]>
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> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
593060d7 | 26 | #include <asm/div64.h> |
4b4e25f2 | 27 | #include "compat.h" |
0b86a832 CM |
28 | #include "ctree.h" |
29 | #include "extent_map.h" | |
30 | #include "disk-io.h" | |
31 | #include "transaction.h" | |
32 | #include "print-tree.h" | |
33 | #include "volumes.h" | |
8b712842 | 34 | #include "async-thread.h" |
0b86a832 | 35 | |
593060d7 CM |
36 | struct map_lookup { |
37 | u64 type; | |
38 | int io_align; | |
39 | int io_width; | |
40 | int stripe_len; | |
41 | int sector_size; | |
42 | int num_stripes; | |
321aecc6 | 43 | int sub_stripes; |
cea9e445 | 44 | struct btrfs_bio_stripe stripes[]; |
593060d7 CM |
45 | }; |
46 | ||
2b82032c YZ |
47 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
48 | struct btrfs_root *root, | |
49 | struct btrfs_device *device); | |
50 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
51 | ||
593060d7 | 52 | #define map_lookup_size(n) (sizeof(struct map_lookup) + \ |
cea9e445 | 53 | (sizeof(struct btrfs_bio_stripe) * (n))) |
593060d7 | 54 | |
8a4b83cc CM |
55 | static DEFINE_MUTEX(uuid_mutex); |
56 | static LIST_HEAD(fs_uuids); | |
57 | ||
a061fc8d CM |
58 | void btrfs_lock_volumes(void) |
59 | { | |
60 | mutex_lock(&uuid_mutex); | |
61 | } | |
62 | ||
63 | void btrfs_unlock_volumes(void) | |
64 | { | |
65 | mutex_unlock(&uuid_mutex); | |
66 | } | |
67 | ||
7d9eb12c CM |
68 | static void lock_chunks(struct btrfs_root *root) |
69 | { | |
7d9eb12c CM |
70 | mutex_lock(&root->fs_info->chunk_mutex); |
71 | } | |
72 | ||
73 | static void unlock_chunks(struct btrfs_root *root) | |
74 | { | |
7d9eb12c CM |
75 | mutex_unlock(&root->fs_info->chunk_mutex); |
76 | } | |
77 | ||
e4404d6e YZ |
78 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
79 | { | |
80 | struct btrfs_device *device; | |
81 | WARN_ON(fs_devices->opened); | |
82 | while (!list_empty(&fs_devices->devices)) { | |
83 | device = list_entry(fs_devices->devices.next, | |
84 | struct btrfs_device, dev_list); | |
85 | list_del(&device->dev_list); | |
86 | kfree(device->name); | |
87 | kfree(device); | |
88 | } | |
89 | kfree(fs_devices); | |
90 | } | |
91 | ||
8a4b83cc CM |
92 | int btrfs_cleanup_fs_uuids(void) |
93 | { | |
94 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 95 | |
2b82032c YZ |
96 | while (!list_empty(&fs_uuids)) { |
97 | fs_devices = list_entry(fs_uuids.next, | |
98 | struct btrfs_fs_devices, list); | |
99 | list_del(&fs_devices->list); | |
e4404d6e | 100 | free_fs_devices(fs_devices); |
8a4b83cc CM |
101 | } |
102 | return 0; | |
103 | } | |
104 | ||
a1b32a59 CM |
105 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
106 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
107 | { |
108 | struct btrfs_device *dev; | |
8a4b83cc | 109 | |
c6e30871 | 110 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 111 | if (dev->devid == devid && |
8f18cf13 | 112 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 113 | return dev; |
a443755f | 114 | } |
8a4b83cc CM |
115 | } |
116 | return NULL; | |
117 | } | |
118 | ||
a1b32a59 | 119 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 120 | { |
8a4b83cc CM |
121 | struct btrfs_fs_devices *fs_devices; |
122 | ||
c6e30871 | 123 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
124 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
125 | return fs_devices; | |
126 | } | |
127 | return NULL; | |
128 | } | |
129 | ||
ffbd517d CM |
130 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
131 | struct bio *head, struct bio *tail) | |
132 | { | |
133 | ||
134 | struct bio *old_head; | |
135 | ||
136 | old_head = pending_bios->head; | |
137 | pending_bios->head = head; | |
138 | if (pending_bios->tail) | |
139 | tail->bi_next = old_head; | |
140 | else | |
141 | pending_bios->tail = tail; | |
142 | } | |
143 | ||
8b712842 CM |
144 | /* |
145 | * we try to collect pending bios for a device so we don't get a large | |
146 | * number of procs sending bios down to the same device. This greatly | |
147 | * improves the schedulers ability to collect and merge the bios. | |
148 | * | |
149 | * But, it also turns into a long list of bios to process and that is sure | |
150 | * to eventually make the worker thread block. The solution here is to | |
151 | * make some progress and then put this work struct back at the end of | |
152 | * the list if the block device is congested. This way, multiple devices | |
153 | * can make progress from a single worker thread. | |
154 | */ | |
d397712b | 155 | static noinline int run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
156 | { |
157 | struct bio *pending; | |
158 | struct backing_dev_info *bdi; | |
b64a2851 | 159 | struct btrfs_fs_info *fs_info; |
ffbd517d | 160 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
161 | struct bio *tail; |
162 | struct bio *cur; | |
163 | int again = 0; | |
ffbd517d CM |
164 | unsigned long num_run; |
165 | unsigned long num_sync_run; | |
d644d8a1 | 166 | unsigned long batch_run = 0; |
b64a2851 | 167 | unsigned long limit; |
b765ead5 | 168 | unsigned long last_waited = 0; |
d84275c9 | 169 | int force_reg = 0; |
8b712842 | 170 | |
bedf762b | 171 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
172 | fs_info = device->dev_root->fs_info; |
173 | limit = btrfs_async_submit_limit(fs_info); | |
174 | limit = limit * 2 / 3; | |
175 | ||
ffbd517d CM |
176 | /* we want to make sure that every time we switch from the sync |
177 | * list to the normal list, we unplug | |
178 | */ | |
179 | num_sync_run = 0; | |
180 | ||
8b712842 CM |
181 | loop: |
182 | spin_lock(&device->io_lock); | |
183 | ||
a6837051 | 184 | loop_lock: |
d84275c9 | 185 | num_run = 0; |
ffbd517d | 186 | |
8b712842 CM |
187 | /* take all the bios off the list at once and process them |
188 | * later on (without the lock held). But, remember the | |
189 | * tail and other pointers so the bios can be properly reinserted | |
190 | * into the list if we hit congestion | |
191 | */ | |
d84275c9 | 192 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 193 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
194 | force_reg = 1; |
195 | } else { | |
ffbd517d | 196 | pending_bios = &device->pending_bios; |
d84275c9 CM |
197 | force_reg = 0; |
198 | } | |
ffbd517d CM |
199 | |
200 | pending = pending_bios->head; | |
201 | tail = pending_bios->tail; | |
8b712842 | 202 | WARN_ON(pending && !tail); |
8b712842 CM |
203 | |
204 | /* | |
205 | * if pending was null this time around, no bios need processing | |
206 | * at all and we can stop. Otherwise it'll loop back up again | |
207 | * and do an additional check so no bios are missed. | |
208 | * | |
209 | * device->running_pending is used to synchronize with the | |
210 | * schedule_bio code. | |
211 | */ | |
ffbd517d CM |
212 | if (device->pending_sync_bios.head == NULL && |
213 | device->pending_bios.head == NULL) { | |
8b712842 CM |
214 | again = 0; |
215 | device->running_pending = 0; | |
ffbd517d CM |
216 | } else { |
217 | again = 1; | |
218 | device->running_pending = 1; | |
8b712842 | 219 | } |
ffbd517d CM |
220 | |
221 | pending_bios->head = NULL; | |
222 | pending_bios->tail = NULL; | |
223 | ||
8b712842 CM |
224 | spin_unlock(&device->io_lock); |
225 | ||
ffbd517d CM |
226 | /* |
227 | * if we're doing the regular priority list, make sure we unplug | |
228 | * for any high prio bios we've sent down | |
229 | */ | |
230 | if (pending_bios == &device->pending_bios && num_sync_run > 0) { | |
231 | num_sync_run = 0; | |
232 | blk_run_backing_dev(bdi, NULL); | |
233 | } | |
234 | ||
d397712b | 235 | while (pending) { |
ffbd517d CM |
236 | |
237 | rmb(); | |
d84275c9 CM |
238 | /* we want to work on both lists, but do more bios on the |
239 | * sync list than the regular list | |
240 | */ | |
241 | if ((num_run > 32 && | |
242 | pending_bios != &device->pending_sync_bios && | |
243 | device->pending_sync_bios.head) || | |
244 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
245 | device->pending_bios.head)) { | |
ffbd517d CM |
246 | spin_lock(&device->io_lock); |
247 | requeue_list(pending_bios, pending, tail); | |
248 | goto loop_lock; | |
249 | } | |
250 | ||
8b712842 CM |
251 | cur = pending; |
252 | pending = pending->bi_next; | |
253 | cur->bi_next = NULL; | |
b64a2851 CM |
254 | atomic_dec(&fs_info->nr_async_bios); |
255 | ||
256 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
257 | waitqueue_active(&fs_info->async_submit_wait)) | |
258 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
259 | |
260 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 261 | |
7b6d91da | 262 | if (cur->bi_rw & REQ_SYNC) |
ffbd517d CM |
263 | num_sync_run++; |
264 | ||
5ff7ba3a CM |
265 | submit_bio(cur->bi_rw, cur); |
266 | num_run++; | |
267 | batch_run++; | |
ffbd517d CM |
268 | if (need_resched()) { |
269 | if (num_sync_run) { | |
270 | blk_run_backing_dev(bdi, NULL); | |
271 | num_sync_run = 0; | |
272 | } | |
273 | cond_resched(); | |
274 | } | |
8b712842 CM |
275 | |
276 | /* | |
277 | * we made progress, there is more work to do and the bdi | |
278 | * is now congested. Back off and let other work structs | |
279 | * run instead | |
280 | */ | |
57fd5a5f | 281 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 282 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 283 | struct io_context *ioc; |
8b712842 | 284 | |
b765ead5 CM |
285 | ioc = current->io_context; |
286 | ||
287 | /* | |
288 | * the main goal here is that we don't want to | |
289 | * block if we're going to be able to submit | |
290 | * more requests without blocking. | |
291 | * | |
292 | * This code does two great things, it pokes into | |
293 | * the elevator code from a filesystem _and_ | |
294 | * it makes assumptions about how batching works. | |
295 | */ | |
296 | if (ioc && ioc->nr_batch_requests > 0 && | |
297 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
298 | (last_waited == 0 || | |
299 | ioc->last_waited == last_waited)) { | |
300 | /* | |
301 | * we want to go through our batch of | |
302 | * requests and stop. So, we copy out | |
303 | * the ioc->last_waited time and test | |
304 | * against it before looping | |
305 | */ | |
306 | last_waited = ioc->last_waited; | |
ffbd517d CM |
307 | if (need_resched()) { |
308 | if (num_sync_run) { | |
309 | blk_run_backing_dev(bdi, NULL); | |
310 | num_sync_run = 0; | |
311 | } | |
312 | cond_resched(); | |
313 | } | |
b765ead5 CM |
314 | continue; |
315 | } | |
8b712842 | 316 | spin_lock(&device->io_lock); |
ffbd517d | 317 | requeue_list(pending_bios, pending, tail); |
a6837051 | 318 | device->running_pending = 1; |
8b712842 CM |
319 | |
320 | spin_unlock(&device->io_lock); | |
321 | btrfs_requeue_work(&device->work); | |
322 | goto done; | |
323 | } | |
324 | } | |
ffbd517d CM |
325 | |
326 | if (num_sync_run) { | |
327 | num_sync_run = 0; | |
328 | blk_run_backing_dev(bdi, NULL); | |
329 | } | |
bedf762b CM |
330 | /* |
331 | * IO has already been through a long path to get here. Checksumming, | |
332 | * async helper threads, perhaps compression. We've done a pretty | |
333 | * good job of collecting a batch of IO and should just unplug | |
334 | * the device right away. | |
335 | * | |
336 | * This will help anyone who is waiting on the IO, they might have | |
337 | * already unplugged, but managed to do so before the bio they | |
338 | * cared about found its way down here. | |
339 | */ | |
340 | blk_run_backing_dev(bdi, NULL); | |
51684082 CM |
341 | |
342 | cond_resched(); | |
343 | if (again) | |
344 | goto loop; | |
345 | ||
346 | spin_lock(&device->io_lock); | |
347 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
348 | goto loop_lock; | |
349 | spin_unlock(&device->io_lock); | |
350 | ||
8b712842 CM |
351 | done: |
352 | return 0; | |
353 | } | |
354 | ||
b2950863 | 355 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
356 | { |
357 | struct btrfs_device *device; | |
358 | ||
359 | device = container_of(work, struct btrfs_device, work); | |
360 | run_scheduled_bios(device); | |
361 | } | |
362 | ||
a1b32a59 | 363 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
364 | struct btrfs_super_block *disk_super, |
365 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
366 | { | |
367 | struct btrfs_device *device; | |
368 | struct btrfs_fs_devices *fs_devices; | |
369 | u64 found_transid = btrfs_super_generation(disk_super); | |
3a0524dc | 370 | char *name; |
8a4b83cc CM |
371 | |
372 | fs_devices = find_fsid(disk_super->fsid); | |
373 | if (!fs_devices) { | |
515dc322 | 374 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
375 | if (!fs_devices) |
376 | return -ENOMEM; | |
377 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 378 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
379 | list_add(&fs_devices->list, &fs_uuids); |
380 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
381 | fs_devices->latest_devid = devid; | |
382 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 383 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
384 | device = NULL; |
385 | } else { | |
a443755f CM |
386 | device = __find_device(&fs_devices->devices, devid, |
387 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
388 | } |
389 | if (!device) { | |
2b82032c YZ |
390 | if (fs_devices->opened) |
391 | return -EBUSY; | |
392 | ||
8a4b83cc CM |
393 | device = kzalloc(sizeof(*device), GFP_NOFS); |
394 | if (!device) { | |
395 | /* we can safely leave the fs_devices entry around */ | |
396 | return -ENOMEM; | |
397 | } | |
398 | device->devid = devid; | |
8b712842 | 399 | device->work.func = pending_bios_fn; |
a443755f CM |
400 | memcpy(device->uuid, disk_super->dev_item.uuid, |
401 | BTRFS_UUID_SIZE); | |
f2984462 | 402 | device->barriers = 1; |
b248a415 | 403 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
404 | device->name = kstrdup(path, GFP_NOFS); |
405 | if (!device->name) { | |
406 | kfree(device); | |
407 | return -ENOMEM; | |
408 | } | |
2b82032c | 409 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 CM |
410 | |
411 | mutex_lock(&fs_devices->device_list_mutex); | |
8a4b83cc | 412 | list_add(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
413 | mutex_unlock(&fs_devices->device_list_mutex); |
414 | ||
2b82032c | 415 | device->fs_devices = fs_devices; |
8a4b83cc | 416 | fs_devices->num_devices++; |
cd02dca5 | 417 | } else if (!device->name || strcmp(device->name, path)) { |
3a0524dc TH |
418 | name = kstrdup(path, GFP_NOFS); |
419 | if (!name) | |
420 | return -ENOMEM; | |
421 | kfree(device->name); | |
422 | device->name = name; | |
cd02dca5 CM |
423 | if (device->missing) { |
424 | fs_devices->missing_devices--; | |
425 | device->missing = 0; | |
426 | } | |
8a4b83cc CM |
427 | } |
428 | ||
429 | if (found_transid > fs_devices->latest_trans) { | |
430 | fs_devices->latest_devid = devid; | |
431 | fs_devices->latest_trans = found_transid; | |
432 | } | |
8a4b83cc CM |
433 | *fs_devices_ret = fs_devices; |
434 | return 0; | |
435 | } | |
436 | ||
e4404d6e YZ |
437 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
438 | { | |
439 | struct btrfs_fs_devices *fs_devices; | |
440 | struct btrfs_device *device; | |
441 | struct btrfs_device *orig_dev; | |
442 | ||
443 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
444 | if (!fs_devices) | |
445 | return ERR_PTR(-ENOMEM); | |
446 | ||
447 | INIT_LIST_HEAD(&fs_devices->devices); | |
448 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
449 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 450 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
451 | fs_devices->latest_devid = orig->latest_devid; |
452 | fs_devices->latest_trans = orig->latest_trans; | |
453 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); | |
454 | ||
e5e9a520 | 455 | mutex_lock(&orig->device_list_mutex); |
e4404d6e YZ |
456 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
457 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
458 | if (!device) | |
459 | goto error; | |
460 | ||
461 | device->name = kstrdup(orig_dev->name, GFP_NOFS); | |
fd2696f3 JL |
462 | if (!device->name) { |
463 | kfree(device); | |
e4404d6e | 464 | goto error; |
fd2696f3 | 465 | } |
e4404d6e YZ |
466 | |
467 | device->devid = orig_dev->devid; | |
468 | device->work.func = pending_bios_fn; | |
469 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
470 | device->barriers = 1; | |
471 | spin_lock_init(&device->io_lock); | |
472 | INIT_LIST_HEAD(&device->dev_list); | |
473 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
474 | ||
475 | list_add(&device->dev_list, &fs_devices->devices); | |
476 | device->fs_devices = fs_devices; | |
477 | fs_devices->num_devices++; | |
478 | } | |
e5e9a520 | 479 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
480 | return fs_devices; |
481 | error: | |
e5e9a520 | 482 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
483 | free_fs_devices(fs_devices); |
484 | return ERR_PTR(-ENOMEM); | |
485 | } | |
486 | ||
dfe25020 CM |
487 | int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
488 | { | |
c6e30871 | 489 | struct btrfs_device *device, *next; |
dfe25020 CM |
490 | |
491 | mutex_lock(&uuid_mutex); | |
492 | again: | |
e5e9a520 | 493 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 494 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
2b82032c YZ |
495 | if (device->in_fs_metadata) |
496 | continue; | |
497 | ||
498 | if (device->bdev) { | |
15916de8 | 499 | close_bdev_exclusive(device->bdev, device->mode); |
2b82032c YZ |
500 | device->bdev = NULL; |
501 | fs_devices->open_devices--; | |
502 | } | |
503 | if (device->writeable) { | |
504 | list_del_init(&device->dev_alloc_list); | |
505 | device->writeable = 0; | |
506 | fs_devices->rw_devices--; | |
507 | } | |
e4404d6e YZ |
508 | list_del_init(&device->dev_list); |
509 | fs_devices->num_devices--; | |
510 | kfree(device->name); | |
511 | kfree(device); | |
dfe25020 | 512 | } |
e5e9a520 | 513 | mutex_unlock(&fs_devices->device_list_mutex); |
2b82032c YZ |
514 | |
515 | if (fs_devices->seed) { | |
516 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
517 | goto again; |
518 | } | |
519 | ||
dfe25020 CM |
520 | mutex_unlock(&uuid_mutex); |
521 | return 0; | |
522 | } | |
a0af469b | 523 | |
2b82032c | 524 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 525 | { |
8a4b83cc | 526 | struct btrfs_device *device; |
e4404d6e | 527 | |
2b82032c YZ |
528 | if (--fs_devices->opened > 0) |
529 | return 0; | |
8a4b83cc | 530 | |
c6e30871 | 531 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
8a4b83cc | 532 | if (device->bdev) { |
15916de8 | 533 | close_bdev_exclusive(device->bdev, device->mode); |
a0af469b | 534 | fs_devices->open_devices--; |
8a4b83cc | 535 | } |
2b82032c YZ |
536 | if (device->writeable) { |
537 | list_del_init(&device->dev_alloc_list); | |
538 | fs_devices->rw_devices--; | |
539 | } | |
540 | ||
8a4b83cc | 541 | device->bdev = NULL; |
2b82032c | 542 | device->writeable = 0; |
dfe25020 | 543 | device->in_fs_metadata = 0; |
8a4b83cc | 544 | } |
e4404d6e YZ |
545 | WARN_ON(fs_devices->open_devices); |
546 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
547 | fs_devices->opened = 0; |
548 | fs_devices->seeding = 0; | |
2b82032c | 549 | |
8a4b83cc CM |
550 | return 0; |
551 | } | |
552 | ||
2b82032c YZ |
553 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
554 | { | |
e4404d6e | 555 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
556 | int ret; |
557 | ||
558 | mutex_lock(&uuid_mutex); | |
559 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
560 | if (!fs_devices->opened) { |
561 | seed_devices = fs_devices->seed; | |
562 | fs_devices->seed = NULL; | |
563 | } | |
2b82032c | 564 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
565 | |
566 | while (seed_devices) { | |
567 | fs_devices = seed_devices; | |
568 | seed_devices = fs_devices->seed; | |
569 | __btrfs_close_devices(fs_devices); | |
570 | free_fs_devices(fs_devices); | |
571 | } | |
2b82032c YZ |
572 | return ret; |
573 | } | |
574 | ||
e4404d6e YZ |
575 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
576 | fmode_t flags, void *holder) | |
8a4b83cc CM |
577 | { |
578 | struct block_device *bdev; | |
579 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 580 | struct btrfs_device *device; |
a0af469b CM |
581 | struct block_device *latest_bdev = NULL; |
582 | struct buffer_head *bh; | |
583 | struct btrfs_super_block *disk_super; | |
584 | u64 latest_devid = 0; | |
585 | u64 latest_transid = 0; | |
a0af469b | 586 | u64 devid; |
2b82032c | 587 | int seeding = 1; |
a0af469b | 588 | int ret = 0; |
8a4b83cc | 589 | |
c6e30871 | 590 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
591 | if (device->bdev) |
592 | continue; | |
dfe25020 CM |
593 | if (!device->name) |
594 | continue; | |
595 | ||
15916de8 | 596 | bdev = open_bdev_exclusive(device->name, flags, holder); |
8a4b83cc | 597 | if (IS_ERR(bdev)) { |
d397712b | 598 | printk(KERN_INFO "open %s failed\n", device->name); |
a0af469b | 599 | goto error; |
8a4b83cc | 600 | } |
a061fc8d | 601 | set_blocksize(bdev, 4096); |
a0af469b | 602 | |
a512bbf8 | 603 | bh = btrfs_read_dev_super(bdev); |
20b45077 DY |
604 | if (!bh) { |
605 | ret = -EINVAL; | |
a0af469b | 606 | goto error_close; |
20b45077 | 607 | } |
a0af469b CM |
608 | |
609 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 610 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
611 | if (devid != device->devid) |
612 | goto error_brelse; | |
613 | ||
2b82032c YZ |
614 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
615 | BTRFS_UUID_SIZE)) | |
616 | goto error_brelse; | |
617 | ||
618 | device->generation = btrfs_super_generation(disk_super); | |
619 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 620 | latest_devid = devid; |
2b82032c | 621 | latest_transid = device->generation; |
a0af469b CM |
622 | latest_bdev = bdev; |
623 | } | |
624 | ||
2b82032c YZ |
625 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
626 | device->writeable = 0; | |
627 | } else { | |
628 | device->writeable = !bdev_read_only(bdev); | |
629 | seeding = 0; | |
630 | } | |
631 | ||
8a4b83cc | 632 | device->bdev = bdev; |
dfe25020 | 633 | device->in_fs_metadata = 0; |
15916de8 CM |
634 | device->mode = flags; |
635 | ||
c289811c CM |
636 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
637 | fs_devices->rotating = 1; | |
638 | ||
a0af469b | 639 | fs_devices->open_devices++; |
2b82032c YZ |
640 | if (device->writeable) { |
641 | fs_devices->rw_devices++; | |
642 | list_add(&device->dev_alloc_list, | |
643 | &fs_devices->alloc_list); | |
644 | } | |
a0af469b | 645 | continue; |
a061fc8d | 646 | |
a0af469b CM |
647 | error_brelse: |
648 | brelse(bh); | |
649 | error_close: | |
97288f2c | 650 | close_bdev_exclusive(bdev, FMODE_READ); |
a0af469b CM |
651 | error: |
652 | continue; | |
8a4b83cc | 653 | } |
a0af469b CM |
654 | if (fs_devices->open_devices == 0) { |
655 | ret = -EIO; | |
656 | goto out; | |
657 | } | |
2b82032c YZ |
658 | fs_devices->seeding = seeding; |
659 | fs_devices->opened = 1; | |
a0af469b CM |
660 | fs_devices->latest_bdev = latest_bdev; |
661 | fs_devices->latest_devid = latest_devid; | |
662 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 663 | fs_devices->total_rw_bytes = 0; |
a0af469b | 664 | out: |
2b82032c YZ |
665 | return ret; |
666 | } | |
667 | ||
668 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 669 | fmode_t flags, void *holder) |
2b82032c YZ |
670 | { |
671 | int ret; | |
672 | ||
673 | mutex_lock(&uuid_mutex); | |
674 | if (fs_devices->opened) { | |
e4404d6e YZ |
675 | fs_devices->opened++; |
676 | ret = 0; | |
2b82032c | 677 | } else { |
15916de8 | 678 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 679 | } |
8a4b83cc | 680 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
681 | return ret; |
682 | } | |
683 | ||
97288f2c | 684 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
685 | struct btrfs_fs_devices **fs_devices_ret) |
686 | { | |
687 | struct btrfs_super_block *disk_super; | |
688 | struct block_device *bdev; | |
689 | struct buffer_head *bh; | |
690 | int ret; | |
691 | u64 devid; | |
f2984462 | 692 | u64 transid; |
8a4b83cc CM |
693 | |
694 | mutex_lock(&uuid_mutex); | |
695 | ||
15916de8 | 696 | bdev = open_bdev_exclusive(path, flags, holder); |
8a4b83cc CM |
697 | |
698 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
699 | ret = PTR_ERR(bdev); |
700 | goto error; | |
701 | } | |
702 | ||
703 | ret = set_blocksize(bdev, 4096); | |
704 | if (ret) | |
705 | goto error_close; | |
a512bbf8 | 706 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 707 | if (!bh) { |
20b45077 | 708 | ret = -EINVAL; |
8a4b83cc CM |
709 | goto error_close; |
710 | } | |
711 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 712 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 713 | transid = btrfs_super_generation(disk_super); |
7ae9c09d | 714 | if (disk_super->label[0]) |
d397712b | 715 | printk(KERN_INFO "device label %s ", disk_super->label); |
7ae9c09d CM |
716 | else { |
717 | /* FIXME, make a readl uuid parser */ | |
d397712b | 718 | printk(KERN_INFO "device fsid %llx-%llx ", |
7ae9c09d CM |
719 | *(unsigned long long *)disk_super->fsid, |
720 | *(unsigned long long *)(disk_super->fsid + 8)); | |
721 | } | |
119e10cf | 722 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 723 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc CM |
724 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
725 | ||
8a4b83cc CM |
726 | brelse(bh); |
727 | error_close: | |
15916de8 | 728 | close_bdev_exclusive(bdev, flags); |
8a4b83cc CM |
729 | error: |
730 | mutex_unlock(&uuid_mutex); | |
731 | return ret; | |
732 | } | |
0b86a832 | 733 | |
6d07bcec MX |
734 | /* helper to account the used device space in the range */ |
735 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
736 | u64 end, u64 *length) | |
737 | { | |
738 | struct btrfs_key key; | |
739 | struct btrfs_root *root = device->dev_root; | |
740 | struct btrfs_dev_extent *dev_extent; | |
741 | struct btrfs_path *path; | |
742 | u64 extent_end; | |
743 | int ret; | |
744 | int slot; | |
745 | struct extent_buffer *l; | |
746 | ||
747 | *length = 0; | |
748 | ||
749 | if (start >= device->total_bytes) | |
750 | return 0; | |
751 | ||
752 | path = btrfs_alloc_path(); | |
753 | if (!path) | |
754 | return -ENOMEM; | |
755 | path->reada = 2; | |
756 | ||
757 | key.objectid = device->devid; | |
758 | key.offset = start; | |
759 | key.type = BTRFS_DEV_EXTENT_KEY; | |
760 | ||
761 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
762 | if (ret < 0) | |
763 | goto out; | |
764 | if (ret > 0) { | |
765 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
766 | if (ret < 0) | |
767 | goto out; | |
768 | } | |
769 | ||
770 | while (1) { | |
771 | l = path->nodes[0]; | |
772 | slot = path->slots[0]; | |
773 | if (slot >= btrfs_header_nritems(l)) { | |
774 | ret = btrfs_next_leaf(root, path); | |
775 | if (ret == 0) | |
776 | continue; | |
777 | if (ret < 0) | |
778 | goto out; | |
779 | ||
780 | break; | |
781 | } | |
782 | btrfs_item_key_to_cpu(l, &key, slot); | |
783 | ||
784 | if (key.objectid < device->devid) | |
785 | goto next; | |
786 | ||
787 | if (key.objectid > device->devid) | |
788 | break; | |
789 | ||
790 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
791 | goto next; | |
792 | ||
793 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
794 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
795 | dev_extent); | |
796 | if (key.offset <= start && extent_end > end) { | |
797 | *length = end - start + 1; | |
798 | break; | |
799 | } else if (key.offset <= start && extent_end > start) | |
800 | *length += extent_end - start; | |
801 | else if (key.offset > start && extent_end <= end) | |
802 | *length += extent_end - key.offset; | |
803 | else if (key.offset > start && key.offset <= end) { | |
804 | *length += end - key.offset + 1; | |
805 | break; | |
806 | } else if (key.offset > end) | |
807 | break; | |
808 | ||
809 | next: | |
810 | path->slots[0]++; | |
811 | } | |
812 | ret = 0; | |
813 | out: | |
814 | btrfs_free_path(path); | |
815 | return ret; | |
816 | } | |
817 | ||
0b86a832 | 818 | /* |
7bfc837d MX |
819 | * find_free_dev_extent - find free space in the specified device |
820 | * @trans: transaction handler | |
821 | * @device: the device which we search the free space in | |
822 | * @num_bytes: the size of the free space that we need | |
823 | * @start: store the start of the free space. | |
824 | * @len: the size of the free space. that we find, or the size of the max | |
825 | * free space if we don't find suitable free space | |
826 | * | |
0b86a832 CM |
827 | * this uses a pretty simple search, the expectation is that it is |
828 | * called very infrequently and that a given device has a small number | |
829 | * of extents | |
7bfc837d MX |
830 | * |
831 | * @start is used to store the start of the free space if we find. But if we | |
832 | * don't find suitable free space, it will be used to store the start position | |
833 | * of the max free space. | |
834 | * | |
835 | * @len is used to store the size of the free space that we find. | |
836 | * But if we don't find suitable free space, it is used to store the size of | |
837 | * the max free space. | |
0b86a832 | 838 | */ |
ba1bf481 JB |
839 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
840 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 841 | u64 *start, u64 *len) |
0b86a832 CM |
842 | { |
843 | struct btrfs_key key; | |
844 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 845 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 846 | struct btrfs_path *path; |
7bfc837d MX |
847 | u64 hole_size; |
848 | u64 max_hole_start; | |
849 | u64 max_hole_size; | |
850 | u64 extent_end; | |
851 | u64 search_start; | |
0b86a832 CM |
852 | u64 search_end = device->total_bytes; |
853 | int ret; | |
7bfc837d | 854 | int slot; |
0b86a832 CM |
855 | struct extent_buffer *l; |
856 | ||
0b86a832 CM |
857 | /* FIXME use last free of some kind */ |
858 | ||
8a4b83cc CM |
859 | /* we don't want to overwrite the superblock on the drive, |
860 | * so we make sure to start at an offset of at least 1MB | |
861 | */ | |
7bfc837d | 862 | search_start = 1024 * 1024; |
8f18cf13 | 863 | |
7bfc837d | 864 | if (root->fs_info->alloc_start + num_bytes <= search_end) |
8f18cf13 CM |
865 | search_start = max(root->fs_info->alloc_start, search_start); |
866 | ||
7bfc837d MX |
867 | max_hole_start = search_start; |
868 | max_hole_size = 0; | |
869 | ||
870 | if (search_start >= search_end) { | |
871 | ret = -ENOSPC; | |
872 | goto error; | |
873 | } | |
874 | ||
875 | path = btrfs_alloc_path(); | |
876 | if (!path) { | |
877 | ret = -ENOMEM; | |
878 | goto error; | |
879 | } | |
880 | path->reada = 2; | |
881 | ||
0b86a832 CM |
882 | key.objectid = device->devid; |
883 | key.offset = search_start; | |
884 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 885 | |
0b86a832 CM |
886 | ret = btrfs_search_slot(trans, root, &key, path, 0, 0); |
887 | if (ret < 0) | |
7bfc837d | 888 | goto out; |
1fcbac58 YZ |
889 | if (ret > 0) { |
890 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
891 | if (ret < 0) | |
7bfc837d | 892 | goto out; |
1fcbac58 | 893 | } |
7bfc837d | 894 | |
0b86a832 CM |
895 | while (1) { |
896 | l = path->nodes[0]; | |
897 | slot = path->slots[0]; | |
898 | if (slot >= btrfs_header_nritems(l)) { | |
899 | ret = btrfs_next_leaf(root, path); | |
900 | if (ret == 0) | |
901 | continue; | |
902 | if (ret < 0) | |
7bfc837d MX |
903 | goto out; |
904 | ||
905 | break; | |
0b86a832 CM |
906 | } |
907 | btrfs_item_key_to_cpu(l, &key, slot); | |
908 | ||
909 | if (key.objectid < device->devid) | |
910 | goto next; | |
911 | ||
912 | if (key.objectid > device->devid) | |
7bfc837d | 913 | break; |
0b86a832 | 914 | |
7bfc837d MX |
915 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
916 | goto next; | |
9779b72f | 917 | |
7bfc837d MX |
918 | if (key.offset > search_start) { |
919 | hole_size = key.offset - search_start; | |
9779b72f | 920 | |
7bfc837d MX |
921 | if (hole_size > max_hole_size) { |
922 | max_hole_start = search_start; | |
923 | max_hole_size = hole_size; | |
924 | } | |
925 | ||
926 | /* | |
927 | * If this free space is greater than which we need, | |
928 | * it must be the max free space that we have found | |
929 | * until now, so max_hole_start must point to the start | |
930 | * of this free space and the length of this free space | |
931 | * is stored in max_hole_size. Thus, we return | |
932 | * max_hole_start and max_hole_size and go back to the | |
933 | * caller. | |
934 | */ | |
935 | if (hole_size >= num_bytes) { | |
936 | ret = 0; | |
937 | goto out; | |
0b86a832 CM |
938 | } |
939 | } | |
0b86a832 | 940 | |
0b86a832 | 941 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
942 | extent_end = key.offset + btrfs_dev_extent_length(l, |
943 | dev_extent); | |
944 | if (extent_end > search_start) | |
945 | search_start = extent_end; | |
0b86a832 CM |
946 | next: |
947 | path->slots[0]++; | |
948 | cond_resched(); | |
949 | } | |
0b86a832 | 950 | |
7bfc837d MX |
951 | hole_size = search_end- search_start; |
952 | if (hole_size > max_hole_size) { | |
953 | max_hole_start = search_start; | |
954 | max_hole_size = hole_size; | |
0b86a832 | 955 | } |
0b86a832 | 956 | |
7bfc837d MX |
957 | /* See above. */ |
958 | if (hole_size < num_bytes) | |
959 | ret = -ENOSPC; | |
960 | else | |
961 | ret = 0; | |
962 | ||
963 | out: | |
2b82032c | 964 | btrfs_free_path(path); |
7bfc837d MX |
965 | error: |
966 | *start = max_hole_start; | |
b2117a39 | 967 | if (len) |
7bfc837d | 968 | *len = max_hole_size; |
0b86a832 CM |
969 | return ret; |
970 | } | |
971 | ||
b2950863 | 972 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
973 | struct btrfs_device *device, |
974 | u64 start) | |
975 | { | |
976 | int ret; | |
977 | struct btrfs_path *path; | |
978 | struct btrfs_root *root = device->dev_root; | |
979 | struct btrfs_key key; | |
a061fc8d CM |
980 | struct btrfs_key found_key; |
981 | struct extent_buffer *leaf = NULL; | |
982 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
983 | |
984 | path = btrfs_alloc_path(); | |
985 | if (!path) | |
986 | return -ENOMEM; | |
987 | ||
988 | key.objectid = device->devid; | |
989 | key.offset = start; | |
990 | key.type = BTRFS_DEV_EXTENT_KEY; | |
991 | ||
992 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
a061fc8d CM |
993 | if (ret > 0) { |
994 | ret = btrfs_previous_item(root, path, key.objectid, | |
995 | BTRFS_DEV_EXTENT_KEY); | |
996 | BUG_ON(ret); | |
997 | leaf = path->nodes[0]; | |
998 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
999 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1000 | struct btrfs_dev_extent); | |
1001 | BUG_ON(found_key.offset > start || found_key.offset + | |
1002 | btrfs_dev_extent_length(leaf, extent) < start); | |
1003 | ret = 0; | |
1004 | } else if (ret == 0) { | |
1005 | leaf = path->nodes[0]; | |
1006 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1007 | struct btrfs_dev_extent); | |
1008 | } | |
8f18cf13 CM |
1009 | BUG_ON(ret); |
1010 | ||
dfe25020 CM |
1011 | if (device->bytes_used > 0) |
1012 | device->bytes_used -= btrfs_dev_extent_length(leaf, extent); | |
8f18cf13 CM |
1013 | ret = btrfs_del_item(trans, root, path); |
1014 | BUG_ON(ret); | |
1015 | ||
1016 | btrfs_free_path(path); | |
1017 | return ret; | |
1018 | } | |
1019 | ||
2b82032c | 1020 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1021 | struct btrfs_device *device, |
e17cade2 | 1022 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1023 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1024 | { |
1025 | int ret; | |
1026 | struct btrfs_path *path; | |
1027 | struct btrfs_root *root = device->dev_root; | |
1028 | struct btrfs_dev_extent *extent; | |
1029 | struct extent_buffer *leaf; | |
1030 | struct btrfs_key key; | |
1031 | ||
dfe25020 | 1032 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1033 | path = btrfs_alloc_path(); |
1034 | if (!path) | |
1035 | return -ENOMEM; | |
1036 | ||
0b86a832 | 1037 | key.objectid = device->devid; |
2b82032c | 1038 | key.offset = start; |
0b86a832 CM |
1039 | key.type = BTRFS_DEV_EXTENT_KEY; |
1040 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1041 | sizeof(*extent)); | |
1042 | BUG_ON(ret); | |
1043 | ||
1044 | leaf = path->nodes[0]; | |
1045 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1046 | struct btrfs_dev_extent); | |
e17cade2 CM |
1047 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1048 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1049 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1050 | ||
1051 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1052 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1053 | BTRFS_UUID_SIZE); | |
1054 | ||
0b86a832 CM |
1055 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1056 | btrfs_mark_buffer_dirty(leaf); | |
0b86a832 CM |
1057 | btrfs_free_path(path); |
1058 | return ret; | |
1059 | } | |
1060 | ||
a1b32a59 CM |
1061 | static noinline int find_next_chunk(struct btrfs_root *root, |
1062 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1063 | { |
1064 | struct btrfs_path *path; | |
1065 | int ret; | |
1066 | struct btrfs_key key; | |
e17cade2 | 1067 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1068 | struct btrfs_key found_key; |
1069 | ||
1070 | path = btrfs_alloc_path(); | |
1071 | BUG_ON(!path); | |
1072 | ||
e17cade2 | 1073 | key.objectid = objectid; |
0b86a832 CM |
1074 | key.offset = (u64)-1; |
1075 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1076 | ||
1077 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1078 | if (ret < 0) | |
1079 | goto error; | |
1080 | ||
1081 | BUG_ON(ret == 0); | |
1082 | ||
1083 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1084 | if (ret) { | |
e17cade2 | 1085 | *offset = 0; |
0b86a832 CM |
1086 | } else { |
1087 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1088 | path->slots[0]); | |
e17cade2 CM |
1089 | if (found_key.objectid != objectid) |
1090 | *offset = 0; | |
1091 | else { | |
1092 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1093 | struct btrfs_chunk); | |
1094 | *offset = found_key.offset + | |
1095 | btrfs_chunk_length(path->nodes[0], chunk); | |
1096 | } | |
0b86a832 CM |
1097 | } |
1098 | ret = 0; | |
1099 | error: | |
1100 | btrfs_free_path(path); | |
1101 | return ret; | |
1102 | } | |
1103 | ||
2b82032c | 1104 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1105 | { |
1106 | int ret; | |
1107 | struct btrfs_key key; | |
1108 | struct btrfs_key found_key; | |
2b82032c YZ |
1109 | struct btrfs_path *path; |
1110 | ||
1111 | root = root->fs_info->chunk_root; | |
1112 | ||
1113 | path = btrfs_alloc_path(); | |
1114 | if (!path) | |
1115 | return -ENOMEM; | |
0b86a832 CM |
1116 | |
1117 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1118 | key.type = BTRFS_DEV_ITEM_KEY; | |
1119 | key.offset = (u64)-1; | |
1120 | ||
1121 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1122 | if (ret < 0) | |
1123 | goto error; | |
1124 | ||
1125 | BUG_ON(ret == 0); | |
1126 | ||
1127 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1128 | BTRFS_DEV_ITEM_KEY); | |
1129 | if (ret) { | |
1130 | *objectid = 1; | |
1131 | } else { | |
1132 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1133 | path->slots[0]); | |
1134 | *objectid = found_key.offset + 1; | |
1135 | } | |
1136 | ret = 0; | |
1137 | error: | |
2b82032c | 1138 | btrfs_free_path(path); |
0b86a832 CM |
1139 | return ret; |
1140 | } | |
1141 | ||
1142 | /* | |
1143 | * the device information is stored in the chunk root | |
1144 | * the btrfs_device struct should be fully filled in | |
1145 | */ | |
1146 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1147 | struct btrfs_root *root, | |
1148 | struct btrfs_device *device) | |
1149 | { | |
1150 | int ret; | |
1151 | struct btrfs_path *path; | |
1152 | struct btrfs_dev_item *dev_item; | |
1153 | struct extent_buffer *leaf; | |
1154 | struct btrfs_key key; | |
1155 | unsigned long ptr; | |
0b86a832 CM |
1156 | |
1157 | root = root->fs_info->chunk_root; | |
1158 | ||
1159 | path = btrfs_alloc_path(); | |
1160 | if (!path) | |
1161 | return -ENOMEM; | |
1162 | ||
0b86a832 CM |
1163 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1164 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1165 | key.offset = device->devid; |
0b86a832 CM |
1166 | |
1167 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1168 | sizeof(*dev_item)); |
0b86a832 CM |
1169 | if (ret) |
1170 | goto out; | |
1171 | ||
1172 | leaf = path->nodes[0]; | |
1173 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1174 | ||
1175 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1176 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1177 | btrfs_set_device_type(leaf, dev_item, device->type); |
1178 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1179 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1180 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1181 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1182 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1183 | btrfs_set_device_group(leaf, dev_item, 0); |
1184 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1185 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1186 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1187 | |
0b86a832 | 1188 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1189 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1190 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1191 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1192 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1193 | |
2b82032c | 1194 | ret = 0; |
0b86a832 CM |
1195 | out: |
1196 | btrfs_free_path(path); | |
1197 | return ret; | |
1198 | } | |
8f18cf13 | 1199 | |
a061fc8d CM |
1200 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1201 | struct btrfs_device *device) | |
1202 | { | |
1203 | int ret; | |
1204 | struct btrfs_path *path; | |
a061fc8d | 1205 | struct btrfs_key key; |
a061fc8d CM |
1206 | struct btrfs_trans_handle *trans; |
1207 | ||
1208 | root = root->fs_info->chunk_root; | |
1209 | ||
1210 | path = btrfs_alloc_path(); | |
1211 | if (!path) | |
1212 | return -ENOMEM; | |
1213 | ||
a22285a6 | 1214 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1215 | if (IS_ERR(trans)) { |
1216 | btrfs_free_path(path); | |
1217 | return PTR_ERR(trans); | |
1218 | } | |
a061fc8d CM |
1219 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1220 | key.type = BTRFS_DEV_ITEM_KEY; | |
1221 | key.offset = device->devid; | |
7d9eb12c | 1222 | lock_chunks(root); |
a061fc8d CM |
1223 | |
1224 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1225 | if (ret < 0) | |
1226 | goto out; | |
1227 | ||
1228 | if (ret > 0) { | |
1229 | ret = -ENOENT; | |
1230 | goto out; | |
1231 | } | |
1232 | ||
1233 | ret = btrfs_del_item(trans, root, path); | |
1234 | if (ret) | |
1235 | goto out; | |
a061fc8d CM |
1236 | out: |
1237 | btrfs_free_path(path); | |
7d9eb12c | 1238 | unlock_chunks(root); |
a061fc8d CM |
1239 | btrfs_commit_transaction(trans, root); |
1240 | return ret; | |
1241 | } | |
1242 | ||
1243 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1244 | { | |
1245 | struct btrfs_device *device; | |
2b82032c | 1246 | struct btrfs_device *next_device; |
a061fc8d | 1247 | struct block_device *bdev; |
dfe25020 | 1248 | struct buffer_head *bh = NULL; |
a061fc8d CM |
1249 | struct btrfs_super_block *disk_super; |
1250 | u64 all_avail; | |
1251 | u64 devid; | |
2b82032c YZ |
1252 | u64 num_devices; |
1253 | u8 *dev_uuid; | |
a061fc8d CM |
1254 | int ret = 0; |
1255 | ||
a061fc8d | 1256 | mutex_lock(&uuid_mutex); |
7d9eb12c | 1257 | mutex_lock(&root->fs_info->volume_mutex); |
a061fc8d CM |
1258 | |
1259 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1260 | root->fs_info->avail_system_alloc_bits | | |
1261 | root->fs_info->avail_metadata_alloc_bits; | |
1262 | ||
1263 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1264 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1265 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1266 | "on raid10\n"); | |
a061fc8d CM |
1267 | ret = -EINVAL; |
1268 | goto out; | |
1269 | } | |
1270 | ||
1271 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1272 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1273 | printk(KERN_ERR "btrfs: unable to go below two " |
1274 | "devices on raid1\n"); | |
a061fc8d CM |
1275 | ret = -EINVAL; |
1276 | goto out; | |
1277 | } | |
1278 | ||
dfe25020 | 1279 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1280 | struct list_head *devices; |
1281 | struct btrfs_device *tmp; | |
a061fc8d | 1282 | |
dfe25020 CM |
1283 | device = NULL; |
1284 | devices = &root->fs_info->fs_devices->devices; | |
e5e9a520 | 1285 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
c6e30871 | 1286 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1287 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1288 | device = tmp; | |
1289 | break; | |
1290 | } | |
1291 | } | |
e5e9a520 | 1292 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
dfe25020 CM |
1293 | bdev = NULL; |
1294 | bh = NULL; | |
1295 | disk_super = NULL; | |
1296 | if (!device) { | |
d397712b CM |
1297 | printk(KERN_ERR "btrfs: no missing devices found to " |
1298 | "remove\n"); | |
dfe25020 CM |
1299 | goto out; |
1300 | } | |
dfe25020 | 1301 | } else { |
97288f2c | 1302 | bdev = open_bdev_exclusive(device_path, FMODE_READ, |
dfe25020 CM |
1303 | root->fs_info->bdev_holder); |
1304 | if (IS_ERR(bdev)) { | |
1305 | ret = PTR_ERR(bdev); | |
1306 | goto out; | |
1307 | } | |
a061fc8d | 1308 | |
2b82032c | 1309 | set_blocksize(bdev, 4096); |
a512bbf8 | 1310 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1311 | if (!bh) { |
20b45077 | 1312 | ret = -EINVAL; |
dfe25020 CM |
1313 | goto error_close; |
1314 | } | |
1315 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1316 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1317 | dev_uuid = disk_super->dev_item.uuid; |
1318 | device = btrfs_find_device(root, devid, dev_uuid, | |
1319 | disk_super->fsid); | |
dfe25020 CM |
1320 | if (!device) { |
1321 | ret = -ENOENT; | |
1322 | goto error_brelse; | |
1323 | } | |
2b82032c | 1324 | } |
dfe25020 | 1325 | |
2b82032c | 1326 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1327 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1328 | "device\n"); | |
2b82032c YZ |
1329 | ret = -EINVAL; |
1330 | goto error_brelse; | |
1331 | } | |
1332 | ||
1333 | if (device->writeable) { | |
1334 | list_del_init(&device->dev_alloc_list); | |
1335 | root->fs_info->fs_devices->rw_devices--; | |
dfe25020 | 1336 | } |
a061fc8d CM |
1337 | |
1338 | ret = btrfs_shrink_device(device, 0); | |
1339 | if (ret) | |
1340 | goto error_brelse; | |
1341 | ||
a061fc8d CM |
1342 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1343 | if (ret) | |
1344 | goto error_brelse; | |
1345 | ||
2b82032c | 1346 | device->in_fs_metadata = 0; |
e5e9a520 CM |
1347 | |
1348 | /* | |
1349 | * the device list mutex makes sure that we don't change | |
1350 | * the device list while someone else is writing out all | |
1351 | * the device supers. | |
1352 | */ | |
1353 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1354 | list_del_init(&device->dev_list); |
e5e9a520 CM |
1355 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1356 | ||
e4404d6e | 1357 | device->fs_devices->num_devices--; |
2b82032c | 1358 | |
cd02dca5 CM |
1359 | if (device->missing) |
1360 | root->fs_info->fs_devices->missing_devices--; | |
1361 | ||
2b82032c YZ |
1362 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1363 | struct btrfs_device, dev_list); | |
1364 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1365 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1366 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1367 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1368 | ||
e4404d6e YZ |
1369 | if (device->bdev) { |
1370 | close_bdev_exclusive(device->bdev, device->mode); | |
1371 | device->bdev = NULL; | |
1372 | device->fs_devices->open_devices--; | |
1373 | } | |
1374 | ||
2b82032c YZ |
1375 | num_devices = btrfs_super_num_devices(&root->fs_info->super_copy) - 1; |
1376 | btrfs_set_super_num_devices(&root->fs_info->super_copy, num_devices); | |
1377 | ||
e4404d6e YZ |
1378 | if (device->fs_devices->open_devices == 0) { |
1379 | struct btrfs_fs_devices *fs_devices; | |
1380 | fs_devices = root->fs_info->fs_devices; | |
1381 | while (fs_devices) { | |
1382 | if (fs_devices->seed == device->fs_devices) | |
1383 | break; | |
1384 | fs_devices = fs_devices->seed; | |
2b82032c | 1385 | } |
e4404d6e YZ |
1386 | fs_devices->seed = device->fs_devices->seed; |
1387 | device->fs_devices->seed = NULL; | |
1388 | __btrfs_close_devices(device->fs_devices); | |
1389 | free_fs_devices(device->fs_devices); | |
2b82032c YZ |
1390 | } |
1391 | ||
1392 | /* | |
1393 | * at this point, the device is zero sized. We want to | |
1394 | * remove it from the devices list and zero out the old super | |
1395 | */ | |
1396 | if (device->writeable) { | |
dfe25020 CM |
1397 | /* make sure this device isn't detected as part of |
1398 | * the FS anymore | |
1399 | */ | |
1400 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1401 | set_buffer_dirty(bh); | |
1402 | sync_dirty_buffer(bh); | |
dfe25020 | 1403 | } |
a061fc8d CM |
1404 | |
1405 | kfree(device->name); | |
1406 | kfree(device); | |
1407 | ret = 0; | |
a061fc8d CM |
1408 | |
1409 | error_brelse: | |
1410 | brelse(bh); | |
1411 | error_close: | |
dfe25020 | 1412 | if (bdev) |
97288f2c | 1413 | close_bdev_exclusive(bdev, FMODE_READ); |
a061fc8d | 1414 | out: |
7d9eb12c | 1415 | mutex_unlock(&root->fs_info->volume_mutex); |
a061fc8d | 1416 | mutex_unlock(&uuid_mutex); |
a061fc8d CM |
1417 | return ret; |
1418 | } | |
1419 | ||
2b82032c YZ |
1420 | /* |
1421 | * does all the dirty work required for changing file system's UUID. | |
1422 | */ | |
1423 | static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans, | |
1424 | struct btrfs_root *root) | |
1425 | { | |
1426 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1427 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1428 | struct btrfs_fs_devices *seed_devices; |
2b82032c YZ |
1429 | struct btrfs_super_block *disk_super = &root->fs_info->super_copy; |
1430 | struct btrfs_device *device; | |
1431 | u64 super_flags; | |
1432 | ||
1433 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1434 | if (!fs_devices->seeding) |
2b82032c YZ |
1435 | return -EINVAL; |
1436 | ||
e4404d6e YZ |
1437 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1438 | if (!seed_devices) | |
2b82032c YZ |
1439 | return -ENOMEM; |
1440 | ||
e4404d6e YZ |
1441 | old_devices = clone_fs_devices(fs_devices); |
1442 | if (IS_ERR(old_devices)) { | |
1443 | kfree(seed_devices); | |
1444 | return PTR_ERR(old_devices); | |
2b82032c | 1445 | } |
e4404d6e | 1446 | |
2b82032c YZ |
1447 | list_add(&old_devices->list, &fs_uuids); |
1448 | ||
e4404d6e YZ |
1449 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1450 | seed_devices->opened = 1; | |
1451 | INIT_LIST_HEAD(&seed_devices->devices); | |
1452 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1453 | mutex_init(&seed_devices->device_list_mutex); |
e4404d6e YZ |
1454 | list_splice_init(&fs_devices->devices, &seed_devices->devices); |
1455 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); | |
1456 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1457 | device->fs_devices = seed_devices; | |
1458 | } | |
1459 | ||
2b82032c YZ |
1460 | fs_devices->seeding = 0; |
1461 | fs_devices->num_devices = 0; | |
1462 | fs_devices->open_devices = 0; | |
e4404d6e | 1463 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1464 | |
1465 | generate_random_uuid(fs_devices->fsid); | |
1466 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1467 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1468 | super_flags = btrfs_super_flags(disk_super) & | |
1469 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1470 | btrfs_set_super_flags(disk_super, super_flags); | |
1471 | ||
1472 | return 0; | |
1473 | } | |
1474 | ||
1475 | /* | |
1476 | * strore the expected generation for seed devices in device items. | |
1477 | */ | |
1478 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1479 | struct btrfs_root *root) | |
1480 | { | |
1481 | struct btrfs_path *path; | |
1482 | struct extent_buffer *leaf; | |
1483 | struct btrfs_dev_item *dev_item; | |
1484 | struct btrfs_device *device; | |
1485 | struct btrfs_key key; | |
1486 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1487 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1488 | u64 devid; | |
1489 | int ret; | |
1490 | ||
1491 | path = btrfs_alloc_path(); | |
1492 | if (!path) | |
1493 | return -ENOMEM; | |
1494 | ||
1495 | root = root->fs_info->chunk_root; | |
1496 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1497 | key.offset = 0; | |
1498 | key.type = BTRFS_DEV_ITEM_KEY; | |
1499 | ||
1500 | while (1) { | |
1501 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1502 | if (ret < 0) | |
1503 | goto error; | |
1504 | ||
1505 | leaf = path->nodes[0]; | |
1506 | next_slot: | |
1507 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1508 | ret = btrfs_next_leaf(root, path); | |
1509 | if (ret > 0) | |
1510 | break; | |
1511 | if (ret < 0) | |
1512 | goto error; | |
1513 | leaf = path->nodes[0]; | |
1514 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1515 | btrfs_release_path(root, path); | |
1516 | continue; | |
1517 | } | |
1518 | ||
1519 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1520 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1521 | key.type != BTRFS_DEV_ITEM_KEY) | |
1522 | break; | |
1523 | ||
1524 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1525 | struct btrfs_dev_item); | |
1526 | devid = btrfs_device_id(leaf, dev_item); | |
1527 | read_extent_buffer(leaf, dev_uuid, | |
1528 | (unsigned long)btrfs_device_uuid(dev_item), | |
1529 | BTRFS_UUID_SIZE); | |
1530 | read_extent_buffer(leaf, fs_uuid, | |
1531 | (unsigned long)btrfs_device_fsid(dev_item), | |
1532 | BTRFS_UUID_SIZE); | |
1533 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
1534 | BUG_ON(!device); | |
1535 | ||
1536 | if (device->fs_devices->seeding) { | |
1537 | btrfs_set_device_generation(leaf, dev_item, | |
1538 | device->generation); | |
1539 | btrfs_mark_buffer_dirty(leaf); | |
1540 | } | |
1541 | ||
1542 | path->slots[0]++; | |
1543 | goto next_slot; | |
1544 | } | |
1545 | ret = 0; | |
1546 | error: | |
1547 | btrfs_free_path(path); | |
1548 | return ret; | |
1549 | } | |
1550 | ||
788f20eb CM |
1551 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1552 | { | |
1553 | struct btrfs_trans_handle *trans; | |
1554 | struct btrfs_device *device; | |
1555 | struct block_device *bdev; | |
788f20eb | 1556 | struct list_head *devices; |
2b82032c | 1557 | struct super_block *sb = root->fs_info->sb; |
788f20eb | 1558 | u64 total_bytes; |
2b82032c | 1559 | int seeding_dev = 0; |
788f20eb CM |
1560 | int ret = 0; |
1561 | ||
2b82032c YZ |
1562 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
1563 | return -EINVAL; | |
788f20eb | 1564 | |
15916de8 | 1565 | bdev = open_bdev_exclusive(device_path, 0, root->fs_info->bdev_holder); |
7f59203a JB |
1566 | if (IS_ERR(bdev)) |
1567 | return PTR_ERR(bdev); | |
a2135011 | 1568 | |
2b82032c YZ |
1569 | if (root->fs_info->fs_devices->seeding) { |
1570 | seeding_dev = 1; | |
1571 | down_write(&sb->s_umount); | |
1572 | mutex_lock(&uuid_mutex); | |
1573 | } | |
1574 | ||
8c8bee1d | 1575 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
7d9eb12c | 1576 | mutex_lock(&root->fs_info->volume_mutex); |
a2135011 | 1577 | |
788f20eb | 1578 | devices = &root->fs_info->fs_devices->devices; |
e5e9a520 CM |
1579 | /* |
1580 | * we have the volume lock, so we don't need the extra | |
1581 | * device list mutex while reading the list here. | |
1582 | */ | |
c6e30871 | 1583 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1584 | if (device->bdev == bdev) { |
1585 | ret = -EEXIST; | |
2b82032c | 1586 | goto error; |
788f20eb CM |
1587 | } |
1588 | } | |
1589 | ||
1590 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1591 | if (!device) { | |
1592 | /* we can safely leave the fs_devices entry around */ | |
1593 | ret = -ENOMEM; | |
2b82032c | 1594 | goto error; |
788f20eb CM |
1595 | } |
1596 | ||
788f20eb CM |
1597 | device->name = kstrdup(device_path, GFP_NOFS); |
1598 | if (!device->name) { | |
1599 | kfree(device); | |
2b82032c YZ |
1600 | ret = -ENOMEM; |
1601 | goto error; | |
788f20eb | 1602 | } |
2b82032c YZ |
1603 | |
1604 | ret = find_next_devid(root, &device->devid); | |
1605 | if (ret) { | |
67100f25 | 1606 | kfree(device->name); |
2b82032c YZ |
1607 | kfree(device); |
1608 | goto error; | |
1609 | } | |
1610 | ||
a22285a6 | 1611 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1612 | if (IS_ERR(trans)) { |
67100f25 | 1613 | kfree(device->name); |
98d5dc13 TI |
1614 | kfree(device); |
1615 | ret = PTR_ERR(trans); | |
1616 | goto error; | |
1617 | } | |
1618 | ||
2b82032c YZ |
1619 | lock_chunks(root); |
1620 | ||
1621 | device->barriers = 1; | |
1622 | device->writeable = 1; | |
1623 | device->work.func = pending_bios_fn; | |
1624 | generate_random_uuid(device->uuid); | |
1625 | spin_lock_init(&device->io_lock); | |
1626 | device->generation = trans->transid; | |
788f20eb CM |
1627 | device->io_width = root->sectorsize; |
1628 | device->io_align = root->sectorsize; | |
1629 | device->sector_size = root->sectorsize; | |
1630 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1631 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1632 | device->dev_root = root->fs_info->dev_root; |
1633 | device->bdev = bdev; | |
dfe25020 | 1634 | device->in_fs_metadata = 1; |
15916de8 | 1635 | device->mode = 0; |
2b82032c | 1636 | set_blocksize(device->bdev, 4096); |
788f20eb | 1637 | |
2b82032c YZ |
1638 | if (seeding_dev) { |
1639 | sb->s_flags &= ~MS_RDONLY; | |
1640 | ret = btrfs_prepare_sprout(trans, root); | |
1641 | BUG_ON(ret); | |
1642 | } | |
788f20eb | 1643 | |
2b82032c | 1644 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 CM |
1645 | |
1646 | /* | |
1647 | * we don't want write_supers to jump in here with our device | |
1648 | * half setup | |
1649 | */ | |
1650 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2b82032c YZ |
1651 | list_add(&device->dev_list, &root->fs_info->fs_devices->devices); |
1652 | list_add(&device->dev_alloc_list, | |
1653 | &root->fs_info->fs_devices->alloc_list); | |
1654 | root->fs_info->fs_devices->num_devices++; | |
1655 | root->fs_info->fs_devices->open_devices++; | |
1656 | root->fs_info->fs_devices->rw_devices++; | |
1657 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 1658 | |
c289811c CM |
1659 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1660 | root->fs_info->fs_devices->rotating = 1; | |
1661 | ||
788f20eb CM |
1662 | total_bytes = btrfs_super_total_bytes(&root->fs_info->super_copy); |
1663 | btrfs_set_super_total_bytes(&root->fs_info->super_copy, | |
1664 | total_bytes + device->total_bytes); | |
1665 | ||
1666 | total_bytes = btrfs_super_num_devices(&root->fs_info->super_copy); | |
1667 | btrfs_set_super_num_devices(&root->fs_info->super_copy, | |
1668 | total_bytes + 1); | |
e5e9a520 | 1669 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1670 | |
2b82032c YZ |
1671 | if (seeding_dev) { |
1672 | ret = init_first_rw_device(trans, root, device); | |
1673 | BUG_ON(ret); | |
1674 | ret = btrfs_finish_sprout(trans, root); | |
1675 | BUG_ON(ret); | |
1676 | } else { | |
1677 | ret = btrfs_add_device(trans, root, device); | |
1678 | } | |
1679 | ||
913d952e CM |
1680 | /* |
1681 | * we've got more storage, clear any full flags on the space | |
1682 | * infos | |
1683 | */ | |
1684 | btrfs_clear_space_info_full(root->fs_info); | |
1685 | ||
7d9eb12c | 1686 | unlock_chunks(root); |
2b82032c | 1687 | btrfs_commit_transaction(trans, root); |
a2135011 | 1688 | |
2b82032c YZ |
1689 | if (seeding_dev) { |
1690 | mutex_unlock(&uuid_mutex); | |
1691 | up_write(&sb->s_umount); | |
788f20eb | 1692 | |
2b82032c YZ |
1693 | ret = btrfs_relocate_sys_chunks(root); |
1694 | BUG_ON(ret); | |
1695 | } | |
1696 | out: | |
1697 | mutex_unlock(&root->fs_info->volume_mutex); | |
1698 | return ret; | |
1699 | error: | |
15916de8 | 1700 | close_bdev_exclusive(bdev, 0); |
2b82032c YZ |
1701 | if (seeding_dev) { |
1702 | mutex_unlock(&uuid_mutex); | |
1703 | up_write(&sb->s_umount); | |
1704 | } | |
788f20eb CM |
1705 | goto out; |
1706 | } | |
1707 | ||
d397712b CM |
1708 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1709 | struct btrfs_device *device) | |
0b86a832 CM |
1710 | { |
1711 | int ret; | |
1712 | struct btrfs_path *path; | |
1713 | struct btrfs_root *root; | |
1714 | struct btrfs_dev_item *dev_item; | |
1715 | struct extent_buffer *leaf; | |
1716 | struct btrfs_key key; | |
1717 | ||
1718 | root = device->dev_root->fs_info->chunk_root; | |
1719 | ||
1720 | path = btrfs_alloc_path(); | |
1721 | if (!path) | |
1722 | return -ENOMEM; | |
1723 | ||
1724 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1725 | key.type = BTRFS_DEV_ITEM_KEY; | |
1726 | key.offset = device->devid; | |
1727 | ||
1728 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1729 | if (ret < 0) | |
1730 | goto out; | |
1731 | ||
1732 | if (ret > 0) { | |
1733 | ret = -ENOENT; | |
1734 | goto out; | |
1735 | } | |
1736 | ||
1737 | leaf = path->nodes[0]; | |
1738 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1739 | ||
1740 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1741 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1742 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1743 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1744 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1745 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1746 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1747 | btrfs_mark_buffer_dirty(leaf); | |
1748 | ||
1749 | out: | |
1750 | btrfs_free_path(path); | |
1751 | return ret; | |
1752 | } | |
1753 | ||
7d9eb12c | 1754 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1755 | struct btrfs_device *device, u64 new_size) |
1756 | { | |
1757 | struct btrfs_super_block *super_copy = | |
1758 | &device->dev_root->fs_info->super_copy; | |
1759 | u64 old_total = btrfs_super_total_bytes(super_copy); | |
1760 | u64 diff = new_size - device->total_bytes; | |
1761 | ||
2b82032c YZ |
1762 | if (!device->writeable) |
1763 | return -EACCES; | |
1764 | if (new_size <= device->total_bytes) | |
1765 | return -EINVAL; | |
1766 | ||
8f18cf13 | 1767 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1768 | device->fs_devices->total_rw_bytes += diff; |
1769 | ||
1770 | device->total_bytes = new_size; | |
9779b72f | 1771 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1772 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1773 | ||
8f18cf13 CM |
1774 | return btrfs_update_device(trans, device); |
1775 | } | |
1776 | ||
7d9eb12c CM |
1777 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1778 | struct btrfs_device *device, u64 new_size) | |
1779 | { | |
1780 | int ret; | |
1781 | lock_chunks(device->dev_root); | |
1782 | ret = __btrfs_grow_device(trans, device, new_size); | |
1783 | unlock_chunks(device->dev_root); | |
1784 | return ret; | |
1785 | } | |
1786 | ||
8f18cf13 CM |
1787 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1788 | struct btrfs_root *root, | |
1789 | u64 chunk_tree, u64 chunk_objectid, | |
1790 | u64 chunk_offset) | |
1791 | { | |
1792 | int ret; | |
1793 | struct btrfs_path *path; | |
1794 | struct btrfs_key key; | |
1795 | ||
1796 | root = root->fs_info->chunk_root; | |
1797 | path = btrfs_alloc_path(); | |
1798 | if (!path) | |
1799 | return -ENOMEM; | |
1800 | ||
1801 | key.objectid = chunk_objectid; | |
1802 | key.offset = chunk_offset; | |
1803 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1804 | ||
1805 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1806 | BUG_ON(ret); | |
1807 | ||
1808 | ret = btrfs_del_item(trans, root, path); | |
1809 | BUG_ON(ret); | |
1810 | ||
1811 | btrfs_free_path(path); | |
1812 | return 0; | |
1813 | } | |
1814 | ||
b2950863 | 1815 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1816 | chunk_offset) |
1817 | { | |
1818 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
1819 | struct btrfs_disk_key *disk_key; | |
1820 | struct btrfs_chunk *chunk; | |
1821 | u8 *ptr; | |
1822 | int ret = 0; | |
1823 | u32 num_stripes; | |
1824 | u32 array_size; | |
1825 | u32 len = 0; | |
1826 | u32 cur; | |
1827 | struct btrfs_key key; | |
1828 | ||
1829 | array_size = btrfs_super_sys_array_size(super_copy); | |
1830 | ||
1831 | ptr = super_copy->sys_chunk_array; | |
1832 | cur = 0; | |
1833 | ||
1834 | while (cur < array_size) { | |
1835 | disk_key = (struct btrfs_disk_key *)ptr; | |
1836 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1837 | ||
1838 | len = sizeof(*disk_key); | |
1839 | ||
1840 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1841 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1842 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1843 | len += btrfs_chunk_item_size(num_stripes); | |
1844 | } else { | |
1845 | ret = -EIO; | |
1846 | break; | |
1847 | } | |
1848 | if (key.objectid == chunk_objectid && | |
1849 | key.offset == chunk_offset) { | |
1850 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1851 | array_size -= len; | |
1852 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1853 | } else { | |
1854 | ptr += len; | |
1855 | cur += len; | |
1856 | } | |
1857 | } | |
1858 | return ret; | |
1859 | } | |
1860 | ||
b2950863 | 1861 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
1862 | u64 chunk_tree, u64 chunk_objectid, |
1863 | u64 chunk_offset) | |
1864 | { | |
1865 | struct extent_map_tree *em_tree; | |
1866 | struct btrfs_root *extent_root; | |
1867 | struct btrfs_trans_handle *trans; | |
1868 | struct extent_map *em; | |
1869 | struct map_lookup *map; | |
1870 | int ret; | |
1871 | int i; | |
1872 | ||
1873 | root = root->fs_info->chunk_root; | |
1874 | extent_root = root->fs_info->extent_root; | |
1875 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
1876 | ||
ba1bf481 JB |
1877 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
1878 | if (ret) | |
1879 | return -ENOSPC; | |
1880 | ||
8f18cf13 | 1881 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 1882 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
1883 | if (ret) |
1884 | return ret; | |
8f18cf13 | 1885 | |
a22285a6 | 1886 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1887 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 1888 | |
7d9eb12c CM |
1889 | lock_chunks(root); |
1890 | ||
8f18cf13 CM |
1891 | /* |
1892 | * step two, delete the device extents and the | |
1893 | * chunk tree entries | |
1894 | */ | |
890871be | 1895 | read_lock(&em_tree->lock); |
8f18cf13 | 1896 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 1897 | read_unlock(&em_tree->lock); |
8f18cf13 | 1898 | |
a061fc8d CM |
1899 | BUG_ON(em->start > chunk_offset || |
1900 | em->start + em->len < chunk_offset); | |
8f18cf13 CM |
1901 | map = (struct map_lookup *)em->bdev; |
1902 | ||
1903 | for (i = 0; i < map->num_stripes; i++) { | |
1904 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
1905 | map->stripes[i].physical); | |
1906 | BUG_ON(ret); | |
a061fc8d | 1907 | |
dfe25020 CM |
1908 | if (map->stripes[i].dev) { |
1909 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
1910 | BUG_ON(ret); | |
1911 | } | |
8f18cf13 CM |
1912 | } |
1913 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
1914 | chunk_offset); | |
1915 | ||
1916 | BUG_ON(ret); | |
1917 | ||
1918 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
1919 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
1920 | BUG_ON(ret); | |
8f18cf13 CM |
1921 | } |
1922 | ||
2b82032c YZ |
1923 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
1924 | BUG_ON(ret); | |
1925 | ||
890871be | 1926 | write_lock(&em_tree->lock); |
2b82032c | 1927 | remove_extent_mapping(em_tree, em); |
890871be | 1928 | write_unlock(&em_tree->lock); |
2b82032c YZ |
1929 | |
1930 | kfree(map); | |
1931 | em->bdev = NULL; | |
1932 | ||
1933 | /* once for the tree */ | |
1934 | free_extent_map(em); | |
1935 | /* once for us */ | |
1936 | free_extent_map(em); | |
1937 | ||
1938 | unlock_chunks(root); | |
1939 | btrfs_end_transaction(trans, root); | |
1940 | return 0; | |
1941 | } | |
1942 | ||
1943 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
1944 | { | |
1945 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
1946 | struct btrfs_path *path; | |
1947 | struct extent_buffer *leaf; | |
1948 | struct btrfs_chunk *chunk; | |
1949 | struct btrfs_key key; | |
1950 | struct btrfs_key found_key; | |
1951 | u64 chunk_tree = chunk_root->root_key.objectid; | |
1952 | u64 chunk_type; | |
ba1bf481 JB |
1953 | bool retried = false; |
1954 | int failed = 0; | |
2b82032c YZ |
1955 | int ret; |
1956 | ||
1957 | path = btrfs_alloc_path(); | |
1958 | if (!path) | |
1959 | return -ENOMEM; | |
1960 | ||
ba1bf481 | 1961 | again: |
2b82032c YZ |
1962 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
1963 | key.offset = (u64)-1; | |
1964 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1965 | ||
1966 | while (1) { | |
1967 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
1968 | if (ret < 0) | |
1969 | goto error; | |
1970 | BUG_ON(ret == 0); | |
1971 | ||
1972 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
1973 | key.type); | |
1974 | if (ret < 0) | |
1975 | goto error; | |
1976 | if (ret > 0) | |
1977 | break; | |
1a40e23b | 1978 | |
2b82032c YZ |
1979 | leaf = path->nodes[0]; |
1980 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 1981 | |
2b82032c YZ |
1982 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
1983 | struct btrfs_chunk); | |
1984 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
1985 | btrfs_release_path(chunk_root, path); | |
8f18cf13 | 1986 | |
2b82032c YZ |
1987 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
1988 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
1989 | found_key.objectid, | |
1990 | found_key.offset); | |
ba1bf481 JB |
1991 | if (ret == -ENOSPC) |
1992 | failed++; | |
1993 | else if (ret) | |
1994 | BUG(); | |
2b82032c | 1995 | } |
8f18cf13 | 1996 | |
2b82032c YZ |
1997 | if (found_key.offset == 0) |
1998 | break; | |
1999 | key.offset = found_key.offset - 1; | |
2000 | } | |
2001 | ret = 0; | |
ba1bf481 JB |
2002 | if (failed && !retried) { |
2003 | failed = 0; | |
2004 | retried = true; | |
2005 | goto again; | |
2006 | } else if (failed && retried) { | |
2007 | WARN_ON(1); | |
2008 | ret = -ENOSPC; | |
2009 | } | |
2b82032c YZ |
2010 | error: |
2011 | btrfs_free_path(path); | |
2012 | return ret; | |
8f18cf13 CM |
2013 | } |
2014 | ||
ec44a35c CM |
2015 | static u64 div_factor(u64 num, int factor) |
2016 | { | |
2017 | if (factor == 10) | |
2018 | return num; | |
2019 | num *= factor; | |
2020 | do_div(num, 10); | |
2021 | return num; | |
2022 | } | |
2023 | ||
ec44a35c CM |
2024 | int btrfs_balance(struct btrfs_root *dev_root) |
2025 | { | |
2026 | int ret; | |
ec44a35c CM |
2027 | struct list_head *devices = &dev_root->fs_info->fs_devices->devices; |
2028 | struct btrfs_device *device; | |
2029 | u64 old_size; | |
2030 | u64 size_to_free; | |
2031 | struct btrfs_path *path; | |
2032 | struct btrfs_key key; | |
ec44a35c CM |
2033 | struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root; |
2034 | struct btrfs_trans_handle *trans; | |
2035 | struct btrfs_key found_key; | |
2036 | ||
2b82032c YZ |
2037 | if (dev_root->fs_info->sb->s_flags & MS_RDONLY) |
2038 | return -EROFS; | |
ec44a35c | 2039 | |
6f88a440 BH |
2040 | if (!capable(CAP_SYS_ADMIN)) |
2041 | return -EPERM; | |
2042 | ||
7d9eb12c | 2043 | mutex_lock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
2044 | dev_root = dev_root->fs_info->dev_root; |
2045 | ||
ec44a35c | 2046 | /* step one make some room on all the devices */ |
c6e30871 | 2047 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2048 | old_size = device->total_bytes; |
2049 | size_to_free = div_factor(old_size, 1); | |
2050 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2051 | if (!device->writeable || |
2052 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2053 | continue; |
2054 | ||
2055 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2056 | if (ret == -ENOSPC) |
2057 | break; | |
ec44a35c CM |
2058 | BUG_ON(ret); |
2059 | ||
a22285a6 | 2060 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2061 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2062 | |
2063 | ret = btrfs_grow_device(trans, device, old_size); | |
2064 | BUG_ON(ret); | |
2065 | ||
2066 | btrfs_end_transaction(trans, dev_root); | |
2067 | } | |
2068 | ||
2069 | /* step two, relocate all the chunks */ | |
2070 | path = btrfs_alloc_path(); | |
2071 | BUG_ON(!path); | |
2072 | ||
2073 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
2074 | key.offset = (u64)-1; | |
2075 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2076 | ||
d397712b | 2077 | while (1) { |
ec44a35c CM |
2078 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2079 | if (ret < 0) | |
2080 | goto error; | |
2081 | ||
2082 | /* | |
2083 | * this shouldn't happen, it means the last relocate | |
2084 | * failed | |
2085 | */ | |
2086 | if (ret == 0) | |
2087 | break; | |
2088 | ||
2089 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2090 | BTRFS_CHUNK_ITEM_KEY); | |
7d9eb12c | 2091 | if (ret) |
ec44a35c | 2092 | break; |
7d9eb12c | 2093 | |
ec44a35c CM |
2094 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2095 | path->slots[0]); | |
2096 | if (found_key.objectid != key.objectid) | |
2097 | break; | |
7d9eb12c | 2098 | |
ec44a35c | 2099 | /* chunk zero is special */ |
ba1bf481 | 2100 | if (found_key.offset == 0) |
ec44a35c CM |
2101 | break; |
2102 | ||
7d9eb12c | 2103 | btrfs_release_path(chunk_root, path); |
ec44a35c CM |
2104 | ret = btrfs_relocate_chunk(chunk_root, |
2105 | chunk_root->root_key.objectid, | |
2106 | found_key.objectid, | |
2107 | found_key.offset); | |
ba1bf481 JB |
2108 | BUG_ON(ret && ret != -ENOSPC); |
2109 | key.offset = found_key.offset - 1; | |
ec44a35c CM |
2110 | } |
2111 | ret = 0; | |
2112 | error: | |
2113 | btrfs_free_path(path); | |
7d9eb12c | 2114 | mutex_unlock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
2115 | return ret; |
2116 | } | |
2117 | ||
8f18cf13 CM |
2118 | /* |
2119 | * shrinking a device means finding all of the device extents past | |
2120 | * the new size, and then following the back refs to the chunks. | |
2121 | * The chunk relocation code actually frees the device extent | |
2122 | */ | |
2123 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
2124 | { | |
2125 | struct btrfs_trans_handle *trans; | |
2126 | struct btrfs_root *root = device->dev_root; | |
2127 | struct btrfs_dev_extent *dev_extent = NULL; | |
2128 | struct btrfs_path *path; | |
2129 | u64 length; | |
2130 | u64 chunk_tree; | |
2131 | u64 chunk_objectid; | |
2132 | u64 chunk_offset; | |
2133 | int ret; | |
2134 | int slot; | |
ba1bf481 JB |
2135 | int failed = 0; |
2136 | bool retried = false; | |
8f18cf13 CM |
2137 | struct extent_buffer *l; |
2138 | struct btrfs_key key; | |
2139 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
2140 | u64 old_total = btrfs_super_total_bytes(super_copy); | |
ba1bf481 | 2141 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
2142 | u64 diff = device->total_bytes - new_size; |
2143 | ||
2b82032c YZ |
2144 | if (new_size >= device->total_bytes) |
2145 | return -EINVAL; | |
8f18cf13 CM |
2146 | |
2147 | path = btrfs_alloc_path(); | |
2148 | if (!path) | |
2149 | return -ENOMEM; | |
2150 | ||
8f18cf13 CM |
2151 | path->reada = 2; |
2152 | ||
7d9eb12c CM |
2153 | lock_chunks(root); |
2154 | ||
8f18cf13 | 2155 | device->total_bytes = new_size; |
2b82032c YZ |
2156 | if (device->writeable) |
2157 | device->fs_devices->total_rw_bytes -= diff; | |
7d9eb12c | 2158 | unlock_chunks(root); |
8f18cf13 | 2159 | |
ba1bf481 | 2160 | again: |
8f18cf13 CM |
2161 | key.objectid = device->devid; |
2162 | key.offset = (u64)-1; | |
2163 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2164 | ||
2165 | while (1) { | |
2166 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2167 | if (ret < 0) | |
2168 | goto done; | |
2169 | ||
2170 | ret = btrfs_previous_item(root, path, 0, key.type); | |
2171 | if (ret < 0) | |
2172 | goto done; | |
2173 | if (ret) { | |
2174 | ret = 0; | |
ba1bf481 | 2175 | btrfs_release_path(root, path); |
bf1fb512 | 2176 | break; |
8f18cf13 CM |
2177 | } |
2178 | ||
2179 | l = path->nodes[0]; | |
2180 | slot = path->slots[0]; | |
2181 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
2182 | ||
ba1bf481 JB |
2183 | if (key.objectid != device->devid) { |
2184 | btrfs_release_path(root, path); | |
bf1fb512 | 2185 | break; |
ba1bf481 | 2186 | } |
8f18cf13 CM |
2187 | |
2188 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
2189 | length = btrfs_dev_extent_length(l, dev_extent); | |
2190 | ||
ba1bf481 JB |
2191 | if (key.offset + length <= new_size) { |
2192 | btrfs_release_path(root, path); | |
d6397bae | 2193 | break; |
ba1bf481 | 2194 | } |
8f18cf13 CM |
2195 | |
2196 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
2197 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
2198 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
2199 | btrfs_release_path(root, path); | |
2200 | ||
2201 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
2202 | chunk_offset); | |
ba1bf481 | 2203 | if (ret && ret != -ENOSPC) |
8f18cf13 | 2204 | goto done; |
ba1bf481 JB |
2205 | if (ret == -ENOSPC) |
2206 | failed++; | |
2207 | key.offset -= 1; | |
2208 | } | |
2209 | ||
2210 | if (failed && !retried) { | |
2211 | failed = 0; | |
2212 | retried = true; | |
2213 | goto again; | |
2214 | } else if (failed && retried) { | |
2215 | ret = -ENOSPC; | |
2216 | lock_chunks(root); | |
2217 | ||
2218 | device->total_bytes = old_size; | |
2219 | if (device->writeable) | |
2220 | device->fs_devices->total_rw_bytes += diff; | |
2221 | unlock_chunks(root); | |
2222 | goto done; | |
8f18cf13 CM |
2223 | } |
2224 | ||
d6397bae | 2225 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 2226 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
2227 | if (IS_ERR(trans)) { |
2228 | ret = PTR_ERR(trans); | |
2229 | goto done; | |
2230 | } | |
2231 | ||
d6397bae CB |
2232 | lock_chunks(root); |
2233 | ||
2234 | device->disk_total_bytes = new_size; | |
2235 | /* Now btrfs_update_device() will change the on-disk size. */ | |
2236 | ret = btrfs_update_device(trans, device); | |
2237 | if (ret) { | |
2238 | unlock_chunks(root); | |
2239 | btrfs_end_transaction(trans, root); | |
2240 | goto done; | |
2241 | } | |
2242 | WARN_ON(diff > old_total); | |
2243 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
2244 | unlock_chunks(root); | |
2245 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
2246 | done: |
2247 | btrfs_free_path(path); | |
2248 | return ret; | |
2249 | } | |
2250 | ||
b2950863 | 2251 | static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans, |
0b86a832 CM |
2252 | struct btrfs_root *root, |
2253 | struct btrfs_key *key, | |
2254 | struct btrfs_chunk *chunk, int item_size) | |
2255 | { | |
2256 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
2257 | struct btrfs_disk_key disk_key; | |
2258 | u32 array_size; | |
2259 | u8 *ptr; | |
2260 | ||
2261 | array_size = btrfs_super_sys_array_size(super_copy); | |
2262 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
2263 | return -EFBIG; | |
2264 | ||
2265 | ptr = super_copy->sys_chunk_array + array_size; | |
2266 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2267 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
2268 | ptr += sizeof(disk_key); | |
2269 | memcpy(ptr, chunk, item_size); | |
2270 | item_size += sizeof(disk_key); | |
2271 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
2272 | return 0; | |
2273 | } | |
2274 | ||
d397712b | 2275 | static noinline u64 chunk_bytes_by_type(u64 type, u64 calc_size, |
a1b32a59 | 2276 | int num_stripes, int sub_stripes) |
9b3f68b9 CM |
2277 | { |
2278 | if (type & (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_DUP)) | |
2279 | return calc_size; | |
2280 | else if (type & BTRFS_BLOCK_GROUP_RAID10) | |
2281 | return calc_size * (num_stripes / sub_stripes); | |
2282 | else | |
2283 | return calc_size * num_stripes; | |
2284 | } | |
2285 | ||
b2117a39 MX |
2286 | /* Used to sort the devices by max_avail(descending sort) */ |
2287 | int btrfs_cmp_device_free_bytes(const void *dev_info1, const void *dev_info2) | |
0b86a832 | 2288 | { |
b2117a39 MX |
2289 | if (((struct btrfs_device_info *)dev_info1)->max_avail > |
2290 | ((struct btrfs_device_info *)dev_info2)->max_avail) | |
2291 | return -1; | |
2292 | else if (((struct btrfs_device_info *)dev_info1)->max_avail < | |
2293 | ((struct btrfs_device_info *)dev_info2)->max_avail) | |
2294 | return 1; | |
2295 | else | |
2296 | return 0; | |
2297 | } | |
0b86a832 | 2298 | |
b2117a39 MX |
2299 | static int __btrfs_calc_nstripes(struct btrfs_fs_devices *fs_devices, u64 type, |
2300 | int *num_stripes, int *min_stripes, | |
2301 | int *sub_stripes) | |
2302 | { | |
2303 | *num_stripes = 1; | |
2304 | *min_stripes = 1; | |
2305 | *sub_stripes = 0; | |
593060d7 | 2306 | |
a40a90a0 | 2307 | if (type & (BTRFS_BLOCK_GROUP_RAID0)) { |
b2117a39 MX |
2308 | *num_stripes = fs_devices->rw_devices; |
2309 | *min_stripes = 2; | |
a40a90a0 CM |
2310 | } |
2311 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
b2117a39 MX |
2312 | *num_stripes = 2; |
2313 | *min_stripes = 2; | |
a40a90a0 | 2314 | } |
8790d502 | 2315 | if (type & (BTRFS_BLOCK_GROUP_RAID1)) { |
f3eae7e8 | 2316 | if (fs_devices->rw_devices < 2) |
9b3f68b9 | 2317 | return -ENOSPC; |
b2117a39 MX |
2318 | *num_stripes = 2; |
2319 | *min_stripes = 2; | |
8790d502 | 2320 | } |
321aecc6 | 2321 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { |
b2117a39 MX |
2322 | *num_stripes = fs_devices->rw_devices; |
2323 | if (*num_stripes < 4) | |
321aecc6 | 2324 | return -ENOSPC; |
b2117a39 MX |
2325 | *num_stripes &= ~(u32)1; |
2326 | *sub_stripes = 2; | |
2327 | *min_stripes = 4; | |
321aecc6 | 2328 | } |
9b3f68b9 | 2329 | |
b2117a39 MX |
2330 | return 0; |
2331 | } | |
2332 | ||
2333 | static u64 __btrfs_calc_stripe_size(struct btrfs_fs_devices *fs_devices, | |
2334 | u64 proposed_size, u64 type, | |
2335 | int num_stripes, int small_stripe) | |
2336 | { | |
2337 | int min_stripe_size = 1 * 1024 * 1024; | |
2338 | u64 calc_size = proposed_size; | |
2339 | u64 max_chunk_size = calc_size; | |
2340 | int ncopies = 1; | |
2341 | ||
2342 | if (type & (BTRFS_BLOCK_GROUP_RAID1 | | |
2343 | BTRFS_BLOCK_GROUP_DUP | | |
2344 | BTRFS_BLOCK_GROUP_RAID10)) | |
2345 | ncopies = 2; | |
2346 | ||
9b3f68b9 CM |
2347 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
2348 | max_chunk_size = 10 * calc_size; | |
a40a90a0 | 2349 | min_stripe_size = 64 * 1024 * 1024; |
9b3f68b9 | 2350 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
83d3c969 | 2351 | max_chunk_size = 256 * 1024 * 1024; |
a40a90a0 CM |
2352 | min_stripe_size = 32 * 1024 * 1024; |
2353 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2354 | calc_size = 8 * 1024 * 1024; | |
2355 | max_chunk_size = calc_size * 2; | |
2356 | min_stripe_size = 1 * 1024 * 1024; | |
9b3f68b9 CM |
2357 | } |
2358 | ||
2b82032c YZ |
2359 | /* we don't want a chunk larger than 10% of writeable space */ |
2360 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
2361 | max_chunk_size); | |
9b3f68b9 | 2362 | |
1974a3b4 MX |
2363 | if (calc_size * num_stripes > max_chunk_size * ncopies) { |
2364 | calc_size = max_chunk_size * ncopies; | |
9b3f68b9 | 2365 | do_div(calc_size, num_stripes); |
b2117a39 MX |
2366 | do_div(calc_size, BTRFS_STRIPE_LEN); |
2367 | calc_size *= BTRFS_STRIPE_LEN; | |
9b3f68b9 | 2368 | } |
0cad8a11 | 2369 | |
9b3f68b9 | 2370 | /* we don't want tiny stripes */ |
b2117a39 | 2371 | if (!small_stripe) |
0cad8a11 | 2372 | calc_size = max_t(u64, min_stripe_size, calc_size); |
9b3f68b9 | 2373 | |
9f680ce0 | 2374 | /* |
b2117a39 | 2375 | * we're about to do_div by the BTRFS_STRIPE_LEN so lets make sure |
9f680ce0 CM |
2376 | * we end up with something bigger than a stripe |
2377 | */ | |
b2117a39 MX |
2378 | calc_size = max_t(u64, calc_size, BTRFS_STRIPE_LEN); |
2379 | ||
2380 | do_div(calc_size, BTRFS_STRIPE_LEN); | |
2381 | calc_size *= BTRFS_STRIPE_LEN; | |
2382 | ||
2383 | return calc_size; | |
2384 | } | |
2385 | ||
2386 | static struct map_lookup *__shrink_map_lookup_stripes(struct map_lookup *map, | |
2387 | int num_stripes) | |
2388 | { | |
2389 | struct map_lookup *new; | |
2390 | size_t len = map_lookup_size(num_stripes); | |
2391 | ||
2392 | BUG_ON(map->num_stripes < num_stripes); | |
2393 | ||
2394 | if (map->num_stripes == num_stripes) | |
2395 | return map; | |
2396 | ||
2397 | new = kmalloc(len, GFP_NOFS); | |
2398 | if (!new) { | |
2399 | /* just change map->num_stripes */ | |
2400 | map->num_stripes = num_stripes; | |
2401 | return map; | |
2402 | } | |
2403 | ||
2404 | memcpy(new, map, len); | |
2405 | new->num_stripes = num_stripes; | |
2406 | kfree(map); | |
2407 | return new; | |
2408 | } | |
2409 | ||
2410 | /* | |
2411 | * helper to allocate device space from btrfs_device_info, in which we stored | |
2412 | * max free space information of every device. It is used when we can not | |
2413 | * allocate chunks by default size. | |
2414 | * | |
2415 | * By this helper, we can allocate a new chunk as larger as possible. | |
2416 | */ | |
2417 | static int __btrfs_alloc_tiny_space(struct btrfs_trans_handle *trans, | |
2418 | struct btrfs_fs_devices *fs_devices, | |
2419 | struct btrfs_device_info *devices, | |
2420 | int nr_device, u64 type, | |
2421 | struct map_lookup **map_lookup, | |
2422 | int min_stripes, u64 *stripe_size) | |
2423 | { | |
2424 | int i, index, sort_again = 0; | |
2425 | int min_devices = min_stripes; | |
2426 | u64 max_avail, min_free; | |
2427 | struct map_lookup *map = *map_lookup; | |
2428 | int ret; | |
9f680ce0 | 2429 | |
b2117a39 MX |
2430 | if (nr_device < min_stripes) |
2431 | return -ENOSPC; | |
2432 | ||
2433 | btrfs_descending_sort_devices(devices, nr_device); | |
2434 | ||
2435 | max_avail = devices[0].max_avail; | |
2436 | if (!max_avail) | |
2437 | return -ENOSPC; | |
2438 | ||
2439 | for (i = 0; i < nr_device; i++) { | |
2440 | /* | |
2441 | * if dev_offset = 0, it means the free space of this device | |
2442 | * is less than what we need, and we didn't search max avail | |
2443 | * extent on this device, so do it now. | |
2444 | */ | |
2445 | if (!devices[i].dev_offset) { | |
2446 | ret = find_free_dev_extent(trans, devices[i].dev, | |
2447 | max_avail, | |
2448 | &devices[i].dev_offset, | |
2449 | &devices[i].max_avail); | |
2450 | if (ret != 0 && ret != -ENOSPC) | |
2451 | return ret; | |
2452 | sort_again = 1; | |
2453 | } | |
2454 | } | |
2455 | ||
2456 | /* we update the max avail free extent of each devices, sort again */ | |
2457 | if (sort_again) | |
2458 | btrfs_descending_sort_devices(devices, nr_device); | |
2459 | ||
2460 | if (type & BTRFS_BLOCK_GROUP_DUP) | |
2461 | min_devices = 1; | |
2462 | ||
2463 | if (!devices[min_devices - 1].max_avail) | |
2464 | return -ENOSPC; | |
2465 | ||
2466 | max_avail = devices[min_devices - 1].max_avail; | |
2467 | if (type & BTRFS_BLOCK_GROUP_DUP) | |
2468 | do_div(max_avail, 2); | |
2469 | ||
2470 | max_avail = __btrfs_calc_stripe_size(fs_devices, max_avail, type, | |
2471 | min_stripes, 1); | |
2472 | if (type & BTRFS_BLOCK_GROUP_DUP) | |
2473 | min_free = max_avail * 2; | |
2474 | else | |
2475 | min_free = max_avail; | |
2476 | ||
2477 | if (min_free > devices[min_devices - 1].max_avail) | |
2478 | return -ENOSPC; | |
2479 | ||
2480 | map = __shrink_map_lookup_stripes(map, min_stripes); | |
2481 | *stripe_size = max_avail; | |
2482 | ||
2483 | index = 0; | |
2484 | for (i = 0; i < min_stripes; i++) { | |
2485 | map->stripes[i].dev = devices[index].dev; | |
2486 | map->stripes[i].physical = devices[index].dev_offset; | |
2487 | if (type & BTRFS_BLOCK_GROUP_DUP) { | |
2488 | i++; | |
2489 | map->stripes[i].dev = devices[index].dev; | |
2490 | map->stripes[i].physical = devices[index].dev_offset + | |
2491 | max_avail; | |
2492 | } | |
2493 | index++; | |
2494 | } | |
2495 | *map_lookup = map; | |
2496 | ||
2497 | return 0; | |
2498 | } | |
2499 | ||
2500 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2501 | struct btrfs_root *extent_root, | |
2502 | struct map_lookup **map_ret, | |
2503 | u64 *num_bytes, u64 *stripe_size, | |
2504 | u64 start, u64 type) | |
2505 | { | |
2506 | struct btrfs_fs_info *info = extent_root->fs_info; | |
2507 | struct btrfs_device *device = NULL; | |
2508 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
2509 | struct list_head *cur; | |
2510 | struct map_lookup *map; | |
2511 | struct extent_map_tree *em_tree; | |
2512 | struct extent_map *em; | |
2513 | struct btrfs_device_info *devices_info; | |
2514 | struct list_head private_devs; | |
2515 | u64 calc_size = 1024 * 1024 * 1024; | |
2516 | u64 min_free; | |
2517 | u64 avail; | |
2518 | u64 dev_offset; | |
2519 | int num_stripes; | |
2520 | int min_stripes; | |
2521 | int sub_stripes; | |
2522 | int min_devices; /* the min number of devices we need */ | |
2523 | int i; | |
2524 | int ret; | |
2525 | int index; | |
2526 | ||
2527 | if ((type & BTRFS_BLOCK_GROUP_RAID1) && | |
2528 | (type & BTRFS_BLOCK_GROUP_DUP)) { | |
2529 | WARN_ON(1); | |
2530 | type &= ~BTRFS_BLOCK_GROUP_DUP; | |
2531 | } | |
2532 | if (list_empty(&fs_devices->alloc_list)) | |
2533 | return -ENOSPC; | |
2534 | ||
2535 | ret = __btrfs_calc_nstripes(fs_devices, type, &num_stripes, | |
2536 | &min_stripes, &sub_stripes); | |
2537 | if (ret) | |
2538 | return ret; | |
2539 | ||
2540 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, | |
2541 | GFP_NOFS); | |
2542 | if (!devices_info) | |
2543 | return -ENOMEM; | |
2544 | ||
2545 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
2546 | if (!map) { | |
2547 | ret = -ENOMEM; | |
2548 | goto error; | |
2549 | } | |
2550 | map->num_stripes = num_stripes; | |
9b3f68b9 | 2551 | |
2b82032c | 2552 | cur = fs_devices->alloc_list.next; |
6324fbf3 | 2553 | index = 0; |
b2117a39 | 2554 | i = 0; |
611f0e00 | 2555 | |
b2117a39 MX |
2556 | calc_size = __btrfs_calc_stripe_size(fs_devices, calc_size, type, |
2557 | num_stripes, 0); | |
2558 | ||
2559 | if (type & BTRFS_BLOCK_GROUP_DUP) { | |
611f0e00 | 2560 | min_free = calc_size * 2; |
b2117a39 MX |
2561 | min_devices = 1; |
2562 | } else { | |
9b3f68b9 | 2563 | min_free = calc_size; |
b2117a39 MX |
2564 | min_devices = min_stripes; |
2565 | } | |
ad5bd91e | 2566 | |
2b82032c | 2567 | INIT_LIST_HEAD(&private_devs); |
d397712b | 2568 | while (index < num_stripes) { |
b3075717 | 2569 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
2b82032c | 2570 | BUG_ON(!device->writeable); |
dfe25020 CM |
2571 | if (device->total_bytes > device->bytes_used) |
2572 | avail = device->total_bytes - device->bytes_used; | |
2573 | else | |
2574 | avail = 0; | |
6324fbf3 | 2575 | cur = cur->next; |
8f18cf13 | 2576 | |
dfe25020 | 2577 | if (device->in_fs_metadata && avail >= min_free) { |
b2117a39 MX |
2578 | ret = find_free_dev_extent(trans, device, min_free, |
2579 | &devices_info[i].dev_offset, | |
2580 | &devices_info[i].max_avail); | |
8f18cf13 CM |
2581 | if (ret == 0) { |
2582 | list_move_tail(&device->dev_alloc_list, | |
2583 | &private_devs); | |
2b82032c | 2584 | map->stripes[index].dev = device; |
b2117a39 MX |
2585 | map->stripes[index].physical = |
2586 | devices_info[i].dev_offset; | |
611f0e00 | 2587 | index++; |
2b82032c YZ |
2588 | if (type & BTRFS_BLOCK_GROUP_DUP) { |
2589 | map->stripes[index].dev = device; | |
2590 | map->stripes[index].physical = | |
b2117a39 MX |
2591 | devices_info[i].dev_offset + |
2592 | calc_size; | |
8f18cf13 | 2593 | index++; |
2b82032c | 2594 | } |
b2117a39 MX |
2595 | } else if (ret != -ENOSPC) |
2596 | goto error; | |
2597 | ||
2598 | devices_info[i].dev = device; | |
2599 | i++; | |
2600 | } else if (device->in_fs_metadata && | |
2601 | avail >= BTRFS_STRIPE_LEN) { | |
2602 | devices_info[i].dev = device; | |
2603 | devices_info[i].max_avail = avail; | |
2604 | i++; | |
2605 | } | |
2606 | ||
2b82032c | 2607 | if (cur == &fs_devices->alloc_list) |
6324fbf3 CM |
2608 | break; |
2609 | } | |
b2117a39 | 2610 | |
2b82032c | 2611 | list_splice(&private_devs, &fs_devices->alloc_list); |
6324fbf3 | 2612 | if (index < num_stripes) { |
a40a90a0 CM |
2613 | if (index >= min_stripes) { |
2614 | num_stripes = index; | |
2615 | if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
2616 | num_stripes /= sub_stripes; | |
2617 | num_stripes *= sub_stripes; | |
2618 | } | |
b2117a39 MX |
2619 | |
2620 | map = __shrink_map_lookup_stripes(map, num_stripes); | |
2621 | } else if (i >= min_devices) { | |
2622 | ret = __btrfs_alloc_tiny_space(trans, fs_devices, | |
2623 | devices_info, i, type, | |
2624 | &map, min_stripes, | |
2625 | &calc_size); | |
2626 | if (ret) | |
2627 | goto error; | |
2628 | } else { | |
2629 | ret = -ENOSPC; | |
2630 | goto error; | |
6324fbf3 | 2631 | } |
6324fbf3 | 2632 | } |
2b82032c | 2633 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
2634 | map->stripe_len = BTRFS_STRIPE_LEN; |
2635 | map->io_align = BTRFS_STRIPE_LEN; | |
2636 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 2637 | map->type = type; |
2b82032c | 2638 | map->sub_stripes = sub_stripes; |
0b86a832 | 2639 | |
2b82032c YZ |
2640 | *map_ret = map; |
2641 | *stripe_size = calc_size; | |
2642 | *num_bytes = chunk_bytes_by_type(type, calc_size, | |
b2117a39 | 2643 | map->num_stripes, sub_stripes); |
0b86a832 | 2644 | |
2b82032c YZ |
2645 | em = alloc_extent_map(GFP_NOFS); |
2646 | if (!em) { | |
b2117a39 MX |
2647 | ret = -ENOMEM; |
2648 | goto error; | |
593060d7 | 2649 | } |
2b82032c YZ |
2650 | em->bdev = (struct block_device *)map; |
2651 | em->start = start; | |
2652 | em->len = *num_bytes; | |
2653 | em->block_start = 0; | |
2654 | em->block_len = em->len; | |
593060d7 | 2655 | |
2b82032c | 2656 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 2657 | write_lock(&em_tree->lock); |
2b82032c | 2658 | ret = add_extent_mapping(em_tree, em); |
890871be | 2659 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2660 | BUG_ON(ret); |
2661 | free_extent_map(em); | |
0b86a832 | 2662 | |
2b82032c YZ |
2663 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
2664 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
2665 | start, *num_bytes); | |
2666 | BUG_ON(ret); | |
611f0e00 | 2667 | |
2b82032c YZ |
2668 | index = 0; |
2669 | while (index < map->num_stripes) { | |
2670 | device = map->stripes[index].dev; | |
2671 | dev_offset = map->stripes[index].physical; | |
0b86a832 CM |
2672 | |
2673 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
2674 | info->chunk_root->root_key.objectid, |
2675 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
2676 | start, dev_offset, calc_size); | |
0b86a832 | 2677 | BUG_ON(ret); |
2b82032c YZ |
2678 | index++; |
2679 | } | |
2680 | ||
b2117a39 | 2681 | kfree(devices_info); |
2b82032c | 2682 | return 0; |
b2117a39 MX |
2683 | |
2684 | error: | |
2685 | kfree(map); | |
2686 | kfree(devices_info); | |
2687 | return ret; | |
2b82032c YZ |
2688 | } |
2689 | ||
2690 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
2691 | struct btrfs_root *extent_root, | |
2692 | struct map_lookup *map, u64 chunk_offset, | |
2693 | u64 chunk_size, u64 stripe_size) | |
2694 | { | |
2695 | u64 dev_offset; | |
2696 | struct btrfs_key key; | |
2697 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
2698 | struct btrfs_device *device; | |
2699 | struct btrfs_chunk *chunk; | |
2700 | struct btrfs_stripe *stripe; | |
2701 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
2702 | int index = 0; | |
2703 | int ret; | |
2704 | ||
2705 | chunk = kzalloc(item_size, GFP_NOFS); | |
2706 | if (!chunk) | |
2707 | return -ENOMEM; | |
2708 | ||
2709 | index = 0; | |
2710 | while (index < map->num_stripes) { | |
2711 | device = map->stripes[index].dev; | |
2712 | device->bytes_used += stripe_size; | |
0b86a832 CM |
2713 | ret = btrfs_update_device(trans, device); |
2714 | BUG_ON(ret); | |
2b82032c YZ |
2715 | index++; |
2716 | } | |
2717 | ||
2718 | index = 0; | |
2719 | stripe = &chunk->stripe; | |
2720 | while (index < map->num_stripes) { | |
2721 | device = map->stripes[index].dev; | |
2722 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 2723 | |
e17cade2 CM |
2724 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
2725 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
2726 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 2727 | stripe++; |
0b86a832 CM |
2728 | index++; |
2729 | } | |
2730 | ||
2b82032c | 2731 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 2732 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
2733 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
2734 | btrfs_set_stack_chunk_type(chunk, map->type); | |
2735 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
2736 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
2737 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 2738 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 2739 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 2740 | |
2b82032c YZ |
2741 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2742 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2743 | key.offset = chunk_offset; | |
0b86a832 | 2744 | |
2b82032c YZ |
2745 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
2746 | BUG_ON(ret); | |
0b86a832 | 2747 | |
2b82032c YZ |
2748 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2749 | ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk, | |
2750 | item_size); | |
8f18cf13 CM |
2751 | BUG_ON(ret); |
2752 | } | |
0b86a832 | 2753 | kfree(chunk); |
2b82032c YZ |
2754 | return 0; |
2755 | } | |
0b86a832 | 2756 | |
2b82032c YZ |
2757 | /* |
2758 | * Chunk allocation falls into two parts. The first part does works | |
2759 | * that make the new allocated chunk useable, but not do any operation | |
2760 | * that modifies the chunk tree. The second part does the works that | |
2761 | * require modifying the chunk tree. This division is important for the | |
2762 | * bootstrap process of adding storage to a seed btrfs. | |
2763 | */ | |
2764 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2765 | struct btrfs_root *extent_root, u64 type) | |
2766 | { | |
2767 | u64 chunk_offset; | |
2768 | u64 chunk_size; | |
2769 | u64 stripe_size; | |
2770 | struct map_lookup *map; | |
2771 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
2772 | int ret; | |
2773 | ||
2774 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
2775 | &chunk_offset); | |
2776 | if (ret) | |
2777 | return ret; | |
2778 | ||
2779 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
2780 | &stripe_size, chunk_offset, type); | |
2781 | if (ret) | |
2782 | return ret; | |
2783 | ||
2784 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
2785 | chunk_size, stripe_size); | |
2786 | BUG_ON(ret); | |
2787 | return 0; | |
2788 | } | |
2789 | ||
d397712b | 2790 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
2791 | struct btrfs_root *root, |
2792 | struct btrfs_device *device) | |
2793 | { | |
2794 | u64 chunk_offset; | |
2795 | u64 sys_chunk_offset; | |
2796 | u64 chunk_size; | |
2797 | u64 sys_chunk_size; | |
2798 | u64 stripe_size; | |
2799 | u64 sys_stripe_size; | |
2800 | u64 alloc_profile; | |
2801 | struct map_lookup *map; | |
2802 | struct map_lookup *sys_map; | |
2803 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2804 | struct btrfs_root *extent_root = fs_info->extent_root; | |
2805 | int ret; | |
2806 | ||
2807 | ret = find_next_chunk(fs_info->chunk_root, | |
2808 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
2809 | BUG_ON(ret); | |
2810 | ||
2811 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
2812 | (fs_info->metadata_alloc_profile & | |
2813 | fs_info->avail_metadata_alloc_bits); | |
2814 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); | |
2815 | ||
2816 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
2817 | &stripe_size, chunk_offset, alloc_profile); | |
2818 | BUG_ON(ret); | |
2819 | ||
2820 | sys_chunk_offset = chunk_offset + chunk_size; | |
2821 | ||
2822 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
2823 | (fs_info->system_alloc_profile & | |
2824 | fs_info->avail_system_alloc_bits); | |
2825 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); | |
2826 | ||
2827 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
2828 | &sys_chunk_size, &sys_stripe_size, | |
2829 | sys_chunk_offset, alloc_profile); | |
2830 | BUG_ON(ret); | |
2831 | ||
2832 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
2833 | BUG_ON(ret); | |
2834 | ||
2835 | /* | |
2836 | * Modifying chunk tree needs allocating new blocks from both | |
2837 | * system block group and metadata block group. So we only can | |
2838 | * do operations require modifying the chunk tree after both | |
2839 | * block groups were created. | |
2840 | */ | |
2841 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
2842 | chunk_size, stripe_size); | |
2843 | BUG_ON(ret); | |
2844 | ||
2845 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
2846 | sys_chunk_offset, sys_chunk_size, | |
2847 | sys_stripe_size); | |
b248a415 | 2848 | BUG_ON(ret); |
2b82032c YZ |
2849 | return 0; |
2850 | } | |
2851 | ||
2852 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
2853 | { | |
2854 | struct extent_map *em; | |
2855 | struct map_lookup *map; | |
2856 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
2857 | int readonly = 0; | |
2858 | int i; | |
2859 | ||
890871be | 2860 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 2861 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 2862 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
2863 | if (!em) |
2864 | return 1; | |
2865 | ||
f48b9075 JB |
2866 | if (btrfs_test_opt(root, DEGRADED)) { |
2867 | free_extent_map(em); | |
2868 | return 0; | |
2869 | } | |
2870 | ||
2b82032c YZ |
2871 | map = (struct map_lookup *)em->bdev; |
2872 | for (i = 0; i < map->num_stripes; i++) { | |
2873 | if (!map->stripes[i].dev->writeable) { | |
2874 | readonly = 1; | |
2875 | break; | |
2876 | } | |
2877 | } | |
0b86a832 | 2878 | free_extent_map(em); |
2b82032c | 2879 | return readonly; |
0b86a832 CM |
2880 | } |
2881 | ||
2882 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
2883 | { | |
2884 | extent_map_tree_init(&tree->map_tree, GFP_NOFS); | |
2885 | } | |
2886 | ||
2887 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
2888 | { | |
2889 | struct extent_map *em; | |
2890 | ||
d397712b | 2891 | while (1) { |
890871be | 2892 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
2893 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
2894 | if (em) | |
2895 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 2896 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
2897 | if (!em) |
2898 | break; | |
2899 | kfree(em->bdev); | |
2900 | /* once for us */ | |
2901 | free_extent_map(em); | |
2902 | /* once for the tree */ | |
2903 | free_extent_map(em); | |
2904 | } | |
2905 | } | |
2906 | ||
f188591e CM |
2907 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
2908 | { | |
2909 | struct extent_map *em; | |
2910 | struct map_lookup *map; | |
2911 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
2912 | int ret; | |
2913 | ||
890871be | 2914 | read_lock(&em_tree->lock); |
f188591e | 2915 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 2916 | read_unlock(&em_tree->lock); |
f188591e CM |
2917 | BUG_ON(!em); |
2918 | ||
2919 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
2920 | map = (struct map_lookup *)em->bdev; | |
2921 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
2922 | ret = map->num_stripes; | |
321aecc6 CM |
2923 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
2924 | ret = map->sub_stripes; | |
f188591e CM |
2925 | else |
2926 | ret = 1; | |
2927 | free_extent_map(em); | |
f188591e CM |
2928 | return ret; |
2929 | } | |
2930 | ||
dfe25020 CM |
2931 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
2932 | int optimal) | |
2933 | { | |
2934 | int i; | |
2935 | if (map->stripes[optimal].dev->bdev) | |
2936 | return optimal; | |
2937 | for (i = first; i < first + num; i++) { | |
2938 | if (map->stripes[i].dev->bdev) | |
2939 | return i; | |
2940 | } | |
2941 | /* we couldn't find one that doesn't fail. Just return something | |
2942 | * and the io error handling code will clean up eventually | |
2943 | */ | |
2944 | return optimal; | |
2945 | } | |
2946 | ||
f2d8d74d CM |
2947 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
2948 | u64 logical, u64 *length, | |
2949 | struct btrfs_multi_bio **multi_ret, | |
2950 | int mirror_num, struct page *unplug_page) | |
0b86a832 CM |
2951 | { |
2952 | struct extent_map *em; | |
2953 | struct map_lookup *map; | |
2954 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
2955 | u64 offset; | |
593060d7 CM |
2956 | u64 stripe_offset; |
2957 | u64 stripe_nr; | |
cea9e445 | 2958 | int stripes_allocated = 8; |
321aecc6 | 2959 | int stripes_required = 1; |
593060d7 | 2960 | int stripe_index; |
cea9e445 | 2961 | int i; |
f2d8d74d | 2962 | int num_stripes; |
a236aed1 | 2963 | int max_errors = 0; |
cea9e445 | 2964 | struct btrfs_multi_bio *multi = NULL; |
0b86a832 | 2965 | |
7b6d91da | 2966 | if (multi_ret && !(rw & REQ_WRITE)) |
cea9e445 | 2967 | stripes_allocated = 1; |
cea9e445 CM |
2968 | again: |
2969 | if (multi_ret) { | |
2970 | multi = kzalloc(btrfs_multi_bio_size(stripes_allocated), | |
2971 | GFP_NOFS); | |
2972 | if (!multi) | |
2973 | return -ENOMEM; | |
a236aed1 CM |
2974 | |
2975 | atomic_set(&multi->error, 0); | |
cea9e445 | 2976 | } |
0b86a832 | 2977 | |
890871be | 2978 | read_lock(&em_tree->lock); |
0b86a832 | 2979 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 2980 | read_unlock(&em_tree->lock); |
f2d8d74d | 2981 | |
2423fdfb JS |
2982 | if (!em && unplug_page) { |
2983 | kfree(multi); | |
f2d8d74d | 2984 | return 0; |
2423fdfb | 2985 | } |
f2d8d74d | 2986 | |
3b951516 | 2987 | if (!em) { |
d397712b CM |
2988 | printk(KERN_CRIT "unable to find logical %llu len %llu\n", |
2989 | (unsigned long long)logical, | |
2990 | (unsigned long long)*length); | |
f2d8d74d | 2991 | BUG(); |
3b951516 | 2992 | } |
0b86a832 CM |
2993 | |
2994 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
2995 | map = (struct map_lookup *)em->bdev; | |
2996 | offset = logical - em->start; | |
593060d7 | 2997 | |
f188591e CM |
2998 | if (mirror_num > map->num_stripes) |
2999 | mirror_num = 0; | |
3000 | ||
cea9e445 | 3001 | /* if our multi bio struct is too small, back off and try again */ |
7b6d91da | 3002 | if (rw & REQ_WRITE) { |
321aecc6 CM |
3003 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | |
3004 | BTRFS_BLOCK_GROUP_DUP)) { | |
3005 | stripes_required = map->num_stripes; | |
a236aed1 | 3006 | max_errors = 1; |
321aecc6 CM |
3007 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3008 | stripes_required = map->sub_stripes; | |
a236aed1 | 3009 | max_errors = 1; |
321aecc6 CM |
3010 | } |
3011 | } | |
7b6d91da | 3012 | if (multi_ret && (rw & REQ_WRITE) && |
321aecc6 | 3013 | stripes_allocated < stripes_required) { |
cea9e445 | 3014 | stripes_allocated = map->num_stripes; |
cea9e445 CM |
3015 | free_extent_map(em); |
3016 | kfree(multi); | |
3017 | goto again; | |
3018 | } | |
593060d7 CM |
3019 | stripe_nr = offset; |
3020 | /* | |
3021 | * stripe_nr counts the total number of stripes we have to stride | |
3022 | * to get to this block | |
3023 | */ | |
3024 | do_div(stripe_nr, map->stripe_len); | |
3025 | ||
3026 | stripe_offset = stripe_nr * map->stripe_len; | |
3027 | BUG_ON(offset < stripe_offset); | |
3028 | ||
3029 | /* stripe_offset is the offset of this block in its stripe*/ | |
3030 | stripe_offset = offset - stripe_offset; | |
3031 | ||
cea9e445 | 3032 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | |
321aecc6 | 3033 | BTRFS_BLOCK_GROUP_RAID10 | |
cea9e445 CM |
3034 | BTRFS_BLOCK_GROUP_DUP)) { |
3035 | /* we limit the length of each bio to what fits in a stripe */ | |
3036 | *length = min_t(u64, em->len - offset, | |
3037 | map->stripe_len - stripe_offset); | |
3038 | } else { | |
3039 | *length = em->len - offset; | |
3040 | } | |
f2d8d74d CM |
3041 | |
3042 | if (!multi_ret && !unplug_page) | |
cea9e445 CM |
3043 | goto out; |
3044 | ||
f2d8d74d | 3045 | num_stripes = 1; |
cea9e445 | 3046 | stripe_index = 0; |
8790d502 | 3047 | if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
7b6d91da | 3048 | if (unplug_page || (rw & REQ_WRITE)) |
f2d8d74d | 3049 | num_stripes = map->num_stripes; |
2fff734f | 3050 | else if (mirror_num) |
f188591e | 3051 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3052 | else { |
3053 | stripe_index = find_live_mirror(map, 0, | |
3054 | map->num_stripes, | |
3055 | current->pid % map->num_stripes); | |
3056 | } | |
2fff734f | 3057 | |
611f0e00 | 3058 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
7b6d91da | 3059 | if (rw & REQ_WRITE) |
f2d8d74d | 3060 | num_stripes = map->num_stripes; |
f188591e CM |
3061 | else if (mirror_num) |
3062 | stripe_index = mirror_num - 1; | |
2fff734f | 3063 | |
321aecc6 CM |
3064 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3065 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3066 | |
3067 | stripe_index = do_div(stripe_nr, factor); | |
3068 | stripe_index *= map->sub_stripes; | |
3069 | ||
7b6d91da | 3070 | if (unplug_page || (rw & REQ_WRITE)) |
f2d8d74d | 3071 | num_stripes = map->sub_stripes; |
321aecc6 CM |
3072 | else if (mirror_num) |
3073 | stripe_index += mirror_num - 1; | |
dfe25020 CM |
3074 | else { |
3075 | stripe_index = find_live_mirror(map, stripe_index, | |
3076 | map->sub_stripes, stripe_index + | |
3077 | current->pid % map->sub_stripes); | |
3078 | } | |
8790d502 CM |
3079 | } else { |
3080 | /* | |
3081 | * after this do_div call, stripe_nr is the number of stripes | |
3082 | * on this device we have to walk to find the data, and | |
3083 | * stripe_index is the number of our device in the stripe array | |
3084 | */ | |
3085 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3086 | } | |
593060d7 | 3087 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3088 | |
f2d8d74d CM |
3089 | for (i = 0; i < num_stripes; i++) { |
3090 | if (unplug_page) { | |
3091 | struct btrfs_device *device; | |
3092 | struct backing_dev_info *bdi; | |
3093 | ||
3094 | device = map->stripes[stripe_index].dev; | |
dfe25020 CM |
3095 | if (device->bdev) { |
3096 | bdi = blk_get_backing_dev_info(device->bdev); | |
d397712b | 3097 | if (bdi->unplug_io_fn) |
dfe25020 | 3098 | bdi->unplug_io_fn(bdi, unplug_page); |
f2d8d74d CM |
3099 | } |
3100 | } else { | |
3101 | multi->stripes[i].physical = | |
3102 | map->stripes[stripe_index].physical + | |
3103 | stripe_offset + stripe_nr * map->stripe_len; | |
3104 | multi->stripes[i].dev = map->stripes[stripe_index].dev; | |
3105 | } | |
cea9e445 | 3106 | stripe_index++; |
593060d7 | 3107 | } |
f2d8d74d CM |
3108 | if (multi_ret) { |
3109 | *multi_ret = multi; | |
3110 | multi->num_stripes = num_stripes; | |
a236aed1 | 3111 | multi->max_errors = max_errors; |
f2d8d74d | 3112 | } |
cea9e445 | 3113 | out: |
0b86a832 | 3114 | free_extent_map(em); |
0b86a832 CM |
3115 | return 0; |
3116 | } | |
3117 | ||
f2d8d74d CM |
3118 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3119 | u64 logical, u64 *length, | |
3120 | struct btrfs_multi_bio **multi_ret, int mirror_num) | |
3121 | { | |
3122 | return __btrfs_map_block(map_tree, rw, logical, length, multi_ret, | |
3123 | mirror_num, NULL); | |
3124 | } | |
3125 | ||
a512bbf8 YZ |
3126 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
3127 | u64 chunk_start, u64 physical, u64 devid, | |
3128 | u64 **logical, int *naddrs, int *stripe_len) | |
3129 | { | |
3130 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3131 | struct extent_map *em; | |
3132 | struct map_lookup *map; | |
3133 | u64 *buf; | |
3134 | u64 bytenr; | |
3135 | u64 length; | |
3136 | u64 stripe_nr; | |
3137 | int i, j, nr = 0; | |
3138 | ||
890871be | 3139 | read_lock(&em_tree->lock); |
a512bbf8 | 3140 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 3141 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
3142 | |
3143 | BUG_ON(!em || em->start != chunk_start); | |
3144 | map = (struct map_lookup *)em->bdev; | |
3145 | ||
3146 | length = em->len; | |
3147 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
3148 | do_div(length, map->num_stripes / map->sub_stripes); | |
3149 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3150 | do_div(length, map->num_stripes); | |
3151 | ||
3152 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
3153 | BUG_ON(!buf); | |
3154 | ||
3155 | for (i = 0; i < map->num_stripes; i++) { | |
3156 | if (devid && map->stripes[i].dev->devid != devid) | |
3157 | continue; | |
3158 | if (map->stripes[i].physical > physical || | |
3159 | map->stripes[i].physical + length <= physical) | |
3160 | continue; | |
3161 | ||
3162 | stripe_nr = physical - map->stripes[i].physical; | |
3163 | do_div(stripe_nr, map->stripe_len); | |
3164 | ||
3165 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
3166 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3167 | do_div(stripe_nr, map->sub_stripes); | |
3168 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3169 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3170 | } | |
3171 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 3172 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
3173 | for (j = 0; j < nr; j++) { |
3174 | if (buf[j] == bytenr) | |
3175 | break; | |
3176 | } | |
934d375b CM |
3177 | if (j == nr) { |
3178 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 3179 | buf[nr++] = bytenr; |
934d375b | 3180 | } |
a512bbf8 YZ |
3181 | } |
3182 | ||
a512bbf8 YZ |
3183 | *logical = buf; |
3184 | *naddrs = nr; | |
3185 | *stripe_len = map->stripe_len; | |
3186 | ||
3187 | free_extent_map(em); | |
3188 | return 0; | |
3189 | } | |
3190 | ||
f2d8d74d CM |
3191 | int btrfs_unplug_page(struct btrfs_mapping_tree *map_tree, |
3192 | u64 logical, struct page *page) | |
3193 | { | |
3194 | u64 length = PAGE_CACHE_SIZE; | |
3195 | return __btrfs_map_block(map_tree, READ, logical, &length, | |
3196 | NULL, 0, page); | |
3197 | } | |
3198 | ||
8790d502 | 3199 | static void end_bio_multi_stripe(struct bio *bio, int err) |
8790d502 | 3200 | { |
cea9e445 | 3201 | struct btrfs_multi_bio *multi = bio->bi_private; |
7d2b4daa | 3202 | int is_orig_bio = 0; |
8790d502 | 3203 | |
8790d502 | 3204 | if (err) |
a236aed1 | 3205 | atomic_inc(&multi->error); |
8790d502 | 3206 | |
7d2b4daa CM |
3207 | if (bio == multi->orig_bio) |
3208 | is_orig_bio = 1; | |
3209 | ||
cea9e445 | 3210 | if (atomic_dec_and_test(&multi->stripes_pending)) { |
7d2b4daa CM |
3211 | if (!is_orig_bio) { |
3212 | bio_put(bio); | |
3213 | bio = multi->orig_bio; | |
3214 | } | |
8790d502 CM |
3215 | bio->bi_private = multi->private; |
3216 | bio->bi_end_io = multi->end_io; | |
a236aed1 CM |
3217 | /* only send an error to the higher layers if it is |
3218 | * beyond the tolerance of the multi-bio | |
3219 | */ | |
1259ab75 | 3220 | if (atomic_read(&multi->error) > multi->max_errors) { |
a236aed1 | 3221 | err = -EIO; |
1259ab75 CM |
3222 | } else if (err) { |
3223 | /* | |
3224 | * this bio is actually up to date, we didn't | |
3225 | * go over the max number of errors | |
3226 | */ | |
3227 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 3228 | err = 0; |
1259ab75 | 3229 | } |
8790d502 CM |
3230 | kfree(multi); |
3231 | ||
3232 | bio_endio(bio, err); | |
7d2b4daa | 3233 | } else if (!is_orig_bio) { |
8790d502 CM |
3234 | bio_put(bio); |
3235 | } | |
8790d502 CM |
3236 | } |
3237 | ||
8b712842 CM |
3238 | struct async_sched { |
3239 | struct bio *bio; | |
3240 | int rw; | |
3241 | struct btrfs_fs_info *info; | |
3242 | struct btrfs_work work; | |
3243 | }; | |
3244 | ||
3245 | /* | |
3246 | * see run_scheduled_bios for a description of why bios are collected for | |
3247 | * async submit. | |
3248 | * | |
3249 | * This will add one bio to the pending list for a device and make sure | |
3250 | * the work struct is scheduled. | |
3251 | */ | |
d397712b | 3252 | static noinline int schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
3253 | struct btrfs_device *device, |
3254 | int rw, struct bio *bio) | |
8b712842 CM |
3255 | { |
3256 | int should_queue = 1; | |
ffbd517d | 3257 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
3258 | |
3259 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 3260 | if (!(rw & REQ_WRITE)) { |
492bb6de | 3261 | bio_get(bio); |
8b712842 | 3262 | submit_bio(rw, bio); |
492bb6de | 3263 | bio_put(bio); |
8b712842 CM |
3264 | return 0; |
3265 | } | |
3266 | ||
3267 | /* | |
0986fe9e | 3268 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
3269 | * higher layers. Otherwise, the async bio makes it appear we have |
3270 | * made progress against dirty pages when we've really just put it | |
3271 | * on a queue for later | |
3272 | */ | |
0986fe9e | 3273 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 3274 | WARN_ON(bio->bi_next); |
8b712842 CM |
3275 | bio->bi_next = NULL; |
3276 | bio->bi_rw |= rw; | |
3277 | ||
3278 | spin_lock(&device->io_lock); | |
7b6d91da | 3279 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
3280 | pending_bios = &device->pending_sync_bios; |
3281 | else | |
3282 | pending_bios = &device->pending_bios; | |
8b712842 | 3283 | |
ffbd517d CM |
3284 | if (pending_bios->tail) |
3285 | pending_bios->tail->bi_next = bio; | |
8b712842 | 3286 | |
ffbd517d CM |
3287 | pending_bios->tail = bio; |
3288 | if (!pending_bios->head) | |
3289 | pending_bios->head = bio; | |
8b712842 CM |
3290 | if (device->running_pending) |
3291 | should_queue = 0; | |
3292 | ||
3293 | spin_unlock(&device->io_lock); | |
3294 | ||
3295 | if (should_queue) | |
1cc127b5 CM |
3296 | btrfs_queue_worker(&root->fs_info->submit_workers, |
3297 | &device->work); | |
8b712842 CM |
3298 | return 0; |
3299 | } | |
3300 | ||
f188591e | 3301 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 3302 | int mirror_num, int async_submit) |
0b86a832 CM |
3303 | { |
3304 | struct btrfs_mapping_tree *map_tree; | |
3305 | struct btrfs_device *dev; | |
8790d502 | 3306 | struct bio *first_bio = bio; |
a62b9401 | 3307 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
3308 | u64 length = 0; |
3309 | u64 map_length; | |
cea9e445 | 3310 | struct btrfs_multi_bio *multi = NULL; |
0b86a832 | 3311 | int ret; |
8790d502 CM |
3312 | int dev_nr = 0; |
3313 | int total_devs = 1; | |
0b86a832 | 3314 | |
f2d8d74d | 3315 | length = bio->bi_size; |
0b86a832 CM |
3316 | map_tree = &root->fs_info->mapping_tree; |
3317 | map_length = length; | |
cea9e445 | 3318 | |
f188591e CM |
3319 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &multi, |
3320 | mirror_num); | |
cea9e445 CM |
3321 | BUG_ON(ret); |
3322 | ||
3323 | total_devs = multi->num_stripes; | |
3324 | if (map_length < length) { | |
d397712b CM |
3325 | printk(KERN_CRIT "mapping failed logical %llu bio len %llu " |
3326 | "len %llu\n", (unsigned long long)logical, | |
3327 | (unsigned long long)length, | |
3328 | (unsigned long long)map_length); | |
cea9e445 CM |
3329 | BUG(); |
3330 | } | |
3331 | multi->end_io = first_bio->bi_end_io; | |
3332 | multi->private = first_bio->bi_private; | |
7d2b4daa | 3333 | multi->orig_bio = first_bio; |
cea9e445 CM |
3334 | atomic_set(&multi->stripes_pending, multi->num_stripes); |
3335 | ||
d397712b | 3336 | while (dev_nr < total_devs) { |
8790d502 | 3337 | if (total_devs > 1) { |
8790d502 CM |
3338 | if (dev_nr < total_devs - 1) { |
3339 | bio = bio_clone(first_bio, GFP_NOFS); | |
3340 | BUG_ON(!bio); | |
3341 | } else { | |
3342 | bio = first_bio; | |
3343 | } | |
3344 | bio->bi_private = multi; | |
3345 | bio->bi_end_io = end_bio_multi_stripe; | |
3346 | } | |
cea9e445 CM |
3347 | bio->bi_sector = multi->stripes[dev_nr].physical >> 9; |
3348 | dev = multi->stripes[dev_nr].dev; | |
18e503d6 | 3349 | if (dev && dev->bdev && (rw != WRITE || dev->writeable)) { |
dfe25020 | 3350 | bio->bi_bdev = dev->bdev; |
8b712842 CM |
3351 | if (async_submit) |
3352 | schedule_bio(root, dev, rw, bio); | |
3353 | else | |
3354 | submit_bio(rw, bio); | |
dfe25020 CM |
3355 | } else { |
3356 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
3357 | bio->bi_sector = logical >> 9; | |
dfe25020 | 3358 | bio_endio(bio, -EIO); |
dfe25020 | 3359 | } |
8790d502 CM |
3360 | dev_nr++; |
3361 | } | |
cea9e445 CM |
3362 | if (total_devs == 1) |
3363 | kfree(multi); | |
0b86a832 CM |
3364 | return 0; |
3365 | } | |
3366 | ||
a443755f | 3367 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 3368 | u8 *uuid, u8 *fsid) |
0b86a832 | 3369 | { |
2b82032c YZ |
3370 | struct btrfs_device *device; |
3371 | struct btrfs_fs_devices *cur_devices; | |
3372 | ||
3373 | cur_devices = root->fs_info->fs_devices; | |
3374 | while (cur_devices) { | |
3375 | if (!fsid || | |
3376 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
3377 | device = __find_device(&cur_devices->devices, | |
3378 | devid, uuid); | |
3379 | if (device) | |
3380 | return device; | |
3381 | } | |
3382 | cur_devices = cur_devices->seed; | |
3383 | } | |
3384 | return NULL; | |
0b86a832 CM |
3385 | } |
3386 | ||
dfe25020 CM |
3387 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
3388 | u64 devid, u8 *dev_uuid) | |
3389 | { | |
3390 | struct btrfs_device *device; | |
3391 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
3392 | ||
3393 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 3394 | if (!device) |
3395 | return NULL; | |
dfe25020 CM |
3396 | list_add(&device->dev_list, |
3397 | &fs_devices->devices); | |
dfe25020 CM |
3398 | device->barriers = 1; |
3399 | device->dev_root = root->fs_info->dev_root; | |
3400 | device->devid = devid; | |
8b712842 | 3401 | device->work.func = pending_bios_fn; |
e4404d6e | 3402 | device->fs_devices = fs_devices; |
cd02dca5 | 3403 | device->missing = 1; |
dfe25020 | 3404 | fs_devices->num_devices++; |
cd02dca5 | 3405 | fs_devices->missing_devices++; |
dfe25020 | 3406 | spin_lock_init(&device->io_lock); |
d20f7043 | 3407 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
3408 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
3409 | return device; | |
3410 | } | |
3411 | ||
0b86a832 CM |
3412 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
3413 | struct extent_buffer *leaf, | |
3414 | struct btrfs_chunk *chunk) | |
3415 | { | |
3416 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3417 | struct map_lookup *map; | |
3418 | struct extent_map *em; | |
3419 | u64 logical; | |
3420 | u64 length; | |
3421 | u64 devid; | |
a443755f | 3422 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 3423 | int num_stripes; |
0b86a832 | 3424 | int ret; |
593060d7 | 3425 | int i; |
0b86a832 | 3426 | |
e17cade2 CM |
3427 | logical = key->offset; |
3428 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 3429 | |
890871be | 3430 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 3431 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 3432 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
3433 | |
3434 | /* already mapped? */ | |
3435 | if (em && em->start <= logical && em->start + em->len > logical) { | |
3436 | free_extent_map(em); | |
0b86a832 CM |
3437 | return 0; |
3438 | } else if (em) { | |
3439 | free_extent_map(em); | |
3440 | } | |
0b86a832 | 3441 | |
0b86a832 CM |
3442 | em = alloc_extent_map(GFP_NOFS); |
3443 | if (!em) | |
3444 | return -ENOMEM; | |
593060d7 CM |
3445 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
3446 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
3447 | if (!map) { |
3448 | free_extent_map(em); | |
3449 | return -ENOMEM; | |
3450 | } | |
3451 | ||
3452 | em->bdev = (struct block_device *)map; | |
3453 | em->start = logical; | |
3454 | em->len = length; | |
3455 | em->block_start = 0; | |
c8b97818 | 3456 | em->block_len = em->len; |
0b86a832 | 3457 | |
593060d7 CM |
3458 | map->num_stripes = num_stripes; |
3459 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
3460 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
3461 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
3462 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
3463 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 3464 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
3465 | for (i = 0; i < num_stripes; i++) { |
3466 | map->stripes[i].physical = | |
3467 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
3468 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
3469 | read_extent_buffer(leaf, uuid, (unsigned long) |
3470 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
3471 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
3472 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
3473 | NULL); | |
dfe25020 | 3474 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
3475 | kfree(map); |
3476 | free_extent_map(em); | |
3477 | return -EIO; | |
3478 | } | |
dfe25020 CM |
3479 | if (!map->stripes[i].dev) { |
3480 | map->stripes[i].dev = | |
3481 | add_missing_dev(root, devid, uuid); | |
3482 | if (!map->stripes[i].dev) { | |
3483 | kfree(map); | |
3484 | free_extent_map(em); | |
3485 | return -EIO; | |
3486 | } | |
3487 | } | |
3488 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
3489 | } |
3490 | ||
890871be | 3491 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 3492 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 3493 | write_unlock(&map_tree->map_tree.lock); |
b248a415 | 3494 | BUG_ON(ret); |
0b86a832 CM |
3495 | free_extent_map(em); |
3496 | ||
3497 | return 0; | |
3498 | } | |
3499 | ||
3500 | static int fill_device_from_item(struct extent_buffer *leaf, | |
3501 | struct btrfs_dev_item *dev_item, | |
3502 | struct btrfs_device *device) | |
3503 | { | |
3504 | unsigned long ptr; | |
0b86a832 CM |
3505 | |
3506 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
3507 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
3508 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
3509 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
3510 | device->type = btrfs_device_type(leaf, dev_item); | |
3511 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
3512 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
3513 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
3514 | |
3515 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 3516 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 3517 | |
0b86a832 CM |
3518 | return 0; |
3519 | } | |
3520 | ||
2b82032c YZ |
3521 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
3522 | { | |
3523 | struct btrfs_fs_devices *fs_devices; | |
3524 | int ret; | |
3525 | ||
3526 | mutex_lock(&uuid_mutex); | |
3527 | ||
3528 | fs_devices = root->fs_info->fs_devices->seed; | |
3529 | while (fs_devices) { | |
3530 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
3531 | ret = 0; | |
3532 | goto out; | |
3533 | } | |
3534 | fs_devices = fs_devices->seed; | |
3535 | } | |
3536 | ||
3537 | fs_devices = find_fsid(fsid); | |
3538 | if (!fs_devices) { | |
3539 | ret = -ENOENT; | |
3540 | goto out; | |
3541 | } | |
e4404d6e YZ |
3542 | |
3543 | fs_devices = clone_fs_devices(fs_devices); | |
3544 | if (IS_ERR(fs_devices)) { | |
3545 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
3546 | goto out; |
3547 | } | |
3548 | ||
97288f2c | 3549 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 3550 | root->fs_info->bdev_holder); |
2b82032c YZ |
3551 | if (ret) |
3552 | goto out; | |
3553 | ||
3554 | if (!fs_devices->seeding) { | |
3555 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 3556 | free_fs_devices(fs_devices); |
2b82032c YZ |
3557 | ret = -EINVAL; |
3558 | goto out; | |
3559 | } | |
3560 | ||
3561 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
3562 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c YZ |
3563 | out: |
3564 | mutex_unlock(&uuid_mutex); | |
3565 | return ret; | |
3566 | } | |
3567 | ||
0d81ba5d | 3568 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
3569 | struct extent_buffer *leaf, |
3570 | struct btrfs_dev_item *dev_item) | |
3571 | { | |
3572 | struct btrfs_device *device; | |
3573 | u64 devid; | |
3574 | int ret; | |
2b82032c | 3575 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
3576 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
3577 | ||
0b86a832 | 3578 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
3579 | read_extent_buffer(leaf, dev_uuid, |
3580 | (unsigned long)btrfs_device_uuid(dev_item), | |
3581 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
3582 | read_extent_buffer(leaf, fs_uuid, |
3583 | (unsigned long)btrfs_device_fsid(dev_item), | |
3584 | BTRFS_UUID_SIZE); | |
3585 | ||
3586 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
3587 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 3588 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 3589 | return ret; |
2b82032c YZ |
3590 | } |
3591 | ||
3592 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
3593 | if (!device || !device->bdev) { | |
e4404d6e | 3594 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
3595 | return -EIO; |
3596 | ||
3597 | if (!device) { | |
d397712b CM |
3598 | printk(KERN_WARNING "warning devid %llu missing\n", |
3599 | (unsigned long long)devid); | |
2b82032c YZ |
3600 | device = add_missing_dev(root, devid, dev_uuid); |
3601 | if (!device) | |
3602 | return -ENOMEM; | |
cd02dca5 CM |
3603 | } else if (!device->missing) { |
3604 | /* | |
3605 | * this happens when a device that was properly setup | |
3606 | * in the device info lists suddenly goes bad. | |
3607 | * device->bdev is NULL, and so we have to set | |
3608 | * device->missing to one here | |
3609 | */ | |
3610 | root->fs_info->fs_devices->missing_devices++; | |
3611 | device->missing = 1; | |
2b82032c YZ |
3612 | } |
3613 | } | |
3614 | ||
3615 | if (device->fs_devices != root->fs_info->fs_devices) { | |
3616 | BUG_ON(device->writeable); | |
3617 | if (device->generation != | |
3618 | btrfs_device_generation(leaf, dev_item)) | |
3619 | return -EINVAL; | |
6324fbf3 | 3620 | } |
0b86a832 CM |
3621 | |
3622 | fill_device_from_item(leaf, dev_item, device); | |
3623 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 3624 | device->in_fs_metadata = 1; |
2b82032c YZ |
3625 | if (device->writeable) |
3626 | device->fs_devices->total_rw_bytes += device->total_bytes; | |
0b86a832 | 3627 | ret = 0; |
0b86a832 CM |
3628 | return ret; |
3629 | } | |
3630 | ||
0d81ba5d CM |
3631 | int btrfs_read_super_device(struct btrfs_root *root, struct extent_buffer *buf) |
3632 | { | |
3633 | struct btrfs_dev_item *dev_item; | |
3634 | ||
3635 | dev_item = (struct btrfs_dev_item *)offsetof(struct btrfs_super_block, | |
3636 | dev_item); | |
3637 | return read_one_dev(root, buf, dev_item); | |
3638 | } | |
3639 | ||
e4404d6e | 3640 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 CM |
3641 | { |
3642 | struct btrfs_super_block *super_copy = &root->fs_info->super_copy; | |
a061fc8d | 3643 | struct extent_buffer *sb; |
0b86a832 | 3644 | struct btrfs_disk_key *disk_key; |
0b86a832 | 3645 | struct btrfs_chunk *chunk; |
84eed90f CM |
3646 | u8 *ptr; |
3647 | unsigned long sb_ptr; | |
3648 | int ret = 0; | |
0b86a832 CM |
3649 | u32 num_stripes; |
3650 | u32 array_size; | |
3651 | u32 len = 0; | |
0b86a832 | 3652 | u32 cur; |
84eed90f | 3653 | struct btrfs_key key; |
0b86a832 | 3654 | |
e4404d6e | 3655 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
3656 | BTRFS_SUPER_INFO_SIZE); |
3657 | if (!sb) | |
3658 | return -ENOMEM; | |
3659 | btrfs_set_buffer_uptodate(sb); | |
4008c04a CM |
3660 | btrfs_set_buffer_lockdep_class(sb, 0); |
3661 | ||
a061fc8d | 3662 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
3663 | array_size = btrfs_super_sys_array_size(super_copy); |
3664 | ||
0b86a832 CM |
3665 | ptr = super_copy->sys_chunk_array; |
3666 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
3667 | cur = 0; | |
3668 | ||
3669 | while (cur < array_size) { | |
3670 | disk_key = (struct btrfs_disk_key *)ptr; | |
3671 | btrfs_disk_key_to_cpu(&key, disk_key); | |
3672 | ||
a061fc8d | 3673 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
3674 | sb_ptr += len; |
3675 | cur += len; | |
3676 | ||
0d81ba5d | 3677 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 3678 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 3679 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
3680 | if (ret) |
3681 | break; | |
0b86a832 CM |
3682 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
3683 | len = btrfs_chunk_item_size(num_stripes); | |
3684 | } else { | |
84eed90f CM |
3685 | ret = -EIO; |
3686 | break; | |
0b86a832 CM |
3687 | } |
3688 | ptr += len; | |
3689 | sb_ptr += len; | |
3690 | cur += len; | |
3691 | } | |
a061fc8d | 3692 | free_extent_buffer(sb); |
84eed90f | 3693 | return ret; |
0b86a832 CM |
3694 | } |
3695 | ||
3696 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
3697 | { | |
3698 | struct btrfs_path *path; | |
3699 | struct extent_buffer *leaf; | |
3700 | struct btrfs_key key; | |
3701 | struct btrfs_key found_key; | |
3702 | int ret; | |
3703 | int slot; | |
3704 | ||
3705 | root = root->fs_info->chunk_root; | |
3706 | ||
3707 | path = btrfs_alloc_path(); | |
3708 | if (!path) | |
3709 | return -ENOMEM; | |
3710 | ||
3711 | /* first we search for all of the device items, and then we | |
3712 | * read in all of the chunk items. This way we can create chunk | |
3713 | * mappings that reference all of the devices that are afound | |
3714 | */ | |
3715 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
3716 | key.offset = 0; | |
3717 | key.type = 0; | |
3718 | again: | |
3719 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
3720 | if (ret < 0) |
3721 | goto error; | |
d397712b | 3722 | while (1) { |
0b86a832 CM |
3723 | leaf = path->nodes[0]; |
3724 | slot = path->slots[0]; | |
3725 | if (slot >= btrfs_header_nritems(leaf)) { | |
3726 | ret = btrfs_next_leaf(root, path); | |
3727 | if (ret == 0) | |
3728 | continue; | |
3729 | if (ret < 0) | |
3730 | goto error; | |
3731 | break; | |
3732 | } | |
3733 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3734 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
3735 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
3736 | break; | |
3737 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
3738 | struct btrfs_dev_item *dev_item; | |
3739 | dev_item = btrfs_item_ptr(leaf, slot, | |
3740 | struct btrfs_dev_item); | |
0d81ba5d | 3741 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
3742 | if (ret) |
3743 | goto error; | |
0b86a832 CM |
3744 | } |
3745 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
3746 | struct btrfs_chunk *chunk; | |
3747 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
3748 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
3749 | if (ret) |
3750 | goto error; | |
0b86a832 CM |
3751 | } |
3752 | path->slots[0]++; | |
3753 | } | |
3754 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
3755 | key.objectid = 0; | |
3756 | btrfs_release_path(root, path); | |
3757 | goto again; | |
3758 | } | |
0b86a832 CM |
3759 | ret = 0; |
3760 | error: | |
2b82032c | 3761 | btrfs_free_path(path); |
0b86a832 CM |
3762 | return ret; |
3763 | } |