]>
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> |
59641015 | 26 | #include <linux/kthread.h> |
593060d7 | 27 | #include <asm/div64.h> |
4b4e25f2 | 28 | #include "compat.h" |
0b86a832 CM |
29 | #include "ctree.h" |
30 | #include "extent_map.h" | |
31 | #include "disk-io.h" | |
32 | #include "transaction.h" | |
33 | #include "print-tree.h" | |
34 | #include "volumes.h" | |
8b712842 | 35 | #include "async-thread.h" |
21adbd5c | 36 | #include "check-integrity.h" |
0b86a832 | 37 | |
2b82032c YZ |
38 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
39 | struct btrfs_root *root, | |
40 | struct btrfs_device *device); | |
41 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
42 | ||
8a4b83cc CM |
43 | static DEFINE_MUTEX(uuid_mutex); |
44 | static LIST_HEAD(fs_uuids); | |
45 | ||
7d9eb12c CM |
46 | static void lock_chunks(struct btrfs_root *root) |
47 | { | |
7d9eb12c CM |
48 | mutex_lock(&root->fs_info->chunk_mutex); |
49 | } | |
50 | ||
51 | static void unlock_chunks(struct btrfs_root *root) | |
52 | { | |
7d9eb12c CM |
53 | mutex_unlock(&root->fs_info->chunk_mutex); |
54 | } | |
55 | ||
e4404d6e YZ |
56 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
57 | { | |
58 | struct btrfs_device *device; | |
59 | WARN_ON(fs_devices->opened); | |
60 | while (!list_empty(&fs_devices->devices)) { | |
61 | device = list_entry(fs_devices->devices.next, | |
62 | struct btrfs_device, dev_list); | |
63 | list_del(&device->dev_list); | |
64 | kfree(device->name); | |
65 | kfree(device); | |
66 | } | |
67 | kfree(fs_devices); | |
68 | } | |
69 | ||
143bede5 | 70 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
71 | { |
72 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 73 | |
2b82032c YZ |
74 | while (!list_empty(&fs_uuids)) { |
75 | fs_devices = list_entry(fs_uuids.next, | |
76 | struct btrfs_fs_devices, list); | |
77 | list_del(&fs_devices->list); | |
e4404d6e | 78 | free_fs_devices(fs_devices); |
8a4b83cc | 79 | } |
8a4b83cc CM |
80 | } |
81 | ||
a1b32a59 CM |
82 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
83 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
84 | { |
85 | struct btrfs_device *dev; | |
8a4b83cc | 86 | |
c6e30871 | 87 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 88 | if (dev->devid == devid && |
8f18cf13 | 89 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 90 | return dev; |
a443755f | 91 | } |
8a4b83cc CM |
92 | } |
93 | return NULL; | |
94 | } | |
95 | ||
a1b32a59 | 96 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 97 | { |
8a4b83cc CM |
98 | struct btrfs_fs_devices *fs_devices; |
99 | ||
c6e30871 | 100 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
101 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
102 | return fs_devices; | |
103 | } | |
104 | return NULL; | |
105 | } | |
106 | ||
ffbd517d CM |
107 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
108 | struct bio *head, struct bio *tail) | |
109 | { | |
110 | ||
111 | struct bio *old_head; | |
112 | ||
113 | old_head = pending_bios->head; | |
114 | pending_bios->head = head; | |
115 | if (pending_bios->tail) | |
116 | tail->bi_next = old_head; | |
117 | else | |
118 | pending_bios->tail = tail; | |
119 | } | |
120 | ||
8b712842 CM |
121 | /* |
122 | * we try to collect pending bios for a device so we don't get a large | |
123 | * number of procs sending bios down to the same device. This greatly | |
124 | * improves the schedulers ability to collect and merge the bios. | |
125 | * | |
126 | * But, it also turns into a long list of bios to process and that is sure | |
127 | * to eventually make the worker thread block. The solution here is to | |
128 | * make some progress and then put this work struct back at the end of | |
129 | * the list if the block device is congested. This way, multiple devices | |
130 | * can make progress from a single worker thread. | |
131 | */ | |
143bede5 | 132 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
133 | { |
134 | struct bio *pending; | |
135 | struct backing_dev_info *bdi; | |
b64a2851 | 136 | struct btrfs_fs_info *fs_info; |
ffbd517d | 137 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
138 | struct bio *tail; |
139 | struct bio *cur; | |
140 | int again = 0; | |
ffbd517d | 141 | unsigned long num_run; |
d644d8a1 | 142 | unsigned long batch_run = 0; |
b64a2851 | 143 | unsigned long limit; |
b765ead5 | 144 | unsigned long last_waited = 0; |
d84275c9 | 145 | int force_reg = 0; |
0e588859 | 146 | int sync_pending = 0; |
211588ad CM |
147 | struct blk_plug plug; |
148 | ||
149 | /* | |
150 | * this function runs all the bios we've collected for | |
151 | * a particular device. We don't want to wander off to | |
152 | * another device without first sending all of these down. | |
153 | * So, setup a plug here and finish it off before we return | |
154 | */ | |
155 | blk_start_plug(&plug); | |
8b712842 | 156 | |
bedf762b | 157 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
158 | fs_info = device->dev_root->fs_info; |
159 | limit = btrfs_async_submit_limit(fs_info); | |
160 | limit = limit * 2 / 3; | |
161 | ||
8b712842 CM |
162 | loop: |
163 | spin_lock(&device->io_lock); | |
164 | ||
a6837051 | 165 | loop_lock: |
d84275c9 | 166 | num_run = 0; |
ffbd517d | 167 | |
8b712842 CM |
168 | /* take all the bios off the list at once and process them |
169 | * later on (without the lock held). But, remember the | |
170 | * tail and other pointers so the bios can be properly reinserted | |
171 | * into the list if we hit congestion | |
172 | */ | |
d84275c9 | 173 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 174 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
175 | force_reg = 1; |
176 | } else { | |
ffbd517d | 177 | pending_bios = &device->pending_bios; |
d84275c9 CM |
178 | force_reg = 0; |
179 | } | |
ffbd517d CM |
180 | |
181 | pending = pending_bios->head; | |
182 | tail = pending_bios->tail; | |
8b712842 | 183 | WARN_ON(pending && !tail); |
8b712842 CM |
184 | |
185 | /* | |
186 | * if pending was null this time around, no bios need processing | |
187 | * at all and we can stop. Otherwise it'll loop back up again | |
188 | * and do an additional check so no bios are missed. | |
189 | * | |
190 | * device->running_pending is used to synchronize with the | |
191 | * schedule_bio code. | |
192 | */ | |
ffbd517d CM |
193 | if (device->pending_sync_bios.head == NULL && |
194 | device->pending_bios.head == NULL) { | |
8b712842 CM |
195 | again = 0; |
196 | device->running_pending = 0; | |
ffbd517d CM |
197 | } else { |
198 | again = 1; | |
199 | device->running_pending = 1; | |
8b712842 | 200 | } |
ffbd517d CM |
201 | |
202 | pending_bios->head = NULL; | |
203 | pending_bios->tail = NULL; | |
204 | ||
8b712842 CM |
205 | spin_unlock(&device->io_lock); |
206 | ||
d397712b | 207 | while (pending) { |
ffbd517d CM |
208 | |
209 | rmb(); | |
d84275c9 CM |
210 | /* we want to work on both lists, but do more bios on the |
211 | * sync list than the regular list | |
212 | */ | |
213 | if ((num_run > 32 && | |
214 | pending_bios != &device->pending_sync_bios && | |
215 | device->pending_sync_bios.head) || | |
216 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
217 | device->pending_bios.head)) { | |
ffbd517d CM |
218 | spin_lock(&device->io_lock); |
219 | requeue_list(pending_bios, pending, tail); | |
220 | goto loop_lock; | |
221 | } | |
222 | ||
8b712842 CM |
223 | cur = pending; |
224 | pending = pending->bi_next; | |
225 | cur->bi_next = NULL; | |
b64a2851 CM |
226 | atomic_dec(&fs_info->nr_async_bios); |
227 | ||
228 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
229 | waitqueue_active(&fs_info->async_submit_wait)) | |
230 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
231 | |
232 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 233 | |
2ab1ba68 CM |
234 | /* |
235 | * if we're doing the sync list, record that our | |
236 | * plug has some sync requests on it | |
237 | * | |
238 | * If we're doing the regular list and there are | |
239 | * sync requests sitting around, unplug before | |
240 | * we add more | |
241 | */ | |
242 | if (pending_bios == &device->pending_sync_bios) { | |
243 | sync_pending = 1; | |
244 | } else if (sync_pending) { | |
245 | blk_finish_plug(&plug); | |
246 | blk_start_plug(&plug); | |
247 | sync_pending = 0; | |
248 | } | |
249 | ||
21adbd5c | 250 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
251 | num_run++; |
252 | batch_run++; | |
7eaceacc | 253 | if (need_resched()) |
ffbd517d | 254 | cond_resched(); |
8b712842 CM |
255 | |
256 | /* | |
257 | * we made progress, there is more work to do and the bdi | |
258 | * is now congested. Back off and let other work structs | |
259 | * run instead | |
260 | */ | |
57fd5a5f | 261 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 262 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 263 | struct io_context *ioc; |
8b712842 | 264 | |
b765ead5 CM |
265 | ioc = current->io_context; |
266 | ||
267 | /* | |
268 | * the main goal here is that we don't want to | |
269 | * block if we're going to be able to submit | |
270 | * more requests without blocking. | |
271 | * | |
272 | * This code does two great things, it pokes into | |
273 | * the elevator code from a filesystem _and_ | |
274 | * it makes assumptions about how batching works. | |
275 | */ | |
276 | if (ioc && ioc->nr_batch_requests > 0 && | |
277 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
278 | (last_waited == 0 || | |
279 | ioc->last_waited == last_waited)) { | |
280 | /* | |
281 | * we want to go through our batch of | |
282 | * requests and stop. So, we copy out | |
283 | * the ioc->last_waited time and test | |
284 | * against it before looping | |
285 | */ | |
286 | last_waited = ioc->last_waited; | |
7eaceacc | 287 | if (need_resched()) |
ffbd517d | 288 | cond_resched(); |
b765ead5 CM |
289 | continue; |
290 | } | |
8b712842 | 291 | spin_lock(&device->io_lock); |
ffbd517d | 292 | requeue_list(pending_bios, pending, tail); |
a6837051 | 293 | device->running_pending = 1; |
8b712842 CM |
294 | |
295 | spin_unlock(&device->io_lock); | |
296 | btrfs_requeue_work(&device->work); | |
297 | goto done; | |
298 | } | |
d85c8a6f CM |
299 | /* unplug every 64 requests just for good measure */ |
300 | if (batch_run % 64 == 0) { | |
301 | blk_finish_plug(&plug); | |
302 | blk_start_plug(&plug); | |
303 | sync_pending = 0; | |
304 | } | |
8b712842 | 305 | } |
ffbd517d | 306 | |
51684082 CM |
307 | cond_resched(); |
308 | if (again) | |
309 | goto loop; | |
310 | ||
311 | spin_lock(&device->io_lock); | |
312 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
313 | goto loop_lock; | |
314 | spin_unlock(&device->io_lock); | |
315 | ||
8b712842 | 316 | done: |
211588ad | 317 | blk_finish_plug(&plug); |
8b712842 CM |
318 | } |
319 | ||
b2950863 | 320 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
321 | { |
322 | struct btrfs_device *device; | |
323 | ||
324 | device = container_of(work, struct btrfs_device, work); | |
325 | run_scheduled_bios(device); | |
326 | } | |
327 | ||
a1b32a59 | 328 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
329 | struct btrfs_super_block *disk_super, |
330 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
331 | { | |
332 | struct btrfs_device *device; | |
333 | struct btrfs_fs_devices *fs_devices; | |
334 | u64 found_transid = btrfs_super_generation(disk_super); | |
3a0524dc | 335 | char *name; |
8a4b83cc CM |
336 | |
337 | fs_devices = find_fsid(disk_super->fsid); | |
338 | if (!fs_devices) { | |
515dc322 | 339 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
340 | if (!fs_devices) |
341 | return -ENOMEM; | |
342 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 343 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
344 | list_add(&fs_devices->list, &fs_uuids); |
345 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
346 | fs_devices->latest_devid = devid; | |
347 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 348 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
349 | device = NULL; |
350 | } else { | |
a443755f CM |
351 | device = __find_device(&fs_devices->devices, devid, |
352 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
353 | } |
354 | if (!device) { | |
2b82032c YZ |
355 | if (fs_devices->opened) |
356 | return -EBUSY; | |
357 | ||
8a4b83cc CM |
358 | device = kzalloc(sizeof(*device), GFP_NOFS); |
359 | if (!device) { | |
360 | /* we can safely leave the fs_devices entry around */ | |
361 | return -ENOMEM; | |
362 | } | |
363 | device->devid = devid; | |
8b712842 | 364 | device->work.func = pending_bios_fn; |
a443755f CM |
365 | memcpy(device->uuid, disk_super->dev_item.uuid, |
366 | BTRFS_UUID_SIZE); | |
b248a415 | 367 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
368 | device->name = kstrdup(path, GFP_NOFS); |
369 | if (!device->name) { | |
370 | kfree(device); | |
371 | return -ENOMEM; | |
372 | } | |
2b82032c | 373 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 374 | |
90519d66 AJ |
375 | /* init readahead state */ |
376 | spin_lock_init(&device->reada_lock); | |
377 | device->reada_curr_zone = NULL; | |
378 | atomic_set(&device->reada_in_flight, 0); | |
379 | device->reada_next = 0; | |
380 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
381 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
382 | ||
e5e9a520 | 383 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 384 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
385 | mutex_unlock(&fs_devices->device_list_mutex); |
386 | ||
2b82032c | 387 | device->fs_devices = fs_devices; |
8a4b83cc | 388 | fs_devices->num_devices++; |
cd02dca5 | 389 | } else if (!device->name || strcmp(device->name, path)) { |
3a0524dc TH |
390 | name = kstrdup(path, GFP_NOFS); |
391 | if (!name) | |
392 | return -ENOMEM; | |
393 | kfree(device->name); | |
394 | device->name = name; | |
cd02dca5 CM |
395 | if (device->missing) { |
396 | fs_devices->missing_devices--; | |
397 | device->missing = 0; | |
398 | } | |
8a4b83cc CM |
399 | } |
400 | ||
401 | if (found_transid > fs_devices->latest_trans) { | |
402 | fs_devices->latest_devid = devid; | |
403 | fs_devices->latest_trans = found_transid; | |
404 | } | |
8a4b83cc CM |
405 | *fs_devices_ret = fs_devices; |
406 | return 0; | |
407 | } | |
408 | ||
e4404d6e YZ |
409 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
410 | { | |
411 | struct btrfs_fs_devices *fs_devices; | |
412 | struct btrfs_device *device; | |
413 | struct btrfs_device *orig_dev; | |
414 | ||
415 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
416 | if (!fs_devices) | |
417 | return ERR_PTR(-ENOMEM); | |
418 | ||
419 | INIT_LIST_HEAD(&fs_devices->devices); | |
420 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
421 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 422 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
423 | fs_devices->latest_devid = orig->latest_devid; |
424 | fs_devices->latest_trans = orig->latest_trans; | |
425 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); | |
426 | ||
46224705 | 427 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e YZ |
428 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
429 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
430 | if (!device) | |
431 | goto error; | |
432 | ||
433 | device->name = kstrdup(orig_dev->name, GFP_NOFS); | |
fd2696f3 JL |
434 | if (!device->name) { |
435 | kfree(device); | |
e4404d6e | 436 | goto error; |
fd2696f3 | 437 | } |
e4404d6e YZ |
438 | |
439 | device->devid = orig_dev->devid; | |
440 | device->work.func = pending_bios_fn; | |
441 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
442 | spin_lock_init(&device->io_lock); |
443 | INIT_LIST_HEAD(&device->dev_list); | |
444 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
445 | ||
446 | list_add(&device->dev_list, &fs_devices->devices); | |
447 | device->fs_devices = fs_devices; | |
448 | fs_devices->num_devices++; | |
449 | } | |
450 | return fs_devices; | |
451 | error: | |
452 | free_fs_devices(fs_devices); | |
453 | return ERR_PTR(-ENOMEM); | |
454 | } | |
455 | ||
143bede5 | 456 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
dfe25020 | 457 | { |
c6e30871 | 458 | struct btrfs_device *device, *next; |
dfe25020 | 459 | |
a6b0d5c8 CM |
460 | struct block_device *latest_bdev = NULL; |
461 | u64 latest_devid = 0; | |
462 | u64 latest_transid = 0; | |
463 | ||
dfe25020 CM |
464 | mutex_lock(&uuid_mutex); |
465 | again: | |
46224705 | 466 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 467 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 CM |
468 | if (device->in_fs_metadata) { |
469 | if (!latest_transid || | |
470 | device->generation > latest_transid) { | |
471 | latest_devid = device->devid; | |
472 | latest_transid = device->generation; | |
473 | latest_bdev = device->bdev; | |
474 | } | |
2b82032c | 475 | continue; |
a6b0d5c8 | 476 | } |
2b82032c YZ |
477 | |
478 | if (device->bdev) { | |
d4d77629 | 479 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
480 | device->bdev = NULL; |
481 | fs_devices->open_devices--; | |
482 | } | |
483 | if (device->writeable) { | |
484 | list_del_init(&device->dev_alloc_list); | |
485 | device->writeable = 0; | |
486 | fs_devices->rw_devices--; | |
487 | } | |
e4404d6e YZ |
488 | list_del_init(&device->dev_list); |
489 | fs_devices->num_devices--; | |
490 | kfree(device->name); | |
491 | kfree(device); | |
dfe25020 | 492 | } |
2b82032c YZ |
493 | |
494 | if (fs_devices->seed) { | |
495 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
496 | goto again; |
497 | } | |
498 | ||
a6b0d5c8 CM |
499 | fs_devices->latest_bdev = latest_bdev; |
500 | fs_devices->latest_devid = latest_devid; | |
501 | fs_devices->latest_trans = latest_transid; | |
502 | ||
dfe25020 | 503 | mutex_unlock(&uuid_mutex); |
dfe25020 | 504 | } |
a0af469b | 505 | |
1f78160c XG |
506 | static void __free_device(struct work_struct *work) |
507 | { | |
508 | struct btrfs_device *device; | |
509 | ||
510 | device = container_of(work, struct btrfs_device, rcu_work); | |
511 | ||
512 | if (device->bdev) | |
513 | blkdev_put(device->bdev, device->mode); | |
514 | ||
515 | kfree(device->name); | |
516 | kfree(device); | |
517 | } | |
518 | ||
519 | static void free_device(struct rcu_head *head) | |
520 | { | |
521 | struct btrfs_device *device; | |
522 | ||
523 | device = container_of(head, struct btrfs_device, rcu); | |
524 | ||
525 | INIT_WORK(&device->rcu_work, __free_device); | |
526 | schedule_work(&device->rcu_work); | |
527 | } | |
528 | ||
2b82032c | 529 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 530 | { |
8a4b83cc | 531 | struct btrfs_device *device; |
e4404d6e | 532 | |
2b82032c YZ |
533 | if (--fs_devices->opened > 0) |
534 | return 0; | |
8a4b83cc | 535 | |
c9513edb | 536 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 537 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c XG |
538 | struct btrfs_device *new_device; |
539 | ||
540 | if (device->bdev) | |
a0af469b | 541 | fs_devices->open_devices--; |
1f78160c | 542 | |
2b82032c YZ |
543 | if (device->writeable) { |
544 | list_del_init(&device->dev_alloc_list); | |
545 | fs_devices->rw_devices--; | |
546 | } | |
547 | ||
d5e2003c JB |
548 | if (device->can_discard) |
549 | fs_devices->num_can_discard--; | |
550 | ||
1f78160c XG |
551 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
552 | BUG_ON(!new_device); | |
553 | memcpy(new_device, device, sizeof(*new_device)); | |
554 | new_device->name = kstrdup(device->name, GFP_NOFS); | |
5f3f302a | 555 | BUG_ON(device->name && !new_device->name); |
1f78160c XG |
556 | new_device->bdev = NULL; |
557 | new_device->writeable = 0; | |
558 | new_device->in_fs_metadata = 0; | |
d5e2003c | 559 | new_device->can_discard = 0; |
1f78160c XG |
560 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
561 | ||
562 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 563 | } |
c9513edb XG |
564 | mutex_unlock(&fs_devices->device_list_mutex); |
565 | ||
e4404d6e YZ |
566 | WARN_ON(fs_devices->open_devices); |
567 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
568 | fs_devices->opened = 0; |
569 | fs_devices->seeding = 0; | |
2b82032c | 570 | |
8a4b83cc CM |
571 | return 0; |
572 | } | |
573 | ||
2b82032c YZ |
574 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
575 | { | |
e4404d6e | 576 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
577 | int ret; |
578 | ||
579 | mutex_lock(&uuid_mutex); | |
580 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
581 | if (!fs_devices->opened) { |
582 | seed_devices = fs_devices->seed; | |
583 | fs_devices->seed = NULL; | |
584 | } | |
2b82032c | 585 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
586 | |
587 | while (seed_devices) { | |
588 | fs_devices = seed_devices; | |
589 | seed_devices = fs_devices->seed; | |
590 | __btrfs_close_devices(fs_devices); | |
591 | free_fs_devices(fs_devices); | |
592 | } | |
2b82032c YZ |
593 | return ret; |
594 | } | |
595 | ||
e4404d6e YZ |
596 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
597 | fmode_t flags, void *holder) | |
8a4b83cc | 598 | { |
d5e2003c | 599 | struct request_queue *q; |
8a4b83cc CM |
600 | struct block_device *bdev; |
601 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 602 | struct btrfs_device *device; |
a0af469b CM |
603 | struct block_device *latest_bdev = NULL; |
604 | struct buffer_head *bh; | |
605 | struct btrfs_super_block *disk_super; | |
606 | u64 latest_devid = 0; | |
607 | u64 latest_transid = 0; | |
a0af469b | 608 | u64 devid; |
2b82032c | 609 | int seeding = 1; |
a0af469b | 610 | int ret = 0; |
8a4b83cc | 611 | |
d4d77629 TH |
612 | flags |= FMODE_EXCL; |
613 | ||
c6e30871 | 614 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
615 | if (device->bdev) |
616 | continue; | |
dfe25020 CM |
617 | if (!device->name) |
618 | continue; | |
619 | ||
d4d77629 | 620 | bdev = blkdev_get_by_path(device->name, flags, holder); |
8a4b83cc | 621 | if (IS_ERR(bdev)) { |
d397712b | 622 | printk(KERN_INFO "open %s failed\n", device->name); |
a0af469b | 623 | goto error; |
8a4b83cc | 624 | } |
a061fc8d | 625 | set_blocksize(bdev, 4096); |
a0af469b | 626 | |
a512bbf8 | 627 | bh = btrfs_read_dev_super(bdev); |
20bcd649 | 628 | if (!bh) |
a0af469b CM |
629 | goto error_close; |
630 | ||
631 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 632 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
633 | if (devid != device->devid) |
634 | goto error_brelse; | |
635 | ||
2b82032c YZ |
636 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
637 | BTRFS_UUID_SIZE)) | |
638 | goto error_brelse; | |
639 | ||
640 | device->generation = btrfs_super_generation(disk_super); | |
641 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 642 | latest_devid = devid; |
2b82032c | 643 | latest_transid = device->generation; |
a0af469b CM |
644 | latest_bdev = bdev; |
645 | } | |
646 | ||
2b82032c YZ |
647 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
648 | device->writeable = 0; | |
649 | } else { | |
650 | device->writeable = !bdev_read_only(bdev); | |
651 | seeding = 0; | |
652 | } | |
653 | ||
d5e2003c JB |
654 | q = bdev_get_queue(bdev); |
655 | if (blk_queue_discard(q)) { | |
656 | device->can_discard = 1; | |
657 | fs_devices->num_can_discard++; | |
658 | } | |
659 | ||
8a4b83cc | 660 | device->bdev = bdev; |
dfe25020 | 661 | device->in_fs_metadata = 0; |
15916de8 CM |
662 | device->mode = flags; |
663 | ||
c289811c CM |
664 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
665 | fs_devices->rotating = 1; | |
666 | ||
a0af469b | 667 | fs_devices->open_devices++; |
2b82032c YZ |
668 | if (device->writeable) { |
669 | fs_devices->rw_devices++; | |
670 | list_add(&device->dev_alloc_list, | |
671 | &fs_devices->alloc_list); | |
672 | } | |
4f6c9328 | 673 | brelse(bh); |
a0af469b | 674 | continue; |
a061fc8d | 675 | |
a0af469b CM |
676 | error_brelse: |
677 | brelse(bh); | |
678 | error_close: | |
d4d77629 | 679 | blkdev_put(bdev, flags); |
a0af469b CM |
680 | error: |
681 | continue; | |
8a4b83cc | 682 | } |
a0af469b | 683 | if (fs_devices->open_devices == 0) { |
20bcd649 | 684 | ret = -EINVAL; |
a0af469b CM |
685 | goto out; |
686 | } | |
2b82032c YZ |
687 | fs_devices->seeding = seeding; |
688 | fs_devices->opened = 1; | |
a0af469b CM |
689 | fs_devices->latest_bdev = latest_bdev; |
690 | fs_devices->latest_devid = latest_devid; | |
691 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 692 | fs_devices->total_rw_bytes = 0; |
a0af469b | 693 | out: |
2b82032c YZ |
694 | return ret; |
695 | } | |
696 | ||
697 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 698 | fmode_t flags, void *holder) |
2b82032c YZ |
699 | { |
700 | int ret; | |
701 | ||
702 | mutex_lock(&uuid_mutex); | |
703 | if (fs_devices->opened) { | |
e4404d6e YZ |
704 | fs_devices->opened++; |
705 | ret = 0; | |
2b82032c | 706 | } else { |
15916de8 | 707 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 708 | } |
8a4b83cc | 709 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
710 | return ret; |
711 | } | |
712 | ||
97288f2c | 713 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
714 | struct btrfs_fs_devices **fs_devices_ret) |
715 | { | |
716 | struct btrfs_super_block *disk_super; | |
717 | struct block_device *bdev; | |
718 | struct buffer_head *bh; | |
719 | int ret; | |
720 | u64 devid; | |
f2984462 | 721 | u64 transid; |
8a4b83cc | 722 | |
d4d77629 TH |
723 | flags |= FMODE_EXCL; |
724 | bdev = blkdev_get_by_path(path, flags, holder); | |
8a4b83cc CM |
725 | |
726 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
727 | ret = PTR_ERR(bdev); |
728 | goto error; | |
729 | } | |
730 | ||
10f6327b | 731 | mutex_lock(&uuid_mutex); |
8a4b83cc CM |
732 | ret = set_blocksize(bdev, 4096); |
733 | if (ret) | |
734 | goto error_close; | |
a512bbf8 | 735 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 736 | if (!bh) { |
20b45077 | 737 | ret = -EINVAL; |
8a4b83cc CM |
738 | goto error_close; |
739 | } | |
740 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 741 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 742 | transid = btrfs_super_generation(disk_super); |
7ae9c09d | 743 | if (disk_super->label[0]) |
d397712b | 744 | printk(KERN_INFO "device label %s ", disk_super->label); |
22b63a29 ID |
745 | else |
746 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); | |
119e10cf | 747 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 748 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc CM |
749 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
750 | ||
8a4b83cc CM |
751 | brelse(bh); |
752 | error_close: | |
10f6327b | 753 | mutex_unlock(&uuid_mutex); |
d4d77629 | 754 | blkdev_put(bdev, flags); |
8a4b83cc | 755 | error: |
8a4b83cc CM |
756 | return ret; |
757 | } | |
0b86a832 | 758 | |
6d07bcec MX |
759 | /* helper to account the used device space in the range */ |
760 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
761 | u64 end, u64 *length) | |
762 | { | |
763 | struct btrfs_key key; | |
764 | struct btrfs_root *root = device->dev_root; | |
765 | struct btrfs_dev_extent *dev_extent; | |
766 | struct btrfs_path *path; | |
767 | u64 extent_end; | |
768 | int ret; | |
769 | int slot; | |
770 | struct extent_buffer *l; | |
771 | ||
772 | *length = 0; | |
773 | ||
774 | if (start >= device->total_bytes) | |
775 | return 0; | |
776 | ||
777 | path = btrfs_alloc_path(); | |
778 | if (!path) | |
779 | return -ENOMEM; | |
780 | path->reada = 2; | |
781 | ||
782 | key.objectid = device->devid; | |
783 | key.offset = start; | |
784 | key.type = BTRFS_DEV_EXTENT_KEY; | |
785 | ||
786 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
787 | if (ret < 0) | |
788 | goto out; | |
789 | if (ret > 0) { | |
790 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
791 | if (ret < 0) | |
792 | goto out; | |
793 | } | |
794 | ||
795 | while (1) { | |
796 | l = path->nodes[0]; | |
797 | slot = path->slots[0]; | |
798 | if (slot >= btrfs_header_nritems(l)) { | |
799 | ret = btrfs_next_leaf(root, path); | |
800 | if (ret == 0) | |
801 | continue; | |
802 | if (ret < 0) | |
803 | goto out; | |
804 | ||
805 | break; | |
806 | } | |
807 | btrfs_item_key_to_cpu(l, &key, slot); | |
808 | ||
809 | if (key.objectid < device->devid) | |
810 | goto next; | |
811 | ||
812 | if (key.objectid > device->devid) | |
813 | break; | |
814 | ||
815 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
816 | goto next; | |
817 | ||
818 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
819 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
820 | dev_extent); | |
821 | if (key.offset <= start && extent_end > end) { | |
822 | *length = end - start + 1; | |
823 | break; | |
824 | } else if (key.offset <= start && extent_end > start) | |
825 | *length += extent_end - start; | |
826 | else if (key.offset > start && extent_end <= end) | |
827 | *length += extent_end - key.offset; | |
828 | else if (key.offset > start && key.offset <= end) { | |
829 | *length += end - key.offset + 1; | |
830 | break; | |
831 | } else if (key.offset > end) | |
832 | break; | |
833 | ||
834 | next: | |
835 | path->slots[0]++; | |
836 | } | |
837 | ret = 0; | |
838 | out: | |
839 | btrfs_free_path(path); | |
840 | return ret; | |
841 | } | |
842 | ||
0b86a832 | 843 | /* |
7bfc837d | 844 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
845 | * @device: the device which we search the free space in |
846 | * @num_bytes: the size of the free space that we need | |
847 | * @start: store the start of the free space. | |
848 | * @len: the size of the free space. that we find, or the size of the max | |
849 | * free space if we don't find suitable free space | |
850 | * | |
0b86a832 CM |
851 | * this uses a pretty simple search, the expectation is that it is |
852 | * called very infrequently and that a given device has a small number | |
853 | * of extents | |
7bfc837d MX |
854 | * |
855 | * @start is used to store the start of the free space if we find. But if we | |
856 | * don't find suitable free space, it will be used to store the start position | |
857 | * of the max free space. | |
858 | * | |
859 | * @len is used to store the size of the free space that we find. | |
860 | * But if we don't find suitable free space, it is used to store the size of | |
861 | * the max free space. | |
0b86a832 | 862 | */ |
125ccb0a | 863 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 864 | u64 *start, u64 *len) |
0b86a832 CM |
865 | { |
866 | struct btrfs_key key; | |
867 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 868 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 869 | struct btrfs_path *path; |
7bfc837d MX |
870 | u64 hole_size; |
871 | u64 max_hole_start; | |
872 | u64 max_hole_size; | |
873 | u64 extent_end; | |
874 | u64 search_start; | |
0b86a832 CM |
875 | u64 search_end = device->total_bytes; |
876 | int ret; | |
7bfc837d | 877 | int slot; |
0b86a832 CM |
878 | struct extent_buffer *l; |
879 | ||
0b86a832 CM |
880 | /* FIXME use last free of some kind */ |
881 | ||
8a4b83cc CM |
882 | /* we don't want to overwrite the superblock on the drive, |
883 | * so we make sure to start at an offset of at least 1MB | |
884 | */ | |
a9c9bf68 | 885 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 886 | |
7bfc837d MX |
887 | max_hole_start = search_start; |
888 | max_hole_size = 0; | |
38c01b96 | 889 | hole_size = 0; |
7bfc837d MX |
890 | |
891 | if (search_start >= search_end) { | |
892 | ret = -ENOSPC; | |
893 | goto error; | |
894 | } | |
895 | ||
896 | path = btrfs_alloc_path(); | |
897 | if (!path) { | |
898 | ret = -ENOMEM; | |
899 | goto error; | |
900 | } | |
901 | path->reada = 2; | |
902 | ||
0b86a832 CM |
903 | key.objectid = device->devid; |
904 | key.offset = search_start; | |
905 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 906 | |
125ccb0a | 907 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 908 | if (ret < 0) |
7bfc837d | 909 | goto out; |
1fcbac58 YZ |
910 | if (ret > 0) { |
911 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
912 | if (ret < 0) | |
7bfc837d | 913 | goto out; |
1fcbac58 | 914 | } |
7bfc837d | 915 | |
0b86a832 CM |
916 | while (1) { |
917 | l = path->nodes[0]; | |
918 | slot = path->slots[0]; | |
919 | if (slot >= btrfs_header_nritems(l)) { | |
920 | ret = btrfs_next_leaf(root, path); | |
921 | if (ret == 0) | |
922 | continue; | |
923 | if (ret < 0) | |
7bfc837d MX |
924 | goto out; |
925 | ||
926 | break; | |
0b86a832 CM |
927 | } |
928 | btrfs_item_key_to_cpu(l, &key, slot); | |
929 | ||
930 | if (key.objectid < device->devid) | |
931 | goto next; | |
932 | ||
933 | if (key.objectid > device->devid) | |
7bfc837d | 934 | break; |
0b86a832 | 935 | |
7bfc837d MX |
936 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
937 | goto next; | |
9779b72f | 938 | |
7bfc837d MX |
939 | if (key.offset > search_start) { |
940 | hole_size = key.offset - search_start; | |
9779b72f | 941 | |
7bfc837d MX |
942 | if (hole_size > max_hole_size) { |
943 | max_hole_start = search_start; | |
944 | max_hole_size = hole_size; | |
945 | } | |
9779b72f | 946 | |
7bfc837d MX |
947 | /* |
948 | * If this free space is greater than which we need, | |
949 | * it must be the max free space that we have found | |
950 | * until now, so max_hole_start must point to the start | |
951 | * of this free space and the length of this free space | |
952 | * is stored in max_hole_size. Thus, we return | |
953 | * max_hole_start and max_hole_size and go back to the | |
954 | * caller. | |
955 | */ | |
956 | if (hole_size >= num_bytes) { | |
957 | ret = 0; | |
958 | goto out; | |
0b86a832 CM |
959 | } |
960 | } | |
0b86a832 | 961 | |
0b86a832 | 962 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
963 | extent_end = key.offset + btrfs_dev_extent_length(l, |
964 | dev_extent); | |
965 | if (extent_end > search_start) | |
966 | search_start = extent_end; | |
0b86a832 CM |
967 | next: |
968 | path->slots[0]++; | |
969 | cond_resched(); | |
970 | } | |
0b86a832 | 971 | |
38c01b96 | 972 | /* |
973 | * At this point, search_start should be the end of | |
974 | * allocated dev extents, and when shrinking the device, | |
975 | * search_end may be smaller than search_start. | |
976 | */ | |
977 | if (search_end > search_start) | |
978 | hole_size = search_end - search_start; | |
979 | ||
7bfc837d MX |
980 | if (hole_size > max_hole_size) { |
981 | max_hole_start = search_start; | |
982 | max_hole_size = hole_size; | |
0b86a832 | 983 | } |
0b86a832 | 984 | |
7bfc837d MX |
985 | /* See above. */ |
986 | if (hole_size < num_bytes) | |
987 | ret = -ENOSPC; | |
988 | else | |
989 | ret = 0; | |
990 | ||
991 | out: | |
2b82032c | 992 | btrfs_free_path(path); |
7bfc837d MX |
993 | error: |
994 | *start = max_hole_start; | |
b2117a39 | 995 | if (len) |
7bfc837d | 996 | *len = max_hole_size; |
0b86a832 CM |
997 | return ret; |
998 | } | |
999 | ||
b2950863 | 1000 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1001 | struct btrfs_device *device, |
1002 | u64 start) | |
1003 | { | |
1004 | int ret; | |
1005 | struct btrfs_path *path; | |
1006 | struct btrfs_root *root = device->dev_root; | |
1007 | struct btrfs_key key; | |
a061fc8d CM |
1008 | struct btrfs_key found_key; |
1009 | struct extent_buffer *leaf = NULL; | |
1010 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1011 | |
1012 | path = btrfs_alloc_path(); | |
1013 | if (!path) | |
1014 | return -ENOMEM; | |
1015 | ||
1016 | key.objectid = device->devid; | |
1017 | key.offset = start; | |
1018 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1019 | again: |
8f18cf13 | 1020 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1021 | if (ret > 0) { |
1022 | ret = btrfs_previous_item(root, path, key.objectid, | |
1023 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1024 | if (ret) |
1025 | goto out; | |
a061fc8d CM |
1026 | leaf = path->nodes[0]; |
1027 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1028 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1029 | struct btrfs_dev_extent); | |
1030 | BUG_ON(found_key.offset > start || found_key.offset + | |
1031 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1032 | key = found_key; |
1033 | btrfs_release_path(path); | |
1034 | goto again; | |
a061fc8d CM |
1035 | } else if (ret == 0) { |
1036 | leaf = path->nodes[0]; | |
1037 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1038 | struct btrfs_dev_extent); | |
1039 | } | |
8f18cf13 CM |
1040 | BUG_ON(ret); |
1041 | ||
2bf64758 JB |
1042 | if (device->bytes_used > 0) { |
1043 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1044 | device->bytes_used -= len; | |
1045 | spin_lock(&root->fs_info->free_chunk_lock); | |
1046 | root->fs_info->free_chunk_space += len; | |
1047 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1048 | } | |
8f18cf13 | 1049 | ret = btrfs_del_item(trans, root, path); |
8f18cf13 | 1050 | |
b0b802d7 | 1051 | out: |
8f18cf13 CM |
1052 | btrfs_free_path(path); |
1053 | return ret; | |
1054 | } | |
1055 | ||
2b82032c | 1056 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1057 | struct btrfs_device *device, |
e17cade2 | 1058 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1059 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1060 | { |
1061 | int ret; | |
1062 | struct btrfs_path *path; | |
1063 | struct btrfs_root *root = device->dev_root; | |
1064 | struct btrfs_dev_extent *extent; | |
1065 | struct extent_buffer *leaf; | |
1066 | struct btrfs_key key; | |
1067 | ||
dfe25020 | 1068 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1069 | path = btrfs_alloc_path(); |
1070 | if (!path) | |
1071 | return -ENOMEM; | |
1072 | ||
0b86a832 | 1073 | key.objectid = device->devid; |
2b82032c | 1074 | key.offset = start; |
0b86a832 CM |
1075 | key.type = BTRFS_DEV_EXTENT_KEY; |
1076 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1077 | sizeof(*extent)); | |
2cdcecbc MF |
1078 | if (ret) |
1079 | goto out; | |
0b86a832 CM |
1080 | |
1081 | leaf = path->nodes[0]; | |
1082 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1083 | struct btrfs_dev_extent); | |
e17cade2 CM |
1084 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1085 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1086 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1087 | ||
1088 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1089 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1090 | BTRFS_UUID_SIZE); | |
1091 | ||
0b86a832 CM |
1092 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1093 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1094 | out: |
0b86a832 CM |
1095 | btrfs_free_path(path); |
1096 | return ret; | |
1097 | } | |
1098 | ||
a1b32a59 CM |
1099 | static noinline int find_next_chunk(struct btrfs_root *root, |
1100 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1101 | { |
1102 | struct btrfs_path *path; | |
1103 | int ret; | |
1104 | struct btrfs_key key; | |
e17cade2 | 1105 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1106 | struct btrfs_key found_key; |
1107 | ||
1108 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1109 | if (!path) |
1110 | return -ENOMEM; | |
0b86a832 | 1111 | |
e17cade2 | 1112 | key.objectid = objectid; |
0b86a832 CM |
1113 | key.offset = (u64)-1; |
1114 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1115 | ||
1116 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1117 | if (ret < 0) | |
1118 | goto error; | |
1119 | ||
1120 | BUG_ON(ret == 0); | |
1121 | ||
1122 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1123 | if (ret) { | |
e17cade2 | 1124 | *offset = 0; |
0b86a832 CM |
1125 | } else { |
1126 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1127 | path->slots[0]); | |
e17cade2 CM |
1128 | if (found_key.objectid != objectid) |
1129 | *offset = 0; | |
1130 | else { | |
1131 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1132 | struct btrfs_chunk); | |
1133 | *offset = found_key.offset + | |
1134 | btrfs_chunk_length(path->nodes[0], chunk); | |
1135 | } | |
0b86a832 CM |
1136 | } |
1137 | ret = 0; | |
1138 | error: | |
1139 | btrfs_free_path(path); | |
1140 | return ret; | |
1141 | } | |
1142 | ||
2b82032c | 1143 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1144 | { |
1145 | int ret; | |
1146 | struct btrfs_key key; | |
1147 | struct btrfs_key found_key; | |
2b82032c YZ |
1148 | struct btrfs_path *path; |
1149 | ||
1150 | root = root->fs_info->chunk_root; | |
1151 | ||
1152 | path = btrfs_alloc_path(); | |
1153 | if (!path) | |
1154 | return -ENOMEM; | |
0b86a832 CM |
1155 | |
1156 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1157 | key.type = BTRFS_DEV_ITEM_KEY; | |
1158 | key.offset = (u64)-1; | |
1159 | ||
1160 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1161 | if (ret < 0) | |
1162 | goto error; | |
1163 | ||
1164 | BUG_ON(ret == 0); | |
1165 | ||
1166 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1167 | BTRFS_DEV_ITEM_KEY); | |
1168 | if (ret) { | |
1169 | *objectid = 1; | |
1170 | } else { | |
1171 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1172 | path->slots[0]); | |
1173 | *objectid = found_key.offset + 1; | |
1174 | } | |
1175 | ret = 0; | |
1176 | error: | |
2b82032c | 1177 | btrfs_free_path(path); |
0b86a832 CM |
1178 | return ret; |
1179 | } | |
1180 | ||
1181 | /* | |
1182 | * the device information is stored in the chunk root | |
1183 | * the btrfs_device struct should be fully filled in | |
1184 | */ | |
1185 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1186 | struct btrfs_root *root, | |
1187 | struct btrfs_device *device) | |
1188 | { | |
1189 | int ret; | |
1190 | struct btrfs_path *path; | |
1191 | struct btrfs_dev_item *dev_item; | |
1192 | struct extent_buffer *leaf; | |
1193 | struct btrfs_key key; | |
1194 | unsigned long ptr; | |
0b86a832 CM |
1195 | |
1196 | root = root->fs_info->chunk_root; | |
1197 | ||
1198 | path = btrfs_alloc_path(); | |
1199 | if (!path) | |
1200 | return -ENOMEM; | |
1201 | ||
0b86a832 CM |
1202 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1203 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1204 | key.offset = device->devid; |
0b86a832 CM |
1205 | |
1206 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1207 | sizeof(*dev_item)); |
0b86a832 CM |
1208 | if (ret) |
1209 | goto out; | |
1210 | ||
1211 | leaf = path->nodes[0]; | |
1212 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1213 | ||
1214 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1215 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1216 | btrfs_set_device_type(leaf, dev_item, device->type); |
1217 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1218 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1219 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1220 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1221 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1222 | btrfs_set_device_group(leaf, dev_item, 0); |
1223 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1224 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1225 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1226 | |
0b86a832 | 1227 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1228 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1229 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1230 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1231 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1232 | |
2b82032c | 1233 | ret = 0; |
0b86a832 CM |
1234 | out: |
1235 | btrfs_free_path(path); | |
1236 | return ret; | |
1237 | } | |
8f18cf13 | 1238 | |
a061fc8d CM |
1239 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1240 | struct btrfs_device *device) | |
1241 | { | |
1242 | int ret; | |
1243 | struct btrfs_path *path; | |
a061fc8d | 1244 | struct btrfs_key key; |
a061fc8d CM |
1245 | struct btrfs_trans_handle *trans; |
1246 | ||
1247 | root = root->fs_info->chunk_root; | |
1248 | ||
1249 | path = btrfs_alloc_path(); | |
1250 | if (!path) | |
1251 | return -ENOMEM; | |
1252 | ||
a22285a6 | 1253 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1254 | if (IS_ERR(trans)) { |
1255 | btrfs_free_path(path); | |
1256 | return PTR_ERR(trans); | |
1257 | } | |
a061fc8d CM |
1258 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1259 | key.type = BTRFS_DEV_ITEM_KEY; | |
1260 | key.offset = device->devid; | |
7d9eb12c | 1261 | lock_chunks(root); |
a061fc8d CM |
1262 | |
1263 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1264 | if (ret < 0) | |
1265 | goto out; | |
1266 | ||
1267 | if (ret > 0) { | |
1268 | ret = -ENOENT; | |
1269 | goto out; | |
1270 | } | |
1271 | ||
1272 | ret = btrfs_del_item(trans, root, path); | |
1273 | if (ret) | |
1274 | goto out; | |
a061fc8d CM |
1275 | out: |
1276 | btrfs_free_path(path); | |
7d9eb12c | 1277 | unlock_chunks(root); |
a061fc8d CM |
1278 | btrfs_commit_transaction(trans, root); |
1279 | return ret; | |
1280 | } | |
1281 | ||
1282 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1283 | { | |
1284 | struct btrfs_device *device; | |
2b82032c | 1285 | struct btrfs_device *next_device; |
a061fc8d | 1286 | struct block_device *bdev; |
dfe25020 | 1287 | struct buffer_head *bh = NULL; |
a061fc8d | 1288 | struct btrfs_super_block *disk_super; |
1f78160c | 1289 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1290 | u64 all_avail; |
1291 | u64 devid; | |
2b82032c YZ |
1292 | u64 num_devices; |
1293 | u8 *dev_uuid; | |
a061fc8d | 1294 | int ret = 0; |
1f78160c | 1295 | bool clear_super = false; |
a061fc8d | 1296 | |
a061fc8d CM |
1297 | mutex_lock(&uuid_mutex); |
1298 | ||
1299 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1300 | root->fs_info->avail_system_alloc_bits | | |
1301 | root->fs_info->avail_metadata_alloc_bits; | |
1302 | ||
1303 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1304 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1305 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1306 | "on raid10\n"); | |
a061fc8d CM |
1307 | ret = -EINVAL; |
1308 | goto out; | |
1309 | } | |
1310 | ||
1311 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1312 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1313 | printk(KERN_ERR "btrfs: unable to go below two " |
1314 | "devices on raid1\n"); | |
a061fc8d CM |
1315 | ret = -EINVAL; |
1316 | goto out; | |
1317 | } | |
1318 | ||
dfe25020 | 1319 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1320 | struct list_head *devices; |
1321 | struct btrfs_device *tmp; | |
a061fc8d | 1322 | |
dfe25020 CM |
1323 | device = NULL; |
1324 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1325 | /* |
1326 | * It is safe to read the devices since the volume_mutex | |
1327 | * is held. | |
1328 | */ | |
c6e30871 | 1329 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1330 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1331 | device = tmp; | |
1332 | break; | |
1333 | } | |
1334 | } | |
1335 | bdev = NULL; | |
1336 | bh = NULL; | |
1337 | disk_super = NULL; | |
1338 | if (!device) { | |
d397712b CM |
1339 | printk(KERN_ERR "btrfs: no missing devices found to " |
1340 | "remove\n"); | |
dfe25020 CM |
1341 | goto out; |
1342 | } | |
dfe25020 | 1343 | } else { |
d4d77629 TH |
1344 | bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL, |
1345 | root->fs_info->bdev_holder); | |
dfe25020 CM |
1346 | if (IS_ERR(bdev)) { |
1347 | ret = PTR_ERR(bdev); | |
1348 | goto out; | |
1349 | } | |
a061fc8d | 1350 | |
2b82032c | 1351 | set_blocksize(bdev, 4096); |
a512bbf8 | 1352 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1353 | if (!bh) { |
20b45077 | 1354 | ret = -EINVAL; |
dfe25020 CM |
1355 | goto error_close; |
1356 | } | |
1357 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1358 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1359 | dev_uuid = disk_super->dev_item.uuid; |
1360 | device = btrfs_find_device(root, devid, dev_uuid, | |
1361 | disk_super->fsid); | |
dfe25020 CM |
1362 | if (!device) { |
1363 | ret = -ENOENT; | |
1364 | goto error_brelse; | |
1365 | } | |
2b82032c | 1366 | } |
dfe25020 | 1367 | |
2b82032c | 1368 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1369 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1370 | "device\n"); | |
2b82032c YZ |
1371 | ret = -EINVAL; |
1372 | goto error_brelse; | |
1373 | } | |
1374 | ||
1375 | if (device->writeable) { | |
0c1daee0 | 1376 | lock_chunks(root); |
2b82032c | 1377 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1378 | unlock_chunks(root); |
2b82032c | 1379 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1380 | clear_super = true; |
dfe25020 | 1381 | } |
a061fc8d CM |
1382 | |
1383 | ret = btrfs_shrink_device(device, 0); | |
1384 | if (ret) | |
9b3517e9 | 1385 | goto error_undo; |
a061fc8d | 1386 | |
a061fc8d CM |
1387 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1388 | if (ret) | |
9b3517e9 | 1389 | goto error_undo; |
a061fc8d | 1390 | |
2bf64758 JB |
1391 | spin_lock(&root->fs_info->free_chunk_lock); |
1392 | root->fs_info->free_chunk_space = device->total_bytes - | |
1393 | device->bytes_used; | |
1394 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1395 | ||
2b82032c | 1396 | device->in_fs_metadata = 0; |
a2de733c | 1397 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1398 | |
1399 | /* | |
1400 | * the device list mutex makes sure that we don't change | |
1401 | * the device list while someone else is writing out all | |
1402 | * the device supers. | |
1403 | */ | |
1f78160c XG |
1404 | |
1405 | cur_devices = device->fs_devices; | |
e5e9a520 | 1406 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1407 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1408 | |
e4404d6e | 1409 | device->fs_devices->num_devices--; |
2b82032c | 1410 | |
cd02dca5 CM |
1411 | if (device->missing) |
1412 | root->fs_info->fs_devices->missing_devices--; | |
1413 | ||
2b82032c YZ |
1414 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1415 | struct btrfs_device, dev_list); | |
1416 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1417 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1418 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1419 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1420 | ||
1f78160c | 1421 | if (device->bdev) |
e4404d6e | 1422 | device->fs_devices->open_devices--; |
1f78160c XG |
1423 | |
1424 | call_rcu(&device->rcu, free_device); | |
1425 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1426 | |
6c41761f DS |
1427 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1428 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1429 | |
1f78160c | 1430 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1431 | struct btrfs_fs_devices *fs_devices; |
1432 | fs_devices = root->fs_info->fs_devices; | |
1433 | while (fs_devices) { | |
1f78160c | 1434 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1435 | break; |
1436 | fs_devices = fs_devices->seed; | |
2b82032c | 1437 | } |
1f78160c XG |
1438 | fs_devices->seed = cur_devices->seed; |
1439 | cur_devices->seed = NULL; | |
0c1daee0 | 1440 | lock_chunks(root); |
1f78160c | 1441 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1442 | unlock_chunks(root); |
1f78160c | 1443 | free_fs_devices(cur_devices); |
2b82032c YZ |
1444 | } |
1445 | ||
1446 | /* | |
1447 | * at this point, the device is zero sized. We want to | |
1448 | * remove it from the devices list and zero out the old super | |
1449 | */ | |
1f78160c | 1450 | if (clear_super) { |
dfe25020 CM |
1451 | /* make sure this device isn't detected as part of |
1452 | * the FS anymore | |
1453 | */ | |
1454 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1455 | set_buffer_dirty(bh); | |
1456 | sync_dirty_buffer(bh); | |
dfe25020 | 1457 | } |
a061fc8d | 1458 | |
a061fc8d | 1459 | ret = 0; |
a061fc8d CM |
1460 | |
1461 | error_brelse: | |
1462 | brelse(bh); | |
1463 | error_close: | |
dfe25020 | 1464 | if (bdev) |
e525fd89 | 1465 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1466 | out: |
1467 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1468 | return ret; |
9b3517e9 ID |
1469 | error_undo: |
1470 | if (device->writeable) { | |
0c1daee0 | 1471 | lock_chunks(root); |
9b3517e9 ID |
1472 | list_add(&device->dev_alloc_list, |
1473 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1474 | unlock_chunks(root); |
9b3517e9 ID |
1475 | root->fs_info->fs_devices->rw_devices++; |
1476 | } | |
1477 | goto error_brelse; | |
a061fc8d CM |
1478 | } |
1479 | ||
2b82032c YZ |
1480 | /* |
1481 | * does all the dirty work required for changing file system's UUID. | |
1482 | */ | |
125ccb0a | 1483 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1484 | { |
1485 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1486 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1487 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1488 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1489 | struct btrfs_device *device; |
1490 | u64 super_flags; | |
1491 | ||
1492 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1493 | if (!fs_devices->seeding) |
2b82032c YZ |
1494 | return -EINVAL; |
1495 | ||
e4404d6e YZ |
1496 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1497 | if (!seed_devices) | |
2b82032c YZ |
1498 | return -ENOMEM; |
1499 | ||
e4404d6e YZ |
1500 | old_devices = clone_fs_devices(fs_devices); |
1501 | if (IS_ERR(old_devices)) { | |
1502 | kfree(seed_devices); | |
1503 | return PTR_ERR(old_devices); | |
2b82032c | 1504 | } |
e4404d6e | 1505 | |
2b82032c YZ |
1506 | list_add(&old_devices->list, &fs_uuids); |
1507 | ||
e4404d6e YZ |
1508 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1509 | seed_devices->opened = 1; | |
1510 | INIT_LIST_HEAD(&seed_devices->devices); | |
1511 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1512 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1513 | |
1514 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1515 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1516 | synchronize_rcu); | |
c9513edb XG |
1517 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1518 | ||
e4404d6e YZ |
1519 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1520 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1521 | device->fs_devices = seed_devices; | |
1522 | } | |
1523 | ||
2b82032c YZ |
1524 | fs_devices->seeding = 0; |
1525 | fs_devices->num_devices = 0; | |
1526 | fs_devices->open_devices = 0; | |
e4404d6e | 1527 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1528 | |
1529 | generate_random_uuid(fs_devices->fsid); | |
1530 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1531 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1532 | super_flags = btrfs_super_flags(disk_super) & | |
1533 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1534 | btrfs_set_super_flags(disk_super, super_flags); | |
1535 | ||
1536 | return 0; | |
1537 | } | |
1538 | ||
1539 | /* | |
1540 | * strore the expected generation for seed devices in device items. | |
1541 | */ | |
1542 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1543 | struct btrfs_root *root) | |
1544 | { | |
1545 | struct btrfs_path *path; | |
1546 | struct extent_buffer *leaf; | |
1547 | struct btrfs_dev_item *dev_item; | |
1548 | struct btrfs_device *device; | |
1549 | struct btrfs_key key; | |
1550 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1551 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1552 | u64 devid; | |
1553 | int ret; | |
1554 | ||
1555 | path = btrfs_alloc_path(); | |
1556 | if (!path) | |
1557 | return -ENOMEM; | |
1558 | ||
1559 | root = root->fs_info->chunk_root; | |
1560 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1561 | key.offset = 0; | |
1562 | key.type = BTRFS_DEV_ITEM_KEY; | |
1563 | ||
1564 | while (1) { | |
1565 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1566 | if (ret < 0) | |
1567 | goto error; | |
1568 | ||
1569 | leaf = path->nodes[0]; | |
1570 | next_slot: | |
1571 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1572 | ret = btrfs_next_leaf(root, path); | |
1573 | if (ret > 0) | |
1574 | break; | |
1575 | if (ret < 0) | |
1576 | goto error; | |
1577 | leaf = path->nodes[0]; | |
1578 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1579 | btrfs_release_path(path); |
2b82032c YZ |
1580 | continue; |
1581 | } | |
1582 | ||
1583 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1584 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1585 | key.type != BTRFS_DEV_ITEM_KEY) | |
1586 | break; | |
1587 | ||
1588 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1589 | struct btrfs_dev_item); | |
1590 | devid = btrfs_device_id(leaf, dev_item); | |
1591 | read_extent_buffer(leaf, dev_uuid, | |
1592 | (unsigned long)btrfs_device_uuid(dev_item), | |
1593 | BTRFS_UUID_SIZE); | |
1594 | read_extent_buffer(leaf, fs_uuid, | |
1595 | (unsigned long)btrfs_device_fsid(dev_item), | |
1596 | BTRFS_UUID_SIZE); | |
1597 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
1598 | BUG_ON(!device); | |
1599 | ||
1600 | if (device->fs_devices->seeding) { | |
1601 | btrfs_set_device_generation(leaf, dev_item, | |
1602 | device->generation); | |
1603 | btrfs_mark_buffer_dirty(leaf); | |
1604 | } | |
1605 | ||
1606 | path->slots[0]++; | |
1607 | goto next_slot; | |
1608 | } | |
1609 | ret = 0; | |
1610 | error: | |
1611 | btrfs_free_path(path); | |
1612 | return ret; | |
1613 | } | |
1614 | ||
788f20eb CM |
1615 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1616 | { | |
d5e2003c | 1617 | struct request_queue *q; |
788f20eb CM |
1618 | struct btrfs_trans_handle *trans; |
1619 | struct btrfs_device *device; | |
1620 | struct block_device *bdev; | |
788f20eb | 1621 | struct list_head *devices; |
2b82032c | 1622 | struct super_block *sb = root->fs_info->sb; |
788f20eb | 1623 | u64 total_bytes; |
2b82032c | 1624 | int seeding_dev = 0; |
788f20eb CM |
1625 | int ret = 0; |
1626 | ||
2b82032c YZ |
1627 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
1628 | return -EINVAL; | |
788f20eb | 1629 | |
a5d16333 | 1630 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1631 | root->fs_info->bdev_holder); |
7f59203a JB |
1632 | if (IS_ERR(bdev)) |
1633 | return PTR_ERR(bdev); | |
a2135011 | 1634 | |
2b82032c YZ |
1635 | if (root->fs_info->fs_devices->seeding) { |
1636 | seeding_dev = 1; | |
1637 | down_write(&sb->s_umount); | |
1638 | mutex_lock(&uuid_mutex); | |
1639 | } | |
1640 | ||
8c8bee1d | 1641 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1642 | |
788f20eb | 1643 | devices = &root->fs_info->fs_devices->devices; |
e5e9a520 CM |
1644 | /* |
1645 | * we have the volume lock, so we don't need the extra | |
1646 | * device list mutex while reading the list here. | |
1647 | */ | |
c6e30871 | 1648 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1649 | if (device->bdev == bdev) { |
1650 | ret = -EEXIST; | |
2b82032c | 1651 | goto error; |
788f20eb CM |
1652 | } |
1653 | } | |
1654 | ||
1655 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1656 | if (!device) { | |
1657 | /* we can safely leave the fs_devices entry around */ | |
1658 | ret = -ENOMEM; | |
2b82032c | 1659 | goto error; |
788f20eb CM |
1660 | } |
1661 | ||
788f20eb CM |
1662 | device->name = kstrdup(device_path, GFP_NOFS); |
1663 | if (!device->name) { | |
1664 | kfree(device); | |
2b82032c YZ |
1665 | ret = -ENOMEM; |
1666 | goto error; | |
788f20eb | 1667 | } |
2b82032c YZ |
1668 | |
1669 | ret = find_next_devid(root, &device->devid); | |
1670 | if (ret) { | |
67100f25 | 1671 | kfree(device->name); |
2b82032c YZ |
1672 | kfree(device); |
1673 | goto error; | |
1674 | } | |
1675 | ||
a22285a6 | 1676 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1677 | if (IS_ERR(trans)) { |
67100f25 | 1678 | kfree(device->name); |
98d5dc13 TI |
1679 | kfree(device); |
1680 | ret = PTR_ERR(trans); | |
1681 | goto error; | |
1682 | } | |
1683 | ||
2b82032c YZ |
1684 | lock_chunks(root); |
1685 | ||
d5e2003c JB |
1686 | q = bdev_get_queue(bdev); |
1687 | if (blk_queue_discard(q)) | |
1688 | device->can_discard = 1; | |
2b82032c YZ |
1689 | device->writeable = 1; |
1690 | device->work.func = pending_bios_fn; | |
1691 | generate_random_uuid(device->uuid); | |
1692 | spin_lock_init(&device->io_lock); | |
1693 | device->generation = trans->transid; | |
788f20eb CM |
1694 | device->io_width = root->sectorsize; |
1695 | device->io_align = root->sectorsize; | |
1696 | device->sector_size = root->sectorsize; | |
1697 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1698 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1699 | device->dev_root = root->fs_info->dev_root; |
1700 | device->bdev = bdev; | |
dfe25020 | 1701 | device->in_fs_metadata = 1; |
fb01aa85 | 1702 | device->mode = FMODE_EXCL; |
2b82032c | 1703 | set_blocksize(device->bdev, 4096); |
788f20eb | 1704 | |
2b82032c YZ |
1705 | if (seeding_dev) { |
1706 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 1707 | ret = btrfs_prepare_sprout(root); |
2b82032c YZ |
1708 | BUG_ON(ret); |
1709 | } | |
788f20eb | 1710 | |
2b82032c | 1711 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 CM |
1712 | |
1713 | /* | |
1714 | * we don't want write_supers to jump in here with our device | |
1715 | * half setup | |
1716 | */ | |
1717 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c | 1718 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1719 | list_add(&device->dev_alloc_list, |
1720 | &root->fs_info->fs_devices->alloc_list); | |
1721 | root->fs_info->fs_devices->num_devices++; | |
1722 | root->fs_info->fs_devices->open_devices++; | |
1723 | root->fs_info->fs_devices->rw_devices++; | |
d5e2003c JB |
1724 | if (device->can_discard) |
1725 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1726 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1727 | |
2bf64758 JB |
1728 | spin_lock(&root->fs_info->free_chunk_lock); |
1729 | root->fs_info->free_chunk_space += device->total_bytes; | |
1730 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1731 | ||
c289811c CM |
1732 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1733 | root->fs_info->fs_devices->rotating = 1; | |
1734 | ||
6c41761f DS |
1735 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1736 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1737 | total_bytes + device->total_bytes); |
1738 | ||
6c41761f DS |
1739 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1740 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1741 | total_bytes + 1); |
e5e9a520 | 1742 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1743 | |
2b82032c YZ |
1744 | if (seeding_dev) { |
1745 | ret = init_first_rw_device(trans, root, device); | |
1746 | BUG_ON(ret); | |
1747 | ret = btrfs_finish_sprout(trans, root); | |
1748 | BUG_ON(ret); | |
1749 | } else { | |
1750 | ret = btrfs_add_device(trans, root, device); | |
1751 | } | |
1752 | ||
913d952e CM |
1753 | /* |
1754 | * we've got more storage, clear any full flags on the space | |
1755 | * infos | |
1756 | */ | |
1757 | btrfs_clear_space_info_full(root->fs_info); | |
1758 | ||
7d9eb12c | 1759 | unlock_chunks(root); |
2b82032c | 1760 | btrfs_commit_transaction(trans, root); |
a2135011 | 1761 | |
2b82032c YZ |
1762 | if (seeding_dev) { |
1763 | mutex_unlock(&uuid_mutex); | |
1764 | up_write(&sb->s_umount); | |
788f20eb | 1765 | |
2b82032c YZ |
1766 | ret = btrfs_relocate_sys_chunks(root); |
1767 | BUG_ON(ret); | |
1768 | } | |
c9e9f97b | 1769 | |
2b82032c YZ |
1770 | return ret; |
1771 | error: | |
e525fd89 | 1772 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1773 | if (seeding_dev) { |
1774 | mutex_unlock(&uuid_mutex); | |
1775 | up_write(&sb->s_umount); | |
1776 | } | |
c9e9f97b | 1777 | return ret; |
788f20eb CM |
1778 | } |
1779 | ||
d397712b CM |
1780 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1781 | struct btrfs_device *device) | |
0b86a832 CM |
1782 | { |
1783 | int ret; | |
1784 | struct btrfs_path *path; | |
1785 | struct btrfs_root *root; | |
1786 | struct btrfs_dev_item *dev_item; | |
1787 | struct extent_buffer *leaf; | |
1788 | struct btrfs_key key; | |
1789 | ||
1790 | root = device->dev_root->fs_info->chunk_root; | |
1791 | ||
1792 | path = btrfs_alloc_path(); | |
1793 | if (!path) | |
1794 | return -ENOMEM; | |
1795 | ||
1796 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1797 | key.type = BTRFS_DEV_ITEM_KEY; | |
1798 | key.offset = device->devid; | |
1799 | ||
1800 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1801 | if (ret < 0) | |
1802 | goto out; | |
1803 | ||
1804 | if (ret > 0) { | |
1805 | ret = -ENOENT; | |
1806 | goto out; | |
1807 | } | |
1808 | ||
1809 | leaf = path->nodes[0]; | |
1810 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1811 | ||
1812 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1813 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1814 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1815 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1816 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1817 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1818 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1819 | btrfs_mark_buffer_dirty(leaf); | |
1820 | ||
1821 | out: | |
1822 | btrfs_free_path(path); | |
1823 | return ret; | |
1824 | } | |
1825 | ||
7d9eb12c | 1826 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1827 | struct btrfs_device *device, u64 new_size) |
1828 | { | |
1829 | struct btrfs_super_block *super_copy = | |
6c41761f | 1830 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1831 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1832 | u64 diff = new_size - device->total_bytes; | |
1833 | ||
2b82032c YZ |
1834 | if (!device->writeable) |
1835 | return -EACCES; | |
1836 | if (new_size <= device->total_bytes) | |
1837 | return -EINVAL; | |
1838 | ||
8f18cf13 | 1839 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1840 | device->fs_devices->total_rw_bytes += diff; |
1841 | ||
1842 | device->total_bytes = new_size; | |
9779b72f | 1843 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1844 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1845 | ||
8f18cf13 CM |
1846 | return btrfs_update_device(trans, device); |
1847 | } | |
1848 | ||
7d9eb12c CM |
1849 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1850 | struct btrfs_device *device, u64 new_size) | |
1851 | { | |
1852 | int ret; | |
1853 | lock_chunks(device->dev_root); | |
1854 | ret = __btrfs_grow_device(trans, device, new_size); | |
1855 | unlock_chunks(device->dev_root); | |
1856 | return ret; | |
1857 | } | |
1858 | ||
8f18cf13 CM |
1859 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1860 | struct btrfs_root *root, | |
1861 | u64 chunk_tree, u64 chunk_objectid, | |
1862 | u64 chunk_offset) | |
1863 | { | |
1864 | int ret; | |
1865 | struct btrfs_path *path; | |
1866 | struct btrfs_key key; | |
1867 | ||
1868 | root = root->fs_info->chunk_root; | |
1869 | path = btrfs_alloc_path(); | |
1870 | if (!path) | |
1871 | return -ENOMEM; | |
1872 | ||
1873 | key.objectid = chunk_objectid; | |
1874 | key.offset = chunk_offset; | |
1875 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1876 | ||
1877 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1878 | BUG_ON(ret); | |
1879 | ||
1880 | ret = btrfs_del_item(trans, root, path); | |
8f18cf13 CM |
1881 | |
1882 | btrfs_free_path(path); | |
65a246c5 | 1883 | return ret; |
8f18cf13 CM |
1884 | } |
1885 | ||
b2950863 | 1886 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1887 | chunk_offset) |
1888 | { | |
6c41761f | 1889 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
1890 | struct btrfs_disk_key *disk_key; |
1891 | struct btrfs_chunk *chunk; | |
1892 | u8 *ptr; | |
1893 | int ret = 0; | |
1894 | u32 num_stripes; | |
1895 | u32 array_size; | |
1896 | u32 len = 0; | |
1897 | u32 cur; | |
1898 | struct btrfs_key key; | |
1899 | ||
1900 | array_size = btrfs_super_sys_array_size(super_copy); | |
1901 | ||
1902 | ptr = super_copy->sys_chunk_array; | |
1903 | cur = 0; | |
1904 | ||
1905 | while (cur < array_size) { | |
1906 | disk_key = (struct btrfs_disk_key *)ptr; | |
1907 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1908 | ||
1909 | len = sizeof(*disk_key); | |
1910 | ||
1911 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1912 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1913 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1914 | len += btrfs_chunk_item_size(num_stripes); | |
1915 | } else { | |
1916 | ret = -EIO; | |
1917 | break; | |
1918 | } | |
1919 | if (key.objectid == chunk_objectid && | |
1920 | key.offset == chunk_offset) { | |
1921 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1922 | array_size -= len; | |
1923 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1924 | } else { | |
1925 | ptr += len; | |
1926 | cur += len; | |
1927 | } | |
1928 | } | |
1929 | return ret; | |
1930 | } | |
1931 | ||
b2950863 | 1932 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
1933 | u64 chunk_tree, u64 chunk_objectid, |
1934 | u64 chunk_offset) | |
1935 | { | |
1936 | struct extent_map_tree *em_tree; | |
1937 | struct btrfs_root *extent_root; | |
1938 | struct btrfs_trans_handle *trans; | |
1939 | struct extent_map *em; | |
1940 | struct map_lookup *map; | |
1941 | int ret; | |
1942 | int i; | |
1943 | ||
1944 | root = root->fs_info->chunk_root; | |
1945 | extent_root = root->fs_info->extent_root; | |
1946 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
1947 | ||
ba1bf481 JB |
1948 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
1949 | if (ret) | |
1950 | return -ENOSPC; | |
1951 | ||
8f18cf13 | 1952 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 1953 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
1954 | if (ret) |
1955 | return ret; | |
8f18cf13 | 1956 | |
a22285a6 | 1957 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1958 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 1959 | |
7d9eb12c CM |
1960 | lock_chunks(root); |
1961 | ||
8f18cf13 CM |
1962 | /* |
1963 | * step two, delete the device extents and the | |
1964 | * chunk tree entries | |
1965 | */ | |
890871be | 1966 | read_lock(&em_tree->lock); |
8f18cf13 | 1967 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 1968 | read_unlock(&em_tree->lock); |
8f18cf13 | 1969 | |
285190d9 | 1970 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 1971 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
1972 | map = (struct map_lookup *)em->bdev; |
1973 | ||
1974 | for (i = 0; i < map->num_stripes; i++) { | |
1975 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
1976 | map->stripes[i].physical); | |
1977 | BUG_ON(ret); | |
a061fc8d | 1978 | |
dfe25020 CM |
1979 | if (map->stripes[i].dev) { |
1980 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
1981 | BUG_ON(ret); | |
1982 | } | |
8f18cf13 CM |
1983 | } |
1984 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
1985 | chunk_offset); | |
1986 | ||
1987 | BUG_ON(ret); | |
1988 | ||
1abe9b8a | 1989 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
1990 | ||
8f18cf13 CM |
1991 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
1992 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
1993 | BUG_ON(ret); | |
8f18cf13 CM |
1994 | } |
1995 | ||
2b82032c YZ |
1996 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
1997 | BUG_ON(ret); | |
1998 | ||
890871be | 1999 | write_lock(&em_tree->lock); |
2b82032c | 2000 | remove_extent_mapping(em_tree, em); |
890871be | 2001 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2002 | |
2003 | kfree(map); | |
2004 | em->bdev = NULL; | |
2005 | ||
2006 | /* once for the tree */ | |
2007 | free_extent_map(em); | |
2008 | /* once for us */ | |
2009 | free_extent_map(em); | |
2010 | ||
2011 | unlock_chunks(root); | |
2012 | btrfs_end_transaction(trans, root); | |
2013 | return 0; | |
2014 | } | |
2015 | ||
2016 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2017 | { | |
2018 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2019 | struct btrfs_path *path; | |
2020 | struct extent_buffer *leaf; | |
2021 | struct btrfs_chunk *chunk; | |
2022 | struct btrfs_key key; | |
2023 | struct btrfs_key found_key; | |
2024 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2025 | u64 chunk_type; | |
ba1bf481 JB |
2026 | bool retried = false; |
2027 | int failed = 0; | |
2b82032c YZ |
2028 | int ret; |
2029 | ||
2030 | path = btrfs_alloc_path(); | |
2031 | if (!path) | |
2032 | return -ENOMEM; | |
2033 | ||
ba1bf481 | 2034 | again: |
2b82032c YZ |
2035 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2036 | key.offset = (u64)-1; | |
2037 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2038 | ||
2039 | while (1) { | |
2040 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2041 | if (ret < 0) | |
2042 | goto error; | |
2043 | BUG_ON(ret == 0); | |
2044 | ||
2045 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2046 | key.type); | |
2047 | if (ret < 0) | |
2048 | goto error; | |
2049 | if (ret > 0) | |
2050 | break; | |
1a40e23b | 2051 | |
2b82032c YZ |
2052 | leaf = path->nodes[0]; |
2053 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2054 | |
2b82032c YZ |
2055 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2056 | struct btrfs_chunk); | |
2057 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2058 | btrfs_release_path(path); |
8f18cf13 | 2059 | |
2b82032c YZ |
2060 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2061 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2062 | found_key.objectid, | |
2063 | found_key.offset); | |
ba1bf481 JB |
2064 | if (ret == -ENOSPC) |
2065 | failed++; | |
2066 | else if (ret) | |
2067 | BUG(); | |
2b82032c | 2068 | } |
8f18cf13 | 2069 | |
2b82032c YZ |
2070 | if (found_key.offset == 0) |
2071 | break; | |
2072 | key.offset = found_key.offset - 1; | |
2073 | } | |
2074 | ret = 0; | |
ba1bf481 JB |
2075 | if (failed && !retried) { |
2076 | failed = 0; | |
2077 | retried = true; | |
2078 | goto again; | |
2079 | } else if (failed && retried) { | |
2080 | WARN_ON(1); | |
2081 | ret = -ENOSPC; | |
2082 | } | |
2b82032c YZ |
2083 | error: |
2084 | btrfs_free_path(path); | |
2085 | return ret; | |
8f18cf13 CM |
2086 | } |
2087 | ||
0940ebf6 ID |
2088 | static int insert_balance_item(struct btrfs_root *root, |
2089 | struct btrfs_balance_control *bctl) | |
2090 | { | |
2091 | struct btrfs_trans_handle *trans; | |
2092 | struct btrfs_balance_item *item; | |
2093 | struct btrfs_disk_balance_args disk_bargs; | |
2094 | struct btrfs_path *path; | |
2095 | struct extent_buffer *leaf; | |
2096 | struct btrfs_key key; | |
2097 | int ret, err; | |
2098 | ||
2099 | path = btrfs_alloc_path(); | |
2100 | if (!path) | |
2101 | return -ENOMEM; | |
2102 | ||
2103 | trans = btrfs_start_transaction(root, 0); | |
2104 | if (IS_ERR(trans)) { | |
2105 | btrfs_free_path(path); | |
2106 | return PTR_ERR(trans); | |
2107 | } | |
2108 | ||
2109 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2110 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2111 | key.offset = 0; | |
2112 | ||
2113 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2114 | sizeof(*item)); | |
2115 | if (ret) | |
2116 | goto out; | |
2117 | ||
2118 | leaf = path->nodes[0]; | |
2119 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2120 | ||
2121 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2122 | ||
2123 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2124 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2125 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2126 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2127 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2128 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2129 | ||
2130 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2131 | ||
2132 | btrfs_mark_buffer_dirty(leaf); | |
2133 | out: | |
2134 | btrfs_free_path(path); | |
2135 | err = btrfs_commit_transaction(trans, root); | |
2136 | if (err && !ret) | |
2137 | ret = err; | |
2138 | return ret; | |
2139 | } | |
2140 | ||
2141 | static int del_balance_item(struct btrfs_root *root) | |
2142 | { | |
2143 | struct btrfs_trans_handle *trans; | |
2144 | struct btrfs_path *path; | |
2145 | struct btrfs_key key; | |
2146 | int ret, err; | |
2147 | ||
2148 | path = btrfs_alloc_path(); | |
2149 | if (!path) | |
2150 | return -ENOMEM; | |
2151 | ||
2152 | trans = btrfs_start_transaction(root, 0); | |
2153 | if (IS_ERR(trans)) { | |
2154 | btrfs_free_path(path); | |
2155 | return PTR_ERR(trans); | |
2156 | } | |
2157 | ||
2158 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2159 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2160 | key.offset = 0; | |
2161 | ||
2162 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2163 | if (ret < 0) | |
2164 | goto out; | |
2165 | if (ret > 0) { | |
2166 | ret = -ENOENT; | |
2167 | goto out; | |
2168 | } | |
2169 | ||
2170 | ret = btrfs_del_item(trans, root, path); | |
2171 | out: | |
2172 | btrfs_free_path(path); | |
2173 | err = btrfs_commit_transaction(trans, root); | |
2174 | if (err && !ret) | |
2175 | ret = err; | |
2176 | return ret; | |
2177 | } | |
2178 | ||
59641015 ID |
2179 | /* |
2180 | * This is a heuristic used to reduce the number of chunks balanced on | |
2181 | * resume after balance was interrupted. | |
2182 | */ | |
2183 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2184 | { | |
2185 | /* | |
2186 | * Turn on soft mode for chunk types that were being converted. | |
2187 | */ | |
2188 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2189 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2190 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2191 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2192 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2193 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2194 | ||
2195 | /* | |
2196 | * Turn on usage filter if is not already used. The idea is | |
2197 | * that chunks that we have already balanced should be | |
2198 | * reasonably full. Don't do it for chunks that are being | |
2199 | * converted - that will keep us from relocating unconverted | |
2200 | * (albeit full) chunks. | |
2201 | */ | |
2202 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2203 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2204 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2205 | bctl->data.usage = 90; | |
2206 | } | |
2207 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2208 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2209 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2210 | bctl->sys.usage = 90; | |
2211 | } | |
2212 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2213 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2214 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2215 | bctl->meta.usage = 90; | |
2216 | } | |
2217 | } | |
2218 | ||
c9e9f97b ID |
2219 | /* |
2220 | * Should be called with both balance and volume mutexes held to | |
2221 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2222 | * restriper. Same goes for unset_balance_control. | |
2223 | */ | |
2224 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2225 | { | |
2226 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2227 | ||
2228 | BUG_ON(fs_info->balance_ctl); | |
2229 | ||
2230 | spin_lock(&fs_info->balance_lock); | |
2231 | fs_info->balance_ctl = bctl; | |
2232 | spin_unlock(&fs_info->balance_lock); | |
2233 | } | |
2234 | ||
2235 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2236 | { | |
2237 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2238 | ||
2239 | BUG_ON(!fs_info->balance_ctl); | |
2240 | ||
2241 | spin_lock(&fs_info->balance_lock); | |
2242 | fs_info->balance_ctl = NULL; | |
2243 | spin_unlock(&fs_info->balance_lock); | |
2244 | ||
2245 | kfree(bctl); | |
2246 | } | |
2247 | ||
ed25e9b2 ID |
2248 | /* |
2249 | * Balance filters. Return 1 if chunk should be filtered out | |
2250 | * (should not be balanced). | |
2251 | */ | |
2252 | static int chunk_profiles_filter(u64 chunk_profile, | |
2253 | struct btrfs_balance_args *bargs) | |
2254 | { | |
2255 | chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
2256 | ||
2257 | if (chunk_profile == 0) | |
2258 | chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
2259 | ||
2260 | if (bargs->profiles & chunk_profile) | |
2261 | return 0; | |
2262 | ||
2263 | return 1; | |
2264 | } | |
2265 | ||
5ce5b3c0 ID |
2266 | static u64 div_factor_fine(u64 num, int factor) |
2267 | { | |
2268 | if (factor <= 0) | |
2269 | return 0; | |
2270 | if (factor >= 100) | |
2271 | return num; | |
2272 | ||
2273 | num *= factor; | |
2274 | do_div(num, 100); | |
2275 | return num; | |
2276 | } | |
2277 | ||
2278 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, | |
2279 | struct btrfs_balance_args *bargs) | |
2280 | { | |
2281 | struct btrfs_block_group_cache *cache; | |
2282 | u64 chunk_used, user_thresh; | |
2283 | int ret = 1; | |
2284 | ||
2285 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2286 | chunk_used = btrfs_block_group_used(&cache->item); | |
2287 | ||
2288 | user_thresh = div_factor_fine(cache->key.offset, bargs->usage); | |
2289 | if (chunk_used < user_thresh) | |
2290 | ret = 0; | |
2291 | ||
2292 | btrfs_put_block_group(cache); | |
2293 | return ret; | |
2294 | } | |
2295 | ||
409d404b ID |
2296 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2297 | struct btrfs_chunk *chunk, | |
2298 | struct btrfs_balance_args *bargs) | |
2299 | { | |
2300 | struct btrfs_stripe *stripe; | |
2301 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2302 | int i; | |
2303 | ||
2304 | for (i = 0; i < num_stripes; i++) { | |
2305 | stripe = btrfs_stripe_nr(chunk, i); | |
2306 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2307 | return 0; | |
2308 | } | |
2309 | ||
2310 | return 1; | |
2311 | } | |
2312 | ||
94e60d5a ID |
2313 | /* [pstart, pend) */ |
2314 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2315 | struct btrfs_chunk *chunk, | |
2316 | u64 chunk_offset, | |
2317 | struct btrfs_balance_args *bargs) | |
2318 | { | |
2319 | struct btrfs_stripe *stripe; | |
2320 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2321 | u64 stripe_offset; | |
2322 | u64 stripe_length; | |
2323 | int factor; | |
2324 | int i; | |
2325 | ||
2326 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2327 | return 0; | |
2328 | ||
2329 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
2330 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) | |
2331 | factor = 2; | |
2332 | else | |
2333 | factor = 1; | |
2334 | factor = num_stripes / factor; | |
2335 | ||
2336 | for (i = 0; i < num_stripes; i++) { | |
2337 | stripe = btrfs_stripe_nr(chunk, i); | |
2338 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2339 | continue; | |
2340 | ||
2341 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2342 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2343 | do_div(stripe_length, factor); | |
2344 | ||
2345 | if (stripe_offset < bargs->pend && | |
2346 | stripe_offset + stripe_length > bargs->pstart) | |
2347 | return 0; | |
2348 | } | |
2349 | ||
2350 | return 1; | |
2351 | } | |
2352 | ||
ea67176a ID |
2353 | /* [vstart, vend) */ |
2354 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2355 | struct btrfs_chunk *chunk, | |
2356 | u64 chunk_offset, | |
2357 | struct btrfs_balance_args *bargs) | |
2358 | { | |
2359 | if (chunk_offset < bargs->vend && | |
2360 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2361 | /* at least part of the chunk is inside this vrange */ | |
2362 | return 0; | |
2363 | ||
2364 | return 1; | |
2365 | } | |
2366 | ||
cfa4c961 ID |
2367 | static int chunk_soft_convert_filter(u64 chunk_profile, |
2368 | struct btrfs_balance_args *bargs) | |
2369 | { | |
2370 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2371 | return 0; | |
2372 | ||
2373 | chunk_profile &= BTRFS_BLOCK_GROUP_PROFILE_MASK; | |
2374 | ||
2375 | if (chunk_profile == 0) | |
2376 | chunk_profile = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
2377 | ||
2378 | if (bargs->target & chunk_profile) | |
2379 | return 1; | |
2380 | ||
2381 | return 0; | |
2382 | } | |
2383 | ||
f43ffb60 ID |
2384 | static int should_balance_chunk(struct btrfs_root *root, |
2385 | struct extent_buffer *leaf, | |
2386 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2387 | { | |
2388 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2389 | struct btrfs_balance_args *bargs = NULL; | |
2390 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2391 | ||
2392 | /* type filter */ | |
2393 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2394 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2395 | return 0; | |
2396 | } | |
2397 | ||
2398 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2399 | bargs = &bctl->data; | |
2400 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2401 | bargs = &bctl->sys; | |
2402 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2403 | bargs = &bctl->meta; | |
2404 | ||
ed25e9b2 ID |
2405 | /* profiles filter */ |
2406 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2407 | chunk_profiles_filter(chunk_type, bargs)) { | |
2408 | return 0; | |
5ce5b3c0 ID |
2409 | } |
2410 | ||
2411 | /* usage filter */ | |
2412 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2413 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2414 | return 0; | |
409d404b ID |
2415 | } |
2416 | ||
2417 | /* devid filter */ | |
2418 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2419 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2420 | return 0; | |
94e60d5a ID |
2421 | } |
2422 | ||
2423 | /* drange filter, makes sense only with devid filter */ | |
2424 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2425 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2426 | return 0; | |
ea67176a ID |
2427 | } |
2428 | ||
2429 | /* vrange filter */ | |
2430 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2431 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2432 | return 0; | |
ed25e9b2 ID |
2433 | } |
2434 | ||
cfa4c961 ID |
2435 | /* soft profile changing mode */ |
2436 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2437 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2438 | return 0; | |
2439 | } | |
2440 | ||
f43ffb60 ID |
2441 | return 1; |
2442 | } | |
2443 | ||
ec44a35c CM |
2444 | static u64 div_factor(u64 num, int factor) |
2445 | { | |
2446 | if (factor == 10) | |
2447 | return num; | |
2448 | num *= factor; | |
2449 | do_div(num, 10); | |
2450 | return num; | |
2451 | } | |
2452 | ||
c9e9f97b | 2453 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2454 | { |
19a39dce | 2455 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2456 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2457 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2458 | struct list_head *devices; | |
ec44a35c CM |
2459 | struct btrfs_device *device; |
2460 | u64 old_size; | |
2461 | u64 size_to_free; | |
f43ffb60 | 2462 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2463 | struct btrfs_path *path; |
2464 | struct btrfs_key key; | |
ec44a35c | 2465 | struct btrfs_key found_key; |
c9e9f97b | 2466 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2467 | struct extent_buffer *leaf; |
2468 | int slot; | |
c9e9f97b ID |
2469 | int ret; |
2470 | int enospc_errors = 0; | |
19a39dce | 2471 | bool counting = true; |
ec44a35c | 2472 | |
ec44a35c | 2473 | /* step one make some room on all the devices */ |
c9e9f97b | 2474 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2475 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2476 | old_size = device->total_bytes; |
2477 | size_to_free = div_factor(old_size, 1); | |
2478 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2479 | if (!device->writeable || |
2480 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2481 | continue; |
2482 | ||
2483 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2484 | if (ret == -ENOSPC) |
2485 | break; | |
ec44a35c CM |
2486 | BUG_ON(ret); |
2487 | ||
a22285a6 | 2488 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2489 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2490 | |
2491 | ret = btrfs_grow_device(trans, device, old_size); | |
2492 | BUG_ON(ret); | |
2493 | ||
2494 | btrfs_end_transaction(trans, dev_root); | |
2495 | } | |
2496 | ||
2497 | /* step two, relocate all the chunks */ | |
2498 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2499 | if (!path) { |
2500 | ret = -ENOMEM; | |
2501 | goto error; | |
2502 | } | |
19a39dce ID |
2503 | |
2504 | /* zero out stat counters */ | |
2505 | spin_lock(&fs_info->balance_lock); | |
2506 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2507 | spin_unlock(&fs_info->balance_lock); | |
2508 | again: | |
ec44a35c CM |
2509 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2510 | key.offset = (u64)-1; | |
2511 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2512 | ||
d397712b | 2513 | while (1) { |
19a39dce | 2514 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2515 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2516 | ret = -ECANCELED; |
2517 | goto error; | |
2518 | } | |
2519 | ||
ec44a35c CM |
2520 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2521 | if (ret < 0) | |
2522 | goto error; | |
2523 | ||
2524 | /* | |
2525 | * this shouldn't happen, it means the last relocate | |
2526 | * failed | |
2527 | */ | |
2528 | if (ret == 0) | |
c9e9f97b | 2529 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2530 | |
2531 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2532 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2533 | if (ret) { |
2534 | ret = 0; | |
ec44a35c | 2535 | break; |
c9e9f97b | 2536 | } |
7d9eb12c | 2537 | |
f43ffb60 ID |
2538 | leaf = path->nodes[0]; |
2539 | slot = path->slots[0]; | |
2540 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2541 | |
ec44a35c CM |
2542 | if (found_key.objectid != key.objectid) |
2543 | break; | |
7d9eb12c | 2544 | |
ec44a35c | 2545 | /* chunk zero is special */ |
ba1bf481 | 2546 | if (found_key.offset == 0) |
ec44a35c CM |
2547 | break; |
2548 | ||
f43ffb60 ID |
2549 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2550 | ||
19a39dce ID |
2551 | if (!counting) { |
2552 | spin_lock(&fs_info->balance_lock); | |
2553 | bctl->stat.considered++; | |
2554 | spin_unlock(&fs_info->balance_lock); | |
2555 | } | |
2556 | ||
f43ffb60 ID |
2557 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2558 | found_key.offset); | |
b3b4aa74 | 2559 | btrfs_release_path(path); |
f43ffb60 ID |
2560 | if (!ret) |
2561 | goto loop; | |
2562 | ||
19a39dce ID |
2563 | if (counting) { |
2564 | spin_lock(&fs_info->balance_lock); | |
2565 | bctl->stat.expected++; | |
2566 | spin_unlock(&fs_info->balance_lock); | |
2567 | goto loop; | |
2568 | } | |
2569 | ||
ec44a35c CM |
2570 | ret = btrfs_relocate_chunk(chunk_root, |
2571 | chunk_root->root_key.objectid, | |
2572 | found_key.objectid, | |
2573 | found_key.offset); | |
508794eb JB |
2574 | if (ret && ret != -ENOSPC) |
2575 | goto error; | |
19a39dce | 2576 | if (ret == -ENOSPC) { |
c9e9f97b | 2577 | enospc_errors++; |
19a39dce ID |
2578 | } else { |
2579 | spin_lock(&fs_info->balance_lock); | |
2580 | bctl->stat.completed++; | |
2581 | spin_unlock(&fs_info->balance_lock); | |
2582 | } | |
f43ffb60 | 2583 | loop: |
ba1bf481 | 2584 | key.offset = found_key.offset - 1; |
ec44a35c | 2585 | } |
c9e9f97b | 2586 | |
19a39dce ID |
2587 | if (counting) { |
2588 | btrfs_release_path(path); | |
2589 | counting = false; | |
2590 | goto again; | |
2591 | } | |
ec44a35c CM |
2592 | error: |
2593 | btrfs_free_path(path); | |
c9e9f97b ID |
2594 | if (enospc_errors) { |
2595 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
2596 | enospc_errors); | |
2597 | if (!ret) | |
2598 | ret = -ENOSPC; | |
2599 | } | |
2600 | ||
ec44a35c CM |
2601 | return ret; |
2602 | } | |
2603 | ||
837d5b6e ID |
2604 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
2605 | { | |
a7e99c69 ID |
2606 | /* cancel requested || normal exit path */ |
2607 | return atomic_read(&fs_info->balance_cancel_req) || | |
2608 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
2609 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
2610 | } |
2611 | ||
c9e9f97b ID |
2612 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
2613 | { | |
0940ebf6 ID |
2614 | int ret; |
2615 | ||
c9e9f97b | 2616 | unset_balance_control(fs_info); |
0940ebf6 ID |
2617 | ret = del_balance_item(fs_info->tree_root); |
2618 | BUG_ON(ret); | |
c9e9f97b ID |
2619 | } |
2620 | ||
19a39dce | 2621 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
2622 | struct btrfs_ioctl_balance_args *bargs); |
2623 | ||
2624 | /* | |
2625 | * Should be called with both balance and volume mutexes held | |
2626 | */ | |
2627 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
2628 | struct btrfs_ioctl_balance_args *bargs) | |
2629 | { | |
2630 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 2631 | u64 allowed; |
c9e9f97b ID |
2632 | int ret; |
2633 | ||
837d5b6e | 2634 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
2635 | atomic_read(&fs_info->balance_pause_req) || |
2636 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
2637 | ret = -EINVAL; |
2638 | goto out; | |
2639 | } | |
2640 | ||
f43ffb60 ID |
2641 | /* |
2642 | * In case of mixed groups both data and meta should be picked, | |
2643 | * and identical options should be given for both of them. | |
2644 | */ | |
2645 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); | |
2646 | if ((allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) && | |
2647 | (bctl->flags & (BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA))) { | |
2648 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || | |
2649 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
2650 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
2651 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
2652 | "metadata balance options must be the same\n"); | |
2653 | ret = -EINVAL; | |
2654 | goto out; | |
2655 | } | |
2656 | } | |
2657 | ||
e4d8ec0f ID |
2658 | /* |
2659 | * Profile changing sanity checks. Skip them if a simple | |
2660 | * balance is requested. | |
2661 | */ | |
2662 | if (!((bctl->data.flags | bctl->sys.flags | bctl->meta.flags) & | |
2663 | BTRFS_BALANCE_ARGS_CONVERT)) | |
2664 | goto do_balance; | |
2665 | ||
2666 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
2667 | if (fs_info->fs_devices->num_devices == 1) | |
2668 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
2669 | else if (fs_info->fs_devices->num_devices < 4) | |
2670 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
2671 | else | |
2672 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
2673 | BTRFS_BLOCK_GROUP_RAID10); | |
2674 | ||
2675 | if (!profile_is_valid(bctl->data.target, 1) || | |
2676 | bctl->data.target & ~allowed) { | |
2677 | printk(KERN_ERR "btrfs: unable to start balance with target " | |
2678 | "data profile %llu\n", | |
2679 | (unsigned long long)bctl->data.target); | |
2680 | ret = -EINVAL; | |
2681 | goto out; | |
2682 | } | |
2683 | if (!profile_is_valid(bctl->meta.target, 1) || | |
2684 | bctl->meta.target & ~allowed) { | |
2685 | printk(KERN_ERR "btrfs: unable to start balance with target " | |
2686 | "metadata profile %llu\n", | |
2687 | (unsigned long long)bctl->meta.target); | |
2688 | ret = -EINVAL; | |
2689 | goto out; | |
2690 | } | |
2691 | if (!profile_is_valid(bctl->sys.target, 1) || | |
2692 | bctl->sys.target & ~allowed) { | |
2693 | printk(KERN_ERR "btrfs: unable to start balance with target " | |
2694 | "system profile %llu\n", | |
2695 | (unsigned long long)bctl->sys.target); | |
2696 | ret = -EINVAL; | |
2697 | goto out; | |
2698 | } | |
2699 | ||
2700 | if (bctl->data.target & BTRFS_BLOCK_GROUP_DUP) { | |
2701 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); | |
2702 | ret = -EINVAL; | |
2703 | goto out; | |
2704 | } | |
2705 | ||
2706 | /* allow to reduce meta or sys integrity only if force set */ | |
2707 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2708 | BTRFS_BLOCK_GROUP_RAID10; | |
2709 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2710 | (fs_info->avail_system_alloc_bits & allowed) && | |
2711 | !(bctl->sys.target & allowed)) || | |
2712 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2713 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
2714 | !(bctl->meta.target & allowed))) { | |
2715 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
2716 | printk(KERN_INFO "btrfs: force reducing metadata " | |
2717 | "integrity\n"); | |
2718 | } else { | |
2719 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
2720 | "integrity, use force if you want this\n"); | |
2721 | ret = -EINVAL; | |
2722 | goto out; | |
2723 | } | |
2724 | } | |
2725 | ||
2726 | do_balance: | |
0940ebf6 | 2727 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 2728 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
2729 | goto out; |
2730 | ||
59641015 ID |
2731 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
2732 | BUG_ON(ret == -EEXIST); | |
2733 | set_balance_control(bctl); | |
2734 | } else { | |
2735 | BUG_ON(ret != -EEXIST); | |
2736 | spin_lock(&fs_info->balance_lock); | |
2737 | update_balance_args(bctl); | |
2738 | spin_unlock(&fs_info->balance_lock); | |
2739 | } | |
c9e9f97b | 2740 | |
837d5b6e | 2741 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
2742 | mutex_unlock(&fs_info->balance_mutex); |
2743 | ||
2744 | ret = __btrfs_balance(fs_info); | |
2745 | ||
2746 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 2747 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
2748 | |
2749 | if (bargs) { | |
2750 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 2751 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
2752 | } |
2753 | ||
837d5b6e ID |
2754 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
2755 | balance_need_close(fs_info)) { | |
2756 | __cancel_balance(fs_info); | |
2757 | } | |
2758 | ||
2759 | wake_up(&fs_info->balance_wait_q); | |
c9e9f97b ID |
2760 | |
2761 | return ret; | |
2762 | out: | |
59641015 ID |
2763 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
2764 | __cancel_balance(fs_info); | |
2765 | else | |
2766 | kfree(bctl); | |
2767 | return ret; | |
2768 | } | |
2769 | ||
2770 | static int balance_kthread(void *data) | |
2771 | { | |
2772 | struct btrfs_balance_control *bctl = | |
2773 | (struct btrfs_balance_control *)data; | |
2774 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
9555c6c1 | 2775 | int ret = 0; |
59641015 ID |
2776 | |
2777 | mutex_lock(&fs_info->volume_mutex); | |
2778 | mutex_lock(&fs_info->balance_mutex); | |
2779 | ||
2780 | set_balance_control(bctl); | |
2781 | ||
9555c6c1 ID |
2782 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { |
2783 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
2784 | } else { | |
2785 | printk(KERN_INFO "btrfs: continuing balance\n"); | |
2786 | ret = btrfs_balance(bctl, NULL); | |
2787 | } | |
59641015 ID |
2788 | |
2789 | mutex_unlock(&fs_info->balance_mutex); | |
2790 | mutex_unlock(&fs_info->volume_mutex); | |
2791 | return ret; | |
2792 | } | |
2793 | ||
2794 | int btrfs_recover_balance(struct btrfs_root *tree_root) | |
2795 | { | |
2796 | struct task_struct *tsk; | |
2797 | struct btrfs_balance_control *bctl; | |
2798 | struct btrfs_balance_item *item; | |
2799 | struct btrfs_disk_balance_args disk_bargs; | |
2800 | struct btrfs_path *path; | |
2801 | struct extent_buffer *leaf; | |
2802 | struct btrfs_key key; | |
2803 | int ret; | |
2804 | ||
2805 | path = btrfs_alloc_path(); | |
2806 | if (!path) | |
2807 | return -ENOMEM; | |
2808 | ||
2809 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
2810 | if (!bctl) { | |
2811 | ret = -ENOMEM; | |
2812 | goto out; | |
2813 | } | |
2814 | ||
2815 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2816 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2817 | key.offset = 0; | |
2818 | ||
2819 | ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); | |
2820 | if (ret < 0) | |
2821 | goto out_bctl; | |
2822 | if (ret > 0) { /* ret = -ENOENT; */ | |
2823 | ret = 0; | |
2824 | goto out_bctl; | |
2825 | } | |
2826 | ||
2827 | leaf = path->nodes[0]; | |
2828 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2829 | ||
2830 | bctl->fs_info = tree_root->fs_info; | |
2831 | bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME; | |
2832 | ||
2833 | btrfs_balance_data(leaf, item, &disk_bargs); | |
2834 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
2835 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
2836 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
2837 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
2838 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
2839 | ||
2840 | tsk = kthread_run(balance_kthread, bctl, "btrfs-balance"); | |
2841 | if (IS_ERR(tsk)) | |
2842 | ret = PTR_ERR(tsk); | |
2843 | else | |
2844 | goto out; | |
2845 | ||
2846 | out_bctl: | |
c9e9f97b | 2847 | kfree(bctl); |
59641015 ID |
2848 | out: |
2849 | btrfs_free_path(path); | |
ec44a35c CM |
2850 | return ret; |
2851 | } | |
2852 | ||
837d5b6e ID |
2853 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
2854 | { | |
2855 | int ret = 0; | |
2856 | ||
2857 | mutex_lock(&fs_info->balance_mutex); | |
2858 | if (!fs_info->balance_ctl) { | |
2859 | mutex_unlock(&fs_info->balance_mutex); | |
2860 | return -ENOTCONN; | |
2861 | } | |
2862 | ||
2863 | if (atomic_read(&fs_info->balance_running)) { | |
2864 | atomic_inc(&fs_info->balance_pause_req); | |
2865 | mutex_unlock(&fs_info->balance_mutex); | |
2866 | ||
2867 | wait_event(fs_info->balance_wait_q, | |
2868 | atomic_read(&fs_info->balance_running) == 0); | |
2869 | ||
2870 | mutex_lock(&fs_info->balance_mutex); | |
2871 | /* we are good with balance_ctl ripped off from under us */ | |
2872 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
2873 | atomic_dec(&fs_info->balance_pause_req); | |
2874 | } else { | |
2875 | ret = -ENOTCONN; | |
2876 | } | |
2877 | ||
2878 | mutex_unlock(&fs_info->balance_mutex); | |
2879 | return ret; | |
2880 | } | |
2881 | ||
a7e99c69 ID |
2882 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
2883 | { | |
2884 | mutex_lock(&fs_info->balance_mutex); | |
2885 | if (!fs_info->balance_ctl) { | |
2886 | mutex_unlock(&fs_info->balance_mutex); | |
2887 | return -ENOTCONN; | |
2888 | } | |
2889 | ||
2890 | atomic_inc(&fs_info->balance_cancel_req); | |
2891 | /* | |
2892 | * if we are running just wait and return, balance item is | |
2893 | * deleted in btrfs_balance in this case | |
2894 | */ | |
2895 | if (atomic_read(&fs_info->balance_running)) { | |
2896 | mutex_unlock(&fs_info->balance_mutex); | |
2897 | wait_event(fs_info->balance_wait_q, | |
2898 | atomic_read(&fs_info->balance_running) == 0); | |
2899 | mutex_lock(&fs_info->balance_mutex); | |
2900 | } else { | |
2901 | /* __cancel_balance needs volume_mutex */ | |
2902 | mutex_unlock(&fs_info->balance_mutex); | |
2903 | mutex_lock(&fs_info->volume_mutex); | |
2904 | mutex_lock(&fs_info->balance_mutex); | |
2905 | ||
2906 | if (fs_info->balance_ctl) | |
2907 | __cancel_balance(fs_info); | |
2908 | ||
2909 | mutex_unlock(&fs_info->volume_mutex); | |
2910 | } | |
2911 | ||
2912 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
2913 | atomic_dec(&fs_info->balance_cancel_req); | |
2914 | mutex_unlock(&fs_info->balance_mutex); | |
2915 | return 0; | |
2916 | } | |
2917 | ||
8f18cf13 CM |
2918 | /* |
2919 | * shrinking a device means finding all of the device extents past | |
2920 | * the new size, and then following the back refs to the chunks. | |
2921 | * The chunk relocation code actually frees the device extent | |
2922 | */ | |
2923 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
2924 | { | |
2925 | struct btrfs_trans_handle *trans; | |
2926 | struct btrfs_root *root = device->dev_root; | |
2927 | struct btrfs_dev_extent *dev_extent = NULL; | |
2928 | struct btrfs_path *path; | |
2929 | u64 length; | |
2930 | u64 chunk_tree; | |
2931 | u64 chunk_objectid; | |
2932 | u64 chunk_offset; | |
2933 | int ret; | |
2934 | int slot; | |
ba1bf481 JB |
2935 | int failed = 0; |
2936 | bool retried = false; | |
8f18cf13 CM |
2937 | struct extent_buffer *l; |
2938 | struct btrfs_key key; | |
6c41761f | 2939 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 2940 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 2941 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
2942 | u64 diff = device->total_bytes - new_size; |
2943 | ||
2b82032c YZ |
2944 | if (new_size >= device->total_bytes) |
2945 | return -EINVAL; | |
8f18cf13 CM |
2946 | |
2947 | path = btrfs_alloc_path(); | |
2948 | if (!path) | |
2949 | return -ENOMEM; | |
2950 | ||
8f18cf13 CM |
2951 | path->reada = 2; |
2952 | ||
7d9eb12c CM |
2953 | lock_chunks(root); |
2954 | ||
8f18cf13 | 2955 | device->total_bytes = new_size; |
2bf64758 | 2956 | if (device->writeable) { |
2b82032c | 2957 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
2958 | spin_lock(&root->fs_info->free_chunk_lock); |
2959 | root->fs_info->free_chunk_space -= diff; | |
2960 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2961 | } | |
7d9eb12c | 2962 | unlock_chunks(root); |
8f18cf13 | 2963 | |
ba1bf481 | 2964 | again: |
8f18cf13 CM |
2965 | key.objectid = device->devid; |
2966 | key.offset = (u64)-1; | |
2967 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2968 | ||
2969 | while (1) { | |
2970 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2971 | if (ret < 0) | |
2972 | goto done; | |
2973 | ||
2974 | ret = btrfs_previous_item(root, path, 0, key.type); | |
2975 | if (ret < 0) | |
2976 | goto done; | |
2977 | if (ret) { | |
2978 | ret = 0; | |
b3b4aa74 | 2979 | btrfs_release_path(path); |
bf1fb512 | 2980 | break; |
8f18cf13 CM |
2981 | } |
2982 | ||
2983 | l = path->nodes[0]; | |
2984 | slot = path->slots[0]; | |
2985 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
2986 | ||
ba1bf481 | 2987 | if (key.objectid != device->devid) { |
b3b4aa74 | 2988 | btrfs_release_path(path); |
bf1fb512 | 2989 | break; |
ba1bf481 | 2990 | } |
8f18cf13 CM |
2991 | |
2992 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
2993 | length = btrfs_dev_extent_length(l, dev_extent); | |
2994 | ||
ba1bf481 | 2995 | if (key.offset + length <= new_size) { |
b3b4aa74 | 2996 | btrfs_release_path(path); |
d6397bae | 2997 | break; |
ba1bf481 | 2998 | } |
8f18cf13 CM |
2999 | |
3000 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3001 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3002 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3003 | btrfs_release_path(path); |
8f18cf13 CM |
3004 | |
3005 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3006 | chunk_offset); | |
ba1bf481 | 3007 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3008 | goto done; |
ba1bf481 JB |
3009 | if (ret == -ENOSPC) |
3010 | failed++; | |
3011 | key.offset -= 1; | |
3012 | } | |
3013 | ||
3014 | if (failed && !retried) { | |
3015 | failed = 0; | |
3016 | retried = true; | |
3017 | goto again; | |
3018 | } else if (failed && retried) { | |
3019 | ret = -ENOSPC; | |
3020 | lock_chunks(root); | |
3021 | ||
3022 | device->total_bytes = old_size; | |
3023 | if (device->writeable) | |
3024 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3025 | spin_lock(&root->fs_info->free_chunk_lock); |
3026 | root->fs_info->free_chunk_space += diff; | |
3027 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3028 | unlock_chunks(root); |
3029 | goto done; | |
8f18cf13 CM |
3030 | } |
3031 | ||
d6397bae | 3032 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3033 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3034 | if (IS_ERR(trans)) { |
3035 | ret = PTR_ERR(trans); | |
3036 | goto done; | |
3037 | } | |
3038 | ||
d6397bae CB |
3039 | lock_chunks(root); |
3040 | ||
3041 | device->disk_total_bytes = new_size; | |
3042 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3043 | ret = btrfs_update_device(trans, device); | |
3044 | if (ret) { | |
3045 | unlock_chunks(root); | |
3046 | btrfs_end_transaction(trans, root); | |
3047 | goto done; | |
3048 | } | |
3049 | WARN_ON(diff > old_total); | |
3050 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3051 | unlock_chunks(root); | |
3052 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3053 | done: |
3054 | btrfs_free_path(path); | |
3055 | return ret; | |
3056 | } | |
3057 | ||
125ccb0a | 3058 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3059 | struct btrfs_key *key, |
3060 | struct btrfs_chunk *chunk, int item_size) | |
3061 | { | |
6c41761f | 3062 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3063 | struct btrfs_disk_key disk_key; |
3064 | u32 array_size; | |
3065 | u8 *ptr; | |
3066 | ||
3067 | array_size = btrfs_super_sys_array_size(super_copy); | |
3068 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3069 | return -EFBIG; | |
3070 | ||
3071 | ptr = super_copy->sys_chunk_array + array_size; | |
3072 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3073 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3074 | ptr += sizeof(disk_key); | |
3075 | memcpy(ptr, chunk, item_size); | |
3076 | item_size += sizeof(disk_key); | |
3077 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3078 | return 0; | |
3079 | } | |
3080 | ||
73c5de00 AJ |
3081 | /* |
3082 | * sort the devices in descending order by max_avail, total_avail | |
3083 | */ | |
3084 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3085 | { |
73c5de00 AJ |
3086 | const struct btrfs_device_info *di_a = a; |
3087 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3088 | |
73c5de00 | 3089 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3090 | return -1; |
73c5de00 | 3091 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3092 | return 1; |
73c5de00 AJ |
3093 | if (di_a->total_avail > di_b->total_avail) |
3094 | return -1; | |
3095 | if (di_a->total_avail < di_b->total_avail) | |
3096 | return 1; | |
3097 | return 0; | |
b2117a39 | 3098 | } |
0b86a832 | 3099 | |
73c5de00 AJ |
3100 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3101 | struct btrfs_root *extent_root, | |
3102 | struct map_lookup **map_ret, | |
3103 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3104 | u64 start, u64 type) | |
b2117a39 | 3105 | { |
73c5de00 AJ |
3106 | struct btrfs_fs_info *info = extent_root->fs_info; |
3107 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3108 | struct list_head *cur; | |
3109 | struct map_lookup *map = NULL; | |
3110 | struct extent_map_tree *em_tree; | |
3111 | struct extent_map *em; | |
3112 | struct btrfs_device_info *devices_info = NULL; | |
3113 | u64 total_avail; | |
3114 | int num_stripes; /* total number of stripes to allocate */ | |
3115 | int sub_stripes; /* sub_stripes info for map */ | |
3116 | int dev_stripes; /* stripes per dev */ | |
3117 | int devs_max; /* max devs to use */ | |
3118 | int devs_min; /* min devs needed */ | |
3119 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3120 | int ncopies; /* how many copies to data has */ | |
3121 | int ret; | |
3122 | u64 max_stripe_size; | |
3123 | u64 max_chunk_size; | |
3124 | u64 stripe_size; | |
3125 | u64 num_bytes; | |
3126 | int ndevs; | |
3127 | int i; | |
3128 | int j; | |
593060d7 | 3129 | |
73c5de00 AJ |
3130 | if ((type & BTRFS_BLOCK_GROUP_RAID1) && |
3131 | (type & BTRFS_BLOCK_GROUP_DUP)) { | |
3132 | WARN_ON(1); | |
3133 | type &= ~BTRFS_BLOCK_GROUP_DUP; | |
321aecc6 | 3134 | } |
9b3f68b9 | 3135 | |
73c5de00 AJ |
3136 | if (list_empty(&fs_devices->alloc_list)) |
3137 | return -ENOSPC; | |
b2117a39 | 3138 | |
73c5de00 AJ |
3139 | sub_stripes = 1; |
3140 | dev_stripes = 1; | |
3141 | devs_increment = 1; | |
3142 | ncopies = 1; | |
3143 | devs_max = 0; /* 0 == as many as possible */ | |
3144 | devs_min = 1; | |
3145 | ||
3146 | /* | |
3147 | * define the properties of each RAID type. | |
3148 | * FIXME: move this to a global table and use it in all RAID | |
3149 | * calculation code | |
3150 | */ | |
3151 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
3152 | dev_stripes = 2; | |
b2117a39 | 3153 | ncopies = 2; |
73c5de00 AJ |
3154 | devs_max = 1; |
3155 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
3156 | devs_min = 2; | |
3157 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
3158 | devs_increment = 2; | |
b2117a39 | 3159 | ncopies = 2; |
73c5de00 AJ |
3160 | devs_max = 2; |
3161 | devs_min = 2; | |
3162 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
3163 | sub_stripes = 2; | |
3164 | devs_increment = 2; | |
3165 | ncopies = 2; | |
3166 | devs_min = 4; | |
3167 | } else { | |
3168 | devs_max = 1; | |
3169 | } | |
b2117a39 | 3170 | |
9b3f68b9 | 3171 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3172 | max_stripe_size = 1024 * 1024 * 1024; |
3173 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3174 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3175 | /* for larger filesystems, use larger metadata chunks */ |
3176 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3177 | max_stripe_size = 1024 * 1024 * 1024; | |
3178 | else | |
3179 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3180 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3181 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3182 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3183 | max_chunk_size = 2 * max_stripe_size; |
3184 | } else { | |
3185 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3186 | type); | |
3187 | BUG_ON(1); | |
9b3f68b9 CM |
3188 | } |
3189 | ||
2b82032c YZ |
3190 | /* we don't want a chunk larger than 10% of writeable space */ |
3191 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3192 | max_chunk_size); | |
9b3f68b9 | 3193 | |
73c5de00 AJ |
3194 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3195 | GFP_NOFS); | |
3196 | if (!devices_info) | |
3197 | return -ENOMEM; | |
0cad8a11 | 3198 | |
73c5de00 | 3199 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3200 | |
9f680ce0 | 3201 | /* |
73c5de00 AJ |
3202 | * in the first pass through the devices list, we gather information |
3203 | * about the available holes on each device. | |
9f680ce0 | 3204 | */ |
73c5de00 AJ |
3205 | ndevs = 0; |
3206 | while (cur != &fs_devices->alloc_list) { | |
3207 | struct btrfs_device *device; | |
3208 | u64 max_avail; | |
3209 | u64 dev_offset; | |
b2117a39 | 3210 | |
73c5de00 | 3211 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3212 | |
73c5de00 | 3213 | cur = cur->next; |
b2117a39 | 3214 | |
73c5de00 AJ |
3215 | if (!device->writeable) { |
3216 | printk(KERN_ERR | |
3217 | "btrfs: read-only device in alloc_list\n"); | |
3218 | WARN_ON(1); | |
3219 | continue; | |
3220 | } | |
b2117a39 | 3221 | |
73c5de00 AJ |
3222 | if (!device->in_fs_metadata) |
3223 | continue; | |
b2117a39 | 3224 | |
73c5de00 AJ |
3225 | if (device->total_bytes > device->bytes_used) |
3226 | total_avail = device->total_bytes - device->bytes_used; | |
3227 | else | |
3228 | total_avail = 0; | |
38c01b96 | 3229 | |
3230 | /* If there is no space on this device, skip it. */ | |
3231 | if (total_avail == 0) | |
3232 | continue; | |
b2117a39 | 3233 | |
125ccb0a | 3234 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3235 | max_stripe_size * dev_stripes, |
3236 | &dev_offset, &max_avail); | |
3237 | if (ret && ret != -ENOSPC) | |
3238 | goto error; | |
b2117a39 | 3239 | |
73c5de00 AJ |
3240 | if (ret == 0) |
3241 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3242 | |
73c5de00 AJ |
3243 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3244 | continue; | |
b2117a39 | 3245 | |
73c5de00 AJ |
3246 | devices_info[ndevs].dev_offset = dev_offset; |
3247 | devices_info[ndevs].max_avail = max_avail; | |
3248 | devices_info[ndevs].total_avail = total_avail; | |
3249 | devices_info[ndevs].dev = device; | |
3250 | ++ndevs; | |
3251 | } | |
b2117a39 | 3252 | |
73c5de00 AJ |
3253 | /* |
3254 | * now sort the devices by hole size / available space | |
3255 | */ | |
3256 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3257 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3258 | |
73c5de00 AJ |
3259 | /* round down to number of usable stripes */ |
3260 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3261 | |
73c5de00 AJ |
3262 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3263 | ret = -ENOSPC; | |
3264 | goto error; | |
b2117a39 | 3265 | } |
9f680ce0 | 3266 | |
73c5de00 AJ |
3267 | if (devs_max && ndevs > devs_max) |
3268 | ndevs = devs_max; | |
3269 | /* | |
3270 | * the primary goal is to maximize the number of stripes, so use as many | |
3271 | * devices as possible, even if the stripes are not maximum sized. | |
3272 | */ | |
3273 | stripe_size = devices_info[ndevs-1].max_avail; | |
3274 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3275 | |
73c5de00 AJ |
3276 | if (stripe_size * num_stripes > max_chunk_size * ncopies) { |
3277 | stripe_size = max_chunk_size * ncopies; | |
3278 | do_div(stripe_size, num_stripes); | |
b2117a39 | 3279 | } |
b2117a39 | 3280 | |
73c5de00 AJ |
3281 | do_div(stripe_size, dev_stripes); |
3282 | do_div(stripe_size, BTRFS_STRIPE_LEN); | |
3283 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
3284 | |
3285 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3286 | if (!map) { | |
3287 | ret = -ENOMEM; | |
3288 | goto error; | |
3289 | } | |
3290 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3291 | |
73c5de00 AJ |
3292 | for (i = 0; i < ndevs; ++i) { |
3293 | for (j = 0; j < dev_stripes; ++j) { | |
3294 | int s = i * dev_stripes + j; | |
3295 | map->stripes[s].dev = devices_info[i].dev; | |
3296 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3297 | j * stripe_size; | |
6324fbf3 | 3298 | } |
6324fbf3 | 3299 | } |
2b82032c | 3300 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
3301 | map->stripe_len = BTRFS_STRIPE_LEN; |
3302 | map->io_align = BTRFS_STRIPE_LEN; | |
3303 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 3304 | map->type = type; |
2b82032c | 3305 | map->sub_stripes = sub_stripes; |
0b86a832 | 3306 | |
2b82032c | 3307 | *map_ret = map; |
73c5de00 | 3308 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 3309 | |
73c5de00 AJ |
3310 | *stripe_size_out = stripe_size; |
3311 | *num_bytes_out = num_bytes; | |
0b86a832 | 3312 | |
73c5de00 | 3313 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3314 | |
172ddd60 | 3315 | em = alloc_extent_map(); |
2b82032c | 3316 | if (!em) { |
b2117a39 MX |
3317 | ret = -ENOMEM; |
3318 | goto error; | |
593060d7 | 3319 | } |
2b82032c YZ |
3320 | em->bdev = (struct block_device *)map; |
3321 | em->start = start; | |
73c5de00 | 3322 | em->len = num_bytes; |
2b82032c YZ |
3323 | em->block_start = 0; |
3324 | em->block_len = em->len; | |
593060d7 | 3325 | |
2b82032c | 3326 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3327 | write_lock(&em_tree->lock); |
2b82032c | 3328 | ret = add_extent_mapping(em_tree, em); |
890871be | 3329 | write_unlock(&em_tree->lock); |
2b82032c YZ |
3330 | BUG_ON(ret); |
3331 | free_extent_map(em); | |
0b86a832 | 3332 | |
2b82032c YZ |
3333 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
3334 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3335 | start, num_bytes); |
2b82032c | 3336 | BUG_ON(ret); |
611f0e00 | 3337 | |
73c5de00 AJ |
3338 | for (i = 0; i < map->num_stripes; ++i) { |
3339 | struct btrfs_device *device; | |
3340 | u64 dev_offset; | |
3341 | ||
3342 | device = map->stripes[i].dev; | |
3343 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3344 | |
3345 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3346 | info->chunk_root->root_key.objectid, |
3347 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3348 | start, dev_offset, stripe_size); |
0b86a832 | 3349 | BUG_ON(ret); |
2b82032c YZ |
3350 | } |
3351 | ||
b2117a39 | 3352 | kfree(devices_info); |
2b82032c | 3353 | return 0; |
b2117a39 MX |
3354 | |
3355 | error: | |
3356 | kfree(map); | |
3357 | kfree(devices_info); | |
3358 | return ret; | |
2b82032c YZ |
3359 | } |
3360 | ||
3361 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3362 | struct btrfs_root *extent_root, | |
3363 | struct map_lookup *map, u64 chunk_offset, | |
3364 | u64 chunk_size, u64 stripe_size) | |
3365 | { | |
3366 | u64 dev_offset; | |
3367 | struct btrfs_key key; | |
3368 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3369 | struct btrfs_device *device; | |
3370 | struct btrfs_chunk *chunk; | |
3371 | struct btrfs_stripe *stripe; | |
3372 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
3373 | int index = 0; | |
3374 | int ret; | |
3375 | ||
3376 | chunk = kzalloc(item_size, GFP_NOFS); | |
3377 | if (!chunk) | |
3378 | return -ENOMEM; | |
3379 | ||
3380 | index = 0; | |
3381 | while (index < map->num_stripes) { | |
3382 | device = map->stripes[index].dev; | |
3383 | device->bytes_used += stripe_size; | |
0b86a832 CM |
3384 | ret = btrfs_update_device(trans, device); |
3385 | BUG_ON(ret); | |
2b82032c YZ |
3386 | index++; |
3387 | } | |
3388 | ||
2bf64758 JB |
3389 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
3390 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
3391 | map->num_stripes); | |
3392 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
3393 | ||
2b82032c YZ |
3394 | index = 0; |
3395 | stripe = &chunk->stripe; | |
3396 | while (index < map->num_stripes) { | |
3397 | device = map->stripes[index].dev; | |
3398 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 3399 | |
e17cade2 CM |
3400 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
3401 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
3402 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 3403 | stripe++; |
0b86a832 CM |
3404 | index++; |
3405 | } | |
3406 | ||
2b82032c | 3407 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 3408 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
3409 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
3410 | btrfs_set_stack_chunk_type(chunk, map->type); | |
3411 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
3412 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
3413 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 3414 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 3415 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 3416 | |
2b82032c YZ |
3417 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3418 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3419 | key.offset = chunk_offset; | |
0b86a832 | 3420 | |
2b82032c | 3421 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
0b86a832 | 3422 | |
4ed1d16e MF |
3423 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
3424 | /* | |
3425 | * TODO: Cleanup of inserted chunk root in case of | |
3426 | * failure. | |
3427 | */ | |
125ccb0a | 3428 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 3429 | item_size); |
8f18cf13 | 3430 | } |
1abe9b8a | 3431 | |
0b86a832 | 3432 | kfree(chunk); |
4ed1d16e | 3433 | return ret; |
2b82032c | 3434 | } |
0b86a832 | 3435 | |
2b82032c YZ |
3436 | /* |
3437 | * Chunk allocation falls into two parts. The first part does works | |
3438 | * that make the new allocated chunk useable, but not do any operation | |
3439 | * that modifies the chunk tree. The second part does the works that | |
3440 | * require modifying the chunk tree. This division is important for the | |
3441 | * bootstrap process of adding storage to a seed btrfs. | |
3442 | */ | |
3443 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
3444 | struct btrfs_root *extent_root, u64 type) | |
3445 | { | |
3446 | u64 chunk_offset; | |
3447 | u64 chunk_size; | |
3448 | u64 stripe_size; | |
3449 | struct map_lookup *map; | |
3450 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3451 | int ret; | |
3452 | ||
3453 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3454 | &chunk_offset); | |
3455 | if (ret) | |
3456 | return ret; | |
3457 | ||
3458 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3459 | &stripe_size, chunk_offset, type); | |
3460 | if (ret) | |
3461 | return ret; | |
3462 | ||
3463 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3464 | chunk_size, stripe_size); | |
3465 | BUG_ON(ret); | |
3466 | return 0; | |
3467 | } | |
3468 | ||
d397712b | 3469 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
3470 | struct btrfs_root *root, |
3471 | struct btrfs_device *device) | |
3472 | { | |
3473 | u64 chunk_offset; | |
3474 | u64 sys_chunk_offset; | |
3475 | u64 chunk_size; | |
3476 | u64 sys_chunk_size; | |
3477 | u64 stripe_size; | |
3478 | u64 sys_stripe_size; | |
3479 | u64 alloc_profile; | |
3480 | struct map_lookup *map; | |
3481 | struct map_lookup *sys_map; | |
3482 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3483 | struct btrfs_root *extent_root = fs_info->extent_root; | |
3484 | int ret; | |
3485 | ||
3486 | ret = find_next_chunk(fs_info->chunk_root, | |
3487 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
3488 | if (ret) |
3489 | return ret; | |
2b82032c YZ |
3490 | |
3491 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
6fef8df1 | 3492 | fs_info->avail_metadata_alloc_bits; |
2b82032c YZ |
3493 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3494 | ||
3495 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3496 | &stripe_size, chunk_offset, alloc_profile); | |
3497 | BUG_ON(ret); | |
3498 | ||
3499 | sys_chunk_offset = chunk_offset + chunk_size; | |
3500 | ||
3501 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
6fef8df1 | 3502 | fs_info->avail_system_alloc_bits; |
2b82032c YZ |
3503 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3504 | ||
3505 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
3506 | &sys_chunk_size, &sys_stripe_size, | |
3507 | sys_chunk_offset, alloc_profile); | |
3508 | BUG_ON(ret); | |
3509 | ||
3510 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
3511 | BUG_ON(ret); | |
3512 | ||
3513 | /* | |
3514 | * Modifying chunk tree needs allocating new blocks from both | |
3515 | * system block group and metadata block group. So we only can | |
3516 | * do operations require modifying the chunk tree after both | |
3517 | * block groups were created. | |
3518 | */ | |
3519 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3520 | chunk_size, stripe_size); | |
3521 | BUG_ON(ret); | |
3522 | ||
3523 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
3524 | sys_chunk_offset, sys_chunk_size, | |
3525 | sys_stripe_size); | |
b248a415 | 3526 | BUG_ON(ret); |
2b82032c YZ |
3527 | return 0; |
3528 | } | |
3529 | ||
3530 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
3531 | { | |
3532 | struct extent_map *em; | |
3533 | struct map_lookup *map; | |
3534 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3535 | int readonly = 0; | |
3536 | int i; | |
3537 | ||
890871be | 3538 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 3539 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 3540 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
3541 | if (!em) |
3542 | return 1; | |
3543 | ||
f48b9075 JB |
3544 | if (btrfs_test_opt(root, DEGRADED)) { |
3545 | free_extent_map(em); | |
3546 | return 0; | |
3547 | } | |
3548 | ||
2b82032c YZ |
3549 | map = (struct map_lookup *)em->bdev; |
3550 | for (i = 0; i < map->num_stripes; i++) { | |
3551 | if (!map->stripes[i].dev->writeable) { | |
3552 | readonly = 1; | |
3553 | break; | |
3554 | } | |
3555 | } | |
0b86a832 | 3556 | free_extent_map(em); |
2b82032c | 3557 | return readonly; |
0b86a832 CM |
3558 | } |
3559 | ||
3560 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
3561 | { | |
a8067e02 | 3562 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
3563 | } |
3564 | ||
3565 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
3566 | { | |
3567 | struct extent_map *em; | |
3568 | ||
d397712b | 3569 | while (1) { |
890871be | 3570 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
3571 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
3572 | if (em) | |
3573 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 3574 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
3575 | if (!em) |
3576 | break; | |
3577 | kfree(em->bdev); | |
3578 | /* once for us */ | |
3579 | free_extent_map(em); | |
3580 | /* once for the tree */ | |
3581 | free_extent_map(em); | |
3582 | } | |
3583 | } | |
3584 | ||
f188591e CM |
3585 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
3586 | { | |
3587 | struct extent_map *em; | |
3588 | struct map_lookup *map; | |
3589 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3590 | int ret; | |
3591 | ||
890871be | 3592 | read_lock(&em_tree->lock); |
f188591e | 3593 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 3594 | read_unlock(&em_tree->lock); |
f188591e CM |
3595 | BUG_ON(!em); |
3596 | ||
3597 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3598 | map = (struct map_lookup *)em->bdev; | |
3599 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
3600 | ret = map->num_stripes; | |
321aecc6 CM |
3601 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
3602 | ret = map->sub_stripes; | |
f188591e CM |
3603 | else |
3604 | ret = 1; | |
3605 | free_extent_map(em); | |
f188591e CM |
3606 | return ret; |
3607 | } | |
3608 | ||
dfe25020 CM |
3609 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
3610 | int optimal) | |
3611 | { | |
3612 | int i; | |
3613 | if (map->stripes[optimal].dev->bdev) | |
3614 | return optimal; | |
3615 | for (i = first; i < first + num; i++) { | |
3616 | if (map->stripes[i].dev->bdev) | |
3617 | return i; | |
3618 | } | |
3619 | /* we couldn't find one that doesn't fail. Just return something | |
3620 | * and the io error handling code will clean up eventually | |
3621 | */ | |
3622 | return optimal; | |
3623 | } | |
3624 | ||
f2d8d74d CM |
3625 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3626 | u64 logical, u64 *length, | |
a1d3c478 | 3627 | struct btrfs_bio **bbio_ret, |
7eaceacc | 3628 | int mirror_num) |
0b86a832 CM |
3629 | { |
3630 | struct extent_map *em; | |
3631 | struct map_lookup *map; | |
3632 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3633 | u64 offset; | |
593060d7 | 3634 | u64 stripe_offset; |
fce3bb9a | 3635 | u64 stripe_end_offset; |
593060d7 | 3636 | u64 stripe_nr; |
fce3bb9a LD |
3637 | u64 stripe_nr_orig; |
3638 | u64 stripe_nr_end; | |
593060d7 | 3639 | int stripe_index; |
cea9e445 | 3640 | int i; |
de11cc12 | 3641 | int ret = 0; |
f2d8d74d | 3642 | int num_stripes; |
a236aed1 | 3643 | int max_errors = 0; |
a1d3c478 | 3644 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3645 | |
890871be | 3646 | read_lock(&em_tree->lock); |
0b86a832 | 3647 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 3648 | read_unlock(&em_tree->lock); |
f2d8d74d | 3649 | |
3b951516 | 3650 | if (!em) { |
d397712b CM |
3651 | printk(KERN_CRIT "unable to find logical %llu len %llu\n", |
3652 | (unsigned long long)logical, | |
3653 | (unsigned long long)*length); | |
f2d8d74d | 3654 | BUG(); |
3b951516 | 3655 | } |
0b86a832 CM |
3656 | |
3657 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3658 | map = (struct map_lookup *)em->bdev; | |
3659 | offset = logical - em->start; | |
593060d7 | 3660 | |
f188591e CM |
3661 | if (mirror_num > map->num_stripes) |
3662 | mirror_num = 0; | |
3663 | ||
593060d7 CM |
3664 | stripe_nr = offset; |
3665 | /* | |
3666 | * stripe_nr counts the total number of stripes we have to stride | |
3667 | * to get to this block | |
3668 | */ | |
3669 | do_div(stripe_nr, map->stripe_len); | |
3670 | ||
3671 | stripe_offset = stripe_nr * map->stripe_len; | |
3672 | BUG_ON(offset < stripe_offset); | |
3673 | ||
3674 | /* stripe_offset is the offset of this block in its stripe*/ | |
3675 | stripe_offset = offset - stripe_offset; | |
3676 | ||
fce3bb9a LD |
3677 | if (rw & REQ_DISCARD) |
3678 | *length = min_t(u64, em->len - offset, *length); | |
52ba6929 | 3679 | else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
cea9e445 CM |
3680 | /* we limit the length of each bio to what fits in a stripe */ |
3681 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 3682 | map->stripe_len - stripe_offset); |
cea9e445 CM |
3683 | } else { |
3684 | *length = em->len - offset; | |
3685 | } | |
f2d8d74d | 3686 | |
a1d3c478 | 3687 | if (!bbio_ret) |
cea9e445 CM |
3688 | goto out; |
3689 | ||
f2d8d74d | 3690 | num_stripes = 1; |
cea9e445 | 3691 | stripe_index = 0; |
fce3bb9a LD |
3692 | stripe_nr_orig = stripe_nr; |
3693 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
3694 | (~(map->stripe_len - 1)); | |
3695 | do_div(stripe_nr_end, map->stripe_len); | |
3696 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
3697 | (offset + *length); | |
3698 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3699 | if (rw & REQ_DISCARD) | |
3700 | num_stripes = min_t(u64, map->num_stripes, | |
3701 | stripe_nr_end - stripe_nr_orig); | |
3702 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3703 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3704 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3705 | num_stripes = map->num_stripes; |
2fff734f | 3706 | else if (mirror_num) |
f188591e | 3707 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3708 | else { |
3709 | stripe_index = find_live_mirror(map, 0, | |
3710 | map->num_stripes, | |
3711 | current->pid % map->num_stripes); | |
a1d3c478 | 3712 | mirror_num = stripe_index + 1; |
dfe25020 | 3713 | } |
2fff734f | 3714 | |
611f0e00 | 3715 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3716 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3717 | num_stripes = map->num_stripes; |
a1d3c478 | 3718 | } else if (mirror_num) { |
f188591e | 3719 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3720 | } else { |
3721 | mirror_num = 1; | |
3722 | } | |
2fff734f | 3723 | |
321aecc6 CM |
3724 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3725 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3726 | |
3727 | stripe_index = do_div(stripe_nr, factor); | |
3728 | stripe_index *= map->sub_stripes; | |
3729 | ||
7eaceacc | 3730 | if (rw & REQ_WRITE) |
f2d8d74d | 3731 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3732 | else if (rw & REQ_DISCARD) |
3733 | num_stripes = min_t(u64, map->sub_stripes * | |
3734 | (stripe_nr_end - stripe_nr_orig), | |
3735 | map->num_stripes); | |
321aecc6 CM |
3736 | else if (mirror_num) |
3737 | stripe_index += mirror_num - 1; | |
dfe25020 CM |
3738 | else { |
3739 | stripe_index = find_live_mirror(map, stripe_index, | |
3740 | map->sub_stripes, stripe_index + | |
3741 | current->pid % map->sub_stripes); | |
a1d3c478 | 3742 | mirror_num = stripe_index + 1; |
dfe25020 | 3743 | } |
8790d502 CM |
3744 | } else { |
3745 | /* | |
3746 | * after this do_div call, stripe_nr is the number of stripes | |
3747 | * on this device we have to walk to find the data, and | |
3748 | * stripe_index is the number of our device in the stripe array | |
3749 | */ | |
3750 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3751 | mirror_num = stripe_index + 1; |
8790d502 | 3752 | } |
593060d7 | 3753 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3754 | |
de11cc12 LZ |
3755 | bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS); |
3756 | if (!bbio) { | |
3757 | ret = -ENOMEM; | |
3758 | goto out; | |
3759 | } | |
3760 | atomic_set(&bbio->error, 0); | |
3761 | ||
fce3bb9a | 3762 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
3763 | int factor = 0; |
3764 | int sub_stripes = 0; | |
3765 | u64 stripes_per_dev = 0; | |
3766 | u32 remaining_stripes = 0; | |
3767 | ||
3768 | if (map->type & | |
3769 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3770 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3771 | sub_stripes = 1; | |
3772 | else | |
3773 | sub_stripes = map->sub_stripes; | |
3774 | ||
3775 | factor = map->num_stripes / sub_stripes; | |
3776 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
3777 | stripe_nr_orig, | |
3778 | factor, | |
3779 | &remaining_stripes); | |
3780 | } | |
3781 | ||
fce3bb9a | 3782 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 3783 | bbio->stripes[i].physical = |
f2d8d74d CM |
3784 | map->stripes[stripe_index].physical + |
3785 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3786 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 3787 | |
ec9ef7a1 LZ |
3788 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
3789 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3790 | bbio->stripes[i].length = stripes_per_dev * | |
3791 | map->stripe_len; | |
3792 | if (i / sub_stripes < remaining_stripes) | |
3793 | bbio->stripes[i].length += | |
3794 | map->stripe_len; | |
3795 | if (i < sub_stripes) | |
a1d3c478 | 3796 | bbio->stripes[i].length -= |
fce3bb9a | 3797 | stripe_offset; |
ec9ef7a1 LZ |
3798 | if ((i / sub_stripes + 1) % |
3799 | sub_stripes == remaining_stripes) | |
a1d3c478 | 3800 | bbio->stripes[i].length -= |
fce3bb9a | 3801 | stripe_end_offset; |
ec9ef7a1 LZ |
3802 | if (i == sub_stripes - 1) |
3803 | stripe_offset = 0; | |
fce3bb9a | 3804 | } else |
a1d3c478 | 3805 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
3806 | |
3807 | stripe_index++; | |
3808 | if (stripe_index == map->num_stripes) { | |
3809 | /* This could only happen for RAID0/10 */ | |
3810 | stripe_index = 0; | |
3811 | stripe_nr++; | |
3812 | } | |
3813 | } | |
3814 | } else { | |
3815 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3816 | bbio->stripes[i].physical = |
212a17ab LT |
3817 | map->stripes[stripe_index].physical + |
3818 | stripe_offset + | |
3819 | stripe_nr * map->stripe_len; | |
a1d3c478 | 3820 | bbio->stripes[i].dev = |
212a17ab | 3821 | map->stripes[stripe_index].dev; |
fce3bb9a | 3822 | stripe_index++; |
f2d8d74d | 3823 | } |
593060d7 | 3824 | } |
de11cc12 LZ |
3825 | |
3826 | if (rw & REQ_WRITE) { | |
3827 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
3828 | BTRFS_BLOCK_GROUP_RAID10 | | |
3829 | BTRFS_BLOCK_GROUP_DUP)) { | |
3830 | max_errors = 1; | |
3831 | } | |
f2d8d74d | 3832 | } |
de11cc12 LZ |
3833 | |
3834 | *bbio_ret = bbio; | |
3835 | bbio->num_stripes = num_stripes; | |
3836 | bbio->max_errors = max_errors; | |
3837 | bbio->mirror_num = mirror_num; | |
cea9e445 | 3838 | out: |
0b86a832 | 3839 | free_extent_map(em); |
de11cc12 | 3840 | return ret; |
0b86a832 CM |
3841 | } |
3842 | ||
f2d8d74d CM |
3843 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3844 | u64 logical, u64 *length, | |
a1d3c478 | 3845 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 3846 | { |
a1d3c478 | 3847 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 3848 | mirror_num); |
f2d8d74d CM |
3849 | } |
3850 | ||
a512bbf8 YZ |
3851 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
3852 | u64 chunk_start, u64 physical, u64 devid, | |
3853 | u64 **logical, int *naddrs, int *stripe_len) | |
3854 | { | |
3855 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3856 | struct extent_map *em; | |
3857 | struct map_lookup *map; | |
3858 | u64 *buf; | |
3859 | u64 bytenr; | |
3860 | u64 length; | |
3861 | u64 stripe_nr; | |
3862 | int i, j, nr = 0; | |
3863 | ||
890871be | 3864 | read_lock(&em_tree->lock); |
a512bbf8 | 3865 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 3866 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
3867 | |
3868 | BUG_ON(!em || em->start != chunk_start); | |
3869 | map = (struct map_lookup *)em->bdev; | |
3870 | ||
3871 | length = em->len; | |
3872 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
3873 | do_div(length, map->num_stripes / map->sub_stripes); | |
3874 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3875 | do_div(length, map->num_stripes); | |
3876 | ||
3877 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
3878 | BUG_ON(!buf); | |
3879 | ||
3880 | for (i = 0; i < map->num_stripes; i++) { | |
3881 | if (devid && map->stripes[i].dev->devid != devid) | |
3882 | continue; | |
3883 | if (map->stripes[i].physical > physical || | |
3884 | map->stripes[i].physical + length <= physical) | |
3885 | continue; | |
3886 | ||
3887 | stripe_nr = physical - map->stripes[i].physical; | |
3888 | do_div(stripe_nr, map->stripe_len); | |
3889 | ||
3890 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
3891 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3892 | do_div(stripe_nr, map->sub_stripes); | |
3893 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3894 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3895 | } | |
3896 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 3897 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
3898 | for (j = 0; j < nr; j++) { |
3899 | if (buf[j] == bytenr) | |
3900 | break; | |
3901 | } | |
934d375b CM |
3902 | if (j == nr) { |
3903 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 3904 | buf[nr++] = bytenr; |
934d375b | 3905 | } |
a512bbf8 YZ |
3906 | } |
3907 | ||
a512bbf8 YZ |
3908 | *logical = buf; |
3909 | *naddrs = nr; | |
3910 | *stripe_len = map->stripe_len; | |
3911 | ||
3912 | free_extent_map(em); | |
3913 | return 0; | |
f2d8d74d CM |
3914 | } |
3915 | ||
a1d3c478 | 3916 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 3917 | { |
a1d3c478 | 3918 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 3919 | int is_orig_bio = 0; |
8790d502 | 3920 | |
8790d502 | 3921 | if (err) |
a1d3c478 | 3922 | atomic_inc(&bbio->error); |
8790d502 | 3923 | |
a1d3c478 | 3924 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
3925 | is_orig_bio = 1; |
3926 | ||
a1d3c478 | 3927 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
3928 | if (!is_orig_bio) { |
3929 | bio_put(bio); | |
a1d3c478 | 3930 | bio = bbio->orig_bio; |
7d2b4daa | 3931 | } |
a1d3c478 JS |
3932 | bio->bi_private = bbio->private; |
3933 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
3934 | bio->bi_bdev = (struct block_device *) |
3935 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
3936 | /* only send an error to the higher layers if it is |
3937 | * beyond the tolerance of the multi-bio | |
3938 | */ | |
a1d3c478 | 3939 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 3940 | err = -EIO; |
5dbc8fca | 3941 | } else { |
1259ab75 CM |
3942 | /* |
3943 | * this bio is actually up to date, we didn't | |
3944 | * go over the max number of errors | |
3945 | */ | |
3946 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 3947 | err = 0; |
1259ab75 | 3948 | } |
a1d3c478 | 3949 | kfree(bbio); |
8790d502 CM |
3950 | |
3951 | bio_endio(bio, err); | |
7d2b4daa | 3952 | } else if (!is_orig_bio) { |
8790d502 CM |
3953 | bio_put(bio); |
3954 | } | |
8790d502 CM |
3955 | } |
3956 | ||
8b712842 CM |
3957 | struct async_sched { |
3958 | struct bio *bio; | |
3959 | int rw; | |
3960 | struct btrfs_fs_info *info; | |
3961 | struct btrfs_work work; | |
3962 | }; | |
3963 | ||
3964 | /* | |
3965 | * see run_scheduled_bios for a description of why bios are collected for | |
3966 | * async submit. | |
3967 | * | |
3968 | * This will add one bio to the pending list for a device and make sure | |
3969 | * the work struct is scheduled. | |
3970 | */ | |
143bede5 | 3971 | static noinline void schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
3972 | struct btrfs_device *device, |
3973 | int rw, struct bio *bio) | |
8b712842 CM |
3974 | { |
3975 | int should_queue = 1; | |
ffbd517d | 3976 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
3977 | |
3978 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 3979 | if (!(rw & REQ_WRITE)) { |
492bb6de | 3980 | bio_get(bio); |
21adbd5c | 3981 | btrfsic_submit_bio(rw, bio); |
492bb6de | 3982 | bio_put(bio); |
143bede5 | 3983 | return; |
8b712842 CM |
3984 | } |
3985 | ||
3986 | /* | |
0986fe9e | 3987 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
3988 | * higher layers. Otherwise, the async bio makes it appear we have |
3989 | * made progress against dirty pages when we've really just put it | |
3990 | * on a queue for later | |
3991 | */ | |
0986fe9e | 3992 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 3993 | WARN_ON(bio->bi_next); |
8b712842 CM |
3994 | bio->bi_next = NULL; |
3995 | bio->bi_rw |= rw; | |
3996 | ||
3997 | spin_lock(&device->io_lock); | |
7b6d91da | 3998 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
3999 | pending_bios = &device->pending_sync_bios; |
4000 | else | |
4001 | pending_bios = &device->pending_bios; | |
8b712842 | 4002 | |
ffbd517d CM |
4003 | if (pending_bios->tail) |
4004 | pending_bios->tail->bi_next = bio; | |
8b712842 | 4005 | |
ffbd517d CM |
4006 | pending_bios->tail = bio; |
4007 | if (!pending_bios->head) | |
4008 | pending_bios->head = bio; | |
8b712842 CM |
4009 | if (device->running_pending) |
4010 | should_queue = 0; | |
4011 | ||
4012 | spin_unlock(&device->io_lock); | |
4013 | ||
4014 | if (should_queue) | |
1cc127b5 CM |
4015 | btrfs_queue_worker(&root->fs_info->submit_workers, |
4016 | &device->work); | |
8b712842 CM |
4017 | } |
4018 | ||
f188591e | 4019 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 4020 | int mirror_num, int async_submit) |
0b86a832 CM |
4021 | { |
4022 | struct btrfs_mapping_tree *map_tree; | |
4023 | struct btrfs_device *dev; | |
8790d502 | 4024 | struct bio *first_bio = bio; |
a62b9401 | 4025 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
4026 | u64 length = 0; |
4027 | u64 map_length; | |
0b86a832 | 4028 | int ret; |
8790d502 CM |
4029 | int dev_nr = 0; |
4030 | int total_devs = 1; | |
a1d3c478 | 4031 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 4032 | |
f2d8d74d | 4033 | length = bio->bi_size; |
0b86a832 CM |
4034 | map_tree = &root->fs_info->mapping_tree; |
4035 | map_length = length; | |
cea9e445 | 4036 | |
a1d3c478 | 4037 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 4038 | mirror_num); |
cea9e445 CM |
4039 | BUG_ON(ret); |
4040 | ||
a1d3c478 | 4041 | total_devs = bbio->num_stripes; |
cea9e445 | 4042 | if (map_length < length) { |
d397712b CM |
4043 | printk(KERN_CRIT "mapping failed logical %llu bio len %llu " |
4044 | "len %llu\n", (unsigned long long)logical, | |
4045 | (unsigned long long)length, | |
4046 | (unsigned long long)map_length); | |
cea9e445 CM |
4047 | BUG(); |
4048 | } | |
a1d3c478 JS |
4049 | |
4050 | bbio->orig_bio = first_bio; | |
4051 | bbio->private = first_bio->bi_private; | |
4052 | bbio->end_io = first_bio->bi_end_io; | |
4053 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 4054 | |
d397712b | 4055 | while (dev_nr < total_devs) { |
a1d3c478 JS |
4056 | if (dev_nr < total_devs - 1) { |
4057 | bio = bio_clone(first_bio, GFP_NOFS); | |
4058 | BUG_ON(!bio); | |
4059 | } else { | |
4060 | bio = first_bio; | |
8790d502 | 4061 | } |
a1d3c478 JS |
4062 | bio->bi_private = bbio; |
4063 | bio->bi_end_io = btrfs_end_bio; | |
4064 | bio->bi_sector = bbio->stripes[dev_nr].physical >> 9; | |
4065 | dev = bbio->stripes[dev_nr].dev; | |
18e503d6 | 4066 | if (dev && dev->bdev && (rw != WRITE || dev->writeable)) { |
a1d3c478 JS |
4067 | pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu " |
4068 | "(%s id %llu), size=%u\n", rw, | |
4069 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
4070 | dev->name, dev->devid, bio->bi_size); | |
dfe25020 | 4071 | bio->bi_bdev = dev->bdev; |
8b712842 CM |
4072 | if (async_submit) |
4073 | schedule_bio(root, dev, rw, bio); | |
4074 | else | |
21adbd5c | 4075 | btrfsic_submit_bio(rw, bio); |
dfe25020 CM |
4076 | } else { |
4077 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
4078 | bio->bi_sector = logical >> 9; | |
dfe25020 | 4079 | bio_endio(bio, -EIO); |
dfe25020 | 4080 | } |
8790d502 CM |
4081 | dev_nr++; |
4082 | } | |
0b86a832 CM |
4083 | return 0; |
4084 | } | |
4085 | ||
a443755f | 4086 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 4087 | u8 *uuid, u8 *fsid) |
0b86a832 | 4088 | { |
2b82032c YZ |
4089 | struct btrfs_device *device; |
4090 | struct btrfs_fs_devices *cur_devices; | |
4091 | ||
4092 | cur_devices = root->fs_info->fs_devices; | |
4093 | while (cur_devices) { | |
4094 | if (!fsid || | |
4095 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4096 | device = __find_device(&cur_devices->devices, | |
4097 | devid, uuid); | |
4098 | if (device) | |
4099 | return device; | |
4100 | } | |
4101 | cur_devices = cur_devices->seed; | |
4102 | } | |
4103 | return NULL; | |
0b86a832 CM |
4104 | } |
4105 | ||
dfe25020 CM |
4106 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
4107 | u64 devid, u8 *dev_uuid) | |
4108 | { | |
4109 | struct btrfs_device *device; | |
4110 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4111 | ||
4112 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 4113 | if (!device) |
4114 | return NULL; | |
dfe25020 CM |
4115 | list_add(&device->dev_list, |
4116 | &fs_devices->devices); | |
dfe25020 CM |
4117 | device->dev_root = root->fs_info->dev_root; |
4118 | device->devid = devid; | |
8b712842 | 4119 | device->work.func = pending_bios_fn; |
e4404d6e | 4120 | device->fs_devices = fs_devices; |
cd02dca5 | 4121 | device->missing = 1; |
dfe25020 | 4122 | fs_devices->num_devices++; |
cd02dca5 | 4123 | fs_devices->missing_devices++; |
dfe25020 | 4124 | spin_lock_init(&device->io_lock); |
d20f7043 | 4125 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
4126 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
4127 | return device; | |
4128 | } | |
4129 | ||
0b86a832 CM |
4130 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
4131 | struct extent_buffer *leaf, | |
4132 | struct btrfs_chunk *chunk) | |
4133 | { | |
4134 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4135 | struct map_lookup *map; | |
4136 | struct extent_map *em; | |
4137 | u64 logical; | |
4138 | u64 length; | |
4139 | u64 devid; | |
a443755f | 4140 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 4141 | int num_stripes; |
0b86a832 | 4142 | int ret; |
593060d7 | 4143 | int i; |
0b86a832 | 4144 | |
e17cade2 CM |
4145 | logical = key->offset; |
4146 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 4147 | |
890871be | 4148 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 4149 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 4150 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
4151 | |
4152 | /* already mapped? */ | |
4153 | if (em && em->start <= logical && em->start + em->len > logical) { | |
4154 | free_extent_map(em); | |
0b86a832 CM |
4155 | return 0; |
4156 | } else if (em) { | |
4157 | free_extent_map(em); | |
4158 | } | |
0b86a832 | 4159 | |
172ddd60 | 4160 | em = alloc_extent_map(); |
0b86a832 CM |
4161 | if (!em) |
4162 | return -ENOMEM; | |
593060d7 CM |
4163 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
4164 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
4165 | if (!map) { |
4166 | free_extent_map(em); | |
4167 | return -ENOMEM; | |
4168 | } | |
4169 | ||
4170 | em->bdev = (struct block_device *)map; | |
4171 | em->start = logical; | |
4172 | em->len = length; | |
4173 | em->block_start = 0; | |
c8b97818 | 4174 | em->block_len = em->len; |
0b86a832 | 4175 | |
593060d7 CM |
4176 | map->num_stripes = num_stripes; |
4177 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
4178 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
4179 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
4180 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
4181 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 4182 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
4183 | for (i = 0; i < num_stripes; i++) { |
4184 | map->stripes[i].physical = | |
4185 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
4186 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
4187 | read_extent_buffer(leaf, uuid, (unsigned long) |
4188 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
4189 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4190 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
4191 | NULL); | |
dfe25020 | 4192 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
4193 | kfree(map); |
4194 | free_extent_map(em); | |
4195 | return -EIO; | |
4196 | } | |
dfe25020 CM |
4197 | if (!map->stripes[i].dev) { |
4198 | map->stripes[i].dev = | |
4199 | add_missing_dev(root, devid, uuid); | |
4200 | if (!map->stripes[i].dev) { | |
4201 | kfree(map); | |
4202 | free_extent_map(em); | |
4203 | return -EIO; | |
4204 | } | |
4205 | } | |
4206 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
4207 | } |
4208 | ||
890871be | 4209 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 4210 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 4211 | write_unlock(&map_tree->map_tree.lock); |
b248a415 | 4212 | BUG_ON(ret); |
0b86a832 CM |
4213 | free_extent_map(em); |
4214 | ||
4215 | return 0; | |
4216 | } | |
4217 | ||
143bede5 | 4218 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
4219 | struct btrfs_dev_item *dev_item, |
4220 | struct btrfs_device *device) | |
4221 | { | |
4222 | unsigned long ptr; | |
0b86a832 CM |
4223 | |
4224 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
4225 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
4226 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
4227 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
4228 | device->type = btrfs_device_type(leaf, dev_item); | |
4229 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
4230 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
4231 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
4232 | |
4233 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 4234 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
4235 | } |
4236 | ||
2b82032c YZ |
4237 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
4238 | { | |
4239 | struct btrfs_fs_devices *fs_devices; | |
4240 | int ret; | |
4241 | ||
b367e47f | 4242 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
4243 | |
4244 | fs_devices = root->fs_info->fs_devices->seed; | |
4245 | while (fs_devices) { | |
4246 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4247 | ret = 0; | |
4248 | goto out; | |
4249 | } | |
4250 | fs_devices = fs_devices->seed; | |
4251 | } | |
4252 | ||
4253 | fs_devices = find_fsid(fsid); | |
4254 | if (!fs_devices) { | |
4255 | ret = -ENOENT; | |
4256 | goto out; | |
4257 | } | |
e4404d6e YZ |
4258 | |
4259 | fs_devices = clone_fs_devices(fs_devices); | |
4260 | if (IS_ERR(fs_devices)) { | |
4261 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
4262 | goto out; |
4263 | } | |
4264 | ||
97288f2c | 4265 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 4266 | root->fs_info->bdev_holder); |
2b82032c YZ |
4267 | if (ret) |
4268 | goto out; | |
4269 | ||
4270 | if (!fs_devices->seeding) { | |
4271 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 4272 | free_fs_devices(fs_devices); |
2b82032c YZ |
4273 | ret = -EINVAL; |
4274 | goto out; | |
4275 | } | |
4276 | ||
4277 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
4278 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 4279 | out: |
2b82032c YZ |
4280 | return ret; |
4281 | } | |
4282 | ||
0d81ba5d | 4283 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
4284 | struct extent_buffer *leaf, |
4285 | struct btrfs_dev_item *dev_item) | |
4286 | { | |
4287 | struct btrfs_device *device; | |
4288 | u64 devid; | |
4289 | int ret; | |
2b82032c | 4290 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
4291 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
4292 | ||
0b86a832 | 4293 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
4294 | read_extent_buffer(leaf, dev_uuid, |
4295 | (unsigned long)btrfs_device_uuid(dev_item), | |
4296 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4297 | read_extent_buffer(leaf, fs_uuid, |
4298 | (unsigned long)btrfs_device_fsid(dev_item), | |
4299 | BTRFS_UUID_SIZE); | |
4300 | ||
4301 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
4302 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 4303 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 4304 | return ret; |
2b82032c YZ |
4305 | } |
4306 | ||
4307 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
4308 | if (!device || !device->bdev) { | |
e4404d6e | 4309 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
4310 | return -EIO; |
4311 | ||
4312 | if (!device) { | |
d397712b CM |
4313 | printk(KERN_WARNING "warning devid %llu missing\n", |
4314 | (unsigned long long)devid); | |
2b82032c YZ |
4315 | device = add_missing_dev(root, devid, dev_uuid); |
4316 | if (!device) | |
4317 | return -ENOMEM; | |
cd02dca5 CM |
4318 | } else if (!device->missing) { |
4319 | /* | |
4320 | * this happens when a device that was properly setup | |
4321 | * in the device info lists suddenly goes bad. | |
4322 | * device->bdev is NULL, and so we have to set | |
4323 | * device->missing to one here | |
4324 | */ | |
4325 | root->fs_info->fs_devices->missing_devices++; | |
4326 | device->missing = 1; | |
2b82032c YZ |
4327 | } |
4328 | } | |
4329 | ||
4330 | if (device->fs_devices != root->fs_info->fs_devices) { | |
4331 | BUG_ON(device->writeable); | |
4332 | if (device->generation != | |
4333 | btrfs_device_generation(leaf, dev_item)) | |
4334 | return -EINVAL; | |
6324fbf3 | 4335 | } |
0b86a832 CM |
4336 | |
4337 | fill_device_from_item(leaf, dev_item, device); | |
4338 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 4339 | device->in_fs_metadata = 1; |
2bf64758 | 4340 | if (device->writeable) { |
2b82032c | 4341 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
4342 | spin_lock(&root->fs_info->free_chunk_lock); |
4343 | root->fs_info->free_chunk_space += device->total_bytes - | |
4344 | device->bytes_used; | |
4345 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4346 | } | |
0b86a832 | 4347 | ret = 0; |
0b86a832 CM |
4348 | return ret; |
4349 | } | |
4350 | ||
e4404d6e | 4351 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 4352 | { |
6c41761f | 4353 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 4354 | struct extent_buffer *sb; |
0b86a832 | 4355 | struct btrfs_disk_key *disk_key; |
0b86a832 | 4356 | struct btrfs_chunk *chunk; |
84eed90f CM |
4357 | u8 *ptr; |
4358 | unsigned long sb_ptr; | |
4359 | int ret = 0; | |
0b86a832 CM |
4360 | u32 num_stripes; |
4361 | u32 array_size; | |
4362 | u32 len = 0; | |
0b86a832 | 4363 | u32 cur; |
84eed90f | 4364 | struct btrfs_key key; |
0b86a832 | 4365 | |
e4404d6e | 4366 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
4367 | BTRFS_SUPER_INFO_SIZE); |
4368 | if (!sb) | |
4369 | return -ENOMEM; | |
4370 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 4371 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
4372 | /* |
4373 | * The sb extent buffer is artifical and just used to read the system array. | |
4374 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
4375 | * pages up-to-date when the page is larger: extent does not cover the | |
4376 | * whole page and consequently check_page_uptodate does not find all | |
4377 | * the page's extents up-to-date (the hole beyond sb), | |
4378 | * write_extent_buffer then triggers a WARN_ON. | |
4379 | * | |
4380 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
4381 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
4382 | * to silence the warning eg. on PowerPC 64. | |
4383 | */ | |
4384 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
4385 | SetPageUptodate(sb->first_page); | |
4008c04a | 4386 | |
a061fc8d | 4387 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
4388 | array_size = btrfs_super_sys_array_size(super_copy); |
4389 | ||
0b86a832 CM |
4390 | ptr = super_copy->sys_chunk_array; |
4391 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
4392 | cur = 0; | |
4393 | ||
4394 | while (cur < array_size) { | |
4395 | disk_key = (struct btrfs_disk_key *)ptr; | |
4396 | btrfs_disk_key_to_cpu(&key, disk_key); | |
4397 | ||
a061fc8d | 4398 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
4399 | sb_ptr += len; |
4400 | cur += len; | |
4401 | ||
0d81ba5d | 4402 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 4403 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 4404 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
4405 | if (ret) |
4406 | break; | |
0b86a832 CM |
4407 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
4408 | len = btrfs_chunk_item_size(num_stripes); | |
4409 | } else { | |
84eed90f CM |
4410 | ret = -EIO; |
4411 | break; | |
0b86a832 CM |
4412 | } |
4413 | ptr += len; | |
4414 | sb_ptr += len; | |
4415 | cur += len; | |
4416 | } | |
a061fc8d | 4417 | free_extent_buffer(sb); |
84eed90f | 4418 | return ret; |
0b86a832 CM |
4419 | } |
4420 | ||
4421 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
4422 | { | |
4423 | struct btrfs_path *path; | |
4424 | struct extent_buffer *leaf; | |
4425 | struct btrfs_key key; | |
4426 | struct btrfs_key found_key; | |
4427 | int ret; | |
4428 | int slot; | |
4429 | ||
4430 | root = root->fs_info->chunk_root; | |
4431 | ||
4432 | path = btrfs_alloc_path(); | |
4433 | if (!path) | |
4434 | return -ENOMEM; | |
4435 | ||
b367e47f LZ |
4436 | mutex_lock(&uuid_mutex); |
4437 | lock_chunks(root); | |
4438 | ||
0b86a832 CM |
4439 | /* first we search for all of the device items, and then we |
4440 | * read in all of the chunk items. This way we can create chunk | |
4441 | * mappings that reference all of the devices that are afound | |
4442 | */ | |
4443 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
4444 | key.offset = 0; | |
4445 | key.type = 0; | |
4446 | again: | |
4447 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
4448 | if (ret < 0) |
4449 | goto error; | |
d397712b | 4450 | while (1) { |
0b86a832 CM |
4451 | leaf = path->nodes[0]; |
4452 | slot = path->slots[0]; | |
4453 | if (slot >= btrfs_header_nritems(leaf)) { | |
4454 | ret = btrfs_next_leaf(root, path); | |
4455 | if (ret == 0) | |
4456 | continue; | |
4457 | if (ret < 0) | |
4458 | goto error; | |
4459 | break; | |
4460 | } | |
4461 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4462 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4463 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
4464 | break; | |
4465 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
4466 | struct btrfs_dev_item *dev_item; | |
4467 | dev_item = btrfs_item_ptr(leaf, slot, | |
4468 | struct btrfs_dev_item); | |
0d81ba5d | 4469 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
4470 | if (ret) |
4471 | goto error; | |
0b86a832 CM |
4472 | } |
4473 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
4474 | struct btrfs_chunk *chunk; | |
4475 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
4476 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
4477 | if (ret) |
4478 | goto error; | |
0b86a832 CM |
4479 | } |
4480 | path->slots[0]++; | |
4481 | } | |
4482 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4483 | key.objectid = 0; | |
b3b4aa74 | 4484 | btrfs_release_path(path); |
0b86a832 CM |
4485 | goto again; |
4486 | } | |
0b86a832 CM |
4487 | ret = 0; |
4488 | error: | |
b367e47f LZ |
4489 | unlock_chunks(root); |
4490 | mutex_unlock(&uuid_mutex); | |
4491 | ||
2b82032c | 4492 | btrfs_free_path(path); |
0b86a832 CM |
4493 | return ret; |
4494 | } |