 0) | |
1392 | goto error; | |
1393 | ||
1394 | /* | |
1395 | * this shouldn't happen, it means the last relocate | |
1396 | * failed | |
1397 | */ | |
1398 | if (ret == 0) | |
1399 | break; | |
1400 | ||
1401 | ret = btrfs_previous_item(chunk_root, path, 0, | |
1402 | BTRFS_CHUNK_ITEM_KEY); | |
7d9eb12c | 1403 | if (ret) |
ec44a35c | 1404 | break; |
7d9eb12c | 1405 | |
ec44a35c CM |
1406 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
1407 | path->slots[0]); | |
1408 | if (found_key.objectid != key.objectid) | |
1409 | break; | |
7d9eb12c | 1410 | |
ec44a35c CM |
1411 | chunk = btrfs_item_ptr(path->nodes[0], |
1412 | path->slots[0], | |
1413 | struct btrfs_chunk); | |
1414 | key.offset = found_key.offset; | |
1415 | /* chunk zero is special */ | |
1416 | if (key.offset == 0) | |
1417 | break; | |
1418 | ||
7d9eb12c | 1419 | btrfs_release_path(chunk_root, path); |
ec44a35c CM |
1420 | ret = btrfs_relocate_chunk(chunk_root, |
1421 | chunk_root->root_key.objectid, | |
1422 | found_key.objectid, | |
1423 | found_key.offset); | |
1424 | BUG_ON(ret); | |
ec44a35c CM |
1425 | } |
1426 | ret = 0; | |
1427 | error: | |
1428 | btrfs_free_path(path); | |
7d9eb12c | 1429 | mutex_unlock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
1430 | return ret; |
1431 | } | |
1432 | ||
8f18cf13 CM |
1433 | /* |
1434 | * shrinking a device means finding all of the device extents past | |
1435 | * the new size, and then following the back refs to the chunks. | |
1436 | * The chunk relocation code actually frees the device extent | |
1437 | */ | |
1438 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
1439 | { | |
1440 | struct btrfs_trans_handle *trans; | |
1441 | struct btrfs_root *root = device->dev_root; | |
1442 | struct btrfs_dev_extent *dev_extent = NULL; | |
1443 | struct btrfs_path *path; | |
1444 | u64 length; | |
1445 | u64 chunk_tree; | |
1446 | u64 chunk_objectid; | |
1447 | u64 chunk_offset; | |
1448 | int ret; | |
1449 | int slot; | |
1450 | struct extent_buffer *l; | |
1451 | struct btrfs_key key; | |
1452 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
1453 | u64 old_total = btrfs_super_total_bytes(super_copy); | |
1454 | u64 diff = device->total_bytes - new_size; | |
1455 | ||
1456 | ||
1457 | path = btrfs_alloc_path(); | |
1458 | if (!path) | |
1459 | return -ENOMEM; | |
1460 | ||
1461 | trans = btrfs_start_transaction(root, 1); | |
1462 | if (!trans) { | |
1463 | ret = -ENOMEM; | |
1464 | goto done; | |
1465 | } | |
1466 | ||
1467 | path->reada = 2; | |
1468 | ||
7d9eb12c CM |
1469 | lock_chunks(root); |
1470 | ||
8f18cf13 CM |
1471 | device->total_bytes = new_size; |
1472 | ret = btrfs_update_device(trans, device); | |
1473 | if (ret) { | |
7d9eb12c | 1474 | unlock_chunks(root); |
8f18cf13 CM |
1475 | btrfs_end_transaction(trans, root); |
1476 | goto done; | |
1477 | } | |
1478 | WARN_ON(diff > old_total); | |
1479 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
7d9eb12c | 1480 | unlock_chunks(root); |
8f18cf13 CM |
1481 | btrfs_end_transaction(trans, root); |
1482 | ||
1483 | key.objectid = device->devid; | |
1484 | key.offset = (u64)-1; | |
1485 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1486 | ||
1487 | while (1) { | |
1488 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1489 | if (ret < 0) | |
1490 | goto done; | |
1491 | ||
1492 | ret = btrfs_previous_item(root, path, 0, key.type); | |
1493 | if (ret < 0) | |
1494 | goto done; | |
1495 | if (ret) { | |
1496 | ret = 0; | |
1497 | goto done; | |
1498 | } | |
1499 | ||
1500 | l = path->nodes[0]; | |
1501 | slot = path->slots[0]; | |
1502 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
1503 | ||
1504 | if (key.objectid != device->devid) | |
1505 | goto done; | |
1506 | ||
1507 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1508 | length = btrfs_dev_extent_length(l, dev_extent); | |
1509 | ||
1510 | if (key.offset + length <= new_size) | |
1511 | goto done; | |
1512 | ||
1513 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
1514 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
1515 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
1516 | btrfs_release_path(root, path); | |
1517 | ||
1518 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
1519 | chunk_offset); | |
1520 | if (ret) | |
1521 | goto done; | |
1522 | } | |
1523 | ||
1524 | done: | |
1525 | btrfs_free_path(path); | |
1526 | return ret; | |
1527 | } | |
1528 | ||
0b86a832 CM |
1529 | int btrfs_add_system_chunk(struct btrfs_trans_handle *trans, |
1530 | struct btrfs_root *root, | |
1531 | struct btrfs_key *key, | |
1532 | struct btrfs_chunk *chunk, int item_size) | |
1533 | { | |
1534 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
1535 | struct btrfs_disk_key disk_key; | |
1536 | u32 array_size; | |
1537 | u8 *ptr; | |
1538 | ||
1539 | array_size = btrfs_super_sys_array_size(super_copy); | |
1540 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
1541 | return -EFBIG; | |
1542 | ||
1543 | ptr = super_copy->sys_chunk_array + array_size; | |
1544 | btrfs_cpu_key_to_disk(&disk_key, key); | |
1545 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
1546 | ptr += sizeof(disk_key); | |
1547 | memcpy(ptr, chunk, item_size); | |
1548 | item_size += sizeof(disk_key); | |
1549 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
1550 | return 0; | |
1551 | } | |
1552 | ||
a1b32a59 CM |
1553 | static u64 noinline chunk_bytes_by_type(u64 type, u64 calc_size, |
1554 | int num_stripes, int sub_stripes) | |
9b3f68b9 CM |
1555 | { |
1556 | if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP)) | |
1557 | return calc_size; | |
1558 | else if (type & BTRFS_BLOCK_GROUP_RAID10) | |
1559 | return calc_size * (num_stripes / sub_stripes); | |
1560 | else | |
1561 | return calc_size * num_stripes; | |
1562 | } | |
1563 | ||
1564 | ||
0b86a832 CM |
1565 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
1566 | struct btrfs_root *extent_root, u64 *start, | |
6324fbf3 | 1567 | u64 *num_bytes, u64 type) |
0b86a832 CM |
1568 | { |
1569 | u64 dev_offset; | |
593060d7 | 1570 | struct btrfs_fs_info *info = extent_root->fs_info; |
0b86a832 | 1571 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; |
8f18cf13 | 1572 | struct btrfs_path *path; |
0b86a832 CM |
1573 | struct btrfs_stripe *stripes; |
1574 | struct btrfs_device *device = NULL; | |
1575 | struct btrfs_chunk *chunk; | |
6324fbf3 | 1576 | struct list_head private_devs; |
b3075717 | 1577 | struct list_head *dev_list; |
6324fbf3 | 1578 | struct list_head *cur; |
0b86a832 CM |
1579 | struct extent_map_tree *em_tree; |
1580 | struct map_lookup *map; | |
1581 | struct extent_map *em; | |
a40a90a0 | 1582 | int min_stripe_size = 1 * 1024 * 1024; |
0b86a832 CM |
1583 | u64 physical; |
1584 | u64 calc_size = 1024 * 1024 * 1024; | |
9b3f68b9 CM |
1585 | u64 max_chunk_size = calc_size; |
1586 | u64 min_free; | |
6324fbf3 CM |
1587 | u64 avail; |
1588 | u64 max_avail = 0; | |
9b3f68b9 | 1589 | u64 percent_max; |
6324fbf3 | 1590 | int num_stripes = 1; |
a40a90a0 | 1591 | int min_stripes = 1; |
321aecc6 | 1592 | int sub_stripes = 0; |
6324fbf3 | 1593 | int looped = 0; |
0b86a832 | 1594 | int ret; |
6324fbf3 | 1595 | int index; |
593060d7 | 1596 | int stripe_len = 64 * 1024; |
0b86a832 CM |
1597 | struct btrfs_key key; |
1598 | ||
ec44a35c CM |
1599 | if ((type & BTRFS_BLOCK_GROUP_RAID1) && |
1600 | (type & BTRFS_BLOCK_GROUP_DUP)) { | |
1601 | WARN_ON(1); | |
1602 | type &= ~BTRFS_BLOCK_GROUP_DUP; | |
1603 | } | |
b3075717 | 1604 | dev_list = &extent_root->fs_info->fs_devices->alloc_list; |
6324fbf3 CM |
1605 | if (list_empty(dev_list)) |
1606 | return -ENOSPC; | |
593060d7 | 1607 | |
a40a90a0 | 1608 | if (type & (BTRFS_BLOCK_GROUP_RAID0)) { |
0ef3e66b | 1609 | num_stripes = extent_root->fs_info->fs_devices->open_devices; |
a40a90a0 CM |
1610 | min_stripes = 2; |
1611 | } | |
1612 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
611f0e00 | 1613 | num_stripes = 2; |
a40a90a0 CM |
1614 | min_stripes = 2; |
1615 | } | |
8790d502 CM |
1616 | if (type & (BTRFS_BLOCK_GROUP_RAID1)) { |
1617 | num_stripes = min_t(u64, 2, | |
0ef3e66b | 1618 | extent_root->fs_info->fs_devices->open_devices); |
9b3f68b9 CM |
1619 | if (num_stripes < 2) |
1620 | return -ENOSPC; | |
a40a90a0 | 1621 | min_stripes = 2; |
8790d502 | 1622 | } |
321aecc6 | 1623 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { |
0ef3e66b | 1624 | num_stripes = extent_root->fs_info->fs_devices->open_devices; |
321aecc6 CM |
1625 | if (num_stripes < 4) |
1626 | return -ENOSPC; | |
1627 | num_stripes &= ~(u32)1; | |
1628 | sub_stripes = 2; | |
a40a90a0 | 1629 | min_stripes = 4; |
321aecc6 | 1630 | } |
9b3f68b9 CM |
1631 | |
1632 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
1633 | max_chunk_size = 10 * calc_size; | |
a40a90a0 | 1634 | min_stripe_size = 64 * 1024 * 1024; |
9b3f68b9 CM |
1635 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1636 | max_chunk_size = 4 * calc_size; | |
a40a90a0 CM |
1637 | min_stripe_size = 32 * 1024 * 1024; |
1638 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
1639 | calc_size = 8 * 1024 * 1024; | |
1640 | max_chunk_size = calc_size * 2; | |
1641 | min_stripe_size = 1 * 1024 * 1024; | |
9b3f68b9 CM |
1642 | } |
1643 | ||
8f18cf13 CM |
1644 | path = btrfs_alloc_path(); |
1645 | if (!path) | |
1646 | return -ENOMEM; | |
1647 | ||
9b3f68b9 CM |
1648 | /* we don't want a chunk larger than 10% of the FS */ |
1649 | percent_max = div_factor(btrfs_super_total_bytes(&info->super_copy), 1); | |
1650 | max_chunk_size = min(percent_max, max_chunk_size); | |
1651 | ||
a40a90a0 | 1652 | again: |
9b3f68b9 CM |
1653 | if (calc_size * num_stripes > max_chunk_size) { |
1654 | calc_size = max_chunk_size; | |
1655 | do_div(calc_size, num_stripes); | |
1656 | do_div(calc_size, stripe_len); | |
1657 | calc_size *= stripe_len; | |
1658 | } | |
1659 | /* we don't want tiny stripes */ | |
a40a90a0 | 1660 | calc_size = max_t(u64, min_stripe_size, calc_size); |
9b3f68b9 | 1661 | |
9b3f68b9 CM |
1662 | do_div(calc_size, stripe_len); |
1663 | calc_size *= stripe_len; | |
1664 | ||
6324fbf3 CM |
1665 | INIT_LIST_HEAD(&private_devs); |
1666 | cur = dev_list->next; | |
1667 | index = 0; | |
611f0e00 CM |
1668 | |
1669 | if (type & BTRFS_BLOCK_GROUP_DUP) | |
1670 | min_free = calc_size * 2; | |
9b3f68b9 CM |
1671 | else |
1672 | min_free = calc_size; | |
611f0e00 | 1673 | |
0f9dd46c JB |
1674 | /* |
1675 | * we add 1MB because we never use the first 1MB of the device, unless | |
1676 | * we've looped, then we are likely allocating the maximum amount of | |
1677 | * space left already | |
1678 | */ | |
1679 | if (!looped) | |
1680 | min_free += 1024 * 1024; | |
ad5bd91e | 1681 | |
6324fbf3 CM |
1682 | /* build a private list of devices we will allocate from */ |
1683 | while(index < num_stripes) { | |
b3075717 | 1684 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
611f0e00 | 1685 | |
dfe25020 CM |
1686 | if (device->total_bytes > device->bytes_used) |
1687 | avail = device->total_bytes - device->bytes_used; | |
1688 | else | |
1689 | avail = 0; | |
6324fbf3 | 1690 | cur = cur->next; |
8f18cf13 | 1691 | |
dfe25020 | 1692 | if (device->in_fs_metadata && avail >= min_free) { |
8f18cf13 CM |
1693 | u64 ignored_start = 0; |
1694 | ret = find_free_dev_extent(trans, device, path, | |
1695 | min_free, | |
1696 | &ignored_start); | |
1697 | if (ret == 0) { | |
1698 | list_move_tail(&device->dev_alloc_list, | |
1699 | &private_devs); | |
611f0e00 | 1700 | index++; |
8f18cf13 CM |
1701 | if (type & BTRFS_BLOCK_GROUP_DUP) |
1702 | index++; | |
1703 | } | |
dfe25020 | 1704 | } else if (device->in_fs_metadata && avail > max_avail) |
a40a90a0 | 1705 | max_avail = avail; |
6324fbf3 CM |
1706 | if (cur == dev_list) |
1707 | break; | |
1708 | } | |
1709 | if (index < num_stripes) { | |
1710 | list_splice(&private_devs, dev_list); | |
a40a90a0 CM |
1711 | if (index >= min_stripes) { |
1712 | num_stripes = index; | |
1713 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
1714 | num_stripes /= sub_stripes; | |
1715 | num_stripes *= sub_stripes; | |
1716 | } | |
1717 | looped = 1; | |
1718 | goto again; | |
1719 | } | |
6324fbf3 CM |
1720 | if (!looped && max_avail > 0) { |
1721 | looped = 1; | |
1722 | calc_size = max_avail; | |
1723 | goto again; | |
1724 | } | |
8f18cf13 | 1725 | btrfs_free_path(path); |
6324fbf3 CM |
1726 | return -ENOSPC; |
1727 | } | |
e17cade2 CM |
1728 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
1729 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1730 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
1731 | &key.offset); | |
8f18cf13 CM |
1732 | if (ret) { |
1733 | btrfs_free_path(path); | |
0b86a832 | 1734 | return ret; |
8f18cf13 | 1735 | } |
0b86a832 | 1736 | |
0b86a832 | 1737 | chunk = kmalloc(btrfs_chunk_item_size(num_stripes), GFP_NOFS); |
8f18cf13 CM |
1738 | if (!chunk) { |
1739 | btrfs_free_path(path); | |
0b86a832 | 1740 | return -ENOMEM; |
8f18cf13 | 1741 | } |
0b86a832 | 1742 | |
593060d7 CM |
1743 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
1744 | if (!map) { | |
1745 | kfree(chunk); | |
8f18cf13 | 1746 | btrfs_free_path(path); |
593060d7 CM |
1747 | return -ENOMEM; |
1748 | } | |
8f18cf13 CM |
1749 | btrfs_free_path(path); |
1750 | path = NULL; | |
593060d7 | 1751 | |
0b86a832 | 1752 | stripes = &chunk->stripe; |
9b3f68b9 CM |
1753 | *num_bytes = chunk_bytes_by_type(type, calc_size, |
1754 | num_stripes, sub_stripes); | |
0b86a832 | 1755 | |
6324fbf3 | 1756 | index = 0; |
0b86a832 | 1757 | while(index < num_stripes) { |
e17cade2 | 1758 | struct btrfs_stripe *stripe; |
6324fbf3 CM |
1759 | BUG_ON(list_empty(&private_devs)); |
1760 | cur = private_devs.next; | |
b3075717 | 1761 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
611f0e00 CM |
1762 | |
1763 | /* loop over this device again if we're doing a dup group */ | |
1764 | if (!(type & BTRFS_BLOCK_GROUP_DUP) || | |
1765 | (index == num_stripes - 1)) | |
b3075717 | 1766 | list_move_tail(&device->dev_alloc_list, dev_list); |
0b86a832 CM |
1767 | |
1768 | ret = btrfs_alloc_dev_extent(trans, device, | |
e17cade2 CM |
1769 | info->chunk_root->root_key.objectid, |
1770 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, key.offset, | |
1771 | calc_size, &dev_offset); | |
0b86a832 | 1772 | BUG_ON(ret); |
0b86a832 CM |
1773 | device->bytes_used += calc_size; |
1774 | ret = btrfs_update_device(trans, device); | |
1775 | BUG_ON(ret); | |
1776 | ||
593060d7 CM |
1777 | map->stripes[index].dev = device; |
1778 | map->stripes[index].physical = dev_offset; | |
e17cade2 CM |
1779 | stripe = stripes + index; |
1780 | btrfs_set_stack_stripe_devid(stripe, device->devid); | |
1781 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
1782 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
0b86a832 CM |
1783 | physical = dev_offset; |
1784 | index++; | |
1785 | } | |
6324fbf3 | 1786 | BUG_ON(!list_empty(&private_devs)); |
0b86a832 | 1787 | |
e17cade2 CM |
1788 | /* key was set above */ |
1789 | btrfs_set_stack_chunk_length(chunk, *num_bytes); | |
0b86a832 | 1790 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
593060d7 | 1791 | btrfs_set_stack_chunk_stripe_len(chunk, stripe_len); |
0b86a832 CM |
1792 | btrfs_set_stack_chunk_type(chunk, type); |
1793 | btrfs_set_stack_chunk_num_stripes(chunk, num_stripes); | |
593060d7 CM |
1794 | btrfs_set_stack_chunk_io_align(chunk, stripe_len); |
1795 | btrfs_set_stack_chunk_io_width(chunk, stripe_len); | |
0b86a832 | 1796 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
321aecc6 | 1797 | btrfs_set_stack_chunk_sub_stripes(chunk, sub_stripes); |
593060d7 CM |
1798 | map->sector_size = extent_root->sectorsize; |
1799 | map->stripe_len = stripe_len; | |
1800 | map->io_align = stripe_len; | |
1801 | map->io_width = stripe_len; | |
1802 | map->type = type; | |
1803 | map->num_stripes = num_stripes; | |
321aecc6 | 1804 | map->sub_stripes = sub_stripes; |
0b86a832 CM |
1805 | |
1806 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, | |
1807 | btrfs_chunk_item_size(num_stripes)); | |
1808 | BUG_ON(ret); | |
e17cade2 | 1809 | *start = key.offset;; |
0b86a832 CM |
1810 | |
1811 | em = alloc_extent_map(GFP_NOFS); | |
1812 | if (!em) | |
1813 | return -ENOMEM; | |
0b86a832 | 1814 | em->bdev = (struct block_device *)map; |
e17cade2 CM |
1815 | em->start = key.offset; |
1816 | em->len = *num_bytes; | |
0b86a832 | 1817 | em->block_start = 0; |
c8b97818 | 1818 | em->block_len = em->len; |
0b86a832 | 1819 | |
8f18cf13 CM |
1820 | if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
1821 | ret = btrfs_add_system_chunk(trans, chunk_root, &key, | |
1822 | chunk, btrfs_chunk_item_size(num_stripes)); | |
1823 | BUG_ON(ret); | |
1824 | } | |
0b86a832 CM |
1825 | kfree(chunk); |
1826 | ||
1827 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; | |
1828 | spin_lock(&em_tree->lock); | |
1829 | ret = add_extent_mapping(em_tree, em); | |
0b86a832 | 1830 | spin_unlock(&em_tree->lock); |
b248a415 | 1831 | BUG_ON(ret); |
0b86a832 CM |
1832 | free_extent_map(em); |
1833 | return ret; | |
1834 | } | |
1835 | ||
1836 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
1837 | { | |
1838 | extent_map_tree_init(&tree->map_tree, GFP_NOFS); | |
1839 | } | |
1840 | ||
1841 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
1842 | { | |
1843 | struct extent_map *em; | |
1844 | ||
1845 | while(1) { | |
1846 | spin_lock(&tree->map_tree.lock); | |
1847 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); | |
1848 | if (em) | |
1849 | remove_extent_mapping(&tree->map_tree, em); | |
1850 | spin_unlock(&tree->map_tree.lock); | |
1851 | if (!em) | |
1852 | break; | |
1853 | kfree(em->bdev); | |
1854 | /* once for us */ | |
1855 | free_extent_map(em); | |
1856 | /* once for the tree */ | |
1857 | free_extent_map(em); | |
1858 | } | |
1859 | } | |
1860 | ||
f188591e CM |
1861 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
1862 | { | |
1863 | struct extent_map *em; | |
1864 | struct map_lookup *map; | |
1865 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
1866 | int ret; | |
1867 | ||
1868 | spin_lock(&em_tree->lock); | |
1869 | em = lookup_extent_mapping(em_tree, logical, len); | |
b248a415 | 1870 | spin_unlock(&em_tree->lock); |
f188591e CM |
1871 | BUG_ON(!em); |
1872 | ||
1873 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
1874 | map = (struct map_lookup *)em->bdev; | |
1875 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
1876 | ret = map->num_stripes; | |
321aecc6 CM |
1877 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
1878 | ret = map->sub_stripes; | |
f188591e CM |
1879 | else |
1880 | ret = 1; | |
1881 | free_extent_map(em); | |
f188591e CM |
1882 | return ret; |
1883 | } | |
1884 | ||
dfe25020 CM |
1885 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
1886 | int optimal) | |
1887 | { | |
1888 | int i; | |
1889 | if (map->stripes[optimal].dev->bdev) | |
1890 | return optimal; | |
1891 | for (i = first; i < first + num; i++) { | |
1892 | if (map->stripes[i].dev->bdev) | |
1893 | return i; | |
1894 | } | |
1895 | /* we couldn't find one that doesn't fail. Just return something | |
1896 | * and the io error handling code will clean up eventually | |
1897 | */ | |
1898 | return optimal; | |
1899 | } | |
1900 | ||
f2d8d74d CM |
1901 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
1902 | u64 logical, u64 *length, | |
1903 | struct btrfs_multi_bio **multi_ret, | |
1904 | int mirror_num, struct page *unplug_page) | |
0b86a832 CM |
1905 | { |
1906 | struct extent_map *em; | |
1907 | struct map_lookup *map; | |
1908 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
1909 | u64 offset; | |
593060d7 CM |
1910 | u64 stripe_offset; |
1911 | u64 stripe_nr; | |
cea9e445 | 1912 | int stripes_allocated = 8; |
321aecc6 | 1913 | int stripes_required = 1; |
593060d7 | 1914 | int stripe_index; |
cea9e445 | 1915 | int i; |
f2d8d74d | 1916 | int num_stripes; |
a236aed1 | 1917 | int max_errors = 0; |
cea9e445 | 1918 | struct btrfs_multi_bio *multi = NULL; |
0b86a832 | 1919 | |
cea9e445 CM |
1920 | if (multi_ret && !(rw & (1 << BIO_RW))) { |
1921 | stripes_allocated = 1; | |
1922 | } | |
1923 | again: | |
1924 | if (multi_ret) { | |
1925 | multi = kzalloc(btrfs_multi_bio_size(stripes_allocated), | |
1926 | GFP_NOFS); | |
1927 | if (!multi) | |
1928 | return -ENOMEM; | |
a236aed1 CM |
1929 | |
1930 | atomic_set(&multi->error, 0); | |
cea9e445 | 1931 | } |
0b86a832 CM |
1932 | |
1933 | spin_lock(&em_tree->lock); | |
1934 | em = lookup_extent_mapping(em_tree, logical, *length); | |
b248a415 | 1935 | spin_unlock(&em_tree->lock); |
f2d8d74d CM |
1936 | |
1937 | if (!em && unplug_page) | |
1938 | return 0; | |
1939 | ||
3b951516 | 1940 | if (!em) { |
a061fc8d | 1941 | printk("unable to find logical %Lu len %Lu\n", logical, *length); |
f2d8d74d | 1942 | BUG(); |
3b951516 | 1943 | } |
0b86a832 CM |
1944 | |
1945 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
1946 | map = (struct map_lookup *)em->bdev; | |
1947 | offset = logical - em->start; | |
593060d7 | 1948 | |
f188591e CM |
1949 | if (mirror_num > map->num_stripes) |
1950 | mirror_num = 0; | |
1951 | ||
cea9e445 | 1952 | /* if our multi bio struct is too small, back off and try again */ |
321aecc6 CM |
1953 | if (rw & (1 << BIO_RW)) { |
1954 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
1955 | BTRFS_BLOCK_GROUP_DUP)) { | |
1956 | stripes_required = map->num_stripes; | |
a236aed1 | 1957 | max_errors = 1; |
321aecc6 CM |
1958 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
1959 | stripes_required = map->sub_stripes; | |
a236aed1 | 1960 | max_errors = 1; |
321aecc6 CM |
1961 | } |
1962 | } | |
1963 | if (multi_ret && rw == WRITE && | |
1964 | stripes_allocated < stripes_required) { | |
cea9e445 | 1965 | stripes_allocated = map->num_stripes; |
cea9e445 CM |
1966 | free_extent_map(em); |
1967 | kfree(multi); | |
1968 | goto again; | |
1969 | } | |
593060d7 CM |
1970 | stripe_nr = offset; |
1971 | /* | |
1972 | * stripe_nr counts the total number of stripes we have to stride | |
1973 | * to get to this block | |
1974 | */ | |
1975 | do_div(stripe_nr, map->stripe_len); | |
1976 | ||
1977 | stripe_offset = stripe_nr * map->stripe_len; | |
1978 | BUG_ON(offset < stripe_offset); | |
1979 | ||
1980 | /* stripe_offset is the offset of this block in its stripe*/ | |
1981 | stripe_offset = offset - stripe_offset; | |
1982 | ||
cea9e445 | 1983 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 1984 | BTRFS_BLOCK_GROUP_RAID10 | |
cea9e445 CM |
1985 | BTRFS_BLOCK_GROUP_DUP)) { |
1986 | /* we limit the length of each bio to what fits in a stripe */ | |
1987 | *length = min_t(u64, em->len - offset, | |
1988 | map->stripe_len - stripe_offset); | |
1989 | } else { | |
1990 | *length = em->len - offset; | |
1991 | } | |
f2d8d74d CM |
1992 | |
1993 | if (!multi_ret && !unplug_page) | |
cea9e445 CM |
1994 | goto out; |
1995 | ||
f2d8d74d | 1996 | num_stripes = 1; |
cea9e445 | 1997 | stripe_index = 0; |
8790d502 | 1998 | if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
f2d8d74d CM |
1999 | if (unplug_page || (rw & (1 << BIO_RW))) |
2000 | num_stripes = map->num_stripes; | |
2fff734f | 2001 | else if (mirror_num) |
f188591e | 2002 | stripe_index = mirror_num - 1; |
dfe25020 CM |
2003 | else { |
2004 | stripe_index = find_live_mirror(map, 0, | |
2005 | map->num_stripes, | |
2006 | current->pid % map->num_stripes); | |
2007 | } | |
2fff734f | 2008 | |
611f0e00 | 2009 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
cea9e445 | 2010 | if (rw & (1 << BIO_RW)) |
f2d8d74d | 2011 | num_stripes = map->num_stripes; |
f188591e CM |
2012 | else if (mirror_num) |
2013 | stripe_index = mirror_num - 1; | |
2fff734f | 2014 | |
321aecc6 CM |
2015 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
2016 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
2017 | |
2018 | stripe_index = do_div(stripe_nr, factor); | |
2019 | stripe_index *= map->sub_stripes; | |
2020 | ||
f2d8d74d CM |
2021 | if (unplug_page || (rw & (1 << BIO_RW))) |
2022 | num_stripes = map->sub_stripes; | |
321aecc6 CM |
2023 | else if (mirror_num) |
2024 | stripe_index += mirror_num - 1; | |
dfe25020 CM |
2025 | else { |
2026 | stripe_index = find_live_mirror(map, stripe_index, | |
2027 | map->sub_stripes, stripe_index + | |
2028 | current->pid % map->sub_stripes); | |
2029 | } | |
8790d502 CM |
2030 | } else { |
2031 | /* | |
2032 | * after this do_div call, stripe_nr is the number of stripes | |
2033 | * on this device we have to walk to find the data, and | |
2034 | * stripe_index is the number of our device in the stripe array | |
2035 | */ | |
2036 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
2037 | } | |
593060d7 | 2038 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 2039 | |
f2d8d74d CM |
2040 | for (i = 0; i < num_stripes; i++) { |
2041 | if (unplug_page) { | |
2042 | struct btrfs_device *device; | |
2043 | struct backing_dev_info *bdi; | |
2044 | ||
2045 | device = map->stripes[stripe_index].dev; | |
dfe25020 CM |
2046 | if (device->bdev) { |
2047 | bdi = blk_get_backing_dev_info(device->bdev); | |
2048 | if (bdi->unplug_io_fn) { | |
2049 | bdi->unplug_io_fn(bdi, unplug_page); | |
2050 | } | |
f2d8d74d CM |
2051 | } |
2052 | } else { | |
2053 | multi->stripes[i].physical = | |
2054 | map->stripes[stripe_index].physical + | |
2055 | stripe_offset + stripe_nr * map->stripe_len; | |
2056 | multi->stripes[i].dev = map->stripes[stripe_index].dev; | |
2057 | } | |
cea9e445 | 2058 | stripe_index++; |
593060d7 | 2059 | } |
f2d8d74d CM |
2060 | if (multi_ret) { |
2061 | *multi_ret = multi; | |
2062 | multi->num_stripes = num_stripes; | |
a236aed1 | 2063 | multi->max_errors = max_errors; |
f2d8d74d | 2064 | } |
cea9e445 | 2065 | out: |
0b86a832 | 2066 | free_extent_map(em); |
0b86a832 CM |
2067 | return 0; |
2068 | } | |
2069 | ||
f2d8d74d CM |
2070 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
2071 | u64 logical, u64 *length, | |
2072 | struct btrfs_multi_bio **multi_ret, int mirror_num) | |
2073 | { | |
2074 | return __btrfs_map_block(map_tree, rw, logical, length, multi_ret, | |
2075 | mirror_num, NULL); | |
2076 | } | |
2077 | ||
2078 | int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree, | |
2079 | u64 logical, struct page *page) | |
2080 | { | |
2081 | u64 length = PAGE_CACHE_SIZE; | |
2082 | return __btrfs_map_block(map_tree, READ, logical, &length, | |
2083 | NULL, 0, page); | |
2084 | } | |
2085 | ||
2086 | ||
8790d502 | 2087 | static void end_bio_multi_stripe(struct bio *bio, int err) |
8790d502 | 2088 | { |
cea9e445 | 2089 | struct btrfs_multi_bio *multi = bio->bi_private; |
7d2b4daa | 2090 | int is_orig_bio = 0; |
8790d502 | 2091 | |
8790d502 | 2092 | if (err) |
a236aed1 | 2093 | atomic_inc(&multi->error); |
8790d502 | 2094 | |
7d2b4daa CM |
2095 | if (bio == multi->orig_bio) |
2096 | is_orig_bio = 1; | |
2097 | ||
cea9e445 | 2098 | if (atomic_dec_and_test(&multi->stripes_pending)) { |
7d2b4daa CM |
2099 | if (!is_orig_bio) { |
2100 | bio_put(bio); | |
2101 | bio = multi->orig_bio; | |
2102 | } | |
8790d502 CM |
2103 | bio->bi_private = multi->private; |
2104 | bio->bi_end_io = multi->end_io; | |
a236aed1 CM |
2105 | /* only send an error to the higher layers if it is |
2106 | * beyond the tolerance of the multi-bio | |
2107 | */ | |
1259ab75 | 2108 | if (atomic_read(&multi->error) > multi->max_errors) { |
a236aed1 | 2109 | err = -EIO; |
1259ab75 CM |
2110 | } else if (err) { |
2111 | /* | |
2112 | * this bio is actually up to date, we didn't | |
2113 | * go over the max number of errors | |
2114 | */ | |
2115 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 2116 | err = 0; |
1259ab75 | 2117 | } |
8790d502 CM |
2118 | kfree(multi); |
2119 | ||
2120 | bio_endio(bio, err); | |
7d2b4daa | 2121 | } else if (!is_orig_bio) { |
8790d502 CM |
2122 | bio_put(bio); |
2123 | } | |
8790d502 CM |
2124 | } |
2125 | ||
8b712842 CM |
2126 | struct async_sched { |
2127 | struct bio *bio; | |
2128 | int rw; | |
2129 | struct btrfs_fs_info *info; | |
2130 | struct btrfs_work work; | |
2131 | }; | |
2132 | ||
2133 | /* | |
2134 | * see run_scheduled_bios for a description of why bios are collected for | |
2135 | * async submit. | |
2136 | * | |
2137 | * This will add one bio to the pending list for a device and make sure | |
2138 | * the work struct is scheduled. | |
2139 | */ | |
a1b32a59 CM |
2140 | static int noinline schedule_bio(struct btrfs_root *root, |
2141 | struct btrfs_device *device, | |
2142 | int rw, struct bio *bio) | |
8b712842 CM |
2143 | { |
2144 | int should_queue = 1; | |
2145 | ||
2146 | /* don't bother with additional async steps for reads, right now */ | |
2147 | if (!(rw & (1 << BIO_RW))) { | |
492bb6de | 2148 | bio_get(bio); |
8b712842 | 2149 | submit_bio(rw, bio); |
492bb6de | 2150 | bio_put(bio); |
8b712842 CM |
2151 | return 0; |
2152 | } | |
2153 | ||
2154 | /* | |
0986fe9e | 2155 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
2156 | * higher layers. Otherwise, the async bio makes it appear we have |
2157 | * made progress against dirty pages when we've really just put it | |
2158 | * on a queue for later | |
2159 | */ | |
0986fe9e | 2160 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 2161 | WARN_ON(bio->bi_next); |
8b712842 CM |
2162 | bio->bi_next = NULL; |
2163 | bio->bi_rw |= rw; | |
2164 | ||
2165 | spin_lock(&device->io_lock); | |
2166 | ||
2167 | if (device->pending_bio_tail) | |
2168 | device->pending_bio_tail->bi_next = bio; | |
2169 | ||
2170 | device->pending_bio_tail = bio; | |
2171 | if (!device->pending_bios) | |
2172 | device->pending_bios = bio; | |
2173 | if (device->running_pending) | |
2174 | should_queue = 0; | |
2175 | ||
2176 | spin_unlock(&device->io_lock); | |
2177 | ||
2178 | if (should_queue) | |
1cc127b5 CM |
2179 | btrfs_queue_worker(&root->fs_info->submit_workers, |
2180 | &device->work); | |
8b712842 CM |
2181 | return 0; |
2182 | } | |
2183 | ||
f188591e | 2184 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 2185 | int mirror_num, int async_submit) |
0b86a832 CM |
2186 | { |
2187 | struct btrfs_mapping_tree *map_tree; | |
2188 | struct btrfs_device *dev; | |
8790d502 | 2189 | struct bio *first_bio = bio; |
a62b9401 | 2190 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
2191 | u64 length = 0; |
2192 | u64 map_length; | |
cea9e445 | 2193 | struct btrfs_multi_bio *multi = NULL; |
0b86a832 | 2194 | int ret; |
8790d502 CM |
2195 | int dev_nr = 0; |
2196 | int total_devs = 1; | |
0b86a832 | 2197 | |
f2d8d74d | 2198 | length = bio->bi_size; |
0b86a832 CM |
2199 | map_tree = &root->fs_info->mapping_tree; |
2200 | map_length = length; | |
cea9e445 | 2201 | |
f188591e CM |
2202 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi, |
2203 | mirror_num); | |
cea9e445 CM |
2204 | BUG_ON(ret); |
2205 | ||
2206 | total_devs = multi->num_stripes; | |
2207 | if (map_length < length) { | |
2208 | printk("mapping failed logical %Lu bio len %Lu " | |
2209 | "len %Lu\n", logical, length, map_length); | |
2210 | BUG(); | |
2211 | } | |
2212 | multi->end_io = first_bio->bi_end_io; | |
2213 | multi->private = first_bio->bi_private; | |
7d2b4daa | 2214 | multi->orig_bio = first_bio; |
cea9e445 CM |
2215 | atomic_set(&multi->stripes_pending, multi->num_stripes); |
2216 | ||
8790d502 | 2217 | while(dev_nr < total_devs) { |
8790d502 | 2218 | if (total_devs > 1) { |
8790d502 CM |
2219 | if (dev_nr < total_devs - 1) { |
2220 | bio = bio_clone(first_bio, GFP_NOFS); | |
2221 | BUG_ON(!bio); | |
2222 | } else { | |
2223 | bio = first_bio; | |
2224 | } | |
2225 | bio->bi_private = multi; | |
2226 | bio->bi_end_io = end_bio_multi_stripe; | |
2227 | } | |
cea9e445 CM |
2228 | bio->bi_sector = multi->stripes[dev_nr].physical >> 9; |
2229 | dev = multi->stripes[dev_nr].dev; | |
dfe25020 CM |
2230 | if (dev && dev->bdev) { |
2231 | bio->bi_bdev = dev->bdev; | |
8b712842 CM |
2232 | if (async_submit) |
2233 | schedule_bio(root, dev, rw, bio); | |
2234 | else | |
2235 | submit_bio(rw, bio); | |
dfe25020 CM |
2236 | } else { |
2237 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
2238 | bio->bi_sector = logical >> 9; | |
dfe25020 | 2239 | bio_endio(bio, -EIO); |
dfe25020 | 2240 | } |
8790d502 CM |
2241 | dev_nr++; |
2242 | } | |
cea9e445 CM |
2243 | if (total_devs == 1) |
2244 | kfree(multi); | |
0b86a832 CM |
2245 | return 0; |
2246 | } | |
2247 | ||
a443755f CM |
2248 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2249 | u8 *uuid) | |
0b86a832 | 2250 | { |
8a4b83cc | 2251 | struct list_head *head = &root->fs_info->fs_devices->devices; |
0b86a832 | 2252 | |
a443755f | 2253 | return __find_device(head, devid, uuid); |
0b86a832 CM |
2254 | } |
2255 | ||
dfe25020 CM |
2256 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
2257 | u64 devid, u8 *dev_uuid) | |
2258 | { | |
2259 | struct btrfs_device *device; | |
2260 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2261 | ||
2262 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
2263 | list_add(&device->dev_list, | |
2264 | &fs_devices->devices); | |
2265 | list_add(&device->dev_alloc_list, | |
2266 | &fs_devices->alloc_list); | |
2267 | device->barriers = 1; | |
2268 | device->dev_root = root->fs_info->dev_root; | |
2269 | device->devid = devid; | |
8b712842 | 2270 | device->work.func = pending_bios_fn; |
dfe25020 CM |
2271 | fs_devices->num_devices++; |
2272 | spin_lock_init(&device->io_lock); | |
2273 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); | |
2274 | return device; | |
2275 | } | |
2276 | ||
2277 | ||
0b86a832 CM |
2278 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
2279 | struct extent_buffer *leaf, | |
2280 | struct btrfs_chunk *chunk) | |
2281 | { | |
2282 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
2283 | struct map_lookup *map; | |
2284 | struct extent_map *em; | |
2285 | u64 logical; | |
2286 | u64 length; | |
2287 | u64 devid; | |
a443755f | 2288 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 2289 | int num_stripes; |
0b86a832 | 2290 | int ret; |
593060d7 | 2291 | int i; |
0b86a832 | 2292 | |
e17cade2 CM |
2293 | logical = key->offset; |
2294 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 2295 | |
0b86a832 CM |
2296 | spin_lock(&map_tree->map_tree.lock); |
2297 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); | |
b248a415 | 2298 | spin_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
2299 | |
2300 | /* already mapped? */ | |
2301 | if (em && em->start <= logical && em->start + em->len > logical) { | |
2302 | free_extent_map(em); | |
0b86a832 CM |
2303 | return 0; |
2304 | } else if (em) { | |
2305 | free_extent_map(em); | |
2306 | } | |
0b86a832 CM |
2307 | |
2308 | map = kzalloc(sizeof(*map), GFP_NOFS); | |
2309 | if (!map) | |
2310 | return -ENOMEM; | |
2311 | ||
2312 | em = alloc_extent_map(GFP_NOFS); | |
2313 | if (!em) | |
2314 | return -ENOMEM; | |
593060d7 CM |
2315 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
2316 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
2317 | if (!map) { |
2318 | free_extent_map(em); | |
2319 | return -ENOMEM; | |
2320 | } | |
2321 | ||
2322 | em->bdev = (struct block_device *)map; | |
2323 | em->start = logical; | |
2324 | em->len = length; | |
2325 | em->block_start = 0; | |
c8b97818 | 2326 | em->block_len = em->len; |
0b86a832 | 2327 | |
593060d7 CM |
2328 | map->num_stripes = num_stripes; |
2329 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
2330 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
2331 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
2332 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
2333 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 2334 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
2335 | for (i = 0; i < num_stripes; i++) { |
2336 | map->stripes[i].physical = | |
2337 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
2338 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
2339 | read_extent_buffer(leaf, uuid, (unsigned long) |
2340 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
2341 | BTRFS_UUID_SIZE); | |
2342 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid); | |
dfe25020 CM |
2343 | |
2344 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { | |
593060d7 CM |
2345 | kfree(map); |
2346 | free_extent_map(em); | |
2347 | return -EIO; | |
2348 | } | |
dfe25020 CM |
2349 | if (!map->stripes[i].dev) { |
2350 | map->stripes[i].dev = | |
2351 | add_missing_dev(root, devid, uuid); | |
2352 | if (!map->stripes[i].dev) { | |
2353 | kfree(map); | |
2354 | free_extent_map(em); | |
2355 | return -EIO; | |
2356 | } | |
2357 | } | |
2358 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
2359 | } |
2360 | ||
2361 | spin_lock(&map_tree->map_tree.lock); | |
2362 | ret = add_extent_mapping(&map_tree->map_tree, em); | |
0b86a832 | 2363 | spin_unlock(&map_tree->map_tree.lock); |
b248a415 | 2364 | BUG_ON(ret); |
0b86a832 CM |
2365 | free_extent_map(em); |
2366 | ||
2367 | return 0; | |
2368 | } | |
2369 | ||
2370 | static int fill_device_from_item(struct extent_buffer *leaf, | |
2371 | struct btrfs_dev_item *dev_item, | |
2372 | struct btrfs_device *device) | |
2373 | { | |
2374 | unsigned long ptr; | |
0b86a832 CM |
2375 | |
2376 | device->devid = btrfs_device_id(leaf, dev_item); | |
2377 | device->total_bytes = btrfs_device_total_bytes(leaf, dev_item); | |
2378 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); | |
2379 | device->type = btrfs_device_type(leaf, dev_item); | |
2380 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
2381 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
2382 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
2383 | |
2384 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 2385 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 2386 | |
0b86a832 CM |
2387 | return 0; |
2388 | } | |
2389 | ||
0d81ba5d | 2390 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
2391 | struct extent_buffer *leaf, |
2392 | struct btrfs_dev_item *dev_item) | |
2393 | { | |
2394 | struct btrfs_device *device; | |
2395 | u64 devid; | |
2396 | int ret; | |
a443755f CM |
2397 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
2398 | ||
0b86a832 | 2399 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
2400 | read_extent_buffer(leaf, dev_uuid, |
2401 | (unsigned long)btrfs_device_uuid(dev_item), | |
2402 | BTRFS_UUID_SIZE); | |
2403 | device = btrfs_find_device(root, devid, dev_uuid); | |
6324fbf3 | 2404 | if (!device) { |
dfe25020 CM |
2405 | printk("warning devid %Lu missing\n", devid); |
2406 | device = add_missing_dev(root, devid, dev_uuid); | |
6324fbf3 CM |
2407 | if (!device) |
2408 | return -ENOMEM; | |
6324fbf3 | 2409 | } |
0b86a832 CM |
2410 | |
2411 | fill_device_from_item(leaf, dev_item, device); | |
2412 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 2413 | device->in_fs_metadata = 1; |
0b86a832 CM |
2414 | ret = 0; |
2415 | #if 0 | |
2416 | ret = btrfs_open_device(device); | |
2417 | if (ret) { | |
2418 | kfree(device); | |
2419 | } | |
2420 | #endif | |
2421 | return ret; | |
2422 | } | |
2423 | ||
0d81ba5d CM |
2424 | int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf) |
2425 | { | |
2426 | struct btrfs_dev_item *dev_item; | |
2427 | ||
2428 | dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block, | |
2429 | dev_item); | |
2430 | return read_one_dev(root, buf, dev_item); | |
2431 | } | |
2432 | ||
0b86a832 CM |
2433 | int btrfs_read_sys_array(struct btrfs_root *root) |
2434 | { | |
2435 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
a061fc8d | 2436 | struct extent_buffer *sb; |
0b86a832 | 2437 | struct btrfs_disk_key *disk_key; |
0b86a832 | 2438 | struct btrfs_chunk *chunk; |
84eed90f CM |
2439 | u8 *ptr; |
2440 | unsigned long sb_ptr; | |
2441 | int ret = 0; | |
0b86a832 CM |
2442 | u32 num_stripes; |
2443 | u32 array_size; | |
2444 | u32 len = 0; | |
0b86a832 | 2445 | u32 cur; |
84eed90f | 2446 | struct btrfs_key key; |
0b86a832 | 2447 | |
a061fc8d CM |
2448 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
2449 | BTRFS_SUPER_INFO_SIZE); | |
2450 | if (!sb) | |
2451 | return -ENOMEM; | |
2452 | btrfs_set_buffer_uptodate(sb); | |
2453 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); | |
0b86a832 CM |
2454 | array_size = btrfs_super_sys_array_size(super_copy); |
2455 | ||
0b86a832 CM |
2456 | ptr = super_copy->sys_chunk_array; |
2457 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
2458 | cur = 0; | |
2459 | ||
2460 | while (cur < array_size) { | |
2461 | disk_key = (struct btrfs_disk_key *)ptr; | |
2462 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2463 | ||
a061fc8d | 2464 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
2465 | sb_ptr += len; |
2466 | cur += len; | |
2467 | ||
0d81ba5d | 2468 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 2469 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 2470 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
2471 | if (ret) |
2472 | break; | |
0b86a832 CM |
2473 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
2474 | len = btrfs_chunk_item_size(num_stripes); | |
2475 | } else { | |
84eed90f CM |
2476 | ret = -EIO; |
2477 | break; | |
0b86a832 CM |
2478 | } |
2479 | ptr += len; | |
2480 | sb_ptr += len; | |
2481 | cur += len; | |
2482 | } | |
a061fc8d | 2483 | free_extent_buffer(sb); |
84eed90f | 2484 | return ret; |
0b86a832 CM |
2485 | } |
2486 | ||
2487 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
2488 | { | |
2489 | struct btrfs_path *path; | |
2490 | struct extent_buffer *leaf; | |
2491 | struct btrfs_key key; | |
2492 | struct btrfs_key found_key; | |
2493 | int ret; | |
2494 | int slot; | |
2495 | ||
2496 | root = root->fs_info->chunk_root; | |
2497 | ||
2498 | path = btrfs_alloc_path(); | |
2499 | if (!path) | |
2500 | return -ENOMEM; | |
2501 | ||
2502 | /* first we search for all of the device items, and then we | |
2503 | * read in all of the chunk items. This way we can create chunk | |
2504 | * mappings that reference all of the devices that are afound | |
2505 | */ | |
2506 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2507 | key.offset = 0; | |
2508 | key.type = 0; | |
2509 | again: | |
2510 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2511 | while(1) { | |
2512 | leaf = path->nodes[0]; | |
2513 | slot = path->slots[0]; | |
2514 | if (slot >= btrfs_header_nritems(leaf)) { | |
2515 | ret = btrfs_next_leaf(root, path); | |
2516 | if (ret == 0) | |
2517 | continue; | |
2518 | if (ret < 0) | |
2519 | goto error; | |
2520 | break; | |
2521 | } | |
2522 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
2523 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
2524 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
2525 | break; | |
2526 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
2527 | struct btrfs_dev_item *dev_item; | |
2528 | dev_item = btrfs_item_ptr(leaf, slot, | |
2529 | struct btrfs_dev_item); | |
0d81ba5d | 2530 | ret = read_one_dev(root, leaf, dev_item); |
0b86a832 CM |
2531 | BUG_ON(ret); |
2532 | } | |
2533 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2534 | struct btrfs_chunk *chunk; | |
2535 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
2536 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2537 | } | |
2538 | path->slots[0]++; | |
2539 | } | |
2540 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
2541 | key.objectid = 0; | |
2542 | btrfs_release_path(root, path); | |
2543 | goto again; | |
2544 | } | |
2545 | ||
2546 | btrfs_free_path(path); | |
2547 | ret = 0; | |
2548 | error: | |
2549 | return ret; | |
2550 | } |