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 | |
2b82032c YZ |
36 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
37 | struct btrfs_root *root, | |
38 | struct btrfs_device *device); | |
39 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
40 | ||
8a4b83cc CM |
41 | static DEFINE_MUTEX(uuid_mutex); |
42 | static LIST_HEAD(fs_uuids); | |
43 | ||
7d9eb12c CM |
44 | static void lock_chunks(struct btrfs_root *root) |
45 | { | |
7d9eb12c CM |
46 | mutex_lock(&root->fs_info->chunk_mutex); |
47 | } | |
48 | ||
49 | static void unlock_chunks(struct btrfs_root *root) | |
50 | { | |
7d9eb12c CM |
51 | mutex_unlock(&root->fs_info->chunk_mutex); |
52 | } | |
53 | ||
e4404d6e YZ |
54 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
55 | { | |
56 | struct btrfs_device *device; | |
57 | WARN_ON(fs_devices->opened); | |
58 | while (!list_empty(&fs_devices->devices)) { | |
59 | device = list_entry(fs_devices->devices.next, | |
60 | struct btrfs_device, dev_list); | |
61 | list_del(&device->dev_list); | |
62 | kfree(device->name); | |
63 | kfree(device); | |
64 | } | |
65 | kfree(fs_devices); | |
66 | } | |
67 | ||
8a4b83cc CM |
68 | int btrfs_cleanup_fs_uuids(void) |
69 | { | |
70 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 71 | |
2b82032c YZ |
72 | while (!list_empty(&fs_uuids)) { |
73 | fs_devices = list_entry(fs_uuids.next, | |
74 | struct btrfs_fs_devices, list); | |
75 | list_del(&fs_devices->list); | |
e4404d6e | 76 | free_fs_devices(fs_devices); |
8a4b83cc CM |
77 | } |
78 | return 0; | |
79 | } | |
80 | ||
a1b32a59 CM |
81 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
82 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
83 | { |
84 | struct btrfs_device *dev; | |
8a4b83cc | 85 | |
c6e30871 | 86 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 87 | if (dev->devid == devid && |
8f18cf13 | 88 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 89 | return dev; |
a443755f | 90 | } |
8a4b83cc CM |
91 | } |
92 | return NULL; | |
93 | } | |
94 | ||
a1b32a59 | 95 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 96 | { |
8a4b83cc CM |
97 | struct btrfs_fs_devices *fs_devices; |
98 | ||
c6e30871 | 99 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
100 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
101 | return fs_devices; | |
102 | } | |
103 | return NULL; | |
104 | } | |
105 | ||
ffbd517d CM |
106 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
107 | struct bio *head, struct bio *tail) | |
108 | { | |
109 | ||
110 | struct bio *old_head; | |
111 | ||
112 | old_head = pending_bios->head; | |
113 | pending_bios->head = head; | |
114 | if (pending_bios->tail) | |
115 | tail->bi_next = old_head; | |
116 | else | |
117 | pending_bios->tail = tail; | |
118 | } | |
119 | ||
8b712842 CM |
120 | /* |
121 | * we try to collect pending bios for a device so we don't get a large | |
122 | * number of procs sending bios down to the same device. This greatly | |
123 | * improves the schedulers ability to collect and merge the bios. | |
124 | * | |
125 | * But, it also turns into a long list of bios to process and that is sure | |
126 | * to eventually make the worker thread block. The solution here is to | |
127 | * make some progress and then put this work struct back at the end of | |
128 | * the list if the block device is congested. This way, multiple devices | |
129 | * can make progress from a single worker thread. | |
130 | */ | |
d397712b | 131 | static noinline int run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
132 | { |
133 | struct bio *pending; | |
134 | struct backing_dev_info *bdi; | |
b64a2851 | 135 | struct btrfs_fs_info *fs_info; |
ffbd517d | 136 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
137 | struct bio *tail; |
138 | struct bio *cur; | |
139 | int again = 0; | |
ffbd517d | 140 | unsigned long num_run; |
d644d8a1 | 141 | unsigned long batch_run = 0; |
b64a2851 | 142 | unsigned long limit; |
b765ead5 | 143 | unsigned long last_waited = 0; |
d84275c9 | 144 | int force_reg = 0; |
0e588859 | 145 | int sync_pending = 0; |
211588ad CM |
146 | struct blk_plug plug; |
147 | ||
148 | /* | |
149 | * this function runs all the bios we've collected for | |
150 | * a particular device. We don't want to wander off to | |
151 | * another device without first sending all of these down. | |
152 | * So, setup a plug here and finish it off before we return | |
153 | */ | |
154 | blk_start_plug(&plug); | |
8b712842 | 155 | |
bedf762b | 156 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
157 | fs_info = device->dev_root->fs_info; |
158 | limit = btrfs_async_submit_limit(fs_info); | |
159 | limit = limit * 2 / 3; | |
160 | ||
8b712842 CM |
161 | loop: |
162 | spin_lock(&device->io_lock); | |
163 | ||
a6837051 | 164 | loop_lock: |
d84275c9 | 165 | num_run = 0; |
ffbd517d | 166 | |
8b712842 CM |
167 | /* take all the bios off the list at once and process them |
168 | * later on (without the lock held). But, remember the | |
169 | * tail and other pointers so the bios can be properly reinserted | |
170 | * into the list if we hit congestion | |
171 | */ | |
d84275c9 | 172 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 173 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
174 | force_reg = 1; |
175 | } else { | |
ffbd517d | 176 | pending_bios = &device->pending_bios; |
d84275c9 CM |
177 | force_reg = 0; |
178 | } | |
ffbd517d CM |
179 | |
180 | pending = pending_bios->head; | |
181 | tail = pending_bios->tail; | |
8b712842 | 182 | WARN_ON(pending && !tail); |
8b712842 CM |
183 | |
184 | /* | |
185 | * if pending was null this time around, no bios need processing | |
186 | * at all and we can stop. Otherwise it'll loop back up again | |
187 | * and do an additional check so no bios are missed. | |
188 | * | |
189 | * device->running_pending is used to synchronize with the | |
190 | * schedule_bio code. | |
191 | */ | |
ffbd517d CM |
192 | if (device->pending_sync_bios.head == NULL && |
193 | device->pending_bios.head == NULL) { | |
8b712842 CM |
194 | again = 0; |
195 | device->running_pending = 0; | |
ffbd517d CM |
196 | } else { |
197 | again = 1; | |
198 | device->running_pending = 1; | |
8b712842 | 199 | } |
ffbd517d CM |
200 | |
201 | pending_bios->head = NULL; | |
202 | pending_bios->tail = NULL; | |
203 | ||
8b712842 CM |
204 | spin_unlock(&device->io_lock); |
205 | ||
d397712b | 206 | while (pending) { |
ffbd517d CM |
207 | |
208 | rmb(); | |
d84275c9 CM |
209 | /* we want to work on both lists, but do more bios on the |
210 | * sync list than the regular list | |
211 | */ | |
212 | if ((num_run > 32 && | |
213 | pending_bios != &device->pending_sync_bios && | |
214 | device->pending_sync_bios.head) || | |
215 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
216 | device->pending_bios.head)) { | |
ffbd517d CM |
217 | spin_lock(&device->io_lock); |
218 | requeue_list(pending_bios, pending, tail); | |
219 | goto loop_lock; | |
220 | } | |
221 | ||
8b712842 CM |
222 | cur = pending; |
223 | pending = pending->bi_next; | |
224 | cur->bi_next = NULL; | |
b64a2851 CM |
225 | atomic_dec(&fs_info->nr_async_bios); |
226 | ||
227 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
228 | waitqueue_active(&fs_info->async_submit_wait)) | |
229 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
230 | |
231 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 232 | |
2ab1ba68 CM |
233 | /* |
234 | * if we're doing the sync list, record that our | |
235 | * plug has some sync requests on it | |
236 | * | |
237 | * If we're doing the regular list and there are | |
238 | * sync requests sitting around, unplug before | |
239 | * we add more | |
240 | */ | |
241 | if (pending_bios == &device->pending_sync_bios) { | |
242 | sync_pending = 1; | |
243 | } else if (sync_pending) { | |
244 | blk_finish_plug(&plug); | |
245 | blk_start_plug(&plug); | |
246 | sync_pending = 0; | |
247 | } | |
248 | ||
5ff7ba3a CM |
249 | submit_bio(cur->bi_rw, cur); |
250 | num_run++; | |
251 | batch_run++; | |
7eaceacc | 252 | if (need_resched()) |
ffbd517d | 253 | cond_resched(); |
8b712842 CM |
254 | |
255 | /* | |
256 | * we made progress, there is more work to do and the bdi | |
257 | * is now congested. Back off and let other work structs | |
258 | * run instead | |
259 | */ | |
57fd5a5f | 260 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 261 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 262 | struct io_context *ioc; |
8b712842 | 263 | |
b765ead5 CM |
264 | ioc = current->io_context; |
265 | ||
266 | /* | |
267 | * the main goal here is that we don't want to | |
268 | * block if we're going to be able to submit | |
269 | * more requests without blocking. | |
270 | * | |
271 | * This code does two great things, it pokes into | |
272 | * the elevator code from a filesystem _and_ | |
273 | * it makes assumptions about how batching works. | |
274 | */ | |
275 | if (ioc && ioc->nr_batch_requests > 0 && | |
276 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
277 | (last_waited == 0 || | |
278 | ioc->last_waited == last_waited)) { | |
279 | /* | |
280 | * we want to go through our batch of | |
281 | * requests and stop. So, we copy out | |
282 | * the ioc->last_waited time and test | |
283 | * against it before looping | |
284 | */ | |
285 | last_waited = ioc->last_waited; | |
7eaceacc | 286 | if (need_resched()) |
ffbd517d | 287 | cond_resched(); |
b765ead5 CM |
288 | continue; |
289 | } | |
8b712842 | 290 | spin_lock(&device->io_lock); |
ffbd517d | 291 | requeue_list(pending_bios, pending, tail); |
a6837051 | 292 | device->running_pending = 1; |
8b712842 CM |
293 | |
294 | spin_unlock(&device->io_lock); | |
295 | btrfs_requeue_work(&device->work); | |
296 | goto done; | |
297 | } | |
298 | } | |
ffbd517d | 299 | |
51684082 CM |
300 | cond_resched(); |
301 | if (again) | |
302 | goto loop; | |
303 | ||
304 | spin_lock(&device->io_lock); | |
305 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
306 | goto loop_lock; | |
307 | spin_unlock(&device->io_lock); | |
308 | ||
8b712842 | 309 | done: |
211588ad | 310 | blk_finish_plug(&plug); |
8b712842 CM |
311 | return 0; |
312 | } | |
313 | ||
b2950863 | 314 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
315 | { |
316 | struct btrfs_device *device; | |
317 | ||
318 | device = container_of(work, struct btrfs_device, work); | |
319 | run_scheduled_bios(device); | |
320 | } | |
321 | ||
a1b32a59 | 322 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
323 | struct btrfs_super_block *disk_super, |
324 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
325 | { | |
326 | struct btrfs_device *device; | |
327 | struct btrfs_fs_devices *fs_devices; | |
328 | u64 found_transid = btrfs_super_generation(disk_super); | |
3a0524dc | 329 | char *name; |
8a4b83cc CM |
330 | |
331 | fs_devices = find_fsid(disk_super->fsid); | |
332 | if (!fs_devices) { | |
515dc322 | 333 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
334 | if (!fs_devices) |
335 | return -ENOMEM; | |
336 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 337 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
338 | list_add(&fs_devices->list, &fs_uuids); |
339 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
340 | fs_devices->latest_devid = devid; | |
341 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 342 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
343 | device = NULL; |
344 | } else { | |
a443755f CM |
345 | device = __find_device(&fs_devices->devices, devid, |
346 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
347 | } |
348 | if (!device) { | |
2b82032c YZ |
349 | if (fs_devices->opened) |
350 | return -EBUSY; | |
351 | ||
8a4b83cc CM |
352 | device = kzalloc(sizeof(*device), GFP_NOFS); |
353 | if (!device) { | |
354 | /* we can safely leave the fs_devices entry around */ | |
355 | return -ENOMEM; | |
356 | } | |
357 | device->devid = devid; | |
8b712842 | 358 | device->work.func = pending_bios_fn; |
a443755f CM |
359 | memcpy(device->uuid, disk_super->dev_item.uuid, |
360 | BTRFS_UUID_SIZE); | |
b248a415 | 361 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
362 | device->name = kstrdup(path, GFP_NOFS); |
363 | if (!device->name) { | |
364 | kfree(device); | |
365 | return -ENOMEM; | |
366 | } | |
2b82032c | 367 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 368 | |
90519d66 AJ |
369 | /* init readahead state */ |
370 | spin_lock_init(&device->reada_lock); | |
371 | device->reada_curr_zone = NULL; | |
372 | atomic_set(&device->reada_in_flight, 0); | |
373 | device->reada_next = 0; | |
374 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
375 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
376 | ||
e5e9a520 | 377 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 378 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
379 | mutex_unlock(&fs_devices->device_list_mutex); |
380 | ||
2b82032c | 381 | device->fs_devices = fs_devices; |
8a4b83cc | 382 | fs_devices->num_devices++; |
cd02dca5 | 383 | } else if (!device->name || strcmp(device->name, path)) { |
3a0524dc TH |
384 | name = kstrdup(path, GFP_NOFS); |
385 | if (!name) | |
386 | return -ENOMEM; | |
387 | kfree(device->name); | |
388 | device->name = name; | |
cd02dca5 CM |
389 | if (device->missing) { |
390 | fs_devices->missing_devices--; | |
391 | device->missing = 0; | |
392 | } | |
8a4b83cc CM |
393 | } |
394 | ||
395 | if (found_transid > fs_devices->latest_trans) { | |
396 | fs_devices->latest_devid = devid; | |
397 | fs_devices->latest_trans = found_transid; | |
398 | } | |
8a4b83cc CM |
399 | *fs_devices_ret = fs_devices; |
400 | return 0; | |
401 | } | |
402 | ||
e4404d6e YZ |
403 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
404 | { | |
405 | struct btrfs_fs_devices *fs_devices; | |
406 | struct btrfs_device *device; | |
407 | struct btrfs_device *orig_dev; | |
408 | ||
409 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
410 | if (!fs_devices) | |
411 | return ERR_PTR(-ENOMEM); | |
412 | ||
413 | INIT_LIST_HEAD(&fs_devices->devices); | |
414 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
415 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 416 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
417 | fs_devices->latest_devid = orig->latest_devid; |
418 | fs_devices->latest_trans = orig->latest_trans; | |
419 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); | |
420 | ||
46224705 | 421 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e YZ |
422 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
423 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
424 | if (!device) | |
425 | goto error; | |
426 | ||
427 | device->name = kstrdup(orig_dev->name, GFP_NOFS); | |
fd2696f3 JL |
428 | if (!device->name) { |
429 | kfree(device); | |
e4404d6e | 430 | goto error; |
fd2696f3 | 431 | } |
e4404d6e YZ |
432 | |
433 | device->devid = orig_dev->devid; | |
434 | device->work.func = pending_bios_fn; | |
435 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
436 | spin_lock_init(&device->io_lock); |
437 | INIT_LIST_HEAD(&device->dev_list); | |
438 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
439 | ||
440 | list_add(&device->dev_list, &fs_devices->devices); | |
441 | device->fs_devices = fs_devices; | |
442 | fs_devices->num_devices++; | |
443 | } | |
444 | return fs_devices; | |
445 | error: | |
446 | free_fs_devices(fs_devices); | |
447 | return ERR_PTR(-ENOMEM); | |
448 | } | |
449 | ||
dfe25020 CM |
450 | int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
451 | { | |
c6e30871 | 452 | struct btrfs_device *device, *next; |
dfe25020 CM |
453 | |
454 | mutex_lock(&uuid_mutex); | |
455 | again: | |
46224705 | 456 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 457 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
2b82032c YZ |
458 | if (device->in_fs_metadata) |
459 | continue; | |
460 | ||
461 | if (device->bdev) { | |
d4d77629 | 462 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
463 | device->bdev = NULL; |
464 | fs_devices->open_devices--; | |
465 | } | |
466 | if (device->writeable) { | |
467 | list_del_init(&device->dev_alloc_list); | |
468 | device->writeable = 0; | |
469 | fs_devices->rw_devices--; | |
470 | } | |
e4404d6e YZ |
471 | list_del_init(&device->dev_list); |
472 | fs_devices->num_devices--; | |
473 | kfree(device->name); | |
474 | kfree(device); | |
dfe25020 | 475 | } |
2b82032c YZ |
476 | |
477 | if (fs_devices->seed) { | |
478 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
479 | goto again; |
480 | } | |
481 | ||
dfe25020 CM |
482 | mutex_unlock(&uuid_mutex); |
483 | return 0; | |
484 | } | |
a0af469b | 485 | |
1f78160c XG |
486 | static void __free_device(struct work_struct *work) |
487 | { | |
488 | struct btrfs_device *device; | |
489 | ||
490 | device = container_of(work, struct btrfs_device, rcu_work); | |
491 | ||
492 | if (device->bdev) | |
493 | blkdev_put(device->bdev, device->mode); | |
494 | ||
495 | kfree(device->name); | |
496 | kfree(device); | |
497 | } | |
498 | ||
499 | static void free_device(struct rcu_head *head) | |
500 | { | |
501 | struct btrfs_device *device; | |
502 | ||
503 | device = container_of(head, struct btrfs_device, rcu); | |
504 | ||
505 | INIT_WORK(&device->rcu_work, __free_device); | |
506 | schedule_work(&device->rcu_work); | |
507 | } | |
508 | ||
2b82032c | 509 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 510 | { |
8a4b83cc | 511 | struct btrfs_device *device; |
e4404d6e | 512 | |
2b82032c YZ |
513 | if (--fs_devices->opened > 0) |
514 | return 0; | |
8a4b83cc | 515 | |
c9513edb | 516 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 517 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c XG |
518 | struct btrfs_device *new_device; |
519 | ||
520 | if (device->bdev) | |
a0af469b | 521 | fs_devices->open_devices--; |
1f78160c | 522 | |
2b82032c YZ |
523 | if (device->writeable) { |
524 | list_del_init(&device->dev_alloc_list); | |
525 | fs_devices->rw_devices--; | |
526 | } | |
527 | ||
d5e2003c JB |
528 | if (device->can_discard) |
529 | fs_devices->num_can_discard--; | |
530 | ||
1f78160c XG |
531 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
532 | BUG_ON(!new_device); | |
533 | memcpy(new_device, device, sizeof(*new_device)); | |
534 | new_device->name = kstrdup(device->name, GFP_NOFS); | |
5f3f302a | 535 | BUG_ON(device->name && !new_device->name); |
1f78160c XG |
536 | new_device->bdev = NULL; |
537 | new_device->writeable = 0; | |
538 | new_device->in_fs_metadata = 0; | |
d5e2003c | 539 | new_device->can_discard = 0; |
1f78160c XG |
540 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
541 | ||
542 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 543 | } |
c9513edb XG |
544 | mutex_unlock(&fs_devices->device_list_mutex); |
545 | ||
e4404d6e YZ |
546 | WARN_ON(fs_devices->open_devices); |
547 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
548 | fs_devices->opened = 0; |
549 | fs_devices->seeding = 0; | |
2b82032c | 550 | |
8a4b83cc CM |
551 | return 0; |
552 | } | |
553 | ||
2b82032c YZ |
554 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
555 | { | |
e4404d6e | 556 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
557 | int ret; |
558 | ||
559 | mutex_lock(&uuid_mutex); | |
560 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
561 | if (!fs_devices->opened) { |
562 | seed_devices = fs_devices->seed; | |
563 | fs_devices->seed = NULL; | |
564 | } | |
2b82032c | 565 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
566 | |
567 | while (seed_devices) { | |
568 | fs_devices = seed_devices; | |
569 | seed_devices = fs_devices->seed; | |
570 | __btrfs_close_devices(fs_devices); | |
571 | free_fs_devices(fs_devices); | |
572 | } | |
2b82032c YZ |
573 | return ret; |
574 | } | |
575 | ||
e4404d6e YZ |
576 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
577 | fmode_t flags, void *holder) | |
8a4b83cc | 578 | { |
d5e2003c | 579 | struct request_queue *q; |
8a4b83cc CM |
580 | struct block_device *bdev; |
581 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 582 | struct btrfs_device *device; |
a0af469b CM |
583 | struct block_device *latest_bdev = NULL; |
584 | struct buffer_head *bh; | |
585 | struct btrfs_super_block *disk_super; | |
586 | u64 latest_devid = 0; | |
587 | u64 latest_transid = 0; | |
a0af469b | 588 | u64 devid; |
2b82032c | 589 | int seeding = 1; |
a0af469b | 590 | int ret = 0; |
8a4b83cc | 591 | |
d4d77629 TH |
592 | flags |= FMODE_EXCL; |
593 | ||
c6e30871 | 594 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
595 | if (device->bdev) |
596 | continue; | |
dfe25020 CM |
597 | if (!device->name) |
598 | continue; | |
599 | ||
d4d77629 | 600 | bdev = blkdev_get_by_path(device->name, flags, holder); |
8a4b83cc | 601 | if (IS_ERR(bdev)) { |
d397712b | 602 | printk(KERN_INFO "open %s failed\n", device->name); |
a0af469b | 603 | goto error; |
8a4b83cc | 604 | } |
a061fc8d | 605 | set_blocksize(bdev, 4096); |
a0af469b | 606 | |
a512bbf8 | 607 | bh = btrfs_read_dev_super(bdev); |
20bcd649 | 608 | if (!bh) |
a0af469b CM |
609 | goto error_close; |
610 | ||
611 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 612 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
613 | if (devid != device->devid) |
614 | goto error_brelse; | |
615 | ||
2b82032c YZ |
616 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
617 | BTRFS_UUID_SIZE)) | |
618 | goto error_brelse; | |
619 | ||
620 | device->generation = btrfs_super_generation(disk_super); | |
621 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 622 | latest_devid = devid; |
2b82032c | 623 | latest_transid = device->generation; |
a0af469b CM |
624 | latest_bdev = bdev; |
625 | } | |
626 | ||
2b82032c YZ |
627 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
628 | device->writeable = 0; | |
629 | } else { | |
630 | device->writeable = !bdev_read_only(bdev); | |
631 | seeding = 0; | |
632 | } | |
633 | ||
d5e2003c JB |
634 | q = bdev_get_queue(bdev); |
635 | if (blk_queue_discard(q)) { | |
636 | device->can_discard = 1; | |
637 | fs_devices->num_can_discard++; | |
638 | } | |
639 | ||
8a4b83cc | 640 | device->bdev = bdev; |
dfe25020 | 641 | device->in_fs_metadata = 0; |
15916de8 CM |
642 | device->mode = flags; |
643 | ||
c289811c CM |
644 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
645 | fs_devices->rotating = 1; | |
646 | ||
a0af469b | 647 | fs_devices->open_devices++; |
2b82032c YZ |
648 | if (device->writeable) { |
649 | fs_devices->rw_devices++; | |
650 | list_add(&device->dev_alloc_list, | |
651 | &fs_devices->alloc_list); | |
652 | } | |
4f6c9328 | 653 | brelse(bh); |
a0af469b | 654 | continue; |
a061fc8d | 655 | |
a0af469b CM |
656 | error_brelse: |
657 | brelse(bh); | |
658 | error_close: | |
d4d77629 | 659 | blkdev_put(bdev, flags); |
a0af469b CM |
660 | error: |
661 | continue; | |
8a4b83cc | 662 | } |
a0af469b | 663 | if (fs_devices->open_devices == 0) { |
20bcd649 | 664 | ret = -EINVAL; |
a0af469b CM |
665 | goto out; |
666 | } | |
2b82032c YZ |
667 | fs_devices->seeding = seeding; |
668 | fs_devices->opened = 1; | |
a0af469b CM |
669 | fs_devices->latest_bdev = latest_bdev; |
670 | fs_devices->latest_devid = latest_devid; | |
671 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 672 | fs_devices->total_rw_bytes = 0; |
a0af469b | 673 | out: |
2b82032c YZ |
674 | return ret; |
675 | } | |
676 | ||
677 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 678 | fmode_t flags, void *holder) |
2b82032c YZ |
679 | { |
680 | int ret; | |
681 | ||
682 | mutex_lock(&uuid_mutex); | |
683 | if (fs_devices->opened) { | |
e4404d6e YZ |
684 | fs_devices->opened++; |
685 | ret = 0; | |
2b82032c | 686 | } else { |
15916de8 | 687 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 688 | } |
8a4b83cc | 689 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
690 | return ret; |
691 | } | |
692 | ||
97288f2c | 693 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
694 | struct btrfs_fs_devices **fs_devices_ret) |
695 | { | |
696 | struct btrfs_super_block *disk_super; | |
697 | struct block_device *bdev; | |
698 | struct buffer_head *bh; | |
699 | int ret; | |
700 | u64 devid; | |
f2984462 | 701 | u64 transid; |
8a4b83cc CM |
702 | |
703 | mutex_lock(&uuid_mutex); | |
704 | ||
d4d77629 TH |
705 | flags |= FMODE_EXCL; |
706 | bdev = blkdev_get_by_path(path, flags, holder); | |
8a4b83cc CM |
707 | |
708 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
709 | ret = PTR_ERR(bdev); |
710 | goto error; | |
711 | } | |
712 | ||
713 | ret = set_blocksize(bdev, 4096); | |
714 | if (ret) | |
715 | goto error_close; | |
a512bbf8 | 716 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 717 | if (!bh) { |
20b45077 | 718 | ret = -EINVAL; |
8a4b83cc CM |
719 | goto error_close; |
720 | } | |
721 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 722 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 723 | transid = btrfs_super_generation(disk_super); |
7ae9c09d | 724 | if (disk_super->label[0]) |
d397712b | 725 | printk(KERN_INFO "device label %s ", disk_super->label); |
22b63a29 ID |
726 | else |
727 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); | |
119e10cf | 728 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 729 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc CM |
730 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
731 | ||
8a4b83cc CM |
732 | brelse(bh); |
733 | error_close: | |
d4d77629 | 734 | blkdev_put(bdev, flags); |
8a4b83cc CM |
735 | error: |
736 | mutex_unlock(&uuid_mutex); | |
737 | return ret; | |
738 | } | |
0b86a832 | 739 | |
6d07bcec MX |
740 | /* helper to account the used device space in the range */ |
741 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
742 | u64 end, u64 *length) | |
743 | { | |
744 | struct btrfs_key key; | |
745 | struct btrfs_root *root = device->dev_root; | |
746 | struct btrfs_dev_extent *dev_extent; | |
747 | struct btrfs_path *path; | |
748 | u64 extent_end; | |
749 | int ret; | |
750 | int slot; | |
751 | struct extent_buffer *l; | |
752 | ||
753 | *length = 0; | |
754 | ||
755 | if (start >= device->total_bytes) | |
756 | return 0; | |
757 | ||
758 | path = btrfs_alloc_path(); | |
759 | if (!path) | |
760 | return -ENOMEM; | |
761 | path->reada = 2; | |
762 | ||
763 | key.objectid = device->devid; | |
764 | key.offset = start; | |
765 | key.type = BTRFS_DEV_EXTENT_KEY; | |
766 | ||
767 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
768 | if (ret < 0) | |
769 | goto out; | |
770 | if (ret > 0) { | |
771 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
772 | if (ret < 0) | |
773 | goto out; | |
774 | } | |
775 | ||
776 | while (1) { | |
777 | l = path->nodes[0]; | |
778 | slot = path->slots[0]; | |
779 | if (slot >= btrfs_header_nritems(l)) { | |
780 | ret = btrfs_next_leaf(root, path); | |
781 | if (ret == 0) | |
782 | continue; | |
783 | if (ret < 0) | |
784 | goto out; | |
785 | ||
786 | break; | |
787 | } | |
788 | btrfs_item_key_to_cpu(l, &key, slot); | |
789 | ||
790 | if (key.objectid < device->devid) | |
791 | goto next; | |
792 | ||
793 | if (key.objectid > device->devid) | |
794 | break; | |
795 | ||
796 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
797 | goto next; | |
798 | ||
799 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
800 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
801 | dev_extent); | |
802 | if (key.offset <= start && extent_end > end) { | |
803 | *length = end - start + 1; | |
804 | break; | |
805 | } else if (key.offset <= start && extent_end > start) | |
806 | *length += extent_end - start; | |
807 | else if (key.offset > start && extent_end <= end) | |
808 | *length += extent_end - key.offset; | |
809 | else if (key.offset > start && key.offset <= end) { | |
810 | *length += end - key.offset + 1; | |
811 | break; | |
812 | } else if (key.offset > end) | |
813 | break; | |
814 | ||
815 | next: | |
816 | path->slots[0]++; | |
817 | } | |
818 | ret = 0; | |
819 | out: | |
820 | btrfs_free_path(path); | |
821 | return ret; | |
822 | } | |
823 | ||
0b86a832 | 824 | /* |
7bfc837d MX |
825 | * find_free_dev_extent - find free space in the specified device |
826 | * @trans: transaction handler | |
827 | * @device: the device which we search the free space in | |
828 | * @num_bytes: the size of the free space that we need | |
829 | * @start: store the start of the free space. | |
830 | * @len: the size of the free space. that we find, or the size of the max | |
831 | * free space if we don't find suitable free space | |
832 | * | |
0b86a832 CM |
833 | * this uses a pretty simple search, the expectation is that it is |
834 | * called very infrequently and that a given device has a small number | |
835 | * of extents | |
7bfc837d MX |
836 | * |
837 | * @start is used to store the start of the free space if we find. But if we | |
838 | * don't find suitable free space, it will be used to store the start position | |
839 | * of the max free space. | |
840 | * | |
841 | * @len is used to store the size of the free space that we find. | |
842 | * But if we don't find suitable free space, it is used to store the size of | |
843 | * the max free space. | |
0b86a832 | 844 | */ |
ba1bf481 JB |
845 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
846 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 847 | u64 *start, u64 *len) |
0b86a832 CM |
848 | { |
849 | struct btrfs_key key; | |
850 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 851 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 852 | struct btrfs_path *path; |
7bfc837d MX |
853 | u64 hole_size; |
854 | u64 max_hole_start; | |
855 | u64 max_hole_size; | |
856 | u64 extent_end; | |
857 | u64 search_start; | |
0b86a832 CM |
858 | u64 search_end = device->total_bytes; |
859 | int ret; | |
7bfc837d | 860 | int slot; |
0b86a832 CM |
861 | struct extent_buffer *l; |
862 | ||
0b86a832 CM |
863 | /* FIXME use last free of some kind */ |
864 | ||
8a4b83cc CM |
865 | /* we don't want to overwrite the superblock on the drive, |
866 | * so we make sure to start at an offset of at least 1MB | |
867 | */ | |
a9c9bf68 | 868 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 869 | |
7bfc837d MX |
870 | max_hole_start = search_start; |
871 | max_hole_size = 0; | |
38c01b96 | 872 | hole_size = 0; |
7bfc837d MX |
873 | |
874 | if (search_start >= search_end) { | |
875 | ret = -ENOSPC; | |
876 | goto error; | |
877 | } | |
878 | ||
879 | path = btrfs_alloc_path(); | |
880 | if (!path) { | |
881 | ret = -ENOMEM; | |
882 | goto error; | |
883 | } | |
884 | path->reada = 2; | |
885 | ||
0b86a832 CM |
886 | key.objectid = device->devid; |
887 | key.offset = search_start; | |
888 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 889 | |
0b86a832 CM |
890 | ret = btrfs_search_slot(trans, root, &key, path, 0, 0); |
891 | if (ret < 0) | |
7bfc837d | 892 | goto out; |
1fcbac58 YZ |
893 | if (ret > 0) { |
894 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
895 | if (ret < 0) | |
7bfc837d | 896 | goto out; |
1fcbac58 | 897 | } |
7bfc837d | 898 | |
0b86a832 CM |
899 | while (1) { |
900 | l = path->nodes[0]; | |
901 | slot = path->slots[0]; | |
902 | if (slot >= btrfs_header_nritems(l)) { | |
903 | ret = btrfs_next_leaf(root, path); | |
904 | if (ret == 0) | |
905 | continue; | |
906 | if (ret < 0) | |
7bfc837d MX |
907 | goto out; |
908 | ||
909 | break; | |
0b86a832 CM |
910 | } |
911 | btrfs_item_key_to_cpu(l, &key, slot); | |
912 | ||
913 | if (key.objectid < device->devid) | |
914 | goto next; | |
915 | ||
916 | if (key.objectid > device->devid) | |
7bfc837d | 917 | break; |
0b86a832 | 918 | |
7bfc837d MX |
919 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
920 | goto next; | |
9779b72f | 921 | |
7bfc837d MX |
922 | if (key.offset > search_start) { |
923 | hole_size = key.offset - search_start; | |
9779b72f | 924 | |
7bfc837d MX |
925 | if (hole_size > max_hole_size) { |
926 | max_hole_start = search_start; | |
927 | max_hole_size = hole_size; | |
928 | } | |
9779b72f | 929 | |
7bfc837d MX |
930 | /* |
931 | * If this free space is greater than which we need, | |
932 | * it must be the max free space that we have found | |
933 | * until now, so max_hole_start must point to the start | |
934 | * of this free space and the length of this free space | |
935 | * is stored in max_hole_size. Thus, we return | |
936 | * max_hole_start and max_hole_size and go back to the | |
937 | * caller. | |
938 | */ | |
939 | if (hole_size >= num_bytes) { | |
940 | ret = 0; | |
941 | goto out; | |
0b86a832 CM |
942 | } |
943 | } | |
0b86a832 | 944 | |
0b86a832 | 945 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
946 | extent_end = key.offset + btrfs_dev_extent_length(l, |
947 | dev_extent); | |
948 | if (extent_end > search_start) | |
949 | search_start = extent_end; | |
0b86a832 CM |
950 | next: |
951 | path->slots[0]++; | |
952 | cond_resched(); | |
953 | } | |
0b86a832 | 954 | |
38c01b96 | 955 | /* |
956 | * At this point, search_start should be the end of | |
957 | * allocated dev extents, and when shrinking the device, | |
958 | * search_end may be smaller than search_start. | |
959 | */ | |
960 | if (search_end > search_start) | |
961 | hole_size = search_end - search_start; | |
962 | ||
7bfc837d MX |
963 | if (hole_size > max_hole_size) { |
964 | max_hole_start = search_start; | |
965 | max_hole_size = hole_size; | |
0b86a832 | 966 | } |
0b86a832 | 967 | |
7bfc837d MX |
968 | /* See above. */ |
969 | if (hole_size < num_bytes) | |
970 | ret = -ENOSPC; | |
971 | else | |
972 | ret = 0; | |
973 | ||
974 | out: | |
2b82032c | 975 | btrfs_free_path(path); |
7bfc837d MX |
976 | error: |
977 | *start = max_hole_start; | |
b2117a39 | 978 | if (len) |
7bfc837d | 979 | *len = max_hole_size; |
0b86a832 CM |
980 | return ret; |
981 | } | |
982 | ||
b2950863 | 983 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
984 | struct btrfs_device *device, |
985 | u64 start) | |
986 | { | |
987 | int ret; | |
988 | struct btrfs_path *path; | |
989 | struct btrfs_root *root = device->dev_root; | |
990 | struct btrfs_key key; | |
a061fc8d CM |
991 | struct btrfs_key found_key; |
992 | struct extent_buffer *leaf = NULL; | |
993 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
994 | |
995 | path = btrfs_alloc_path(); | |
996 | if (!path) | |
997 | return -ENOMEM; | |
998 | ||
999 | key.objectid = device->devid; | |
1000 | key.offset = start; | |
1001 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1002 | again: |
8f18cf13 | 1003 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1004 | if (ret > 0) { |
1005 | ret = btrfs_previous_item(root, path, key.objectid, | |
1006 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1007 | if (ret) |
1008 | goto out; | |
a061fc8d CM |
1009 | leaf = path->nodes[0]; |
1010 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1011 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1012 | struct btrfs_dev_extent); | |
1013 | BUG_ON(found_key.offset > start || found_key.offset + | |
1014 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1015 | key = found_key; |
1016 | btrfs_release_path(path); | |
1017 | goto again; | |
a061fc8d CM |
1018 | } else if (ret == 0) { |
1019 | leaf = path->nodes[0]; | |
1020 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1021 | struct btrfs_dev_extent); | |
1022 | } | |
8f18cf13 CM |
1023 | BUG_ON(ret); |
1024 | ||
2bf64758 JB |
1025 | if (device->bytes_used > 0) { |
1026 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1027 | device->bytes_used -= len; | |
1028 | spin_lock(&root->fs_info->free_chunk_lock); | |
1029 | root->fs_info->free_chunk_space += len; | |
1030 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1031 | } | |
8f18cf13 | 1032 | ret = btrfs_del_item(trans, root, path); |
8f18cf13 | 1033 | |
b0b802d7 | 1034 | out: |
8f18cf13 CM |
1035 | btrfs_free_path(path); |
1036 | return ret; | |
1037 | } | |
1038 | ||
2b82032c | 1039 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1040 | struct btrfs_device *device, |
e17cade2 | 1041 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1042 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1043 | { |
1044 | int ret; | |
1045 | struct btrfs_path *path; | |
1046 | struct btrfs_root *root = device->dev_root; | |
1047 | struct btrfs_dev_extent *extent; | |
1048 | struct extent_buffer *leaf; | |
1049 | struct btrfs_key key; | |
1050 | ||
dfe25020 | 1051 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1052 | path = btrfs_alloc_path(); |
1053 | if (!path) | |
1054 | return -ENOMEM; | |
1055 | ||
0b86a832 | 1056 | key.objectid = device->devid; |
2b82032c | 1057 | key.offset = start; |
0b86a832 CM |
1058 | key.type = BTRFS_DEV_EXTENT_KEY; |
1059 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1060 | sizeof(*extent)); | |
1061 | BUG_ON(ret); | |
1062 | ||
1063 | leaf = path->nodes[0]; | |
1064 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1065 | struct btrfs_dev_extent); | |
e17cade2 CM |
1066 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1067 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1068 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1069 | ||
1070 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1071 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1072 | BTRFS_UUID_SIZE); | |
1073 | ||
0b86a832 CM |
1074 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1075 | btrfs_mark_buffer_dirty(leaf); | |
0b86a832 CM |
1076 | btrfs_free_path(path); |
1077 | return ret; | |
1078 | } | |
1079 | ||
a1b32a59 CM |
1080 | static noinline int find_next_chunk(struct btrfs_root *root, |
1081 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1082 | { |
1083 | struct btrfs_path *path; | |
1084 | int ret; | |
1085 | struct btrfs_key key; | |
e17cade2 | 1086 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1087 | struct btrfs_key found_key; |
1088 | ||
1089 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1090 | if (!path) |
1091 | return -ENOMEM; | |
0b86a832 | 1092 | |
e17cade2 | 1093 | key.objectid = objectid; |
0b86a832 CM |
1094 | key.offset = (u64)-1; |
1095 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1096 | ||
1097 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1098 | if (ret < 0) | |
1099 | goto error; | |
1100 | ||
1101 | BUG_ON(ret == 0); | |
1102 | ||
1103 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1104 | if (ret) { | |
e17cade2 | 1105 | *offset = 0; |
0b86a832 CM |
1106 | } else { |
1107 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1108 | path->slots[0]); | |
e17cade2 CM |
1109 | if (found_key.objectid != objectid) |
1110 | *offset = 0; | |
1111 | else { | |
1112 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1113 | struct btrfs_chunk); | |
1114 | *offset = found_key.offset + | |
1115 | btrfs_chunk_length(path->nodes[0], chunk); | |
1116 | } | |
0b86a832 CM |
1117 | } |
1118 | ret = 0; | |
1119 | error: | |
1120 | btrfs_free_path(path); | |
1121 | return ret; | |
1122 | } | |
1123 | ||
2b82032c | 1124 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1125 | { |
1126 | int ret; | |
1127 | struct btrfs_key key; | |
1128 | struct btrfs_key found_key; | |
2b82032c YZ |
1129 | struct btrfs_path *path; |
1130 | ||
1131 | root = root->fs_info->chunk_root; | |
1132 | ||
1133 | path = btrfs_alloc_path(); | |
1134 | if (!path) | |
1135 | return -ENOMEM; | |
0b86a832 CM |
1136 | |
1137 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1138 | key.type = BTRFS_DEV_ITEM_KEY; | |
1139 | key.offset = (u64)-1; | |
1140 | ||
1141 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1142 | if (ret < 0) | |
1143 | goto error; | |
1144 | ||
1145 | BUG_ON(ret == 0); | |
1146 | ||
1147 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1148 | BTRFS_DEV_ITEM_KEY); | |
1149 | if (ret) { | |
1150 | *objectid = 1; | |
1151 | } else { | |
1152 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1153 | path->slots[0]); | |
1154 | *objectid = found_key.offset + 1; | |
1155 | } | |
1156 | ret = 0; | |
1157 | error: | |
2b82032c | 1158 | btrfs_free_path(path); |
0b86a832 CM |
1159 | return ret; |
1160 | } | |
1161 | ||
1162 | /* | |
1163 | * the device information is stored in the chunk root | |
1164 | * the btrfs_device struct should be fully filled in | |
1165 | */ | |
1166 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1167 | struct btrfs_root *root, | |
1168 | struct btrfs_device *device) | |
1169 | { | |
1170 | int ret; | |
1171 | struct btrfs_path *path; | |
1172 | struct btrfs_dev_item *dev_item; | |
1173 | struct extent_buffer *leaf; | |
1174 | struct btrfs_key key; | |
1175 | unsigned long ptr; | |
0b86a832 CM |
1176 | |
1177 | root = root->fs_info->chunk_root; | |
1178 | ||
1179 | path = btrfs_alloc_path(); | |
1180 | if (!path) | |
1181 | return -ENOMEM; | |
1182 | ||
0b86a832 CM |
1183 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1184 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1185 | key.offset = device->devid; |
0b86a832 CM |
1186 | |
1187 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1188 | sizeof(*dev_item)); |
0b86a832 CM |
1189 | if (ret) |
1190 | goto out; | |
1191 | ||
1192 | leaf = path->nodes[0]; | |
1193 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1194 | ||
1195 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1196 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1197 | btrfs_set_device_type(leaf, dev_item, device->type); |
1198 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1199 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1200 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1201 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1202 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1203 | btrfs_set_device_group(leaf, dev_item, 0); |
1204 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1205 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1206 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1207 | |
0b86a832 | 1208 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1209 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1210 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1211 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1212 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1213 | |
2b82032c | 1214 | ret = 0; |
0b86a832 CM |
1215 | out: |
1216 | btrfs_free_path(path); | |
1217 | return ret; | |
1218 | } | |
8f18cf13 | 1219 | |
a061fc8d CM |
1220 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1221 | struct btrfs_device *device) | |
1222 | { | |
1223 | int ret; | |
1224 | struct btrfs_path *path; | |
a061fc8d | 1225 | struct btrfs_key key; |
a061fc8d CM |
1226 | struct btrfs_trans_handle *trans; |
1227 | ||
1228 | root = root->fs_info->chunk_root; | |
1229 | ||
1230 | path = btrfs_alloc_path(); | |
1231 | if (!path) | |
1232 | return -ENOMEM; | |
1233 | ||
a22285a6 | 1234 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1235 | if (IS_ERR(trans)) { |
1236 | btrfs_free_path(path); | |
1237 | return PTR_ERR(trans); | |
1238 | } | |
a061fc8d CM |
1239 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1240 | key.type = BTRFS_DEV_ITEM_KEY; | |
1241 | key.offset = device->devid; | |
7d9eb12c | 1242 | lock_chunks(root); |
a061fc8d CM |
1243 | |
1244 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1245 | if (ret < 0) | |
1246 | goto out; | |
1247 | ||
1248 | if (ret > 0) { | |
1249 | ret = -ENOENT; | |
1250 | goto out; | |
1251 | } | |
1252 | ||
1253 | ret = btrfs_del_item(trans, root, path); | |
1254 | if (ret) | |
1255 | goto out; | |
a061fc8d CM |
1256 | out: |
1257 | btrfs_free_path(path); | |
7d9eb12c | 1258 | unlock_chunks(root); |
a061fc8d CM |
1259 | btrfs_commit_transaction(trans, root); |
1260 | return ret; | |
1261 | } | |
1262 | ||
1263 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1264 | { | |
1265 | struct btrfs_device *device; | |
2b82032c | 1266 | struct btrfs_device *next_device; |
a061fc8d | 1267 | struct block_device *bdev; |
dfe25020 | 1268 | struct buffer_head *bh = NULL; |
a061fc8d | 1269 | struct btrfs_super_block *disk_super; |
1f78160c | 1270 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1271 | u64 all_avail; |
1272 | u64 devid; | |
2b82032c YZ |
1273 | u64 num_devices; |
1274 | u8 *dev_uuid; | |
a061fc8d | 1275 | int ret = 0; |
1f78160c | 1276 | bool clear_super = false; |
a061fc8d | 1277 | |
a061fc8d | 1278 | mutex_lock(&uuid_mutex); |
7d9eb12c | 1279 | mutex_lock(&root->fs_info->volume_mutex); |
a061fc8d CM |
1280 | |
1281 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1282 | root->fs_info->avail_system_alloc_bits | | |
1283 | root->fs_info->avail_metadata_alloc_bits; | |
1284 | ||
1285 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1286 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1287 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1288 | "on raid10\n"); | |
a061fc8d CM |
1289 | ret = -EINVAL; |
1290 | goto out; | |
1291 | } | |
1292 | ||
1293 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1294 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1295 | printk(KERN_ERR "btrfs: unable to go below two " |
1296 | "devices on raid1\n"); | |
a061fc8d CM |
1297 | ret = -EINVAL; |
1298 | goto out; | |
1299 | } | |
1300 | ||
dfe25020 | 1301 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1302 | struct list_head *devices; |
1303 | struct btrfs_device *tmp; | |
a061fc8d | 1304 | |
dfe25020 CM |
1305 | device = NULL; |
1306 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1307 | /* |
1308 | * It is safe to read the devices since the volume_mutex | |
1309 | * is held. | |
1310 | */ | |
c6e30871 | 1311 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1312 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1313 | device = tmp; | |
1314 | break; | |
1315 | } | |
1316 | } | |
1317 | bdev = NULL; | |
1318 | bh = NULL; | |
1319 | disk_super = NULL; | |
1320 | if (!device) { | |
d397712b CM |
1321 | printk(KERN_ERR "btrfs: no missing devices found to " |
1322 | "remove\n"); | |
dfe25020 CM |
1323 | goto out; |
1324 | } | |
dfe25020 | 1325 | } else { |
d4d77629 TH |
1326 | bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL, |
1327 | root->fs_info->bdev_holder); | |
dfe25020 CM |
1328 | if (IS_ERR(bdev)) { |
1329 | ret = PTR_ERR(bdev); | |
1330 | goto out; | |
1331 | } | |
a061fc8d | 1332 | |
2b82032c | 1333 | set_blocksize(bdev, 4096); |
a512bbf8 | 1334 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1335 | if (!bh) { |
20b45077 | 1336 | ret = -EINVAL; |
dfe25020 CM |
1337 | goto error_close; |
1338 | } | |
1339 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1340 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1341 | dev_uuid = disk_super->dev_item.uuid; |
1342 | device = btrfs_find_device(root, devid, dev_uuid, | |
1343 | disk_super->fsid); | |
dfe25020 CM |
1344 | if (!device) { |
1345 | ret = -ENOENT; | |
1346 | goto error_brelse; | |
1347 | } | |
2b82032c | 1348 | } |
dfe25020 | 1349 | |
2b82032c | 1350 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1351 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1352 | "device\n"); | |
2b82032c YZ |
1353 | ret = -EINVAL; |
1354 | goto error_brelse; | |
1355 | } | |
1356 | ||
1357 | if (device->writeable) { | |
0c1daee0 | 1358 | lock_chunks(root); |
2b82032c | 1359 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1360 | unlock_chunks(root); |
2b82032c | 1361 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1362 | clear_super = true; |
dfe25020 | 1363 | } |
a061fc8d CM |
1364 | |
1365 | ret = btrfs_shrink_device(device, 0); | |
1366 | if (ret) | |
9b3517e9 | 1367 | goto error_undo; |
a061fc8d | 1368 | |
a061fc8d CM |
1369 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1370 | if (ret) | |
9b3517e9 | 1371 | goto error_undo; |
a061fc8d | 1372 | |
2bf64758 JB |
1373 | spin_lock(&root->fs_info->free_chunk_lock); |
1374 | root->fs_info->free_chunk_space = device->total_bytes - | |
1375 | device->bytes_used; | |
1376 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1377 | ||
2b82032c | 1378 | device->in_fs_metadata = 0; |
a2de733c | 1379 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1380 | |
1381 | /* | |
1382 | * the device list mutex makes sure that we don't change | |
1383 | * the device list while someone else is writing out all | |
1384 | * the device supers. | |
1385 | */ | |
1f78160c XG |
1386 | |
1387 | cur_devices = device->fs_devices; | |
e5e9a520 | 1388 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1389 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1390 | |
e4404d6e | 1391 | device->fs_devices->num_devices--; |
2b82032c | 1392 | |
cd02dca5 CM |
1393 | if (device->missing) |
1394 | root->fs_info->fs_devices->missing_devices--; | |
1395 | ||
2b82032c YZ |
1396 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1397 | struct btrfs_device, dev_list); | |
1398 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1399 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1400 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1401 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1402 | ||
1f78160c | 1403 | if (device->bdev) |
e4404d6e | 1404 | device->fs_devices->open_devices--; |
1f78160c XG |
1405 | |
1406 | call_rcu(&device->rcu, free_device); | |
1407 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1408 | |
6c41761f DS |
1409 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1410 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1411 | |
1f78160c | 1412 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1413 | struct btrfs_fs_devices *fs_devices; |
1414 | fs_devices = root->fs_info->fs_devices; | |
1415 | while (fs_devices) { | |
1f78160c | 1416 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1417 | break; |
1418 | fs_devices = fs_devices->seed; | |
2b82032c | 1419 | } |
1f78160c XG |
1420 | fs_devices->seed = cur_devices->seed; |
1421 | cur_devices->seed = NULL; | |
0c1daee0 | 1422 | lock_chunks(root); |
1f78160c | 1423 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1424 | unlock_chunks(root); |
1f78160c | 1425 | free_fs_devices(cur_devices); |
2b82032c YZ |
1426 | } |
1427 | ||
1428 | /* | |
1429 | * at this point, the device is zero sized. We want to | |
1430 | * remove it from the devices list and zero out the old super | |
1431 | */ | |
1f78160c | 1432 | if (clear_super) { |
dfe25020 CM |
1433 | /* make sure this device isn't detected as part of |
1434 | * the FS anymore | |
1435 | */ | |
1436 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1437 | set_buffer_dirty(bh); | |
1438 | sync_dirty_buffer(bh); | |
dfe25020 | 1439 | } |
a061fc8d | 1440 | |
a061fc8d | 1441 | ret = 0; |
a061fc8d CM |
1442 | |
1443 | error_brelse: | |
1444 | brelse(bh); | |
1445 | error_close: | |
dfe25020 | 1446 | if (bdev) |
e525fd89 | 1447 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d | 1448 | out: |
7d9eb12c | 1449 | mutex_unlock(&root->fs_info->volume_mutex); |
a061fc8d | 1450 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1451 | return ret; |
9b3517e9 ID |
1452 | error_undo: |
1453 | if (device->writeable) { | |
0c1daee0 | 1454 | lock_chunks(root); |
9b3517e9 ID |
1455 | list_add(&device->dev_alloc_list, |
1456 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1457 | unlock_chunks(root); |
9b3517e9 ID |
1458 | root->fs_info->fs_devices->rw_devices++; |
1459 | } | |
1460 | goto error_brelse; | |
a061fc8d CM |
1461 | } |
1462 | ||
2b82032c YZ |
1463 | /* |
1464 | * does all the dirty work required for changing file system's UUID. | |
1465 | */ | |
1466 | static int btrfs_prepare_sprout(struct btrfs_trans_handle *trans, | |
1467 | struct btrfs_root *root) | |
1468 | { | |
1469 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1470 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1471 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1472 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1473 | struct btrfs_device *device; |
1474 | u64 super_flags; | |
1475 | ||
1476 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1477 | if (!fs_devices->seeding) |
2b82032c YZ |
1478 | return -EINVAL; |
1479 | ||
e4404d6e YZ |
1480 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1481 | if (!seed_devices) | |
2b82032c YZ |
1482 | return -ENOMEM; |
1483 | ||
e4404d6e YZ |
1484 | old_devices = clone_fs_devices(fs_devices); |
1485 | if (IS_ERR(old_devices)) { | |
1486 | kfree(seed_devices); | |
1487 | return PTR_ERR(old_devices); | |
2b82032c | 1488 | } |
e4404d6e | 1489 | |
2b82032c YZ |
1490 | list_add(&old_devices->list, &fs_uuids); |
1491 | ||
e4404d6e YZ |
1492 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1493 | seed_devices->opened = 1; | |
1494 | INIT_LIST_HEAD(&seed_devices->devices); | |
1495 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1496 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1497 | |
1498 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1499 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1500 | synchronize_rcu); | |
c9513edb XG |
1501 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1502 | ||
e4404d6e YZ |
1503 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1504 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1505 | device->fs_devices = seed_devices; | |
1506 | } | |
1507 | ||
2b82032c YZ |
1508 | fs_devices->seeding = 0; |
1509 | fs_devices->num_devices = 0; | |
1510 | fs_devices->open_devices = 0; | |
e4404d6e | 1511 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1512 | |
1513 | generate_random_uuid(fs_devices->fsid); | |
1514 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1515 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1516 | super_flags = btrfs_super_flags(disk_super) & | |
1517 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1518 | btrfs_set_super_flags(disk_super, super_flags); | |
1519 | ||
1520 | return 0; | |
1521 | } | |
1522 | ||
1523 | /* | |
1524 | * strore the expected generation for seed devices in device items. | |
1525 | */ | |
1526 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1527 | struct btrfs_root *root) | |
1528 | { | |
1529 | struct btrfs_path *path; | |
1530 | struct extent_buffer *leaf; | |
1531 | struct btrfs_dev_item *dev_item; | |
1532 | struct btrfs_device *device; | |
1533 | struct btrfs_key key; | |
1534 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1535 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1536 | u64 devid; | |
1537 | int ret; | |
1538 | ||
1539 | path = btrfs_alloc_path(); | |
1540 | if (!path) | |
1541 | return -ENOMEM; | |
1542 | ||
1543 | root = root->fs_info->chunk_root; | |
1544 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1545 | key.offset = 0; | |
1546 | key.type = BTRFS_DEV_ITEM_KEY; | |
1547 | ||
1548 | while (1) { | |
1549 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1550 | if (ret < 0) | |
1551 | goto error; | |
1552 | ||
1553 | leaf = path->nodes[0]; | |
1554 | next_slot: | |
1555 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1556 | ret = btrfs_next_leaf(root, path); | |
1557 | if (ret > 0) | |
1558 | break; | |
1559 | if (ret < 0) | |
1560 | goto error; | |
1561 | leaf = path->nodes[0]; | |
1562 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1563 | btrfs_release_path(path); |
2b82032c YZ |
1564 | continue; |
1565 | } | |
1566 | ||
1567 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1568 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1569 | key.type != BTRFS_DEV_ITEM_KEY) | |
1570 | break; | |
1571 | ||
1572 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1573 | struct btrfs_dev_item); | |
1574 | devid = btrfs_device_id(leaf, dev_item); | |
1575 | read_extent_buffer(leaf, dev_uuid, | |
1576 | (unsigned long)btrfs_device_uuid(dev_item), | |
1577 | BTRFS_UUID_SIZE); | |
1578 | read_extent_buffer(leaf, fs_uuid, | |
1579 | (unsigned long)btrfs_device_fsid(dev_item), | |
1580 | BTRFS_UUID_SIZE); | |
1581 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
1582 | BUG_ON(!device); | |
1583 | ||
1584 | if (device->fs_devices->seeding) { | |
1585 | btrfs_set_device_generation(leaf, dev_item, | |
1586 | device->generation); | |
1587 | btrfs_mark_buffer_dirty(leaf); | |
1588 | } | |
1589 | ||
1590 | path->slots[0]++; | |
1591 | goto next_slot; | |
1592 | } | |
1593 | ret = 0; | |
1594 | error: | |
1595 | btrfs_free_path(path); | |
1596 | return ret; | |
1597 | } | |
1598 | ||
788f20eb CM |
1599 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1600 | { | |
d5e2003c | 1601 | struct request_queue *q; |
788f20eb CM |
1602 | struct btrfs_trans_handle *trans; |
1603 | struct btrfs_device *device; | |
1604 | struct block_device *bdev; | |
788f20eb | 1605 | struct list_head *devices; |
2b82032c | 1606 | struct super_block *sb = root->fs_info->sb; |
788f20eb | 1607 | u64 total_bytes; |
2b82032c | 1608 | int seeding_dev = 0; |
788f20eb CM |
1609 | int ret = 0; |
1610 | ||
2b82032c YZ |
1611 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
1612 | return -EINVAL; | |
788f20eb | 1613 | |
a5d16333 | 1614 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1615 | root->fs_info->bdev_holder); |
7f59203a JB |
1616 | if (IS_ERR(bdev)) |
1617 | return PTR_ERR(bdev); | |
a2135011 | 1618 | |
2b82032c YZ |
1619 | if (root->fs_info->fs_devices->seeding) { |
1620 | seeding_dev = 1; | |
1621 | down_write(&sb->s_umount); | |
1622 | mutex_lock(&uuid_mutex); | |
1623 | } | |
1624 | ||
8c8bee1d | 1625 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
7d9eb12c | 1626 | mutex_lock(&root->fs_info->volume_mutex); |
a2135011 | 1627 | |
788f20eb | 1628 | devices = &root->fs_info->fs_devices->devices; |
e5e9a520 CM |
1629 | /* |
1630 | * we have the volume lock, so we don't need the extra | |
1631 | * device list mutex while reading the list here. | |
1632 | */ | |
c6e30871 | 1633 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1634 | if (device->bdev == bdev) { |
1635 | ret = -EEXIST; | |
2b82032c | 1636 | goto error; |
788f20eb CM |
1637 | } |
1638 | } | |
1639 | ||
1640 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1641 | if (!device) { | |
1642 | /* we can safely leave the fs_devices entry around */ | |
1643 | ret = -ENOMEM; | |
2b82032c | 1644 | goto error; |
788f20eb CM |
1645 | } |
1646 | ||
788f20eb CM |
1647 | device->name = kstrdup(device_path, GFP_NOFS); |
1648 | if (!device->name) { | |
1649 | kfree(device); | |
2b82032c YZ |
1650 | ret = -ENOMEM; |
1651 | goto error; | |
788f20eb | 1652 | } |
2b82032c YZ |
1653 | |
1654 | ret = find_next_devid(root, &device->devid); | |
1655 | if (ret) { | |
67100f25 | 1656 | kfree(device->name); |
2b82032c YZ |
1657 | kfree(device); |
1658 | goto error; | |
1659 | } | |
1660 | ||
a22285a6 | 1661 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1662 | if (IS_ERR(trans)) { |
67100f25 | 1663 | kfree(device->name); |
98d5dc13 TI |
1664 | kfree(device); |
1665 | ret = PTR_ERR(trans); | |
1666 | goto error; | |
1667 | } | |
1668 | ||
2b82032c YZ |
1669 | lock_chunks(root); |
1670 | ||
d5e2003c JB |
1671 | q = bdev_get_queue(bdev); |
1672 | if (blk_queue_discard(q)) | |
1673 | device->can_discard = 1; | |
2b82032c YZ |
1674 | device->writeable = 1; |
1675 | device->work.func = pending_bios_fn; | |
1676 | generate_random_uuid(device->uuid); | |
1677 | spin_lock_init(&device->io_lock); | |
1678 | device->generation = trans->transid; | |
788f20eb CM |
1679 | device->io_width = root->sectorsize; |
1680 | device->io_align = root->sectorsize; | |
1681 | device->sector_size = root->sectorsize; | |
1682 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1683 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1684 | device->dev_root = root->fs_info->dev_root; |
1685 | device->bdev = bdev; | |
dfe25020 | 1686 | device->in_fs_metadata = 1; |
fb01aa85 | 1687 | device->mode = FMODE_EXCL; |
2b82032c | 1688 | set_blocksize(device->bdev, 4096); |
788f20eb | 1689 | |
2b82032c YZ |
1690 | if (seeding_dev) { |
1691 | sb->s_flags &= ~MS_RDONLY; | |
1692 | ret = btrfs_prepare_sprout(trans, root); | |
1693 | BUG_ON(ret); | |
1694 | } | |
788f20eb | 1695 | |
2b82032c | 1696 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 CM |
1697 | |
1698 | /* | |
1699 | * we don't want write_supers to jump in here with our device | |
1700 | * half setup | |
1701 | */ | |
1702 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c | 1703 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1704 | list_add(&device->dev_alloc_list, |
1705 | &root->fs_info->fs_devices->alloc_list); | |
1706 | root->fs_info->fs_devices->num_devices++; | |
1707 | root->fs_info->fs_devices->open_devices++; | |
1708 | root->fs_info->fs_devices->rw_devices++; | |
d5e2003c JB |
1709 | if (device->can_discard) |
1710 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1711 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1712 | |
2bf64758 JB |
1713 | spin_lock(&root->fs_info->free_chunk_lock); |
1714 | root->fs_info->free_chunk_space += device->total_bytes; | |
1715 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1716 | ||
c289811c CM |
1717 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1718 | root->fs_info->fs_devices->rotating = 1; | |
1719 | ||
6c41761f DS |
1720 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1721 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1722 | total_bytes + device->total_bytes); |
1723 | ||
6c41761f DS |
1724 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1725 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1726 | total_bytes + 1); |
e5e9a520 | 1727 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1728 | |
2b82032c YZ |
1729 | if (seeding_dev) { |
1730 | ret = init_first_rw_device(trans, root, device); | |
1731 | BUG_ON(ret); | |
1732 | ret = btrfs_finish_sprout(trans, root); | |
1733 | BUG_ON(ret); | |
1734 | } else { | |
1735 | ret = btrfs_add_device(trans, root, device); | |
1736 | } | |
1737 | ||
913d952e CM |
1738 | /* |
1739 | * we've got more storage, clear any full flags on the space | |
1740 | * infos | |
1741 | */ | |
1742 | btrfs_clear_space_info_full(root->fs_info); | |
1743 | ||
7d9eb12c | 1744 | unlock_chunks(root); |
2b82032c | 1745 | btrfs_commit_transaction(trans, root); |
a2135011 | 1746 | |
2b82032c YZ |
1747 | if (seeding_dev) { |
1748 | mutex_unlock(&uuid_mutex); | |
1749 | up_write(&sb->s_umount); | |
788f20eb | 1750 | |
2b82032c YZ |
1751 | ret = btrfs_relocate_sys_chunks(root); |
1752 | BUG_ON(ret); | |
1753 | } | |
1754 | out: | |
1755 | mutex_unlock(&root->fs_info->volume_mutex); | |
1756 | return ret; | |
1757 | error: | |
e525fd89 | 1758 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1759 | if (seeding_dev) { |
1760 | mutex_unlock(&uuid_mutex); | |
1761 | up_write(&sb->s_umount); | |
1762 | } | |
788f20eb CM |
1763 | goto out; |
1764 | } | |
1765 | ||
d397712b CM |
1766 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1767 | struct btrfs_device *device) | |
0b86a832 CM |
1768 | { |
1769 | int ret; | |
1770 | struct btrfs_path *path; | |
1771 | struct btrfs_root *root; | |
1772 | struct btrfs_dev_item *dev_item; | |
1773 | struct extent_buffer *leaf; | |
1774 | struct btrfs_key key; | |
1775 | ||
1776 | root = device->dev_root->fs_info->chunk_root; | |
1777 | ||
1778 | path = btrfs_alloc_path(); | |
1779 | if (!path) | |
1780 | return -ENOMEM; | |
1781 | ||
1782 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1783 | key.type = BTRFS_DEV_ITEM_KEY; | |
1784 | key.offset = device->devid; | |
1785 | ||
1786 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1787 | if (ret < 0) | |
1788 | goto out; | |
1789 | ||
1790 | if (ret > 0) { | |
1791 | ret = -ENOENT; | |
1792 | goto out; | |
1793 | } | |
1794 | ||
1795 | leaf = path->nodes[0]; | |
1796 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1797 | ||
1798 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1799 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1800 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1801 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1802 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1803 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1804 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1805 | btrfs_mark_buffer_dirty(leaf); | |
1806 | ||
1807 | out: | |
1808 | btrfs_free_path(path); | |
1809 | return ret; | |
1810 | } | |
1811 | ||
7d9eb12c | 1812 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1813 | struct btrfs_device *device, u64 new_size) |
1814 | { | |
1815 | struct btrfs_super_block *super_copy = | |
6c41761f | 1816 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1817 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1818 | u64 diff = new_size - device->total_bytes; | |
1819 | ||
2b82032c YZ |
1820 | if (!device->writeable) |
1821 | return -EACCES; | |
1822 | if (new_size <= device->total_bytes) | |
1823 | return -EINVAL; | |
1824 | ||
8f18cf13 | 1825 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1826 | device->fs_devices->total_rw_bytes += diff; |
1827 | ||
1828 | device->total_bytes = new_size; | |
9779b72f | 1829 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1830 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1831 | ||
8f18cf13 CM |
1832 | return btrfs_update_device(trans, device); |
1833 | } | |
1834 | ||
7d9eb12c CM |
1835 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1836 | struct btrfs_device *device, u64 new_size) | |
1837 | { | |
1838 | int ret; | |
1839 | lock_chunks(device->dev_root); | |
1840 | ret = __btrfs_grow_device(trans, device, new_size); | |
1841 | unlock_chunks(device->dev_root); | |
1842 | return ret; | |
1843 | } | |
1844 | ||
8f18cf13 CM |
1845 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1846 | struct btrfs_root *root, | |
1847 | u64 chunk_tree, u64 chunk_objectid, | |
1848 | u64 chunk_offset) | |
1849 | { | |
1850 | int ret; | |
1851 | struct btrfs_path *path; | |
1852 | struct btrfs_key key; | |
1853 | ||
1854 | root = root->fs_info->chunk_root; | |
1855 | path = btrfs_alloc_path(); | |
1856 | if (!path) | |
1857 | return -ENOMEM; | |
1858 | ||
1859 | key.objectid = chunk_objectid; | |
1860 | key.offset = chunk_offset; | |
1861 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1862 | ||
1863 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1864 | BUG_ON(ret); | |
1865 | ||
1866 | ret = btrfs_del_item(trans, root, path); | |
8f18cf13 CM |
1867 | |
1868 | btrfs_free_path(path); | |
65a246c5 | 1869 | return ret; |
8f18cf13 CM |
1870 | } |
1871 | ||
b2950863 | 1872 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1873 | chunk_offset) |
1874 | { | |
6c41761f | 1875 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
1876 | struct btrfs_disk_key *disk_key; |
1877 | struct btrfs_chunk *chunk; | |
1878 | u8 *ptr; | |
1879 | int ret = 0; | |
1880 | u32 num_stripes; | |
1881 | u32 array_size; | |
1882 | u32 len = 0; | |
1883 | u32 cur; | |
1884 | struct btrfs_key key; | |
1885 | ||
1886 | array_size = btrfs_super_sys_array_size(super_copy); | |
1887 | ||
1888 | ptr = super_copy->sys_chunk_array; | |
1889 | cur = 0; | |
1890 | ||
1891 | while (cur < array_size) { | |
1892 | disk_key = (struct btrfs_disk_key *)ptr; | |
1893 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1894 | ||
1895 | len = sizeof(*disk_key); | |
1896 | ||
1897 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1898 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1899 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1900 | len += btrfs_chunk_item_size(num_stripes); | |
1901 | } else { | |
1902 | ret = -EIO; | |
1903 | break; | |
1904 | } | |
1905 | if (key.objectid == chunk_objectid && | |
1906 | key.offset == chunk_offset) { | |
1907 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1908 | array_size -= len; | |
1909 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1910 | } else { | |
1911 | ptr += len; | |
1912 | cur += len; | |
1913 | } | |
1914 | } | |
1915 | return ret; | |
1916 | } | |
1917 | ||
b2950863 | 1918 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
1919 | u64 chunk_tree, u64 chunk_objectid, |
1920 | u64 chunk_offset) | |
1921 | { | |
1922 | struct extent_map_tree *em_tree; | |
1923 | struct btrfs_root *extent_root; | |
1924 | struct btrfs_trans_handle *trans; | |
1925 | struct extent_map *em; | |
1926 | struct map_lookup *map; | |
1927 | int ret; | |
1928 | int i; | |
1929 | ||
1930 | root = root->fs_info->chunk_root; | |
1931 | extent_root = root->fs_info->extent_root; | |
1932 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
1933 | ||
ba1bf481 JB |
1934 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
1935 | if (ret) | |
1936 | return -ENOSPC; | |
1937 | ||
8f18cf13 | 1938 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 1939 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
1940 | if (ret) |
1941 | return ret; | |
8f18cf13 | 1942 | |
a22285a6 | 1943 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1944 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 1945 | |
7d9eb12c CM |
1946 | lock_chunks(root); |
1947 | ||
8f18cf13 CM |
1948 | /* |
1949 | * step two, delete the device extents and the | |
1950 | * chunk tree entries | |
1951 | */ | |
890871be | 1952 | read_lock(&em_tree->lock); |
8f18cf13 | 1953 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 1954 | read_unlock(&em_tree->lock); |
8f18cf13 | 1955 | |
a061fc8d CM |
1956 | BUG_ON(em->start > chunk_offset || |
1957 | em->start + em->len < chunk_offset); | |
8f18cf13 CM |
1958 | map = (struct map_lookup *)em->bdev; |
1959 | ||
1960 | for (i = 0; i < map->num_stripes; i++) { | |
1961 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
1962 | map->stripes[i].physical); | |
1963 | BUG_ON(ret); | |
a061fc8d | 1964 | |
dfe25020 CM |
1965 | if (map->stripes[i].dev) { |
1966 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
1967 | BUG_ON(ret); | |
1968 | } | |
8f18cf13 CM |
1969 | } |
1970 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
1971 | chunk_offset); | |
1972 | ||
1973 | BUG_ON(ret); | |
1974 | ||
1abe9b8a | 1975 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
1976 | ||
8f18cf13 CM |
1977 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
1978 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
1979 | BUG_ON(ret); | |
8f18cf13 CM |
1980 | } |
1981 | ||
2b82032c YZ |
1982 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
1983 | BUG_ON(ret); | |
1984 | ||
890871be | 1985 | write_lock(&em_tree->lock); |
2b82032c | 1986 | remove_extent_mapping(em_tree, em); |
890871be | 1987 | write_unlock(&em_tree->lock); |
2b82032c YZ |
1988 | |
1989 | kfree(map); | |
1990 | em->bdev = NULL; | |
1991 | ||
1992 | /* once for the tree */ | |
1993 | free_extent_map(em); | |
1994 | /* once for us */ | |
1995 | free_extent_map(em); | |
1996 | ||
1997 | unlock_chunks(root); | |
1998 | btrfs_end_transaction(trans, root); | |
1999 | return 0; | |
2000 | } | |
2001 | ||
2002 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2003 | { | |
2004 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2005 | struct btrfs_path *path; | |
2006 | struct extent_buffer *leaf; | |
2007 | struct btrfs_chunk *chunk; | |
2008 | struct btrfs_key key; | |
2009 | struct btrfs_key found_key; | |
2010 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2011 | u64 chunk_type; | |
ba1bf481 JB |
2012 | bool retried = false; |
2013 | int failed = 0; | |
2b82032c YZ |
2014 | int ret; |
2015 | ||
2016 | path = btrfs_alloc_path(); | |
2017 | if (!path) | |
2018 | return -ENOMEM; | |
2019 | ||
ba1bf481 | 2020 | again: |
2b82032c YZ |
2021 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2022 | key.offset = (u64)-1; | |
2023 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2024 | ||
2025 | while (1) { | |
2026 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2027 | if (ret < 0) | |
2028 | goto error; | |
2029 | BUG_ON(ret == 0); | |
2030 | ||
2031 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2032 | key.type); | |
2033 | if (ret < 0) | |
2034 | goto error; | |
2035 | if (ret > 0) | |
2036 | break; | |
1a40e23b | 2037 | |
2b82032c YZ |
2038 | leaf = path->nodes[0]; |
2039 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2040 | |
2b82032c YZ |
2041 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2042 | struct btrfs_chunk); | |
2043 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2044 | btrfs_release_path(path); |
8f18cf13 | 2045 | |
2b82032c YZ |
2046 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2047 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2048 | found_key.objectid, | |
2049 | found_key.offset); | |
ba1bf481 JB |
2050 | if (ret == -ENOSPC) |
2051 | failed++; | |
2052 | else if (ret) | |
2053 | BUG(); | |
2b82032c | 2054 | } |
8f18cf13 | 2055 | |
2b82032c YZ |
2056 | if (found_key.offset == 0) |
2057 | break; | |
2058 | key.offset = found_key.offset - 1; | |
2059 | } | |
2060 | ret = 0; | |
ba1bf481 JB |
2061 | if (failed && !retried) { |
2062 | failed = 0; | |
2063 | retried = true; | |
2064 | goto again; | |
2065 | } else if (failed && retried) { | |
2066 | WARN_ON(1); | |
2067 | ret = -ENOSPC; | |
2068 | } | |
2b82032c YZ |
2069 | error: |
2070 | btrfs_free_path(path); | |
2071 | return ret; | |
8f18cf13 CM |
2072 | } |
2073 | ||
ec44a35c CM |
2074 | static u64 div_factor(u64 num, int factor) |
2075 | { | |
2076 | if (factor == 10) | |
2077 | return num; | |
2078 | num *= factor; | |
2079 | do_div(num, 10); | |
2080 | return num; | |
2081 | } | |
2082 | ||
ec44a35c CM |
2083 | int btrfs_balance(struct btrfs_root *dev_root) |
2084 | { | |
2085 | int ret; | |
ec44a35c CM |
2086 | struct list_head *devices = &dev_root->fs_info->fs_devices->devices; |
2087 | struct btrfs_device *device; | |
2088 | u64 old_size; | |
2089 | u64 size_to_free; | |
2090 | struct btrfs_path *path; | |
2091 | struct btrfs_key key; | |
ec44a35c CM |
2092 | struct btrfs_root *chunk_root = dev_root->fs_info->chunk_root; |
2093 | struct btrfs_trans_handle *trans; | |
2094 | struct btrfs_key found_key; | |
2095 | ||
2b82032c YZ |
2096 | if (dev_root->fs_info->sb->s_flags & MS_RDONLY) |
2097 | return -EROFS; | |
ec44a35c | 2098 | |
6f88a440 BH |
2099 | if (!capable(CAP_SYS_ADMIN)) |
2100 | return -EPERM; | |
2101 | ||
7d9eb12c | 2102 | mutex_lock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
2103 | dev_root = dev_root->fs_info->dev_root; |
2104 | ||
ec44a35c | 2105 | /* step one make some room on all the devices */ |
c6e30871 | 2106 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2107 | old_size = device->total_bytes; |
2108 | size_to_free = div_factor(old_size, 1); | |
2109 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2110 | if (!device->writeable || |
2111 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2112 | continue; |
2113 | ||
2114 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2115 | if (ret == -ENOSPC) |
2116 | break; | |
ec44a35c CM |
2117 | BUG_ON(ret); |
2118 | ||
a22285a6 | 2119 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2120 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2121 | |
2122 | ret = btrfs_grow_device(trans, device, old_size); | |
2123 | BUG_ON(ret); | |
2124 | ||
2125 | btrfs_end_transaction(trans, dev_root); | |
2126 | } | |
2127 | ||
2128 | /* step two, relocate all the chunks */ | |
2129 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2130 | if (!path) { |
2131 | ret = -ENOMEM; | |
2132 | goto error; | |
2133 | } | |
ec44a35c CM |
2134 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2135 | key.offset = (u64)-1; | |
2136 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2137 | ||
d397712b | 2138 | while (1) { |
ec44a35c CM |
2139 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2140 | if (ret < 0) | |
2141 | goto error; | |
2142 | ||
2143 | /* | |
2144 | * this shouldn't happen, it means the last relocate | |
2145 | * failed | |
2146 | */ | |
2147 | if (ret == 0) | |
2148 | break; | |
2149 | ||
2150 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2151 | BTRFS_CHUNK_ITEM_KEY); | |
7d9eb12c | 2152 | if (ret) |
ec44a35c | 2153 | break; |
7d9eb12c | 2154 | |
ec44a35c CM |
2155 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, |
2156 | path->slots[0]); | |
2157 | if (found_key.objectid != key.objectid) | |
2158 | break; | |
7d9eb12c | 2159 | |
ec44a35c | 2160 | /* chunk zero is special */ |
ba1bf481 | 2161 | if (found_key.offset == 0) |
ec44a35c CM |
2162 | break; |
2163 | ||
b3b4aa74 | 2164 | btrfs_release_path(path); |
ec44a35c CM |
2165 | ret = btrfs_relocate_chunk(chunk_root, |
2166 | chunk_root->root_key.objectid, | |
2167 | found_key.objectid, | |
2168 | found_key.offset); | |
508794eb JB |
2169 | if (ret && ret != -ENOSPC) |
2170 | goto error; | |
ba1bf481 | 2171 | key.offset = found_key.offset - 1; |
ec44a35c CM |
2172 | } |
2173 | ret = 0; | |
2174 | error: | |
2175 | btrfs_free_path(path); | |
7d9eb12c | 2176 | mutex_unlock(&dev_root->fs_info->volume_mutex); |
ec44a35c CM |
2177 | return ret; |
2178 | } | |
2179 | ||
8f18cf13 CM |
2180 | /* |
2181 | * shrinking a device means finding all of the device extents past | |
2182 | * the new size, and then following the back refs to the chunks. | |
2183 | * The chunk relocation code actually frees the device extent | |
2184 | */ | |
2185 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
2186 | { | |
2187 | struct btrfs_trans_handle *trans; | |
2188 | struct btrfs_root *root = device->dev_root; | |
2189 | struct btrfs_dev_extent *dev_extent = NULL; | |
2190 | struct btrfs_path *path; | |
2191 | u64 length; | |
2192 | u64 chunk_tree; | |
2193 | u64 chunk_objectid; | |
2194 | u64 chunk_offset; | |
2195 | int ret; | |
2196 | int slot; | |
ba1bf481 JB |
2197 | int failed = 0; |
2198 | bool retried = false; | |
8f18cf13 CM |
2199 | struct extent_buffer *l; |
2200 | struct btrfs_key key; | |
6c41761f | 2201 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 2202 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 2203 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
2204 | u64 diff = device->total_bytes - new_size; |
2205 | ||
2b82032c YZ |
2206 | if (new_size >= device->total_bytes) |
2207 | return -EINVAL; | |
8f18cf13 CM |
2208 | |
2209 | path = btrfs_alloc_path(); | |
2210 | if (!path) | |
2211 | return -ENOMEM; | |
2212 | ||
8f18cf13 CM |
2213 | path->reada = 2; |
2214 | ||
7d9eb12c CM |
2215 | lock_chunks(root); |
2216 | ||
8f18cf13 | 2217 | device->total_bytes = new_size; |
2bf64758 | 2218 | if (device->writeable) { |
2b82032c | 2219 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
2220 | spin_lock(&root->fs_info->free_chunk_lock); |
2221 | root->fs_info->free_chunk_space -= diff; | |
2222 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2223 | } | |
7d9eb12c | 2224 | unlock_chunks(root); |
8f18cf13 | 2225 | |
ba1bf481 | 2226 | again: |
8f18cf13 CM |
2227 | key.objectid = device->devid; |
2228 | key.offset = (u64)-1; | |
2229 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2230 | ||
2231 | while (1) { | |
2232 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2233 | if (ret < 0) | |
2234 | goto done; | |
2235 | ||
2236 | ret = btrfs_previous_item(root, path, 0, key.type); | |
2237 | if (ret < 0) | |
2238 | goto done; | |
2239 | if (ret) { | |
2240 | ret = 0; | |
b3b4aa74 | 2241 | btrfs_release_path(path); |
bf1fb512 | 2242 | break; |
8f18cf13 CM |
2243 | } |
2244 | ||
2245 | l = path->nodes[0]; | |
2246 | slot = path->slots[0]; | |
2247 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
2248 | ||
ba1bf481 | 2249 | if (key.objectid != device->devid) { |
b3b4aa74 | 2250 | btrfs_release_path(path); |
bf1fb512 | 2251 | break; |
ba1bf481 | 2252 | } |
8f18cf13 CM |
2253 | |
2254 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
2255 | length = btrfs_dev_extent_length(l, dev_extent); | |
2256 | ||
ba1bf481 | 2257 | if (key.offset + length <= new_size) { |
b3b4aa74 | 2258 | btrfs_release_path(path); |
d6397bae | 2259 | break; |
ba1bf481 | 2260 | } |
8f18cf13 CM |
2261 | |
2262 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
2263 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
2264 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 2265 | btrfs_release_path(path); |
8f18cf13 CM |
2266 | |
2267 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
2268 | chunk_offset); | |
ba1bf481 | 2269 | if (ret && ret != -ENOSPC) |
8f18cf13 | 2270 | goto done; |
ba1bf481 JB |
2271 | if (ret == -ENOSPC) |
2272 | failed++; | |
2273 | key.offset -= 1; | |
2274 | } | |
2275 | ||
2276 | if (failed && !retried) { | |
2277 | failed = 0; | |
2278 | retried = true; | |
2279 | goto again; | |
2280 | } else if (failed && retried) { | |
2281 | ret = -ENOSPC; | |
2282 | lock_chunks(root); | |
2283 | ||
2284 | device->total_bytes = old_size; | |
2285 | if (device->writeable) | |
2286 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
2287 | spin_lock(&root->fs_info->free_chunk_lock); |
2288 | root->fs_info->free_chunk_space += diff; | |
2289 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
2290 | unlock_chunks(root); |
2291 | goto done; | |
8f18cf13 CM |
2292 | } |
2293 | ||
d6397bae | 2294 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 2295 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
2296 | if (IS_ERR(trans)) { |
2297 | ret = PTR_ERR(trans); | |
2298 | goto done; | |
2299 | } | |
2300 | ||
d6397bae CB |
2301 | lock_chunks(root); |
2302 | ||
2303 | device->disk_total_bytes = new_size; | |
2304 | /* Now btrfs_update_device() will change the on-disk size. */ | |
2305 | ret = btrfs_update_device(trans, device); | |
2306 | if (ret) { | |
2307 | unlock_chunks(root); | |
2308 | btrfs_end_transaction(trans, root); | |
2309 | goto done; | |
2310 | } | |
2311 | WARN_ON(diff > old_total); | |
2312 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
2313 | unlock_chunks(root); | |
2314 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
2315 | done: |
2316 | btrfs_free_path(path); | |
2317 | return ret; | |
2318 | } | |
2319 | ||
b2950863 | 2320 | static int btrfs_add_system_chunk(struct btrfs_trans_handle *trans, |
0b86a832 CM |
2321 | struct btrfs_root *root, |
2322 | struct btrfs_key *key, | |
2323 | struct btrfs_chunk *chunk, int item_size) | |
2324 | { | |
6c41761f | 2325 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
2326 | struct btrfs_disk_key disk_key; |
2327 | u32 array_size; | |
2328 | u8 *ptr; | |
2329 | ||
2330 | array_size = btrfs_super_sys_array_size(super_copy); | |
2331 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
2332 | return -EFBIG; | |
2333 | ||
2334 | ptr = super_copy->sys_chunk_array + array_size; | |
2335 | btrfs_cpu_key_to_disk(&disk_key, key); | |
2336 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
2337 | ptr += sizeof(disk_key); | |
2338 | memcpy(ptr, chunk, item_size); | |
2339 | item_size += sizeof(disk_key); | |
2340 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
2341 | return 0; | |
2342 | } | |
2343 | ||
73c5de00 AJ |
2344 | /* |
2345 | * sort the devices in descending order by max_avail, total_avail | |
2346 | */ | |
2347 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 2348 | { |
73c5de00 AJ |
2349 | const struct btrfs_device_info *di_a = a; |
2350 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 2351 | |
73c5de00 | 2352 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 2353 | return -1; |
73c5de00 | 2354 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 2355 | return 1; |
73c5de00 AJ |
2356 | if (di_a->total_avail > di_b->total_avail) |
2357 | return -1; | |
2358 | if (di_a->total_avail < di_b->total_avail) | |
2359 | return 1; | |
2360 | return 0; | |
b2117a39 | 2361 | } |
0b86a832 | 2362 | |
73c5de00 AJ |
2363 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
2364 | struct btrfs_root *extent_root, | |
2365 | struct map_lookup **map_ret, | |
2366 | u64 *num_bytes_out, u64 *stripe_size_out, | |
2367 | u64 start, u64 type) | |
b2117a39 | 2368 | { |
73c5de00 AJ |
2369 | struct btrfs_fs_info *info = extent_root->fs_info; |
2370 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
2371 | struct list_head *cur; | |
2372 | struct map_lookup *map = NULL; | |
2373 | struct extent_map_tree *em_tree; | |
2374 | struct extent_map *em; | |
2375 | struct btrfs_device_info *devices_info = NULL; | |
2376 | u64 total_avail; | |
2377 | int num_stripes; /* total number of stripes to allocate */ | |
2378 | int sub_stripes; /* sub_stripes info for map */ | |
2379 | int dev_stripes; /* stripes per dev */ | |
2380 | int devs_max; /* max devs to use */ | |
2381 | int devs_min; /* min devs needed */ | |
2382 | int devs_increment; /* ndevs has to be a multiple of this */ | |
2383 | int ncopies; /* how many copies to data has */ | |
2384 | int ret; | |
2385 | u64 max_stripe_size; | |
2386 | u64 max_chunk_size; | |
2387 | u64 stripe_size; | |
2388 | u64 num_bytes; | |
2389 | int ndevs; | |
2390 | int i; | |
2391 | int j; | |
593060d7 | 2392 | |
73c5de00 AJ |
2393 | if ((type & BTRFS_BLOCK_GROUP_RAID1) && |
2394 | (type & BTRFS_BLOCK_GROUP_DUP)) { | |
2395 | WARN_ON(1); | |
2396 | type &= ~BTRFS_BLOCK_GROUP_DUP; | |
321aecc6 | 2397 | } |
9b3f68b9 | 2398 | |
73c5de00 AJ |
2399 | if (list_empty(&fs_devices->alloc_list)) |
2400 | return -ENOSPC; | |
b2117a39 | 2401 | |
73c5de00 AJ |
2402 | sub_stripes = 1; |
2403 | dev_stripes = 1; | |
2404 | devs_increment = 1; | |
2405 | ncopies = 1; | |
2406 | devs_max = 0; /* 0 == as many as possible */ | |
2407 | devs_min = 1; | |
2408 | ||
2409 | /* | |
2410 | * define the properties of each RAID type. | |
2411 | * FIXME: move this to a global table and use it in all RAID | |
2412 | * calculation code | |
2413 | */ | |
2414 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
2415 | dev_stripes = 2; | |
b2117a39 | 2416 | ncopies = 2; |
73c5de00 AJ |
2417 | devs_max = 1; |
2418 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
2419 | devs_min = 2; | |
2420 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
2421 | devs_increment = 2; | |
b2117a39 | 2422 | ncopies = 2; |
73c5de00 AJ |
2423 | devs_max = 2; |
2424 | devs_min = 2; | |
2425 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
2426 | sub_stripes = 2; | |
2427 | devs_increment = 2; | |
2428 | ncopies = 2; | |
2429 | devs_min = 4; | |
2430 | } else { | |
2431 | devs_max = 1; | |
2432 | } | |
b2117a39 | 2433 | |
9b3f68b9 | 2434 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
2435 | max_stripe_size = 1024 * 1024 * 1024; |
2436 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 2437 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
73c5de00 AJ |
2438 | max_stripe_size = 256 * 1024 * 1024; |
2439 | max_chunk_size = max_stripe_size; | |
a40a90a0 | 2440 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
73c5de00 AJ |
2441 | max_stripe_size = 8 * 1024 * 1024; |
2442 | max_chunk_size = 2 * max_stripe_size; | |
2443 | } else { | |
2444 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
2445 | type); | |
2446 | BUG_ON(1); | |
9b3f68b9 CM |
2447 | } |
2448 | ||
2b82032c YZ |
2449 | /* we don't want a chunk larger than 10% of writeable space */ |
2450 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
2451 | max_chunk_size); | |
9b3f68b9 | 2452 | |
73c5de00 AJ |
2453 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
2454 | GFP_NOFS); | |
2455 | if (!devices_info) | |
2456 | return -ENOMEM; | |
0cad8a11 | 2457 | |
73c5de00 | 2458 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 2459 | |
9f680ce0 | 2460 | /* |
73c5de00 AJ |
2461 | * in the first pass through the devices list, we gather information |
2462 | * about the available holes on each device. | |
9f680ce0 | 2463 | */ |
73c5de00 AJ |
2464 | ndevs = 0; |
2465 | while (cur != &fs_devices->alloc_list) { | |
2466 | struct btrfs_device *device; | |
2467 | u64 max_avail; | |
2468 | u64 dev_offset; | |
b2117a39 | 2469 | |
73c5de00 | 2470 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 2471 | |
73c5de00 | 2472 | cur = cur->next; |
b2117a39 | 2473 | |
73c5de00 AJ |
2474 | if (!device->writeable) { |
2475 | printk(KERN_ERR | |
2476 | "btrfs: read-only device in alloc_list\n"); | |
2477 | WARN_ON(1); | |
2478 | continue; | |
2479 | } | |
b2117a39 | 2480 | |
73c5de00 AJ |
2481 | if (!device->in_fs_metadata) |
2482 | continue; | |
b2117a39 | 2483 | |
73c5de00 AJ |
2484 | if (device->total_bytes > device->bytes_used) |
2485 | total_avail = device->total_bytes - device->bytes_used; | |
2486 | else | |
2487 | total_avail = 0; | |
38c01b96 | 2488 | |
2489 | /* If there is no space on this device, skip it. */ | |
2490 | if (total_avail == 0) | |
2491 | continue; | |
b2117a39 | 2492 | |
73c5de00 AJ |
2493 | ret = find_free_dev_extent(trans, device, |
2494 | max_stripe_size * dev_stripes, | |
2495 | &dev_offset, &max_avail); | |
2496 | if (ret && ret != -ENOSPC) | |
2497 | goto error; | |
b2117a39 | 2498 | |
73c5de00 AJ |
2499 | if (ret == 0) |
2500 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 2501 | |
73c5de00 AJ |
2502 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
2503 | continue; | |
b2117a39 | 2504 | |
73c5de00 AJ |
2505 | devices_info[ndevs].dev_offset = dev_offset; |
2506 | devices_info[ndevs].max_avail = max_avail; | |
2507 | devices_info[ndevs].total_avail = total_avail; | |
2508 | devices_info[ndevs].dev = device; | |
2509 | ++ndevs; | |
2510 | } | |
b2117a39 | 2511 | |
73c5de00 AJ |
2512 | /* |
2513 | * now sort the devices by hole size / available space | |
2514 | */ | |
2515 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
2516 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 2517 | |
73c5de00 AJ |
2518 | /* round down to number of usable stripes */ |
2519 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 2520 | |
73c5de00 AJ |
2521 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
2522 | ret = -ENOSPC; | |
2523 | goto error; | |
b2117a39 | 2524 | } |
9f680ce0 | 2525 | |
73c5de00 AJ |
2526 | if (devs_max && ndevs > devs_max) |
2527 | ndevs = devs_max; | |
2528 | /* | |
2529 | * the primary goal is to maximize the number of stripes, so use as many | |
2530 | * devices as possible, even if the stripes are not maximum sized. | |
2531 | */ | |
2532 | stripe_size = devices_info[ndevs-1].max_avail; | |
2533 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 2534 | |
73c5de00 AJ |
2535 | if (stripe_size * num_stripes > max_chunk_size * ncopies) { |
2536 | stripe_size = max_chunk_size * ncopies; | |
2537 | do_div(stripe_size, num_stripes); | |
b2117a39 | 2538 | } |
b2117a39 | 2539 | |
73c5de00 AJ |
2540 | do_div(stripe_size, dev_stripes); |
2541 | do_div(stripe_size, BTRFS_STRIPE_LEN); | |
2542 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
2543 | |
2544 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
2545 | if (!map) { | |
2546 | ret = -ENOMEM; | |
2547 | goto error; | |
2548 | } | |
2549 | map->num_stripes = num_stripes; | |
9b3f68b9 | 2550 | |
73c5de00 AJ |
2551 | for (i = 0; i < ndevs; ++i) { |
2552 | for (j = 0; j < dev_stripes; ++j) { | |
2553 | int s = i * dev_stripes + j; | |
2554 | map->stripes[s].dev = devices_info[i].dev; | |
2555 | map->stripes[s].physical = devices_info[i].dev_offset + | |
2556 | j * stripe_size; | |
6324fbf3 | 2557 | } |
6324fbf3 | 2558 | } |
2b82032c | 2559 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
2560 | map->stripe_len = BTRFS_STRIPE_LEN; |
2561 | map->io_align = BTRFS_STRIPE_LEN; | |
2562 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 2563 | map->type = type; |
2b82032c | 2564 | map->sub_stripes = sub_stripes; |
0b86a832 | 2565 | |
2b82032c | 2566 | *map_ret = map; |
73c5de00 | 2567 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 2568 | |
73c5de00 AJ |
2569 | *stripe_size_out = stripe_size; |
2570 | *num_bytes_out = num_bytes; | |
0b86a832 | 2571 | |
73c5de00 | 2572 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 2573 | |
172ddd60 | 2574 | em = alloc_extent_map(); |
2b82032c | 2575 | if (!em) { |
b2117a39 MX |
2576 | ret = -ENOMEM; |
2577 | goto error; | |
593060d7 | 2578 | } |
2b82032c YZ |
2579 | em->bdev = (struct block_device *)map; |
2580 | em->start = start; | |
73c5de00 | 2581 | em->len = num_bytes; |
2b82032c YZ |
2582 | em->block_start = 0; |
2583 | em->block_len = em->len; | |
593060d7 | 2584 | |
2b82032c | 2585 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 2586 | write_lock(&em_tree->lock); |
2b82032c | 2587 | ret = add_extent_mapping(em_tree, em); |
890871be | 2588 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2589 | BUG_ON(ret); |
2590 | free_extent_map(em); | |
0b86a832 | 2591 | |
2b82032c YZ |
2592 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
2593 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 2594 | start, num_bytes); |
2b82032c | 2595 | BUG_ON(ret); |
611f0e00 | 2596 | |
73c5de00 AJ |
2597 | for (i = 0; i < map->num_stripes; ++i) { |
2598 | struct btrfs_device *device; | |
2599 | u64 dev_offset; | |
2600 | ||
2601 | device = map->stripes[i].dev; | |
2602 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
2603 | |
2604 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
2605 | info->chunk_root->root_key.objectid, |
2606 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 2607 | start, dev_offset, stripe_size); |
0b86a832 | 2608 | BUG_ON(ret); |
2b82032c YZ |
2609 | } |
2610 | ||
b2117a39 | 2611 | kfree(devices_info); |
2b82032c | 2612 | return 0; |
b2117a39 MX |
2613 | |
2614 | error: | |
2615 | kfree(map); | |
2616 | kfree(devices_info); | |
2617 | return ret; | |
2b82032c YZ |
2618 | } |
2619 | ||
2620 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
2621 | struct btrfs_root *extent_root, | |
2622 | struct map_lookup *map, u64 chunk_offset, | |
2623 | u64 chunk_size, u64 stripe_size) | |
2624 | { | |
2625 | u64 dev_offset; | |
2626 | struct btrfs_key key; | |
2627 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
2628 | struct btrfs_device *device; | |
2629 | struct btrfs_chunk *chunk; | |
2630 | struct btrfs_stripe *stripe; | |
2631 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
2632 | int index = 0; | |
2633 | int ret; | |
2634 | ||
2635 | chunk = kzalloc(item_size, GFP_NOFS); | |
2636 | if (!chunk) | |
2637 | return -ENOMEM; | |
2638 | ||
2639 | index = 0; | |
2640 | while (index < map->num_stripes) { | |
2641 | device = map->stripes[index].dev; | |
2642 | device->bytes_used += stripe_size; | |
0b86a832 CM |
2643 | ret = btrfs_update_device(trans, device); |
2644 | BUG_ON(ret); | |
2b82032c YZ |
2645 | index++; |
2646 | } | |
2647 | ||
2bf64758 JB |
2648 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
2649 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
2650 | map->num_stripes); | |
2651 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
2652 | ||
2b82032c YZ |
2653 | index = 0; |
2654 | stripe = &chunk->stripe; | |
2655 | while (index < map->num_stripes) { | |
2656 | device = map->stripes[index].dev; | |
2657 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 2658 | |
e17cade2 CM |
2659 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
2660 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
2661 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 2662 | stripe++; |
0b86a832 CM |
2663 | index++; |
2664 | } | |
2665 | ||
2b82032c | 2666 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 2667 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
2668 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
2669 | btrfs_set_stack_chunk_type(chunk, map->type); | |
2670 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
2671 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
2672 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 2673 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 2674 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 2675 | |
2b82032c YZ |
2676 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2677 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2678 | key.offset = chunk_offset; | |
0b86a832 | 2679 | |
2b82032c YZ |
2680 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
2681 | BUG_ON(ret); | |
0b86a832 | 2682 | |
2b82032c YZ |
2683 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2684 | ret = btrfs_add_system_chunk(trans, chunk_root, &key, chunk, | |
2685 | item_size); | |
8f18cf13 CM |
2686 | BUG_ON(ret); |
2687 | } | |
1abe9b8a | 2688 | |
0b86a832 | 2689 | kfree(chunk); |
2b82032c YZ |
2690 | return 0; |
2691 | } | |
0b86a832 | 2692 | |
2b82032c YZ |
2693 | /* |
2694 | * Chunk allocation falls into two parts. The first part does works | |
2695 | * that make the new allocated chunk useable, but not do any operation | |
2696 | * that modifies the chunk tree. The second part does the works that | |
2697 | * require modifying the chunk tree. This division is important for the | |
2698 | * bootstrap process of adding storage to a seed btrfs. | |
2699 | */ | |
2700 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
2701 | struct btrfs_root *extent_root, u64 type) | |
2702 | { | |
2703 | u64 chunk_offset; | |
2704 | u64 chunk_size; | |
2705 | u64 stripe_size; | |
2706 | struct map_lookup *map; | |
2707 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
2708 | int ret; | |
2709 | ||
2710 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
2711 | &chunk_offset); | |
2712 | if (ret) | |
2713 | return ret; | |
2714 | ||
2715 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
2716 | &stripe_size, chunk_offset, type); | |
2717 | if (ret) | |
2718 | return ret; | |
2719 | ||
2720 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
2721 | chunk_size, stripe_size); | |
2722 | BUG_ON(ret); | |
2723 | return 0; | |
2724 | } | |
2725 | ||
d397712b | 2726 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
2727 | struct btrfs_root *root, |
2728 | struct btrfs_device *device) | |
2729 | { | |
2730 | u64 chunk_offset; | |
2731 | u64 sys_chunk_offset; | |
2732 | u64 chunk_size; | |
2733 | u64 sys_chunk_size; | |
2734 | u64 stripe_size; | |
2735 | u64 sys_stripe_size; | |
2736 | u64 alloc_profile; | |
2737 | struct map_lookup *map; | |
2738 | struct map_lookup *sys_map; | |
2739 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2740 | struct btrfs_root *extent_root = fs_info->extent_root; | |
2741 | int ret; | |
2742 | ||
2743 | ret = find_next_chunk(fs_info->chunk_root, | |
2744 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
2745 | if (ret) |
2746 | return ret; | |
2b82032c YZ |
2747 | |
2748 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
2749 | (fs_info->metadata_alloc_profile & | |
2750 | fs_info->avail_metadata_alloc_bits); | |
2751 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); | |
2752 | ||
2753 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
2754 | &stripe_size, chunk_offset, alloc_profile); | |
2755 | BUG_ON(ret); | |
2756 | ||
2757 | sys_chunk_offset = chunk_offset + chunk_size; | |
2758 | ||
2759 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
2760 | (fs_info->system_alloc_profile & | |
2761 | fs_info->avail_system_alloc_bits); | |
2762 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); | |
2763 | ||
2764 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
2765 | &sys_chunk_size, &sys_stripe_size, | |
2766 | sys_chunk_offset, alloc_profile); | |
2767 | BUG_ON(ret); | |
2768 | ||
2769 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
2770 | BUG_ON(ret); | |
2771 | ||
2772 | /* | |
2773 | * Modifying chunk tree needs allocating new blocks from both | |
2774 | * system block group and metadata block group. So we only can | |
2775 | * do operations require modifying the chunk tree after both | |
2776 | * block groups were created. | |
2777 | */ | |
2778 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
2779 | chunk_size, stripe_size); | |
2780 | BUG_ON(ret); | |
2781 | ||
2782 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
2783 | sys_chunk_offset, sys_chunk_size, | |
2784 | sys_stripe_size); | |
b248a415 | 2785 | BUG_ON(ret); |
2b82032c YZ |
2786 | return 0; |
2787 | } | |
2788 | ||
2789 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
2790 | { | |
2791 | struct extent_map *em; | |
2792 | struct map_lookup *map; | |
2793 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
2794 | int readonly = 0; | |
2795 | int i; | |
2796 | ||
890871be | 2797 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 2798 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 2799 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
2800 | if (!em) |
2801 | return 1; | |
2802 | ||
f48b9075 JB |
2803 | if (btrfs_test_opt(root, DEGRADED)) { |
2804 | free_extent_map(em); | |
2805 | return 0; | |
2806 | } | |
2807 | ||
2b82032c YZ |
2808 | map = (struct map_lookup *)em->bdev; |
2809 | for (i = 0; i < map->num_stripes; i++) { | |
2810 | if (!map->stripes[i].dev->writeable) { | |
2811 | readonly = 1; | |
2812 | break; | |
2813 | } | |
2814 | } | |
0b86a832 | 2815 | free_extent_map(em); |
2b82032c | 2816 | return readonly; |
0b86a832 CM |
2817 | } |
2818 | ||
2819 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
2820 | { | |
a8067e02 | 2821 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
2822 | } |
2823 | ||
2824 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
2825 | { | |
2826 | struct extent_map *em; | |
2827 | ||
d397712b | 2828 | while (1) { |
890871be | 2829 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
2830 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
2831 | if (em) | |
2832 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 2833 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
2834 | if (!em) |
2835 | break; | |
2836 | kfree(em->bdev); | |
2837 | /* once for us */ | |
2838 | free_extent_map(em); | |
2839 | /* once for the tree */ | |
2840 | free_extent_map(em); | |
2841 | } | |
2842 | } | |
2843 | ||
f188591e CM |
2844 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
2845 | { | |
2846 | struct extent_map *em; | |
2847 | struct map_lookup *map; | |
2848 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
2849 | int ret; | |
2850 | ||
890871be | 2851 | read_lock(&em_tree->lock); |
f188591e | 2852 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 2853 | read_unlock(&em_tree->lock); |
f188591e CM |
2854 | BUG_ON(!em); |
2855 | ||
2856 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
2857 | map = (struct map_lookup *)em->bdev; | |
2858 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
2859 | ret = map->num_stripes; | |
321aecc6 CM |
2860 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
2861 | ret = map->sub_stripes; | |
f188591e CM |
2862 | else |
2863 | ret = 1; | |
2864 | free_extent_map(em); | |
f188591e CM |
2865 | return ret; |
2866 | } | |
2867 | ||
dfe25020 CM |
2868 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
2869 | int optimal) | |
2870 | { | |
2871 | int i; | |
2872 | if (map->stripes[optimal].dev->bdev) | |
2873 | return optimal; | |
2874 | for (i = first; i < first + num; i++) { | |
2875 | if (map->stripes[i].dev->bdev) | |
2876 | return i; | |
2877 | } | |
2878 | /* we couldn't find one that doesn't fail. Just return something | |
2879 | * and the io error handling code will clean up eventually | |
2880 | */ | |
2881 | return optimal; | |
2882 | } | |
2883 | ||
f2d8d74d CM |
2884 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
2885 | u64 logical, u64 *length, | |
a1d3c478 | 2886 | struct btrfs_bio **bbio_ret, |
7eaceacc | 2887 | int mirror_num) |
0b86a832 CM |
2888 | { |
2889 | struct extent_map *em; | |
2890 | struct map_lookup *map; | |
2891 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
2892 | u64 offset; | |
593060d7 | 2893 | u64 stripe_offset; |
fce3bb9a | 2894 | u64 stripe_end_offset; |
593060d7 | 2895 | u64 stripe_nr; |
fce3bb9a LD |
2896 | u64 stripe_nr_orig; |
2897 | u64 stripe_nr_end; | |
cea9e445 | 2898 | int stripes_allocated = 8; |
321aecc6 | 2899 | int stripes_required = 1; |
593060d7 | 2900 | int stripe_index; |
cea9e445 | 2901 | int i; |
f2d8d74d | 2902 | int num_stripes; |
a236aed1 | 2903 | int max_errors = 0; |
a1d3c478 | 2904 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 2905 | |
a1d3c478 | 2906 | if (bbio_ret && !(rw & (REQ_WRITE | REQ_DISCARD))) |
cea9e445 | 2907 | stripes_allocated = 1; |
cea9e445 | 2908 | again: |
a1d3c478 JS |
2909 | if (bbio_ret) { |
2910 | bbio = kzalloc(btrfs_bio_size(stripes_allocated), | |
cea9e445 | 2911 | GFP_NOFS); |
a1d3c478 | 2912 | if (!bbio) |
cea9e445 | 2913 | return -ENOMEM; |
a236aed1 | 2914 | |
a1d3c478 | 2915 | atomic_set(&bbio->error, 0); |
cea9e445 | 2916 | } |
0b86a832 | 2917 | |
890871be | 2918 | read_lock(&em_tree->lock); |
0b86a832 | 2919 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 2920 | read_unlock(&em_tree->lock); |
f2d8d74d | 2921 | |
3b951516 | 2922 | if (!em) { |
d397712b CM |
2923 | printk(KERN_CRIT "unable to find logical %llu len %llu\n", |
2924 | (unsigned long long)logical, | |
2925 | (unsigned long long)*length); | |
f2d8d74d | 2926 | BUG(); |
3b951516 | 2927 | } |
0b86a832 CM |
2928 | |
2929 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
2930 | map = (struct map_lookup *)em->bdev; | |
2931 | offset = logical - em->start; | |
593060d7 | 2932 | |
f188591e CM |
2933 | if (mirror_num > map->num_stripes) |
2934 | mirror_num = 0; | |
2935 | ||
a1d3c478 | 2936 | /* if our btrfs_bio struct is too small, back off and try again */ |
7b6d91da | 2937 | if (rw & REQ_WRITE) { |
321aecc6 CM |
2938 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | |
2939 | BTRFS_BLOCK_GROUP_DUP)) { | |
2940 | stripes_required = map->num_stripes; | |
a236aed1 | 2941 | max_errors = 1; |
321aecc6 CM |
2942 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
2943 | stripes_required = map->sub_stripes; | |
a236aed1 | 2944 | max_errors = 1; |
321aecc6 CM |
2945 | } |
2946 | } | |
fce3bb9a LD |
2947 | if (rw & REQ_DISCARD) { |
2948 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
2949 | BTRFS_BLOCK_GROUP_RAID1 | | |
2950 | BTRFS_BLOCK_GROUP_DUP | | |
2951 | BTRFS_BLOCK_GROUP_RAID10)) { | |
2952 | stripes_required = map->num_stripes; | |
2953 | } | |
2954 | } | |
a1d3c478 | 2955 | if (bbio_ret && (rw & (REQ_WRITE | REQ_DISCARD)) && |
321aecc6 | 2956 | stripes_allocated < stripes_required) { |
cea9e445 | 2957 | stripes_allocated = map->num_stripes; |
cea9e445 | 2958 | free_extent_map(em); |
a1d3c478 | 2959 | kfree(bbio); |
cea9e445 CM |
2960 | goto again; |
2961 | } | |
593060d7 CM |
2962 | stripe_nr = offset; |
2963 | /* | |
2964 | * stripe_nr counts the total number of stripes we have to stride | |
2965 | * to get to this block | |
2966 | */ | |
2967 | do_div(stripe_nr, map->stripe_len); | |
2968 | ||
2969 | stripe_offset = stripe_nr * map->stripe_len; | |
2970 | BUG_ON(offset < stripe_offset); | |
2971 | ||
2972 | /* stripe_offset is the offset of this block in its stripe*/ | |
2973 | stripe_offset = offset - stripe_offset; | |
2974 | ||
fce3bb9a LD |
2975 | if (rw & REQ_DISCARD) |
2976 | *length = min_t(u64, em->len - offset, *length); | |
2977 | else if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
2978 | BTRFS_BLOCK_GROUP_RAID1 | | |
2979 | BTRFS_BLOCK_GROUP_RAID10 | | |
2980 | BTRFS_BLOCK_GROUP_DUP)) { | |
cea9e445 CM |
2981 | /* we limit the length of each bio to what fits in a stripe */ |
2982 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 2983 | map->stripe_len - stripe_offset); |
cea9e445 CM |
2984 | } else { |
2985 | *length = em->len - offset; | |
2986 | } | |
f2d8d74d | 2987 | |
a1d3c478 | 2988 | if (!bbio_ret) |
cea9e445 CM |
2989 | goto out; |
2990 | ||
f2d8d74d | 2991 | num_stripes = 1; |
cea9e445 | 2992 | stripe_index = 0; |
fce3bb9a LD |
2993 | stripe_nr_orig = stripe_nr; |
2994 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
2995 | (~(map->stripe_len - 1)); | |
2996 | do_div(stripe_nr_end, map->stripe_len); | |
2997 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
2998 | (offset + *length); | |
2999 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3000 | if (rw & REQ_DISCARD) | |
3001 | num_stripes = min_t(u64, map->num_stripes, | |
3002 | stripe_nr_end - stripe_nr_orig); | |
3003 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3004 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3005 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3006 | num_stripes = map->num_stripes; |
2fff734f | 3007 | else if (mirror_num) |
f188591e | 3008 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3009 | else { |
3010 | stripe_index = find_live_mirror(map, 0, | |
3011 | map->num_stripes, | |
3012 | current->pid % map->num_stripes); | |
a1d3c478 | 3013 | mirror_num = stripe_index + 1; |
dfe25020 | 3014 | } |
2fff734f | 3015 | |
611f0e00 | 3016 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3017 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3018 | num_stripes = map->num_stripes; |
a1d3c478 | 3019 | } else if (mirror_num) { |
f188591e | 3020 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3021 | } else { |
3022 | mirror_num = 1; | |
3023 | } | |
2fff734f | 3024 | |
321aecc6 CM |
3025 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3026 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3027 | |
3028 | stripe_index = do_div(stripe_nr, factor); | |
3029 | stripe_index *= map->sub_stripes; | |
3030 | ||
7eaceacc | 3031 | if (rw & REQ_WRITE) |
f2d8d74d | 3032 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3033 | else if (rw & REQ_DISCARD) |
3034 | num_stripes = min_t(u64, map->sub_stripes * | |
3035 | (stripe_nr_end - stripe_nr_orig), | |
3036 | map->num_stripes); | |
321aecc6 CM |
3037 | else if (mirror_num) |
3038 | stripe_index += mirror_num - 1; | |
dfe25020 CM |
3039 | else { |
3040 | stripe_index = find_live_mirror(map, stripe_index, | |
3041 | map->sub_stripes, stripe_index + | |
3042 | current->pid % map->sub_stripes); | |
a1d3c478 | 3043 | mirror_num = stripe_index + 1; |
dfe25020 | 3044 | } |
8790d502 CM |
3045 | } else { |
3046 | /* | |
3047 | * after this do_div call, stripe_nr is the number of stripes | |
3048 | * on this device we have to walk to find the data, and | |
3049 | * stripe_index is the number of our device in the stripe array | |
3050 | */ | |
3051 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3052 | mirror_num = stripe_index + 1; |
8790d502 | 3053 | } |
593060d7 | 3054 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3055 | |
fce3bb9a LD |
3056 | if (rw & REQ_DISCARD) { |
3057 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3058 | bbio->stripes[i].physical = |
f2d8d74d CM |
3059 | map->stripes[stripe_index].physical + |
3060 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3061 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a LD |
3062 | |
3063 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3064 | u64 stripes; | |
d9d04879 | 3065 | u32 last_stripe = 0; |
fce3bb9a LD |
3066 | int j; |
3067 | ||
d9d04879 CM |
3068 | div_u64_rem(stripe_nr_end - 1, |
3069 | map->num_stripes, | |
3070 | &last_stripe); | |
3071 | ||
fce3bb9a | 3072 | for (j = 0; j < map->num_stripes; j++) { |
d9d04879 CM |
3073 | u32 test; |
3074 | ||
3075 | div_u64_rem(stripe_nr_end - 1 - j, | |
3076 | map->num_stripes, &test); | |
3077 | if (test == stripe_index) | |
fce3bb9a LD |
3078 | break; |
3079 | } | |
3080 | stripes = stripe_nr_end - 1 - j; | |
3081 | do_div(stripes, map->num_stripes); | |
a1d3c478 | 3082 | bbio->stripes[i].length = map->stripe_len * |
fce3bb9a LD |
3083 | (stripes - stripe_nr + 1); |
3084 | ||
3085 | if (i == 0) { | |
a1d3c478 | 3086 | bbio->stripes[i].length -= |
fce3bb9a LD |
3087 | stripe_offset; |
3088 | stripe_offset = 0; | |
3089 | } | |
3090 | if (stripe_index == last_stripe) | |
a1d3c478 | 3091 | bbio->stripes[i].length -= |
fce3bb9a LD |
3092 | stripe_end_offset; |
3093 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
3094 | u64 stripes; | |
3095 | int j; | |
3096 | int factor = map->num_stripes / | |
3097 | map->sub_stripes; | |
d9d04879 CM |
3098 | u32 last_stripe = 0; |
3099 | ||
3100 | div_u64_rem(stripe_nr_end - 1, | |
3101 | factor, &last_stripe); | |
fce3bb9a LD |
3102 | last_stripe *= map->sub_stripes; |
3103 | ||
3104 | for (j = 0; j < factor; j++) { | |
d9d04879 CM |
3105 | u32 test; |
3106 | ||
3107 | div_u64_rem(stripe_nr_end - 1 - j, | |
3108 | factor, &test); | |
3109 | ||
3110 | if (test == | |
fce3bb9a LD |
3111 | stripe_index / map->sub_stripes) |
3112 | break; | |
3113 | } | |
3114 | stripes = stripe_nr_end - 1 - j; | |
3115 | do_div(stripes, factor); | |
a1d3c478 | 3116 | bbio->stripes[i].length = map->stripe_len * |
fce3bb9a LD |
3117 | (stripes - stripe_nr + 1); |
3118 | ||
3119 | if (i < map->sub_stripes) { | |
a1d3c478 | 3120 | bbio->stripes[i].length -= |
fce3bb9a LD |
3121 | stripe_offset; |
3122 | if (i == map->sub_stripes - 1) | |
3123 | stripe_offset = 0; | |
3124 | } | |
3125 | if (stripe_index >= last_stripe && | |
3126 | stripe_index <= (last_stripe + | |
3127 | map->sub_stripes - 1)) { | |
a1d3c478 | 3128 | bbio->stripes[i].length -= |
fce3bb9a LD |
3129 | stripe_end_offset; |
3130 | } | |
3131 | } else | |
a1d3c478 | 3132 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
3133 | |
3134 | stripe_index++; | |
3135 | if (stripe_index == map->num_stripes) { | |
3136 | /* This could only happen for RAID0/10 */ | |
3137 | stripe_index = 0; | |
3138 | stripe_nr++; | |
3139 | } | |
3140 | } | |
3141 | } else { | |
3142 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3143 | bbio->stripes[i].physical = |
212a17ab LT |
3144 | map->stripes[stripe_index].physical + |
3145 | stripe_offset + | |
3146 | stripe_nr * map->stripe_len; | |
a1d3c478 | 3147 | bbio->stripes[i].dev = |
212a17ab | 3148 | map->stripes[stripe_index].dev; |
fce3bb9a | 3149 | stripe_index++; |
f2d8d74d | 3150 | } |
593060d7 | 3151 | } |
a1d3c478 JS |
3152 | if (bbio_ret) { |
3153 | *bbio_ret = bbio; | |
3154 | bbio->num_stripes = num_stripes; | |
3155 | bbio->max_errors = max_errors; | |
3156 | bbio->mirror_num = mirror_num; | |
f2d8d74d | 3157 | } |
cea9e445 | 3158 | out: |
0b86a832 | 3159 | free_extent_map(em); |
0b86a832 CM |
3160 | return 0; |
3161 | } | |
3162 | ||
f2d8d74d CM |
3163 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3164 | u64 logical, u64 *length, | |
a1d3c478 | 3165 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 3166 | { |
a1d3c478 | 3167 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 3168 | mirror_num); |
f2d8d74d CM |
3169 | } |
3170 | ||
a512bbf8 YZ |
3171 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
3172 | u64 chunk_start, u64 physical, u64 devid, | |
3173 | u64 **logical, int *naddrs, int *stripe_len) | |
3174 | { | |
3175 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3176 | struct extent_map *em; | |
3177 | struct map_lookup *map; | |
3178 | u64 *buf; | |
3179 | u64 bytenr; | |
3180 | u64 length; | |
3181 | u64 stripe_nr; | |
3182 | int i, j, nr = 0; | |
3183 | ||
890871be | 3184 | read_lock(&em_tree->lock); |
a512bbf8 | 3185 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 3186 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
3187 | |
3188 | BUG_ON(!em || em->start != chunk_start); | |
3189 | map = (struct map_lookup *)em->bdev; | |
3190 | ||
3191 | length = em->len; | |
3192 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
3193 | do_div(length, map->num_stripes / map->sub_stripes); | |
3194 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3195 | do_div(length, map->num_stripes); | |
3196 | ||
3197 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
3198 | BUG_ON(!buf); | |
3199 | ||
3200 | for (i = 0; i < map->num_stripes; i++) { | |
3201 | if (devid && map->stripes[i].dev->devid != devid) | |
3202 | continue; | |
3203 | if (map->stripes[i].physical > physical || | |
3204 | map->stripes[i].physical + length <= physical) | |
3205 | continue; | |
3206 | ||
3207 | stripe_nr = physical - map->stripes[i].physical; | |
3208 | do_div(stripe_nr, map->stripe_len); | |
3209 | ||
3210 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
3211 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3212 | do_div(stripe_nr, map->sub_stripes); | |
3213 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3214 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3215 | } | |
3216 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 3217 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
3218 | for (j = 0; j < nr; j++) { |
3219 | if (buf[j] == bytenr) | |
3220 | break; | |
3221 | } | |
934d375b CM |
3222 | if (j == nr) { |
3223 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 3224 | buf[nr++] = bytenr; |
934d375b | 3225 | } |
a512bbf8 YZ |
3226 | } |
3227 | ||
a512bbf8 YZ |
3228 | *logical = buf; |
3229 | *naddrs = nr; | |
3230 | *stripe_len = map->stripe_len; | |
3231 | ||
3232 | free_extent_map(em); | |
3233 | return 0; | |
f2d8d74d CM |
3234 | } |
3235 | ||
a1d3c478 | 3236 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 3237 | { |
a1d3c478 | 3238 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 3239 | int is_orig_bio = 0; |
8790d502 | 3240 | |
8790d502 | 3241 | if (err) |
a1d3c478 | 3242 | atomic_inc(&bbio->error); |
8790d502 | 3243 | |
a1d3c478 | 3244 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
3245 | is_orig_bio = 1; |
3246 | ||
a1d3c478 | 3247 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
3248 | if (!is_orig_bio) { |
3249 | bio_put(bio); | |
a1d3c478 | 3250 | bio = bbio->orig_bio; |
7d2b4daa | 3251 | } |
a1d3c478 JS |
3252 | bio->bi_private = bbio->private; |
3253 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
3254 | bio->bi_bdev = (struct block_device *) |
3255 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
3256 | /* only send an error to the higher layers if it is |
3257 | * beyond the tolerance of the multi-bio | |
3258 | */ | |
a1d3c478 | 3259 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 3260 | err = -EIO; |
1259ab75 CM |
3261 | } else if (err) { |
3262 | /* | |
3263 | * this bio is actually up to date, we didn't | |
3264 | * go over the max number of errors | |
3265 | */ | |
3266 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 3267 | err = 0; |
1259ab75 | 3268 | } |
a1d3c478 | 3269 | kfree(bbio); |
8790d502 CM |
3270 | |
3271 | bio_endio(bio, err); | |
7d2b4daa | 3272 | } else if (!is_orig_bio) { |
8790d502 CM |
3273 | bio_put(bio); |
3274 | } | |
8790d502 CM |
3275 | } |
3276 | ||
8b712842 CM |
3277 | struct async_sched { |
3278 | struct bio *bio; | |
3279 | int rw; | |
3280 | struct btrfs_fs_info *info; | |
3281 | struct btrfs_work work; | |
3282 | }; | |
3283 | ||
3284 | /* | |
3285 | * see run_scheduled_bios for a description of why bios are collected for | |
3286 | * async submit. | |
3287 | * | |
3288 | * This will add one bio to the pending list for a device and make sure | |
3289 | * the work struct is scheduled. | |
3290 | */ | |
d397712b | 3291 | static noinline int schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
3292 | struct btrfs_device *device, |
3293 | int rw, struct bio *bio) | |
8b712842 CM |
3294 | { |
3295 | int should_queue = 1; | |
ffbd517d | 3296 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
3297 | |
3298 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 3299 | if (!(rw & REQ_WRITE)) { |
492bb6de | 3300 | bio_get(bio); |
8b712842 | 3301 | submit_bio(rw, bio); |
492bb6de | 3302 | bio_put(bio); |
8b712842 CM |
3303 | return 0; |
3304 | } | |
3305 | ||
3306 | /* | |
0986fe9e | 3307 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
3308 | * higher layers. Otherwise, the async bio makes it appear we have |
3309 | * made progress against dirty pages when we've really just put it | |
3310 | * on a queue for later | |
3311 | */ | |
0986fe9e | 3312 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 3313 | WARN_ON(bio->bi_next); |
8b712842 CM |
3314 | bio->bi_next = NULL; |
3315 | bio->bi_rw |= rw; | |
3316 | ||
3317 | spin_lock(&device->io_lock); | |
7b6d91da | 3318 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
3319 | pending_bios = &device->pending_sync_bios; |
3320 | else | |
3321 | pending_bios = &device->pending_bios; | |
8b712842 | 3322 | |
ffbd517d CM |
3323 | if (pending_bios->tail) |
3324 | pending_bios->tail->bi_next = bio; | |
8b712842 | 3325 | |
ffbd517d CM |
3326 | pending_bios->tail = bio; |
3327 | if (!pending_bios->head) | |
3328 | pending_bios->head = bio; | |
8b712842 CM |
3329 | if (device->running_pending) |
3330 | should_queue = 0; | |
3331 | ||
3332 | spin_unlock(&device->io_lock); | |
3333 | ||
3334 | if (should_queue) | |
1cc127b5 CM |
3335 | btrfs_queue_worker(&root->fs_info->submit_workers, |
3336 | &device->work); | |
8b712842 CM |
3337 | return 0; |
3338 | } | |
3339 | ||
f188591e | 3340 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 3341 | int mirror_num, int async_submit) |
0b86a832 CM |
3342 | { |
3343 | struct btrfs_mapping_tree *map_tree; | |
3344 | struct btrfs_device *dev; | |
8790d502 | 3345 | struct bio *first_bio = bio; |
a62b9401 | 3346 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
3347 | u64 length = 0; |
3348 | u64 map_length; | |
0b86a832 | 3349 | int ret; |
8790d502 CM |
3350 | int dev_nr = 0; |
3351 | int total_devs = 1; | |
a1d3c478 | 3352 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3353 | |
f2d8d74d | 3354 | length = bio->bi_size; |
0b86a832 CM |
3355 | map_tree = &root->fs_info->mapping_tree; |
3356 | map_length = length; | |
cea9e445 | 3357 | |
a1d3c478 | 3358 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 3359 | mirror_num); |
cea9e445 CM |
3360 | BUG_ON(ret); |
3361 | ||
a1d3c478 | 3362 | total_devs = bbio->num_stripes; |
cea9e445 | 3363 | if (map_length < length) { |
d397712b CM |
3364 | printk(KERN_CRIT "mapping failed logical %llu bio len %llu " |
3365 | "len %llu\n", (unsigned long long)logical, | |
3366 | (unsigned long long)length, | |
3367 | (unsigned long long)map_length); | |
cea9e445 CM |
3368 | BUG(); |
3369 | } | |
a1d3c478 JS |
3370 | |
3371 | bbio->orig_bio = first_bio; | |
3372 | bbio->private = first_bio->bi_private; | |
3373 | bbio->end_io = first_bio->bi_end_io; | |
3374 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 3375 | |
d397712b | 3376 | while (dev_nr < total_devs) { |
a1d3c478 JS |
3377 | if (dev_nr < total_devs - 1) { |
3378 | bio = bio_clone(first_bio, GFP_NOFS); | |
3379 | BUG_ON(!bio); | |
3380 | } else { | |
3381 | bio = first_bio; | |
8790d502 | 3382 | } |
a1d3c478 JS |
3383 | bio->bi_private = bbio; |
3384 | bio->bi_end_io = btrfs_end_bio; | |
3385 | bio->bi_sector = bbio->stripes[dev_nr].physical >> 9; | |
3386 | dev = bbio->stripes[dev_nr].dev; | |
18e503d6 | 3387 | if (dev && dev->bdev && (rw != WRITE || dev->writeable)) { |
a1d3c478 JS |
3388 | pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu " |
3389 | "(%s id %llu), size=%u\n", rw, | |
3390 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
3391 | dev->name, dev->devid, bio->bi_size); | |
dfe25020 | 3392 | bio->bi_bdev = dev->bdev; |
8b712842 CM |
3393 | if (async_submit) |
3394 | schedule_bio(root, dev, rw, bio); | |
3395 | else | |
3396 | submit_bio(rw, bio); | |
dfe25020 CM |
3397 | } else { |
3398 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
3399 | bio->bi_sector = logical >> 9; | |
dfe25020 | 3400 | bio_endio(bio, -EIO); |
dfe25020 | 3401 | } |
8790d502 CM |
3402 | dev_nr++; |
3403 | } | |
0b86a832 CM |
3404 | return 0; |
3405 | } | |
3406 | ||
a443755f | 3407 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 3408 | u8 *uuid, u8 *fsid) |
0b86a832 | 3409 | { |
2b82032c YZ |
3410 | struct btrfs_device *device; |
3411 | struct btrfs_fs_devices *cur_devices; | |
3412 | ||
3413 | cur_devices = root->fs_info->fs_devices; | |
3414 | while (cur_devices) { | |
3415 | if (!fsid || | |
3416 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
3417 | device = __find_device(&cur_devices->devices, | |
3418 | devid, uuid); | |
3419 | if (device) | |
3420 | return device; | |
3421 | } | |
3422 | cur_devices = cur_devices->seed; | |
3423 | } | |
3424 | return NULL; | |
0b86a832 CM |
3425 | } |
3426 | ||
dfe25020 CM |
3427 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
3428 | u64 devid, u8 *dev_uuid) | |
3429 | { | |
3430 | struct btrfs_device *device; | |
3431 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
3432 | ||
3433 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 3434 | if (!device) |
3435 | return NULL; | |
dfe25020 CM |
3436 | list_add(&device->dev_list, |
3437 | &fs_devices->devices); | |
dfe25020 CM |
3438 | device->dev_root = root->fs_info->dev_root; |
3439 | device->devid = devid; | |
8b712842 | 3440 | device->work.func = pending_bios_fn; |
e4404d6e | 3441 | device->fs_devices = fs_devices; |
cd02dca5 | 3442 | device->missing = 1; |
dfe25020 | 3443 | fs_devices->num_devices++; |
cd02dca5 | 3444 | fs_devices->missing_devices++; |
dfe25020 | 3445 | spin_lock_init(&device->io_lock); |
d20f7043 | 3446 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
3447 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
3448 | return device; | |
3449 | } | |
3450 | ||
0b86a832 CM |
3451 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
3452 | struct extent_buffer *leaf, | |
3453 | struct btrfs_chunk *chunk) | |
3454 | { | |
3455 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3456 | struct map_lookup *map; | |
3457 | struct extent_map *em; | |
3458 | u64 logical; | |
3459 | u64 length; | |
3460 | u64 devid; | |
a443755f | 3461 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 3462 | int num_stripes; |
0b86a832 | 3463 | int ret; |
593060d7 | 3464 | int i; |
0b86a832 | 3465 | |
e17cade2 CM |
3466 | logical = key->offset; |
3467 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 3468 | |
890871be | 3469 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 3470 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 3471 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
3472 | |
3473 | /* already mapped? */ | |
3474 | if (em && em->start <= logical && em->start + em->len > logical) { | |
3475 | free_extent_map(em); | |
0b86a832 CM |
3476 | return 0; |
3477 | } else if (em) { | |
3478 | free_extent_map(em); | |
3479 | } | |
0b86a832 | 3480 | |
172ddd60 | 3481 | em = alloc_extent_map(); |
0b86a832 CM |
3482 | if (!em) |
3483 | return -ENOMEM; | |
593060d7 CM |
3484 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
3485 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
3486 | if (!map) { |
3487 | free_extent_map(em); | |
3488 | return -ENOMEM; | |
3489 | } | |
3490 | ||
3491 | em->bdev = (struct block_device *)map; | |
3492 | em->start = logical; | |
3493 | em->len = length; | |
3494 | em->block_start = 0; | |
c8b97818 | 3495 | em->block_len = em->len; |
0b86a832 | 3496 | |
593060d7 CM |
3497 | map->num_stripes = num_stripes; |
3498 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
3499 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
3500 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
3501 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
3502 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 3503 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
3504 | for (i = 0; i < num_stripes; i++) { |
3505 | map->stripes[i].physical = | |
3506 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
3507 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
3508 | read_extent_buffer(leaf, uuid, (unsigned long) |
3509 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
3510 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
3511 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
3512 | NULL); | |
dfe25020 | 3513 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
3514 | kfree(map); |
3515 | free_extent_map(em); | |
3516 | return -EIO; | |
3517 | } | |
dfe25020 CM |
3518 | if (!map->stripes[i].dev) { |
3519 | map->stripes[i].dev = | |
3520 | add_missing_dev(root, devid, uuid); | |
3521 | if (!map->stripes[i].dev) { | |
3522 | kfree(map); | |
3523 | free_extent_map(em); | |
3524 | return -EIO; | |
3525 | } | |
3526 | } | |
3527 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
3528 | } |
3529 | ||
890871be | 3530 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 3531 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 3532 | write_unlock(&map_tree->map_tree.lock); |
b248a415 | 3533 | BUG_ON(ret); |
0b86a832 CM |
3534 | free_extent_map(em); |
3535 | ||
3536 | return 0; | |
3537 | } | |
3538 | ||
3539 | static int fill_device_from_item(struct extent_buffer *leaf, | |
3540 | struct btrfs_dev_item *dev_item, | |
3541 | struct btrfs_device *device) | |
3542 | { | |
3543 | unsigned long ptr; | |
0b86a832 CM |
3544 | |
3545 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
3546 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
3547 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
3548 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
3549 | device->type = btrfs_device_type(leaf, dev_item); | |
3550 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
3551 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
3552 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
3553 | |
3554 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 3555 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 3556 | |
0b86a832 CM |
3557 | return 0; |
3558 | } | |
3559 | ||
2b82032c YZ |
3560 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
3561 | { | |
3562 | struct btrfs_fs_devices *fs_devices; | |
3563 | int ret; | |
3564 | ||
3565 | mutex_lock(&uuid_mutex); | |
3566 | ||
3567 | fs_devices = root->fs_info->fs_devices->seed; | |
3568 | while (fs_devices) { | |
3569 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
3570 | ret = 0; | |
3571 | goto out; | |
3572 | } | |
3573 | fs_devices = fs_devices->seed; | |
3574 | } | |
3575 | ||
3576 | fs_devices = find_fsid(fsid); | |
3577 | if (!fs_devices) { | |
3578 | ret = -ENOENT; | |
3579 | goto out; | |
3580 | } | |
e4404d6e YZ |
3581 | |
3582 | fs_devices = clone_fs_devices(fs_devices); | |
3583 | if (IS_ERR(fs_devices)) { | |
3584 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
3585 | goto out; |
3586 | } | |
3587 | ||
97288f2c | 3588 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 3589 | root->fs_info->bdev_holder); |
2b82032c YZ |
3590 | if (ret) |
3591 | goto out; | |
3592 | ||
3593 | if (!fs_devices->seeding) { | |
3594 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 3595 | free_fs_devices(fs_devices); |
2b82032c YZ |
3596 | ret = -EINVAL; |
3597 | goto out; | |
3598 | } | |
3599 | ||
3600 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
3601 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c YZ |
3602 | out: |
3603 | mutex_unlock(&uuid_mutex); | |
3604 | return ret; | |
3605 | } | |
3606 | ||
0d81ba5d | 3607 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
3608 | struct extent_buffer *leaf, |
3609 | struct btrfs_dev_item *dev_item) | |
3610 | { | |
3611 | struct btrfs_device *device; | |
3612 | u64 devid; | |
3613 | int ret; | |
2b82032c | 3614 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
3615 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
3616 | ||
0b86a832 | 3617 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
3618 | read_extent_buffer(leaf, dev_uuid, |
3619 | (unsigned long)btrfs_device_uuid(dev_item), | |
3620 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
3621 | read_extent_buffer(leaf, fs_uuid, |
3622 | (unsigned long)btrfs_device_fsid(dev_item), | |
3623 | BTRFS_UUID_SIZE); | |
3624 | ||
3625 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
3626 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 3627 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 3628 | return ret; |
2b82032c YZ |
3629 | } |
3630 | ||
3631 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
3632 | if (!device || !device->bdev) { | |
e4404d6e | 3633 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
3634 | return -EIO; |
3635 | ||
3636 | if (!device) { | |
d397712b CM |
3637 | printk(KERN_WARNING "warning devid %llu missing\n", |
3638 | (unsigned long long)devid); | |
2b82032c YZ |
3639 | device = add_missing_dev(root, devid, dev_uuid); |
3640 | if (!device) | |
3641 | return -ENOMEM; | |
cd02dca5 CM |
3642 | } else if (!device->missing) { |
3643 | /* | |
3644 | * this happens when a device that was properly setup | |
3645 | * in the device info lists suddenly goes bad. | |
3646 | * device->bdev is NULL, and so we have to set | |
3647 | * device->missing to one here | |
3648 | */ | |
3649 | root->fs_info->fs_devices->missing_devices++; | |
3650 | device->missing = 1; | |
2b82032c YZ |
3651 | } |
3652 | } | |
3653 | ||
3654 | if (device->fs_devices != root->fs_info->fs_devices) { | |
3655 | BUG_ON(device->writeable); | |
3656 | if (device->generation != | |
3657 | btrfs_device_generation(leaf, dev_item)) | |
3658 | return -EINVAL; | |
6324fbf3 | 3659 | } |
0b86a832 CM |
3660 | |
3661 | fill_device_from_item(leaf, dev_item, device); | |
3662 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 3663 | device->in_fs_metadata = 1; |
2bf64758 | 3664 | if (device->writeable) { |
2b82032c | 3665 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
3666 | spin_lock(&root->fs_info->free_chunk_lock); |
3667 | root->fs_info->free_chunk_space += device->total_bytes - | |
3668 | device->bytes_used; | |
3669 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3670 | } | |
0b86a832 | 3671 | ret = 0; |
0b86a832 CM |
3672 | return ret; |
3673 | } | |
3674 | ||
e4404d6e | 3675 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 3676 | { |
6c41761f | 3677 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 3678 | struct extent_buffer *sb; |
0b86a832 | 3679 | struct btrfs_disk_key *disk_key; |
0b86a832 | 3680 | struct btrfs_chunk *chunk; |
84eed90f CM |
3681 | u8 *ptr; |
3682 | unsigned long sb_ptr; | |
3683 | int ret = 0; | |
0b86a832 CM |
3684 | u32 num_stripes; |
3685 | u32 array_size; | |
3686 | u32 len = 0; | |
0b86a832 | 3687 | u32 cur; |
84eed90f | 3688 | struct btrfs_key key; |
0b86a832 | 3689 | |
e4404d6e | 3690 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
3691 | BTRFS_SUPER_INFO_SIZE); |
3692 | if (!sb) | |
3693 | return -ENOMEM; | |
3694 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 3695 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
4008c04a | 3696 | |
a061fc8d | 3697 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
3698 | array_size = btrfs_super_sys_array_size(super_copy); |
3699 | ||
0b86a832 CM |
3700 | ptr = super_copy->sys_chunk_array; |
3701 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
3702 | cur = 0; | |
3703 | ||
3704 | while (cur < array_size) { | |
3705 | disk_key = (struct btrfs_disk_key *)ptr; | |
3706 | btrfs_disk_key_to_cpu(&key, disk_key); | |
3707 | ||
a061fc8d | 3708 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
3709 | sb_ptr += len; |
3710 | cur += len; | |
3711 | ||
0d81ba5d | 3712 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 3713 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 3714 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
3715 | if (ret) |
3716 | break; | |
0b86a832 CM |
3717 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
3718 | len = btrfs_chunk_item_size(num_stripes); | |
3719 | } else { | |
84eed90f CM |
3720 | ret = -EIO; |
3721 | break; | |
0b86a832 CM |
3722 | } |
3723 | ptr += len; | |
3724 | sb_ptr += len; | |
3725 | cur += len; | |
3726 | } | |
a061fc8d | 3727 | free_extent_buffer(sb); |
84eed90f | 3728 | return ret; |
0b86a832 CM |
3729 | } |
3730 | ||
3731 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
3732 | { | |
3733 | struct btrfs_path *path; | |
3734 | struct extent_buffer *leaf; | |
3735 | struct btrfs_key key; | |
3736 | struct btrfs_key found_key; | |
3737 | int ret; | |
3738 | int slot; | |
3739 | ||
3740 | root = root->fs_info->chunk_root; | |
3741 | ||
3742 | path = btrfs_alloc_path(); | |
3743 | if (!path) | |
3744 | return -ENOMEM; | |
3745 | ||
3746 | /* first we search for all of the device items, and then we | |
3747 | * read in all of the chunk items. This way we can create chunk | |
3748 | * mappings that reference all of the devices that are afound | |
3749 | */ | |
3750 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
3751 | key.offset = 0; | |
3752 | key.type = 0; | |
3753 | again: | |
3754 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
3755 | if (ret < 0) |
3756 | goto error; | |
d397712b | 3757 | while (1) { |
0b86a832 CM |
3758 | leaf = path->nodes[0]; |
3759 | slot = path->slots[0]; | |
3760 | if (slot >= btrfs_header_nritems(leaf)) { | |
3761 | ret = btrfs_next_leaf(root, path); | |
3762 | if (ret == 0) | |
3763 | continue; | |
3764 | if (ret < 0) | |
3765 | goto error; | |
3766 | break; | |
3767 | } | |
3768 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3769 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
3770 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
3771 | break; | |
3772 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
3773 | struct btrfs_dev_item *dev_item; | |
3774 | dev_item = btrfs_item_ptr(leaf, slot, | |
3775 | struct btrfs_dev_item); | |
0d81ba5d | 3776 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
3777 | if (ret) |
3778 | goto error; | |
0b86a832 CM |
3779 | } |
3780 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
3781 | struct btrfs_chunk *chunk; | |
3782 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
3783 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
3784 | if (ret) |
3785 | goto error; | |
0b86a832 CM |
3786 | } |
3787 | path->slots[0]++; | |
3788 | } | |
3789 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
3790 | key.objectid = 0; | |
b3b4aa74 | 3791 | btrfs_release_path(path); |
0b86a832 CM |
3792 | goto again; |
3793 | } | |
0b86a832 CM |
3794 | ret = 0; |
3795 | error: | |
2b82032c | 3796 | btrfs_free_path(path); |
0b86a832 CM |
3797 | return ret; |
3798 | } |