]>
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> |
442a4f63 | 26 | #include <linux/ratelimit.h> |
59641015 | 27 | #include <linux/kthread.h> |
593060d7 | 28 | #include <asm/div64.h> |
4b4e25f2 | 29 | #include "compat.h" |
0b86a832 CM |
30 | #include "ctree.h" |
31 | #include "extent_map.h" | |
32 | #include "disk-io.h" | |
33 | #include "transaction.h" | |
34 | #include "print-tree.h" | |
35 | #include "volumes.h" | |
8b712842 | 36 | #include "async-thread.h" |
21adbd5c | 37 | #include "check-integrity.h" |
606686ee | 38 | #include "rcu-string.h" |
0b86a832 | 39 | |
2b82032c YZ |
40 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
41 | struct btrfs_root *root, | |
42 | struct btrfs_device *device); | |
43 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb SB |
44 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
45 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); | |
2b82032c | 46 | |
8a4b83cc CM |
47 | static DEFINE_MUTEX(uuid_mutex); |
48 | static LIST_HEAD(fs_uuids); | |
49 | ||
7d9eb12c CM |
50 | static void lock_chunks(struct btrfs_root *root) |
51 | { | |
7d9eb12c CM |
52 | mutex_lock(&root->fs_info->chunk_mutex); |
53 | } | |
54 | ||
55 | static void unlock_chunks(struct btrfs_root *root) | |
56 | { | |
7d9eb12c CM |
57 | mutex_unlock(&root->fs_info->chunk_mutex); |
58 | } | |
59 | ||
e4404d6e YZ |
60 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
61 | { | |
62 | struct btrfs_device *device; | |
63 | WARN_ON(fs_devices->opened); | |
64 | while (!list_empty(&fs_devices->devices)) { | |
65 | device = list_entry(fs_devices->devices.next, | |
66 | struct btrfs_device, dev_list); | |
67 | list_del(&device->dev_list); | |
606686ee | 68 | rcu_string_free(device->name); |
e4404d6e YZ |
69 | kfree(device); |
70 | } | |
71 | kfree(fs_devices); | |
72 | } | |
73 | ||
143bede5 | 74 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
75 | { |
76 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 77 | |
2b82032c YZ |
78 | while (!list_empty(&fs_uuids)) { |
79 | fs_devices = list_entry(fs_uuids.next, | |
80 | struct btrfs_fs_devices, list); | |
81 | list_del(&fs_devices->list); | |
e4404d6e | 82 | free_fs_devices(fs_devices); |
8a4b83cc | 83 | } |
8a4b83cc CM |
84 | } |
85 | ||
a1b32a59 CM |
86 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
87 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
88 | { |
89 | struct btrfs_device *dev; | |
8a4b83cc | 90 | |
c6e30871 | 91 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 92 | if (dev->devid == devid && |
8f18cf13 | 93 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 94 | return dev; |
a443755f | 95 | } |
8a4b83cc CM |
96 | } |
97 | return NULL; | |
98 | } | |
99 | ||
a1b32a59 | 100 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 101 | { |
8a4b83cc CM |
102 | struct btrfs_fs_devices *fs_devices; |
103 | ||
c6e30871 | 104 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
105 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
106 | return fs_devices; | |
107 | } | |
108 | return NULL; | |
109 | } | |
110 | ||
ffbd517d CM |
111 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
112 | struct bio *head, struct bio *tail) | |
113 | { | |
114 | ||
115 | struct bio *old_head; | |
116 | ||
117 | old_head = pending_bios->head; | |
118 | pending_bios->head = head; | |
119 | if (pending_bios->tail) | |
120 | tail->bi_next = old_head; | |
121 | else | |
122 | pending_bios->tail = tail; | |
123 | } | |
124 | ||
8b712842 CM |
125 | /* |
126 | * we try to collect pending bios for a device so we don't get a large | |
127 | * number of procs sending bios down to the same device. This greatly | |
128 | * improves the schedulers ability to collect and merge the bios. | |
129 | * | |
130 | * But, it also turns into a long list of bios to process and that is sure | |
131 | * to eventually make the worker thread block. The solution here is to | |
132 | * make some progress and then put this work struct back at the end of | |
133 | * the list if the block device is congested. This way, multiple devices | |
134 | * can make progress from a single worker thread. | |
135 | */ | |
143bede5 | 136 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
137 | { |
138 | struct bio *pending; | |
139 | struct backing_dev_info *bdi; | |
b64a2851 | 140 | struct btrfs_fs_info *fs_info; |
ffbd517d | 141 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
142 | struct bio *tail; |
143 | struct bio *cur; | |
144 | int again = 0; | |
ffbd517d | 145 | unsigned long num_run; |
d644d8a1 | 146 | unsigned long batch_run = 0; |
b64a2851 | 147 | unsigned long limit; |
b765ead5 | 148 | unsigned long last_waited = 0; |
d84275c9 | 149 | int force_reg = 0; |
0e588859 | 150 | int sync_pending = 0; |
211588ad CM |
151 | struct blk_plug plug; |
152 | ||
153 | /* | |
154 | * this function runs all the bios we've collected for | |
155 | * a particular device. We don't want to wander off to | |
156 | * another device without first sending all of these down. | |
157 | * So, setup a plug here and finish it off before we return | |
158 | */ | |
159 | blk_start_plug(&plug); | |
8b712842 | 160 | |
bedf762b | 161 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
162 | fs_info = device->dev_root->fs_info; |
163 | limit = btrfs_async_submit_limit(fs_info); | |
164 | limit = limit * 2 / 3; | |
165 | ||
8b712842 CM |
166 | loop: |
167 | spin_lock(&device->io_lock); | |
168 | ||
a6837051 | 169 | loop_lock: |
d84275c9 | 170 | num_run = 0; |
ffbd517d | 171 | |
8b712842 CM |
172 | /* take all the bios off the list at once and process them |
173 | * later on (without the lock held). But, remember the | |
174 | * tail and other pointers so the bios can be properly reinserted | |
175 | * into the list if we hit congestion | |
176 | */ | |
d84275c9 | 177 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 178 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
179 | force_reg = 1; |
180 | } else { | |
ffbd517d | 181 | pending_bios = &device->pending_bios; |
d84275c9 CM |
182 | force_reg = 0; |
183 | } | |
ffbd517d CM |
184 | |
185 | pending = pending_bios->head; | |
186 | tail = pending_bios->tail; | |
8b712842 | 187 | WARN_ON(pending && !tail); |
8b712842 CM |
188 | |
189 | /* | |
190 | * if pending was null this time around, no bios need processing | |
191 | * at all and we can stop. Otherwise it'll loop back up again | |
192 | * and do an additional check so no bios are missed. | |
193 | * | |
194 | * device->running_pending is used to synchronize with the | |
195 | * schedule_bio code. | |
196 | */ | |
ffbd517d CM |
197 | if (device->pending_sync_bios.head == NULL && |
198 | device->pending_bios.head == NULL) { | |
8b712842 CM |
199 | again = 0; |
200 | device->running_pending = 0; | |
ffbd517d CM |
201 | } else { |
202 | again = 1; | |
203 | device->running_pending = 1; | |
8b712842 | 204 | } |
ffbd517d CM |
205 | |
206 | pending_bios->head = NULL; | |
207 | pending_bios->tail = NULL; | |
208 | ||
8b712842 CM |
209 | spin_unlock(&device->io_lock); |
210 | ||
d397712b | 211 | while (pending) { |
ffbd517d CM |
212 | |
213 | rmb(); | |
d84275c9 CM |
214 | /* we want to work on both lists, but do more bios on the |
215 | * sync list than the regular list | |
216 | */ | |
217 | if ((num_run > 32 && | |
218 | pending_bios != &device->pending_sync_bios && | |
219 | device->pending_sync_bios.head) || | |
220 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
221 | device->pending_bios.head)) { | |
ffbd517d CM |
222 | spin_lock(&device->io_lock); |
223 | requeue_list(pending_bios, pending, tail); | |
224 | goto loop_lock; | |
225 | } | |
226 | ||
8b712842 CM |
227 | cur = pending; |
228 | pending = pending->bi_next; | |
229 | cur->bi_next = NULL; | |
b64a2851 CM |
230 | atomic_dec(&fs_info->nr_async_bios); |
231 | ||
232 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
233 | waitqueue_active(&fs_info->async_submit_wait)) | |
234 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
235 | |
236 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 237 | |
2ab1ba68 CM |
238 | /* |
239 | * if we're doing the sync list, record that our | |
240 | * plug has some sync requests on it | |
241 | * | |
242 | * If we're doing the regular list and there are | |
243 | * sync requests sitting around, unplug before | |
244 | * we add more | |
245 | */ | |
246 | if (pending_bios == &device->pending_sync_bios) { | |
247 | sync_pending = 1; | |
248 | } else if (sync_pending) { | |
249 | blk_finish_plug(&plug); | |
250 | blk_start_plug(&plug); | |
251 | sync_pending = 0; | |
252 | } | |
253 | ||
21adbd5c | 254 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
255 | num_run++; |
256 | batch_run++; | |
7eaceacc | 257 | if (need_resched()) |
ffbd517d | 258 | cond_resched(); |
8b712842 CM |
259 | |
260 | /* | |
261 | * we made progress, there is more work to do and the bdi | |
262 | * is now congested. Back off and let other work structs | |
263 | * run instead | |
264 | */ | |
57fd5a5f | 265 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 266 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 267 | struct io_context *ioc; |
8b712842 | 268 | |
b765ead5 CM |
269 | ioc = current->io_context; |
270 | ||
271 | /* | |
272 | * the main goal here is that we don't want to | |
273 | * block if we're going to be able to submit | |
274 | * more requests without blocking. | |
275 | * | |
276 | * This code does two great things, it pokes into | |
277 | * the elevator code from a filesystem _and_ | |
278 | * it makes assumptions about how batching works. | |
279 | */ | |
280 | if (ioc && ioc->nr_batch_requests > 0 && | |
281 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
282 | (last_waited == 0 || | |
283 | ioc->last_waited == last_waited)) { | |
284 | /* | |
285 | * we want to go through our batch of | |
286 | * requests and stop. So, we copy out | |
287 | * the ioc->last_waited time and test | |
288 | * against it before looping | |
289 | */ | |
290 | last_waited = ioc->last_waited; | |
7eaceacc | 291 | if (need_resched()) |
ffbd517d | 292 | cond_resched(); |
b765ead5 CM |
293 | continue; |
294 | } | |
8b712842 | 295 | spin_lock(&device->io_lock); |
ffbd517d | 296 | requeue_list(pending_bios, pending, tail); |
a6837051 | 297 | device->running_pending = 1; |
8b712842 CM |
298 | |
299 | spin_unlock(&device->io_lock); | |
300 | btrfs_requeue_work(&device->work); | |
301 | goto done; | |
302 | } | |
d85c8a6f CM |
303 | /* unplug every 64 requests just for good measure */ |
304 | if (batch_run % 64 == 0) { | |
305 | blk_finish_plug(&plug); | |
306 | blk_start_plug(&plug); | |
307 | sync_pending = 0; | |
308 | } | |
8b712842 | 309 | } |
ffbd517d | 310 | |
51684082 CM |
311 | cond_resched(); |
312 | if (again) | |
313 | goto loop; | |
314 | ||
315 | spin_lock(&device->io_lock); | |
316 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
317 | goto loop_lock; | |
318 | spin_unlock(&device->io_lock); | |
319 | ||
8b712842 | 320 | done: |
211588ad | 321 | blk_finish_plug(&plug); |
8b712842 CM |
322 | } |
323 | ||
b2950863 | 324 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
325 | { |
326 | struct btrfs_device *device; | |
327 | ||
328 | device = container_of(work, struct btrfs_device, work); | |
329 | run_scheduled_bios(device); | |
330 | } | |
331 | ||
a1b32a59 | 332 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
333 | struct btrfs_super_block *disk_super, |
334 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
335 | { | |
336 | struct btrfs_device *device; | |
337 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 338 | struct rcu_string *name; |
8a4b83cc CM |
339 | u64 found_transid = btrfs_super_generation(disk_super); |
340 | ||
341 | fs_devices = find_fsid(disk_super->fsid); | |
342 | if (!fs_devices) { | |
515dc322 | 343 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
344 | if (!fs_devices) |
345 | return -ENOMEM; | |
346 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 347 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
348 | list_add(&fs_devices->list, &fs_uuids); |
349 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
350 | fs_devices->latest_devid = devid; | |
351 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 352 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
353 | device = NULL; |
354 | } else { | |
a443755f CM |
355 | device = __find_device(&fs_devices->devices, devid, |
356 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
357 | } |
358 | if (!device) { | |
2b82032c YZ |
359 | if (fs_devices->opened) |
360 | return -EBUSY; | |
361 | ||
8a4b83cc CM |
362 | device = kzalloc(sizeof(*device), GFP_NOFS); |
363 | if (!device) { | |
364 | /* we can safely leave the fs_devices entry around */ | |
365 | return -ENOMEM; | |
366 | } | |
367 | device->devid = devid; | |
733f4fbb | 368 | device->dev_stats_valid = 0; |
8b712842 | 369 | device->work.func = pending_bios_fn; |
a443755f CM |
370 | memcpy(device->uuid, disk_super->dev_item.uuid, |
371 | BTRFS_UUID_SIZE); | |
b248a415 | 372 | spin_lock_init(&device->io_lock); |
606686ee JB |
373 | |
374 | name = rcu_string_strdup(path, GFP_NOFS); | |
375 | if (!name) { | |
8a4b83cc CM |
376 | kfree(device); |
377 | return -ENOMEM; | |
378 | } | |
606686ee | 379 | rcu_assign_pointer(device->name, name); |
2b82032c | 380 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 381 | |
90519d66 AJ |
382 | /* init readahead state */ |
383 | spin_lock_init(&device->reada_lock); | |
384 | device->reada_curr_zone = NULL; | |
385 | atomic_set(&device->reada_in_flight, 0); | |
386 | device->reada_next = 0; | |
387 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
388 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
389 | ||
e5e9a520 | 390 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 391 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
392 | mutex_unlock(&fs_devices->device_list_mutex); |
393 | ||
2b82032c | 394 | device->fs_devices = fs_devices; |
8a4b83cc | 395 | fs_devices->num_devices++; |
606686ee JB |
396 | } else if (!device->name || strcmp(device->name->str, path)) { |
397 | name = rcu_string_strdup(path, GFP_NOFS); | |
3a0524dc TH |
398 | if (!name) |
399 | return -ENOMEM; | |
606686ee JB |
400 | rcu_string_free(device->name); |
401 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
402 | if (device->missing) { |
403 | fs_devices->missing_devices--; | |
404 | device->missing = 0; | |
405 | } | |
8a4b83cc CM |
406 | } |
407 | ||
408 | if (found_transid > fs_devices->latest_trans) { | |
409 | fs_devices->latest_devid = devid; | |
410 | fs_devices->latest_trans = found_transid; | |
411 | } | |
8a4b83cc CM |
412 | *fs_devices_ret = fs_devices; |
413 | return 0; | |
414 | } | |
415 | ||
e4404d6e YZ |
416 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
417 | { | |
418 | struct btrfs_fs_devices *fs_devices; | |
419 | struct btrfs_device *device; | |
420 | struct btrfs_device *orig_dev; | |
421 | ||
422 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
423 | if (!fs_devices) | |
424 | return ERR_PTR(-ENOMEM); | |
425 | ||
426 | INIT_LIST_HEAD(&fs_devices->devices); | |
427 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
428 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 429 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
430 | fs_devices->latest_devid = orig->latest_devid; |
431 | fs_devices->latest_trans = orig->latest_trans; | |
432 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); | |
433 | ||
46224705 | 434 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 435 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
436 | struct rcu_string *name; |
437 | ||
e4404d6e YZ |
438 | device = kzalloc(sizeof(*device), GFP_NOFS); |
439 | if (!device) | |
440 | goto error; | |
441 | ||
606686ee JB |
442 | /* |
443 | * This is ok to do without rcu read locked because we hold the | |
444 | * uuid mutex so nothing we touch in here is going to disappear. | |
445 | */ | |
446 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
447 | if (!name) { | |
fd2696f3 | 448 | kfree(device); |
e4404d6e | 449 | goto error; |
fd2696f3 | 450 | } |
606686ee | 451 | rcu_assign_pointer(device->name, name); |
e4404d6e YZ |
452 | |
453 | device->devid = orig_dev->devid; | |
454 | device->work.func = pending_bios_fn; | |
455 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
456 | spin_lock_init(&device->io_lock); |
457 | INIT_LIST_HEAD(&device->dev_list); | |
458 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
459 | ||
460 | list_add(&device->dev_list, &fs_devices->devices); | |
461 | device->fs_devices = fs_devices; | |
462 | fs_devices->num_devices++; | |
463 | } | |
464 | return fs_devices; | |
465 | error: | |
466 | free_fs_devices(fs_devices); | |
467 | return ERR_PTR(-ENOMEM); | |
468 | } | |
469 | ||
143bede5 | 470 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
dfe25020 | 471 | { |
c6e30871 | 472 | struct btrfs_device *device, *next; |
dfe25020 | 473 | |
a6b0d5c8 CM |
474 | struct block_device *latest_bdev = NULL; |
475 | u64 latest_devid = 0; | |
476 | u64 latest_transid = 0; | |
477 | ||
dfe25020 CM |
478 | mutex_lock(&uuid_mutex); |
479 | again: | |
46224705 | 480 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 481 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 CM |
482 | if (device->in_fs_metadata) { |
483 | if (!latest_transid || | |
484 | device->generation > latest_transid) { | |
485 | latest_devid = device->devid; | |
486 | latest_transid = device->generation; | |
487 | latest_bdev = device->bdev; | |
488 | } | |
2b82032c | 489 | continue; |
a6b0d5c8 | 490 | } |
2b82032c YZ |
491 | |
492 | if (device->bdev) { | |
d4d77629 | 493 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
494 | device->bdev = NULL; |
495 | fs_devices->open_devices--; | |
496 | } | |
497 | if (device->writeable) { | |
498 | list_del_init(&device->dev_alloc_list); | |
499 | device->writeable = 0; | |
500 | fs_devices->rw_devices--; | |
501 | } | |
e4404d6e YZ |
502 | list_del_init(&device->dev_list); |
503 | fs_devices->num_devices--; | |
606686ee | 504 | rcu_string_free(device->name); |
e4404d6e | 505 | kfree(device); |
dfe25020 | 506 | } |
2b82032c YZ |
507 | |
508 | if (fs_devices->seed) { | |
509 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
510 | goto again; |
511 | } | |
512 | ||
a6b0d5c8 CM |
513 | fs_devices->latest_bdev = latest_bdev; |
514 | fs_devices->latest_devid = latest_devid; | |
515 | fs_devices->latest_trans = latest_transid; | |
516 | ||
dfe25020 | 517 | mutex_unlock(&uuid_mutex); |
dfe25020 | 518 | } |
a0af469b | 519 | |
1f78160c XG |
520 | static void __free_device(struct work_struct *work) |
521 | { | |
522 | struct btrfs_device *device; | |
523 | ||
524 | device = container_of(work, struct btrfs_device, rcu_work); | |
525 | ||
526 | if (device->bdev) | |
527 | blkdev_put(device->bdev, device->mode); | |
528 | ||
606686ee | 529 | rcu_string_free(device->name); |
1f78160c XG |
530 | kfree(device); |
531 | } | |
532 | ||
533 | static void free_device(struct rcu_head *head) | |
534 | { | |
535 | struct btrfs_device *device; | |
536 | ||
537 | device = container_of(head, struct btrfs_device, rcu); | |
538 | ||
539 | INIT_WORK(&device->rcu_work, __free_device); | |
540 | schedule_work(&device->rcu_work); | |
541 | } | |
542 | ||
2b82032c | 543 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 544 | { |
8a4b83cc | 545 | struct btrfs_device *device; |
e4404d6e | 546 | |
2b82032c YZ |
547 | if (--fs_devices->opened > 0) |
548 | return 0; | |
8a4b83cc | 549 | |
c9513edb | 550 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 551 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 552 | struct btrfs_device *new_device; |
606686ee | 553 | struct rcu_string *name; |
1f78160c XG |
554 | |
555 | if (device->bdev) | |
a0af469b | 556 | fs_devices->open_devices--; |
1f78160c | 557 | |
2b82032c YZ |
558 | if (device->writeable) { |
559 | list_del_init(&device->dev_alloc_list); | |
560 | fs_devices->rw_devices--; | |
561 | } | |
562 | ||
d5e2003c JB |
563 | if (device->can_discard) |
564 | fs_devices->num_can_discard--; | |
565 | ||
1f78160c | 566 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
79787eaa | 567 | BUG_ON(!new_device); /* -ENOMEM */ |
1f78160c | 568 | memcpy(new_device, device, sizeof(*new_device)); |
606686ee JB |
569 | |
570 | /* Safe because we are under uuid_mutex */ | |
571 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
572 | BUG_ON(device->name && !name); /* -ENOMEM */ | |
573 | rcu_assign_pointer(new_device->name, name); | |
1f78160c XG |
574 | new_device->bdev = NULL; |
575 | new_device->writeable = 0; | |
576 | new_device->in_fs_metadata = 0; | |
d5e2003c | 577 | new_device->can_discard = 0; |
1f78160c XG |
578 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
579 | ||
580 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 581 | } |
c9513edb XG |
582 | mutex_unlock(&fs_devices->device_list_mutex); |
583 | ||
e4404d6e YZ |
584 | WARN_ON(fs_devices->open_devices); |
585 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
586 | fs_devices->opened = 0; |
587 | fs_devices->seeding = 0; | |
2b82032c | 588 | |
8a4b83cc CM |
589 | return 0; |
590 | } | |
591 | ||
2b82032c YZ |
592 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
593 | { | |
e4404d6e | 594 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
595 | int ret; |
596 | ||
597 | mutex_lock(&uuid_mutex); | |
598 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
599 | if (!fs_devices->opened) { |
600 | seed_devices = fs_devices->seed; | |
601 | fs_devices->seed = NULL; | |
602 | } | |
2b82032c | 603 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
604 | |
605 | while (seed_devices) { | |
606 | fs_devices = seed_devices; | |
607 | seed_devices = fs_devices->seed; | |
608 | __btrfs_close_devices(fs_devices); | |
609 | free_fs_devices(fs_devices); | |
610 | } | |
2b82032c YZ |
611 | return ret; |
612 | } | |
613 | ||
e4404d6e YZ |
614 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
615 | fmode_t flags, void *holder) | |
8a4b83cc | 616 | { |
d5e2003c | 617 | struct request_queue *q; |
8a4b83cc CM |
618 | struct block_device *bdev; |
619 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 620 | struct btrfs_device *device; |
a0af469b CM |
621 | struct block_device *latest_bdev = NULL; |
622 | struct buffer_head *bh; | |
623 | struct btrfs_super_block *disk_super; | |
624 | u64 latest_devid = 0; | |
625 | u64 latest_transid = 0; | |
a0af469b | 626 | u64 devid; |
2b82032c | 627 | int seeding = 1; |
a0af469b | 628 | int ret = 0; |
8a4b83cc | 629 | |
d4d77629 TH |
630 | flags |= FMODE_EXCL; |
631 | ||
c6e30871 | 632 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
633 | if (device->bdev) |
634 | continue; | |
dfe25020 CM |
635 | if (!device->name) |
636 | continue; | |
637 | ||
606686ee | 638 | bdev = blkdev_get_by_path(device->name->str, flags, holder); |
8a4b83cc | 639 | if (IS_ERR(bdev)) { |
606686ee | 640 | printk(KERN_INFO "open %s failed\n", device->name->str); |
a0af469b | 641 | goto error; |
8a4b83cc | 642 | } |
3c4bb26b CM |
643 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
644 | invalidate_bdev(bdev); | |
a061fc8d | 645 | set_blocksize(bdev, 4096); |
a0af469b | 646 | |
a512bbf8 | 647 | bh = btrfs_read_dev_super(bdev); |
20bcd649 | 648 | if (!bh) |
a0af469b CM |
649 | goto error_close; |
650 | ||
651 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 652 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
653 | if (devid != device->devid) |
654 | goto error_brelse; | |
655 | ||
2b82032c YZ |
656 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
657 | BTRFS_UUID_SIZE)) | |
658 | goto error_brelse; | |
659 | ||
660 | device->generation = btrfs_super_generation(disk_super); | |
661 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 662 | latest_devid = devid; |
2b82032c | 663 | latest_transid = device->generation; |
a0af469b CM |
664 | latest_bdev = bdev; |
665 | } | |
666 | ||
2b82032c YZ |
667 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
668 | device->writeable = 0; | |
669 | } else { | |
670 | device->writeable = !bdev_read_only(bdev); | |
671 | seeding = 0; | |
672 | } | |
673 | ||
d5e2003c JB |
674 | q = bdev_get_queue(bdev); |
675 | if (blk_queue_discard(q)) { | |
676 | device->can_discard = 1; | |
677 | fs_devices->num_can_discard++; | |
678 | } | |
679 | ||
8a4b83cc | 680 | device->bdev = bdev; |
dfe25020 | 681 | device->in_fs_metadata = 0; |
15916de8 CM |
682 | device->mode = flags; |
683 | ||
c289811c CM |
684 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
685 | fs_devices->rotating = 1; | |
686 | ||
a0af469b | 687 | fs_devices->open_devices++; |
2b82032c YZ |
688 | if (device->writeable) { |
689 | fs_devices->rw_devices++; | |
690 | list_add(&device->dev_alloc_list, | |
691 | &fs_devices->alloc_list); | |
692 | } | |
4f6c9328 | 693 | brelse(bh); |
a0af469b | 694 | continue; |
a061fc8d | 695 | |
a0af469b CM |
696 | error_brelse: |
697 | brelse(bh); | |
698 | error_close: | |
d4d77629 | 699 | blkdev_put(bdev, flags); |
a0af469b CM |
700 | error: |
701 | continue; | |
8a4b83cc | 702 | } |
a0af469b | 703 | if (fs_devices->open_devices == 0) { |
20bcd649 | 704 | ret = -EINVAL; |
a0af469b CM |
705 | goto out; |
706 | } | |
2b82032c YZ |
707 | fs_devices->seeding = seeding; |
708 | fs_devices->opened = 1; | |
a0af469b CM |
709 | fs_devices->latest_bdev = latest_bdev; |
710 | fs_devices->latest_devid = latest_devid; | |
711 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 712 | fs_devices->total_rw_bytes = 0; |
a0af469b | 713 | out: |
2b82032c YZ |
714 | return ret; |
715 | } | |
716 | ||
717 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 718 | fmode_t flags, void *holder) |
2b82032c YZ |
719 | { |
720 | int ret; | |
721 | ||
722 | mutex_lock(&uuid_mutex); | |
723 | if (fs_devices->opened) { | |
e4404d6e YZ |
724 | fs_devices->opened++; |
725 | ret = 0; | |
2b82032c | 726 | } else { |
15916de8 | 727 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 728 | } |
8a4b83cc | 729 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
730 | return ret; |
731 | } | |
732 | ||
97288f2c | 733 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
734 | struct btrfs_fs_devices **fs_devices_ret) |
735 | { | |
736 | struct btrfs_super_block *disk_super; | |
737 | struct block_device *bdev; | |
738 | struct buffer_head *bh; | |
739 | int ret; | |
740 | u64 devid; | |
f2984462 | 741 | u64 transid; |
8a4b83cc | 742 | |
d4d77629 TH |
743 | flags |= FMODE_EXCL; |
744 | bdev = blkdev_get_by_path(path, flags, holder); | |
8a4b83cc CM |
745 | |
746 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
747 | ret = PTR_ERR(bdev); |
748 | goto error; | |
749 | } | |
750 | ||
10f6327b | 751 | mutex_lock(&uuid_mutex); |
8a4b83cc CM |
752 | ret = set_blocksize(bdev, 4096); |
753 | if (ret) | |
754 | goto error_close; | |
a512bbf8 | 755 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 756 | if (!bh) { |
20b45077 | 757 | ret = -EINVAL; |
8a4b83cc CM |
758 | goto error_close; |
759 | } | |
760 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 761 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 762 | transid = btrfs_super_generation(disk_super); |
7ae9c09d | 763 | if (disk_super->label[0]) |
d397712b | 764 | printk(KERN_INFO "device label %s ", disk_super->label); |
22b63a29 ID |
765 | else |
766 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); | |
119e10cf | 767 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 768 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc CM |
769 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
770 | ||
8a4b83cc CM |
771 | brelse(bh); |
772 | error_close: | |
10f6327b | 773 | mutex_unlock(&uuid_mutex); |
d4d77629 | 774 | blkdev_put(bdev, flags); |
8a4b83cc | 775 | error: |
8a4b83cc CM |
776 | return ret; |
777 | } | |
0b86a832 | 778 | |
6d07bcec MX |
779 | /* helper to account the used device space in the range */ |
780 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
781 | u64 end, u64 *length) | |
782 | { | |
783 | struct btrfs_key key; | |
784 | struct btrfs_root *root = device->dev_root; | |
785 | struct btrfs_dev_extent *dev_extent; | |
786 | struct btrfs_path *path; | |
787 | u64 extent_end; | |
788 | int ret; | |
789 | int slot; | |
790 | struct extent_buffer *l; | |
791 | ||
792 | *length = 0; | |
793 | ||
794 | if (start >= device->total_bytes) | |
795 | return 0; | |
796 | ||
797 | path = btrfs_alloc_path(); | |
798 | if (!path) | |
799 | return -ENOMEM; | |
800 | path->reada = 2; | |
801 | ||
802 | key.objectid = device->devid; | |
803 | key.offset = start; | |
804 | key.type = BTRFS_DEV_EXTENT_KEY; | |
805 | ||
806 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
807 | if (ret < 0) | |
808 | goto out; | |
809 | if (ret > 0) { | |
810 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
811 | if (ret < 0) | |
812 | goto out; | |
813 | } | |
814 | ||
815 | while (1) { | |
816 | l = path->nodes[0]; | |
817 | slot = path->slots[0]; | |
818 | if (slot >= btrfs_header_nritems(l)) { | |
819 | ret = btrfs_next_leaf(root, path); | |
820 | if (ret == 0) | |
821 | continue; | |
822 | if (ret < 0) | |
823 | goto out; | |
824 | ||
825 | break; | |
826 | } | |
827 | btrfs_item_key_to_cpu(l, &key, slot); | |
828 | ||
829 | if (key.objectid < device->devid) | |
830 | goto next; | |
831 | ||
832 | if (key.objectid > device->devid) | |
833 | break; | |
834 | ||
835 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
836 | goto next; | |
837 | ||
838 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
839 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
840 | dev_extent); | |
841 | if (key.offset <= start && extent_end > end) { | |
842 | *length = end - start + 1; | |
843 | break; | |
844 | } else if (key.offset <= start && extent_end > start) | |
845 | *length += extent_end - start; | |
846 | else if (key.offset > start && extent_end <= end) | |
847 | *length += extent_end - key.offset; | |
848 | else if (key.offset > start && key.offset <= end) { | |
849 | *length += end - key.offset + 1; | |
850 | break; | |
851 | } else if (key.offset > end) | |
852 | break; | |
853 | ||
854 | next: | |
855 | path->slots[0]++; | |
856 | } | |
857 | ret = 0; | |
858 | out: | |
859 | btrfs_free_path(path); | |
860 | return ret; | |
861 | } | |
862 | ||
0b86a832 | 863 | /* |
7bfc837d | 864 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
865 | * @device: the device which we search the free space in |
866 | * @num_bytes: the size of the free space that we need | |
867 | * @start: store the start of the free space. | |
868 | * @len: the size of the free space. that we find, or the size of the max | |
869 | * free space if we don't find suitable free space | |
870 | * | |
0b86a832 CM |
871 | * this uses a pretty simple search, the expectation is that it is |
872 | * called very infrequently and that a given device has a small number | |
873 | * of extents | |
7bfc837d MX |
874 | * |
875 | * @start is used to store the start of the free space if we find. But if we | |
876 | * don't find suitable free space, it will be used to store the start position | |
877 | * of the max free space. | |
878 | * | |
879 | * @len is used to store the size of the free space that we find. | |
880 | * But if we don't find suitable free space, it is used to store the size of | |
881 | * the max free space. | |
0b86a832 | 882 | */ |
125ccb0a | 883 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 884 | u64 *start, u64 *len) |
0b86a832 CM |
885 | { |
886 | struct btrfs_key key; | |
887 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 888 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 889 | struct btrfs_path *path; |
7bfc837d MX |
890 | u64 hole_size; |
891 | u64 max_hole_start; | |
892 | u64 max_hole_size; | |
893 | u64 extent_end; | |
894 | u64 search_start; | |
0b86a832 CM |
895 | u64 search_end = device->total_bytes; |
896 | int ret; | |
7bfc837d | 897 | int slot; |
0b86a832 CM |
898 | struct extent_buffer *l; |
899 | ||
0b86a832 CM |
900 | /* FIXME use last free of some kind */ |
901 | ||
8a4b83cc CM |
902 | /* we don't want to overwrite the superblock on the drive, |
903 | * so we make sure to start at an offset of at least 1MB | |
904 | */ | |
a9c9bf68 | 905 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 906 | |
7bfc837d MX |
907 | max_hole_start = search_start; |
908 | max_hole_size = 0; | |
38c01b96 | 909 | hole_size = 0; |
7bfc837d MX |
910 | |
911 | if (search_start >= search_end) { | |
912 | ret = -ENOSPC; | |
913 | goto error; | |
914 | } | |
915 | ||
916 | path = btrfs_alloc_path(); | |
917 | if (!path) { | |
918 | ret = -ENOMEM; | |
919 | goto error; | |
920 | } | |
921 | path->reada = 2; | |
922 | ||
0b86a832 CM |
923 | key.objectid = device->devid; |
924 | key.offset = search_start; | |
925 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 926 | |
125ccb0a | 927 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 928 | if (ret < 0) |
7bfc837d | 929 | goto out; |
1fcbac58 YZ |
930 | if (ret > 0) { |
931 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
932 | if (ret < 0) | |
7bfc837d | 933 | goto out; |
1fcbac58 | 934 | } |
7bfc837d | 935 | |
0b86a832 CM |
936 | while (1) { |
937 | l = path->nodes[0]; | |
938 | slot = path->slots[0]; | |
939 | if (slot >= btrfs_header_nritems(l)) { | |
940 | ret = btrfs_next_leaf(root, path); | |
941 | if (ret == 0) | |
942 | continue; | |
943 | if (ret < 0) | |
7bfc837d MX |
944 | goto out; |
945 | ||
946 | break; | |
0b86a832 CM |
947 | } |
948 | btrfs_item_key_to_cpu(l, &key, slot); | |
949 | ||
950 | if (key.objectid < device->devid) | |
951 | goto next; | |
952 | ||
953 | if (key.objectid > device->devid) | |
7bfc837d | 954 | break; |
0b86a832 | 955 | |
7bfc837d MX |
956 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
957 | goto next; | |
9779b72f | 958 | |
7bfc837d MX |
959 | if (key.offset > search_start) { |
960 | hole_size = key.offset - search_start; | |
9779b72f | 961 | |
7bfc837d MX |
962 | if (hole_size > max_hole_size) { |
963 | max_hole_start = search_start; | |
964 | max_hole_size = hole_size; | |
965 | } | |
9779b72f | 966 | |
7bfc837d MX |
967 | /* |
968 | * If this free space is greater than which we need, | |
969 | * it must be the max free space that we have found | |
970 | * until now, so max_hole_start must point to the start | |
971 | * of this free space and the length of this free space | |
972 | * is stored in max_hole_size. Thus, we return | |
973 | * max_hole_start and max_hole_size and go back to the | |
974 | * caller. | |
975 | */ | |
976 | if (hole_size >= num_bytes) { | |
977 | ret = 0; | |
978 | goto out; | |
0b86a832 CM |
979 | } |
980 | } | |
0b86a832 | 981 | |
0b86a832 | 982 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
983 | extent_end = key.offset + btrfs_dev_extent_length(l, |
984 | dev_extent); | |
985 | if (extent_end > search_start) | |
986 | search_start = extent_end; | |
0b86a832 CM |
987 | next: |
988 | path->slots[0]++; | |
989 | cond_resched(); | |
990 | } | |
0b86a832 | 991 | |
38c01b96 | 992 | /* |
993 | * At this point, search_start should be the end of | |
994 | * allocated dev extents, and when shrinking the device, | |
995 | * search_end may be smaller than search_start. | |
996 | */ | |
997 | if (search_end > search_start) | |
998 | hole_size = search_end - search_start; | |
999 | ||
7bfc837d MX |
1000 | if (hole_size > max_hole_size) { |
1001 | max_hole_start = search_start; | |
1002 | max_hole_size = hole_size; | |
0b86a832 | 1003 | } |
0b86a832 | 1004 | |
7bfc837d MX |
1005 | /* See above. */ |
1006 | if (hole_size < num_bytes) | |
1007 | ret = -ENOSPC; | |
1008 | else | |
1009 | ret = 0; | |
1010 | ||
1011 | out: | |
2b82032c | 1012 | btrfs_free_path(path); |
7bfc837d MX |
1013 | error: |
1014 | *start = max_hole_start; | |
b2117a39 | 1015 | if (len) |
7bfc837d | 1016 | *len = max_hole_size; |
0b86a832 CM |
1017 | return ret; |
1018 | } | |
1019 | ||
b2950863 | 1020 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1021 | struct btrfs_device *device, |
1022 | u64 start) | |
1023 | { | |
1024 | int ret; | |
1025 | struct btrfs_path *path; | |
1026 | struct btrfs_root *root = device->dev_root; | |
1027 | struct btrfs_key key; | |
a061fc8d CM |
1028 | struct btrfs_key found_key; |
1029 | struct extent_buffer *leaf = NULL; | |
1030 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1031 | |
1032 | path = btrfs_alloc_path(); | |
1033 | if (!path) | |
1034 | return -ENOMEM; | |
1035 | ||
1036 | key.objectid = device->devid; | |
1037 | key.offset = start; | |
1038 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1039 | again: |
8f18cf13 | 1040 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1041 | if (ret > 0) { |
1042 | ret = btrfs_previous_item(root, path, key.objectid, | |
1043 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1044 | if (ret) |
1045 | goto out; | |
a061fc8d CM |
1046 | leaf = path->nodes[0]; |
1047 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1048 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1049 | struct btrfs_dev_extent); | |
1050 | BUG_ON(found_key.offset > start || found_key.offset + | |
1051 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1052 | key = found_key; |
1053 | btrfs_release_path(path); | |
1054 | goto again; | |
a061fc8d CM |
1055 | } else if (ret == 0) { |
1056 | leaf = path->nodes[0]; | |
1057 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1058 | struct btrfs_dev_extent); | |
79787eaa JM |
1059 | } else { |
1060 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1061 | goto out; | |
a061fc8d | 1062 | } |
8f18cf13 | 1063 | |
2bf64758 JB |
1064 | if (device->bytes_used > 0) { |
1065 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1066 | device->bytes_used -= len; | |
1067 | spin_lock(&root->fs_info->free_chunk_lock); | |
1068 | root->fs_info->free_chunk_space += len; | |
1069 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1070 | } | |
8f18cf13 | 1071 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1072 | if (ret) { |
1073 | btrfs_error(root->fs_info, ret, | |
1074 | "Failed to remove dev extent item"); | |
1075 | } | |
b0b802d7 | 1076 | out: |
8f18cf13 CM |
1077 | btrfs_free_path(path); |
1078 | return ret; | |
1079 | } | |
1080 | ||
2b82032c | 1081 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1082 | struct btrfs_device *device, |
e17cade2 | 1083 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1084 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1085 | { |
1086 | int ret; | |
1087 | struct btrfs_path *path; | |
1088 | struct btrfs_root *root = device->dev_root; | |
1089 | struct btrfs_dev_extent *extent; | |
1090 | struct extent_buffer *leaf; | |
1091 | struct btrfs_key key; | |
1092 | ||
dfe25020 | 1093 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1094 | path = btrfs_alloc_path(); |
1095 | if (!path) | |
1096 | return -ENOMEM; | |
1097 | ||
0b86a832 | 1098 | key.objectid = device->devid; |
2b82032c | 1099 | key.offset = start; |
0b86a832 CM |
1100 | key.type = BTRFS_DEV_EXTENT_KEY; |
1101 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1102 | sizeof(*extent)); | |
2cdcecbc MF |
1103 | if (ret) |
1104 | goto out; | |
0b86a832 CM |
1105 | |
1106 | leaf = path->nodes[0]; | |
1107 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1108 | struct btrfs_dev_extent); | |
e17cade2 CM |
1109 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1110 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1111 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1112 | ||
1113 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1114 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1115 | BTRFS_UUID_SIZE); | |
1116 | ||
0b86a832 CM |
1117 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1118 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1119 | out: |
0b86a832 CM |
1120 | btrfs_free_path(path); |
1121 | return ret; | |
1122 | } | |
1123 | ||
a1b32a59 CM |
1124 | static noinline int find_next_chunk(struct btrfs_root *root, |
1125 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1126 | { |
1127 | struct btrfs_path *path; | |
1128 | int ret; | |
1129 | struct btrfs_key key; | |
e17cade2 | 1130 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1131 | struct btrfs_key found_key; |
1132 | ||
1133 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1134 | if (!path) |
1135 | return -ENOMEM; | |
0b86a832 | 1136 | |
e17cade2 | 1137 | key.objectid = objectid; |
0b86a832 CM |
1138 | key.offset = (u64)-1; |
1139 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1140 | ||
1141 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1142 | if (ret < 0) | |
1143 | goto error; | |
1144 | ||
79787eaa | 1145 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1146 | |
1147 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1148 | if (ret) { | |
e17cade2 | 1149 | *offset = 0; |
0b86a832 CM |
1150 | } else { |
1151 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1152 | path->slots[0]); | |
e17cade2 CM |
1153 | if (found_key.objectid != objectid) |
1154 | *offset = 0; | |
1155 | else { | |
1156 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1157 | struct btrfs_chunk); | |
1158 | *offset = found_key.offset + | |
1159 | btrfs_chunk_length(path->nodes[0], chunk); | |
1160 | } | |
0b86a832 CM |
1161 | } |
1162 | ret = 0; | |
1163 | error: | |
1164 | btrfs_free_path(path); | |
1165 | return ret; | |
1166 | } | |
1167 | ||
2b82032c | 1168 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1169 | { |
1170 | int ret; | |
1171 | struct btrfs_key key; | |
1172 | struct btrfs_key found_key; | |
2b82032c YZ |
1173 | struct btrfs_path *path; |
1174 | ||
1175 | root = root->fs_info->chunk_root; | |
1176 | ||
1177 | path = btrfs_alloc_path(); | |
1178 | if (!path) | |
1179 | return -ENOMEM; | |
0b86a832 CM |
1180 | |
1181 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1182 | key.type = BTRFS_DEV_ITEM_KEY; | |
1183 | key.offset = (u64)-1; | |
1184 | ||
1185 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1186 | if (ret < 0) | |
1187 | goto error; | |
1188 | ||
79787eaa | 1189 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1190 | |
1191 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1192 | BTRFS_DEV_ITEM_KEY); | |
1193 | if (ret) { | |
1194 | *objectid = 1; | |
1195 | } else { | |
1196 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1197 | path->slots[0]); | |
1198 | *objectid = found_key.offset + 1; | |
1199 | } | |
1200 | ret = 0; | |
1201 | error: | |
2b82032c | 1202 | btrfs_free_path(path); |
0b86a832 CM |
1203 | return ret; |
1204 | } | |
1205 | ||
1206 | /* | |
1207 | * the device information is stored in the chunk root | |
1208 | * the btrfs_device struct should be fully filled in | |
1209 | */ | |
1210 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1211 | struct btrfs_root *root, | |
1212 | struct btrfs_device *device) | |
1213 | { | |
1214 | int ret; | |
1215 | struct btrfs_path *path; | |
1216 | struct btrfs_dev_item *dev_item; | |
1217 | struct extent_buffer *leaf; | |
1218 | struct btrfs_key key; | |
1219 | unsigned long ptr; | |
0b86a832 CM |
1220 | |
1221 | root = root->fs_info->chunk_root; | |
1222 | ||
1223 | path = btrfs_alloc_path(); | |
1224 | if (!path) | |
1225 | return -ENOMEM; | |
1226 | ||
0b86a832 CM |
1227 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1228 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1229 | key.offset = device->devid; |
0b86a832 CM |
1230 | |
1231 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1232 | sizeof(*dev_item)); |
0b86a832 CM |
1233 | if (ret) |
1234 | goto out; | |
1235 | ||
1236 | leaf = path->nodes[0]; | |
1237 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1238 | ||
1239 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1240 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1241 | btrfs_set_device_type(leaf, dev_item, device->type); |
1242 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1243 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1244 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1245 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1246 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1247 | btrfs_set_device_group(leaf, dev_item, 0); |
1248 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1249 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1250 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1251 | |
0b86a832 | 1252 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1253 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1254 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1255 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1256 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1257 | |
2b82032c | 1258 | ret = 0; |
0b86a832 CM |
1259 | out: |
1260 | btrfs_free_path(path); | |
1261 | return ret; | |
1262 | } | |
8f18cf13 | 1263 | |
a061fc8d CM |
1264 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1265 | struct btrfs_device *device) | |
1266 | { | |
1267 | int ret; | |
1268 | struct btrfs_path *path; | |
a061fc8d | 1269 | struct btrfs_key key; |
a061fc8d CM |
1270 | struct btrfs_trans_handle *trans; |
1271 | ||
1272 | root = root->fs_info->chunk_root; | |
1273 | ||
1274 | path = btrfs_alloc_path(); | |
1275 | if (!path) | |
1276 | return -ENOMEM; | |
1277 | ||
a22285a6 | 1278 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1279 | if (IS_ERR(trans)) { |
1280 | btrfs_free_path(path); | |
1281 | return PTR_ERR(trans); | |
1282 | } | |
a061fc8d CM |
1283 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1284 | key.type = BTRFS_DEV_ITEM_KEY; | |
1285 | key.offset = device->devid; | |
7d9eb12c | 1286 | lock_chunks(root); |
a061fc8d CM |
1287 | |
1288 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1289 | if (ret < 0) | |
1290 | goto out; | |
1291 | ||
1292 | if (ret > 0) { | |
1293 | ret = -ENOENT; | |
1294 | goto out; | |
1295 | } | |
1296 | ||
1297 | ret = btrfs_del_item(trans, root, path); | |
1298 | if (ret) | |
1299 | goto out; | |
a061fc8d CM |
1300 | out: |
1301 | btrfs_free_path(path); | |
7d9eb12c | 1302 | unlock_chunks(root); |
a061fc8d CM |
1303 | btrfs_commit_transaction(trans, root); |
1304 | return ret; | |
1305 | } | |
1306 | ||
1307 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1308 | { | |
1309 | struct btrfs_device *device; | |
2b82032c | 1310 | struct btrfs_device *next_device; |
a061fc8d | 1311 | struct block_device *bdev; |
dfe25020 | 1312 | struct buffer_head *bh = NULL; |
a061fc8d | 1313 | struct btrfs_super_block *disk_super; |
1f78160c | 1314 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1315 | u64 all_avail; |
1316 | u64 devid; | |
2b82032c YZ |
1317 | u64 num_devices; |
1318 | u8 *dev_uuid; | |
a061fc8d | 1319 | int ret = 0; |
1f78160c | 1320 | bool clear_super = false; |
a061fc8d | 1321 | |
a061fc8d CM |
1322 | mutex_lock(&uuid_mutex); |
1323 | ||
1324 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1325 | root->fs_info->avail_system_alloc_bits | | |
1326 | root->fs_info->avail_metadata_alloc_bits; | |
1327 | ||
1328 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1329 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1330 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1331 | "on raid10\n"); | |
a061fc8d CM |
1332 | ret = -EINVAL; |
1333 | goto out; | |
1334 | } | |
1335 | ||
1336 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1337 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1338 | printk(KERN_ERR "btrfs: unable to go below two " |
1339 | "devices on raid1\n"); | |
a061fc8d CM |
1340 | ret = -EINVAL; |
1341 | goto out; | |
1342 | } | |
1343 | ||
dfe25020 | 1344 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1345 | struct list_head *devices; |
1346 | struct btrfs_device *tmp; | |
a061fc8d | 1347 | |
dfe25020 CM |
1348 | device = NULL; |
1349 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1350 | /* |
1351 | * It is safe to read the devices since the volume_mutex | |
1352 | * is held. | |
1353 | */ | |
c6e30871 | 1354 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1355 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1356 | device = tmp; | |
1357 | break; | |
1358 | } | |
1359 | } | |
1360 | bdev = NULL; | |
1361 | bh = NULL; | |
1362 | disk_super = NULL; | |
1363 | if (!device) { | |
d397712b CM |
1364 | printk(KERN_ERR "btrfs: no missing devices found to " |
1365 | "remove\n"); | |
dfe25020 CM |
1366 | goto out; |
1367 | } | |
dfe25020 | 1368 | } else { |
d4d77629 TH |
1369 | bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL, |
1370 | root->fs_info->bdev_holder); | |
dfe25020 CM |
1371 | if (IS_ERR(bdev)) { |
1372 | ret = PTR_ERR(bdev); | |
1373 | goto out; | |
1374 | } | |
a061fc8d | 1375 | |
2b82032c | 1376 | set_blocksize(bdev, 4096); |
3c4bb26b | 1377 | invalidate_bdev(bdev); |
a512bbf8 | 1378 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1379 | if (!bh) { |
20b45077 | 1380 | ret = -EINVAL; |
dfe25020 CM |
1381 | goto error_close; |
1382 | } | |
1383 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1384 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1385 | dev_uuid = disk_super->dev_item.uuid; |
1386 | device = btrfs_find_device(root, devid, dev_uuid, | |
1387 | disk_super->fsid); | |
dfe25020 CM |
1388 | if (!device) { |
1389 | ret = -ENOENT; | |
1390 | goto error_brelse; | |
1391 | } | |
2b82032c | 1392 | } |
dfe25020 | 1393 | |
2b82032c | 1394 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1395 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1396 | "device\n"); | |
2b82032c YZ |
1397 | ret = -EINVAL; |
1398 | goto error_brelse; | |
1399 | } | |
1400 | ||
1401 | if (device->writeable) { | |
0c1daee0 | 1402 | lock_chunks(root); |
2b82032c | 1403 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1404 | unlock_chunks(root); |
2b82032c | 1405 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1406 | clear_super = true; |
dfe25020 | 1407 | } |
a061fc8d CM |
1408 | |
1409 | ret = btrfs_shrink_device(device, 0); | |
1410 | if (ret) | |
9b3517e9 | 1411 | goto error_undo; |
a061fc8d | 1412 | |
a061fc8d CM |
1413 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1414 | if (ret) | |
9b3517e9 | 1415 | goto error_undo; |
a061fc8d | 1416 | |
2bf64758 JB |
1417 | spin_lock(&root->fs_info->free_chunk_lock); |
1418 | root->fs_info->free_chunk_space = device->total_bytes - | |
1419 | device->bytes_used; | |
1420 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1421 | ||
2b82032c | 1422 | device->in_fs_metadata = 0; |
a2de733c | 1423 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1424 | |
1425 | /* | |
1426 | * the device list mutex makes sure that we don't change | |
1427 | * the device list while someone else is writing out all | |
1428 | * the device supers. | |
1429 | */ | |
1f78160c XG |
1430 | |
1431 | cur_devices = device->fs_devices; | |
e5e9a520 | 1432 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1433 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1434 | |
e4404d6e | 1435 | device->fs_devices->num_devices--; |
2b82032c | 1436 | |
cd02dca5 CM |
1437 | if (device->missing) |
1438 | root->fs_info->fs_devices->missing_devices--; | |
1439 | ||
2b82032c YZ |
1440 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1441 | struct btrfs_device, dev_list); | |
1442 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1443 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1444 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1445 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1446 | ||
1f78160c | 1447 | if (device->bdev) |
e4404d6e | 1448 | device->fs_devices->open_devices--; |
1f78160c XG |
1449 | |
1450 | call_rcu(&device->rcu, free_device); | |
1451 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1452 | |
6c41761f DS |
1453 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1454 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1455 | |
1f78160c | 1456 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1457 | struct btrfs_fs_devices *fs_devices; |
1458 | fs_devices = root->fs_info->fs_devices; | |
1459 | while (fs_devices) { | |
1f78160c | 1460 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1461 | break; |
1462 | fs_devices = fs_devices->seed; | |
2b82032c | 1463 | } |
1f78160c XG |
1464 | fs_devices->seed = cur_devices->seed; |
1465 | cur_devices->seed = NULL; | |
0c1daee0 | 1466 | lock_chunks(root); |
1f78160c | 1467 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1468 | unlock_chunks(root); |
1f78160c | 1469 | free_fs_devices(cur_devices); |
2b82032c YZ |
1470 | } |
1471 | ||
1472 | /* | |
1473 | * at this point, the device is zero sized. We want to | |
1474 | * remove it from the devices list and zero out the old super | |
1475 | */ | |
1f78160c | 1476 | if (clear_super) { |
dfe25020 CM |
1477 | /* make sure this device isn't detected as part of |
1478 | * the FS anymore | |
1479 | */ | |
1480 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1481 | set_buffer_dirty(bh); | |
1482 | sync_dirty_buffer(bh); | |
dfe25020 | 1483 | } |
a061fc8d | 1484 | |
a061fc8d | 1485 | ret = 0; |
a061fc8d CM |
1486 | |
1487 | error_brelse: | |
1488 | brelse(bh); | |
1489 | error_close: | |
dfe25020 | 1490 | if (bdev) |
e525fd89 | 1491 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1492 | out: |
1493 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1494 | return ret; |
9b3517e9 ID |
1495 | error_undo: |
1496 | if (device->writeable) { | |
0c1daee0 | 1497 | lock_chunks(root); |
9b3517e9 ID |
1498 | list_add(&device->dev_alloc_list, |
1499 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1500 | unlock_chunks(root); |
9b3517e9 ID |
1501 | root->fs_info->fs_devices->rw_devices++; |
1502 | } | |
1503 | goto error_brelse; | |
a061fc8d CM |
1504 | } |
1505 | ||
2b82032c YZ |
1506 | /* |
1507 | * does all the dirty work required for changing file system's UUID. | |
1508 | */ | |
125ccb0a | 1509 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1510 | { |
1511 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1512 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1513 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1514 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1515 | struct btrfs_device *device; |
1516 | u64 super_flags; | |
1517 | ||
1518 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1519 | if (!fs_devices->seeding) |
2b82032c YZ |
1520 | return -EINVAL; |
1521 | ||
e4404d6e YZ |
1522 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1523 | if (!seed_devices) | |
2b82032c YZ |
1524 | return -ENOMEM; |
1525 | ||
e4404d6e YZ |
1526 | old_devices = clone_fs_devices(fs_devices); |
1527 | if (IS_ERR(old_devices)) { | |
1528 | kfree(seed_devices); | |
1529 | return PTR_ERR(old_devices); | |
2b82032c | 1530 | } |
e4404d6e | 1531 | |
2b82032c YZ |
1532 | list_add(&old_devices->list, &fs_uuids); |
1533 | ||
e4404d6e YZ |
1534 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1535 | seed_devices->opened = 1; | |
1536 | INIT_LIST_HEAD(&seed_devices->devices); | |
1537 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1538 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1539 | |
1540 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1541 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1542 | synchronize_rcu); | |
c9513edb XG |
1543 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1544 | ||
e4404d6e YZ |
1545 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1546 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1547 | device->fs_devices = seed_devices; | |
1548 | } | |
1549 | ||
2b82032c YZ |
1550 | fs_devices->seeding = 0; |
1551 | fs_devices->num_devices = 0; | |
1552 | fs_devices->open_devices = 0; | |
e4404d6e | 1553 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1554 | |
1555 | generate_random_uuid(fs_devices->fsid); | |
1556 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1557 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1558 | super_flags = btrfs_super_flags(disk_super) & | |
1559 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1560 | btrfs_set_super_flags(disk_super, super_flags); | |
1561 | ||
1562 | return 0; | |
1563 | } | |
1564 | ||
1565 | /* | |
1566 | * strore the expected generation for seed devices in device items. | |
1567 | */ | |
1568 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1569 | struct btrfs_root *root) | |
1570 | { | |
1571 | struct btrfs_path *path; | |
1572 | struct extent_buffer *leaf; | |
1573 | struct btrfs_dev_item *dev_item; | |
1574 | struct btrfs_device *device; | |
1575 | struct btrfs_key key; | |
1576 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1577 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1578 | u64 devid; | |
1579 | int ret; | |
1580 | ||
1581 | path = btrfs_alloc_path(); | |
1582 | if (!path) | |
1583 | return -ENOMEM; | |
1584 | ||
1585 | root = root->fs_info->chunk_root; | |
1586 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1587 | key.offset = 0; | |
1588 | key.type = BTRFS_DEV_ITEM_KEY; | |
1589 | ||
1590 | while (1) { | |
1591 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1592 | if (ret < 0) | |
1593 | goto error; | |
1594 | ||
1595 | leaf = path->nodes[0]; | |
1596 | next_slot: | |
1597 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1598 | ret = btrfs_next_leaf(root, path); | |
1599 | if (ret > 0) | |
1600 | break; | |
1601 | if (ret < 0) | |
1602 | goto error; | |
1603 | leaf = path->nodes[0]; | |
1604 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1605 | btrfs_release_path(path); |
2b82032c YZ |
1606 | continue; |
1607 | } | |
1608 | ||
1609 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1610 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1611 | key.type != BTRFS_DEV_ITEM_KEY) | |
1612 | break; | |
1613 | ||
1614 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1615 | struct btrfs_dev_item); | |
1616 | devid = btrfs_device_id(leaf, dev_item); | |
1617 | read_extent_buffer(leaf, dev_uuid, | |
1618 | (unsigned long)btrfs_device_uuid(dev_item), | |
1619 | BTRFS_UUID_SIZE); | |
1620 | read_extent_buffer(leaf, fs_uuid, | |
1621 | (unsigned long)btrfs_device_fsid(dev_item), | |
1622 | BTRFS_UUID_SIZE); | |
1623 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
79787eaa | 1624 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
1625 | |
1626 | if (device->fs_devices->seeding) { | |
1627 | btrfs_set_device_generation(leaf, dev_item, | |
1628 | device->generation); | |
1629 | btrfs_mark_buffer_dirty(leaf); | |
1630 | } | |
1631 | ||
1632 | path->slots[0]++; | |
1633 | goto next_slot; | |
1634 | } | |
1635 | ret = 0; | |
1636 | error: | |
1637 | btrfs_free_path(path); | |
1638 | return ret; | |
1639 | } | |
1640 | ||
788f20eb CM |
1641 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1642 | { | |
d5e2003c | 1643 | struct request_queue *q; |
788f20eb CM |
1644 | struct btrfs_trans_handle *trans; |
1645 | struct btrfs_device *device; | |
1646 | struct block_device *bdev; | |
788f20eb | 1647 | struct list_head *devices; |
2b82032c | 1648 | struct super_block *sb = root->fs_info->sb; |
606686ee | 1649 | struct rcu_string *name; |
788f20eb | 1650 | u64 total_bytes; |
2b82032c | 1651 | int seeding_dev = 0; |
788f20eb CM |
1652 | int ret = 0; |
1653 | ||
2b82032c | 1654 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 1655 | return -EROFS; |
788f20eb | 1656 | |
a5d16333 | 1657 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1658 | root->fs_info->bdev_holder); |
7f59203a JB |
1659 | if (IS_ERR(bdev)) |
1660 | return PTR_ERR(bdev); | |
a2135011 | 1661 | |
2b82032c YZ |
1662 | if (root->fs_info->fs_devices->seeding) { |
1663 | seeding_dev = 1; | |
1664 | down_write(&sb->s_umount); | |
1665 | mutex_lock(&uuid_mutex); | |
1666 | } | |
1667 | ||
8c8bee1d | 1668 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1669 | |
788f20eb | 1670 | devices = &root->fs_info->fs_devices->devices; |
e5e9a520 CM |
1671 | /* |
1672 | * we have the volume lock, so we don't need the extra | |
1673 | * device list mutex while reading the list here. | |
1674 | */ | |
c6e30871 | 1675 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1676 | if (device->bdev == bdev) { |
1677 | ret = -EEXIST; | |
2b82032c | 1678 | goto error; |
788f20eb CM |
1679 | } |
1680 | } | |
1681 | ||
1682 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1683 | if (!device) { | |
1684 | /* we can safely leave the fs_devices entry around */ | |
1685 | ret = -ENOMEM; | |
2b82032c | 1686 | goto error; |
788f20eb CM |
1687 | } |
1688 | ||
606686ee JB |
1689 | name = rcu_string_strdup(device_path, GFP_NOFS); |
1690 | if (!name) { | |
788f20eb | 1691 | kfree(device); |
2b82032c YZ |
1692 | ret = -ENOMEM; |
1693 | goto error; | |
788f20eb | 1694 | } |
606686ee | 1695 | rcu_assign_pointer(device->name, name); |
2b82032c YZ |
1696 | |
1697 | ret = find_next_devid(root, &device->devid); | |
1698 | if (ret) { | |
606686ee | 1699 | rcu_string_free(device->name); |
2b82032c YZ |
1700 | kfree(device); |
1701 | goto error; | |
1702 | } | |
1703 | ||
a22285a6 | 1704 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1705 | if (IS_ERR(trans)) { |
606686ee | 1706 | rcu_string_free(device->name); |
98d5dc13 TI |
1707 | kfree(device); |
1708 | ret = PTR_ERR(trans); | |
1709 | goto error; | |
1710 | } | |
1711 | ||
2b82032c YZ |
1712 | lock_chunks(root); |
1713 | ||
d5e2003c JB |
1714 | q = bdev_get_queue(bdev); |
1715 | if (blk_queue_discard(q)) | |
1716 | device->can_discard = 1; | |
2b82032c YZ |
1717 | device->writeable = 1; |
1718 | device->work.func = pending_bios_fn; | |
1719 | generate_random_uuid(device->uuid); | |
1720 | spin_lock_init(&device->io_lock); | |
1721 | device->generation = trans->transid; | |
788f20eb CM |
1722 | device->io_width = root->sectorsize; |
1723 | device->io_align = root->sectorsize; | |
1724 | device->sector_size = root->sectorsize; | |
1725 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1726 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1727 | device->dev_root = root->fs_info->dev_root; |
1728 | device->bdev = bdev; | |
dfe25020 | 1729 | device->in_fs_metadata = 1; |
fb01aa85 | 1730 | device->mode = FMODE_EXCL; |
2b82032c | 1731 | set_blocksize(device->bdev, 4096); |
788f20eb | 1732 | |
2b82032c YZ |
1733 | if (seeding_dev) { |
1734 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 1735 | ret = btrfs_prepare_sprout(root); |
79787eaa | 1736 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 1737 | } |
788f20eb | 1738 | |
2b82032c | 1739 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 CM |
1740 | |
1741 | /* | |
1742 | * we don't want write_supers to jump in here with our device | |
1743 | * half setup | |
1744 | */ | |
1745 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c | 1746 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1747 | list_add(&device->dev_alloc_list, |
1748 | &root->fs_info->fs_devices->alloc_list); | |
1749 | root->fs_info->fs_devices->num_devices++; | |
1750 | root->fs_info->fs_devices->open_devices++; | |
1751 | root->fs_info->fs_devices->rw_devices++; | |
d5e2003c JB |
1752 | if (device->can_discard) |
1753 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1754 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1755 | |
2bf64758 JB |
1756 | spin_lock(&root->fs_info->free_chunk_lock); |
1757 | root->fs_info->free_chunk_space += device->total_bytes; | |
1758 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1759 | ||
c289811c CM |
1760 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1761 | root->fs_info->fs_devices->rotating = 1; | |
1762 | ||
6c41761f DS |
1763 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1764 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1765 | total_bytes + device->total_bytes); |
1766 | ||
6c41761f DS |
1767 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1768 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1769 | total_bytes + 1); |
e5e9a520 | 1770 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1771 | |
2b82032c YZ |
1772 | if (seeding_dev) { |
1773 | ret = init_first_rw_device(trans, root, device); | |
79787eaa JM |
1774 | if (ret) |
1775 | goto error_trans; | |
2b82032c | 1776 | ret = btrfs_finish_sprout(trans, root); |
79787eaa JM |
1777 | if (ret) |
1778 | goto error_trans; | |
2b82032c YZ |
1779 | } else { |
1780 | ret = btrfs_add_device(trans, root, device); | |
79787eaa JM |
1781 | if (ret) |
1782 | goto error_trans; | |
2b82032c YZ |
1783 | } |
1784 | ||
913d952e CM |
1785 | /* |
1786 | * we've got more storage, clear any full flags on the space | |
1787 | * infos | |
1788 | */ | |
1789 | btrfs_clear_space_info_full(root->fs_info); | |
1790 | ||
7d9eb12c | 1791 | unlock_chunks(root); |
79787eaa | 1792 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 1793 | |
2b82032c YZ |
1794 | if (seeding_dev) { |
1795 | mutex_unlock(&uuid_mutex); | |
1796 | up_write(&sb->s_umount); | |
788f20eb | 1797 | |
79787eaa JM |
1798 | if (ret) /* transaction commit */ |
1799 | return ret; | |
1800 | ||
2b82032c | 1801 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
1802 | if (ret < 0) |
1803 | btrfs_error(root->fs_info, ret, | |
1804 | "Failed to relocate sys chunks after " | |
1805 | "device initialization. This can be fixed " | |
1806 | "using the \"btrfs balance\" command."); | |
2b82032c | 1807 | } |
c9e9f97b | 1808 | |
2b82032c | 1809 | return ret; |
79787eaa JM |
1810 | |
1811 | error_trans: | |
1812 | unlock_chunks(root); | |
1813 | btrfs_abort_transaction(trans, root, ret); | |
1814 | btrfs_end_transaction(trans, root); | |
606686ee | 1815 | rcu_string_free(device->name); |
79787eaa | 1816 | kfree(device); |
2b82032c | 1817 | error: |
e525fd89 | 1818 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1819 | if (seeding_dev) { |
1820 | mutex_unlock(&uuid_mutex); | |
1821 | up_write(&sb->s_umount); | |
1822 | } | |
c9e9f97b | 1823 | return ret; |
788f20eb CM |
1824 | } |
1825 | ||
d397712b CM |
1826 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1827 | struct btrfs_device *device) | |
0b86a832 CM |
1828 | { |
1829 | int ret; | |
1830 | struct btrfs_path *path; | |
1831 | struct btrfs_root *root; | |
1832 | struct btrfs_dev_item *dev_item; | |
1833 | struct extent_buffer *leaf; | |
1834 | struct btrfs_key key; | |
1835 | ||
1836 | root = device->dev_root->fs_info->chunk_root; | |
1837 | ||
1838 | path = btrfs_alloc_path(); | |
1839 | if (!path) | |
1840 | return -ENOMEM; | |
1841 | ||
1842 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1843 | key.type = BTRFS_DEV_ITEM_KEY; | |
1844 | key.offset = device->devid; | |
1845 | ||
1846 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1847 | if (ret < 0) | |
1848 | goto out; | |
1849 | ||
1850 | if (ret > 0) { | |
1851 | ret = -ENOENT; | |
1852 | goto out; | |
1853 | } | |
1854 | ||
1855 | leaf = path->nodes[0]; | |
1856 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1857 | ||
1858 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1859 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1860 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1861 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1862 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1863 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1864 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1865 | btrfs_mark_buffer_dirty(leaf); | |
1866 | ||
1867 | out: | |
1868 | btrfs_free_path(path); | |
1869 | return ret; | |
1870 | } | |
1871 | ||
7d9eb12c | 1872 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1873 | struct btrfs_device *device, u64 new_size) |
1874 | { | |
1875 | struct btrfs_super_block *super_copy = | |
6c41761f | 1876 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1877 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1878 | u64 diff = new_size - device->total_bytes; | |
1879 | ||
2b82032c YZ |
1880 | if (!device->writeable) |
1881 | return -EACCES; | |
1882 | if (new_size <= device->total_bytes) | |
1883 | return -EINVAL; | |
1884 | ||
8f18cf13 | 1885 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1886 | device->fs_devices->total_rw_bytes += diff; |
1887 | ||
1888 | device->total_bytes = new_size; | |
9779b72f | 1889 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1890 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1891 | ||
8f18cf13 CM |
1892 | return btrfs_update_device(trans, device); |
1893 | } | |
1894 | ||
7d9eb12c CM |
1895 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1896 | struct btrfs_device *device, u64 new_size) | |
1897 | { | |
1898 | int ret; | |
1899 | lock_chunks(device->dev_root); | |
1900 | ret = __btrfs_grow_device(trans, device, new_size); | |
1901 | unlock_chunks(device->dev_root); | |
1902 | return ret; | |
1903 | } | |
1904 | ||
8f18cf13 CM |
1905 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1906 | struct btrfs_root *root, | |
1907 | u64 chunk_tree, u64 chunk_objectid, | |
1908 | u64 chunk_offset) | |
1909 | { | |
1910 | int ret; | |
1911 | struct btrfs_path *path; | |
1912 | struct btrfs_key key; | |
1913 | ||
1914 | root = root->fs_info->chunk_root; | |
1915 | path = btrfs_alloc_path(); | |
1916 | if (!path) | |
1917 | return -ENOMEM; | |
1918 | ||
1919 | key.objectid = chunk_objectid; | |
1920 | key.offset = chunk_offset; | |
1921 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1922 | ||
1923 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
1924 | if (ret < 0) |
1925 | goto out; | |
1926 | else if (ret > 0) { /* Logic error or corruption */ | |
1927 | btrfs_error(root->fs_info, -ENOENT, | |
1928 | "Failed lookup while freeing chunk."); | |
1929 | ret = -ENOENT; | |
1930 | goto out; | |
1931 | } | |
8f18cf13 CM |
1932 | |
1933 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
1934 | if (ret < 0) |
1935 | btrfs_error(root->fs_info, ret, | |
1936 | "Failed to delete chunk item."); | |
1937 | out: | |
8f18cf13 | 1938 | btrfs_free_path(path); |
65a246c5 | 1939 | return ret; |
8f18cf13 CM |
1940 | } |
1941 | ||
b2950863 | 1942 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1943 | chunk_offset) |
1944 | { | |
6c41761f | 1945 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
1946 | struct btrfs_disk_key *disk_key; |
1947 | struct btrfs_chunk *chunk; | |
1948 | u8 *ptr; | |
1949 | int ret = 0; | |
1950 | u32 num_stripes; | |
1951 | u32 array_size; | |
1952 | u32 len = 0; | |
1953 | u32 cur; | |
1954 | struct btrfs_key key; | |
1955 | ||
1956 | array_size = btrfs_super_sys_array_size(super_copy); | |
1957 | ||
1958 | ptr = super_copy->sys_chunk_array; | |
1959 | cur = 0; | |
1960 | ||
1961 | while (cur < array_size) { | |
1962 | disk_key = (struct btrfs_disk_key *)ptr; | |
1963 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1964 | ||
1965 | len = sizeof(*disk_key); | |
1966 | ||
1967 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1968 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1969 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1970 | len += btrfs_chunk_item_size(num_stripes); | |
1971 | } else { | |
1972 | ret = -EIO; | |
1973 | break; | |
1974 | } | |
1975 | if (key.objectid == chunk_objectid && | |
1976 | key.offset == chunk_offset) { | |
1977 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1978 | array_size -= len; | |
1979 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1980 | } else { | |
1981 | ptr += len; | |
1982 | cur += len; | |
1983 | } | |
1984 | } | |
1985 | return ret; | |
1986 | } | |
1987 | ||
b2950863 | 1988 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
1989 | u64 chunk_tree, u64 chunk_objectid, |
1990 | u64 chunk_offset) | |
1991 | { | |
1992 | struct extent_map_tree *em_tree; | |
1993 | struct btrfs_root *extent_root; | |
1994 | struct btrfs_trans_handle *trans; | |
1995 | struct extent_map *em; | |
1996 | struct map_lookup *map; | |
1997 | int ret; | |
1998 | int i; | |
1999 | ||
2000 | root = root->fs_info->chunk_root; | |
2001 | extent_root = root->fs_info->extent_root; | |
2002 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2003 | ||
ba1bf481 JB |
2004 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2005 | if (ret) | |
2006 | return -ENOSPC; | |
2007 | ||
8f18cf13 | 2008 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2009 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2010 | if (ret) |
2011 | return ret; | |
8f18cf13 | 2012 | |
a22285a6 | 2013 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2014 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 2015 | |
7d9eb12c CM |
2016 | lock_chunks(root); |
2017 | ||
8f18cf13 CM |
2018 | /* |
2019 | * step two, delete the device extents and the | |
2020 | * chunk tree entries | |
2021 | */ | |
890871be | 2022 | read_lock(&em_tree->lock); |
8f18cf13 | 2023 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2024 | read_unlock(&em_tree->lock); |
8f18cf13 | 2025 | |
285190d9 | 2026 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2027 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2028 | map = (struct map_lookup *)em->bdev; |
2029 | ||
2030 | for (i = 0; i < map->num_stripes; i++) { | |
2031 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
2032 | map->stripes[i].physical); | |
2033 | BUG_ON(ret); | |
a061fc8d | 2034 | |
dfe25020 CM |
2035 | if (map->stripes[i].dev) { |
2036 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2037 | BUG_ON(ret); | |
2038 | } | |
8f18cf13 CM |
2039 | } |
2040 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2041 | chunk_offset); | |
2042 | ||
2043 | BUG_ON(ret); | |
2044 | ||
1abe9b8a | 2045 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2046 | ||
8f18cf13 CM |
2047 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2048 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2049 | BUG_ON(ret); | |
8f18cf13 CM |
2050 | } |
2051 | ||
2b82032c YZ |
2052 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2053 | BUG_ON(ret); | |
2054 | ||
890871be | 2055 | write_lock(&em_tree->lock); |
2b82032c | 2056 | remove_extent_mapping(em_tree, em); |
890871be | 2057 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2058 | |
2059 | kfree(map); | |
2060 | em->bdev = NULL; | |
2061 | ||
2062 | /* once for the tree */ | |
2063 | free_extent_map(em); | |
2064 | /* once for us */ | |
2065 | free_extent_map(em); | |
2066 | ||
2067 | unlock_chunks(root); | |
2068 | btrfs_end_transaction(trans, root); | |
2069 | return 0; | |
2070 | } | |
2071 | ||
2072 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2073 | { | |
2074 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2075 | struct btrfs_path *path; | |
2076 | struct extent_buffer *leaf; | |
2077 | struct btrfs_chunk *chunk; | |
2078 | struct btrfs_key key; | |
2079 | struct btrfs_key found_key; | |
2080 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2081 | u64 chunk_type; | |
ba1bf481 JB |
2082 | bool retried = false; |
2083 | int failed = 0; | |
2b82032c YZ |
2084 | int ret; |
2085 | ||
2086 | path = btrfs_alloc_path(); | |
2087 | if (!path) | |
2088 | return -ENOMEM; | |
2089 | ||
ba1bf481 | 2090 | again: |
2b82032c YZ |
2091 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2092 | key.offset = (u64)-1; | |
2093 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2094 | ||
2095 | while (1) { | |
2096 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2097 | if (ret < 0) | |
2098 | goto error; | |
79787eaa | 2099 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2100 | |
2101 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2102 | key.type); | |
2103 | if (ret < 0) | |
2104 | goto error; | |
2105 | if (ret > 0) | |
2106 | break; | |
1a40e23b | 2107 | |
2b82032c YZ |
2108 | leaf = path->nodes[0]; |
2109 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2110 | |
2b82032c YZ |
2111 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2112 | struct btrfs_chunk); | |
2113 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2114 | btrfs_release_path(path); |
8f18cf13 | 2115 | |
2b82032c YZ |
2116 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2117 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2118 | found_key.objectid, | |
2119 | found_key.offset); | |
ba1bf481 JB |
2120 | if (ret == -ENOSPC) |
2121 | failed++; | |
2122 | else if (ret) | |
2123 | BUG(); | |
2b82032c | 2124 | } |
8f18cf13 | 2125 | |
2b82032c YZ |
2126 | if (found_key.offset == 0) |
2127 | break; | |
2128 | key.offset = found_key.offset - 1; | |
2129 | } | |
2130 | ret = 0; | |
ba1bf481 JB |
2131 | if (failed && !retried) { |
2132 | failed = 0; | |
2133 | retried = true; | |
2134 | goto again; | |
2135 | } else if (failed && retried) { | |
2136 | WARN_ON(1); | |
2137 | ret = -ENOSPC; | |
2138 | } | |
2b82032c YZ |
2139 | error: |
2140 | btrfs_free_path(path); | |
2141 | return ret; | |
8f18cf13 CM |
2142 | } |
2143 | ||
0940ebf6 ID |
2144 | static int insert_balance_item(struct btrfs_root *root, |
2145 | struct btrfs_balance_control *bctl) | |
2146 | { | |
2147 | struct btrfs_trans_handle *trans; | |
2148 | struct btrfs_balance_item *item; | |
2149 | struct btrfs_disk_balance_args disk_bargs; | |
2150 | struct btrfs_path *path; | |
2151 | struct extent_buffer *leaf; | |
2152 | struct btrfs_key key; | |
2153 | int ret, err; | |
2154 | ||
2155 | path = btrfs_alloc_path(); | |
2156 | if (!path) | |
2157 | return -ENOMEM; | |
2158 | ||
2159 | trans = btrfs_start_transaction(root, 0); | |
2160 | if (IS_ERR(trans)) { | |
2161 | btrfs_free_path(path); | |
2162 | return PTR_ERR(trans); | |
2163 | } | |
2164 | ||
2165 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2166 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2167 | key.offset = 0; | |
2168 | ||
2169 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2170 | sizeof(*item)); | |
2171 | if (ret) | |
2172 | goto out; | |
2173 | ||
2174 | leaf = path->nodes[0]; | |
2175 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2176 | ||
2177 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2178 | ||
2179 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2180 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2181 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2182 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2183 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2184 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2185 | ||
2186 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2187 | ||
2188 | btrfs_mark_buffer_dirty(leaf); | |
2189 | out: | |
2190 | btrfs_free_path(path); | |
2191 | err = btrfs_commit_transaction(trans, root); | |
2192 | if (err && !ret) | |
2193 | ret = err; | |
2194 | return ret; | |
2195 | } | |
2196 | ||
2197 | static int del_balance_item(struct btrfs_root *root) | |
2198 | { | |
2199 | struct btrfs_trans_handle *trans; | |
2200 | struct btrfs_path *path; | |
2201 | struct btrfs_key key; | |
2202 | int ret, err; | |
2203 | ||
2204 | path = btrfs_alloc_path(); | |
2205 | if (!path) | |
2206 | return -ENOMEM; | |
2207 | ||
2208 | trans = btrfs_start_transaction(root, 0); | |
2209 | if (IS_ERR(trans)) { | |
2210 | btrfs_free_path(path); | |
2211 | return PTR_ERR(trans); | |
2212 | } | |
2213 | ||
2214 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2215 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2216 | key.offset = 0; | |
2217 | ||
2218 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2219 | if (ret < 0) | |
2220 | goto out; | |
2221 | if (ret > 0) { | |
2222 | ret = -ENOENT; | |
2223 | goto out; | |
2224 | } | |
2225 | ||
2226 | ret = btrfs_del_item(trans, root, path); | |
2227 | out: | |
2228 | btrfs_free_path(path); | |
2229 | err = btrfs_commit_transaction(trans, root); | |
2230 | if (err && !ret) | |
2231 | ret = err; | |
2232 | return ret; | |
2233 | } | |
2234 | ||
59641015 ID |
2235 | /* |
2236 | * This is a heuristic used to reduce the number of chunks balanced on | |
2237 | * resume after balance was interrupted. | |
2238 | */ | |
2239 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2240 | { | |
2241 | /* | |
2242 | * Turn on soft mode for chunk types that were being converted. | |
2243 | */ | |
2244 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2245 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2246 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2247 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2248 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2249 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2250 | ||
2251 | /* | |
2252 | * Turn on usage filter if is not already used. The idea is | |
2253 | * that chunks that we have already balanced should be | |
2254 | * reasonably full. Don't do it for chunks that are being | |
2255 | * converted - that will keep us from relocating unconverted | |
2256 | * (albeit full) chunks. | |
2257 | */ | |
2258 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2259 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2260 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2261 | bctl->data.usage = 90; | |
2262 | } | |
2263 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2264 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2265 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2266 | bctl->sys.usage = 90; | |
2267 | } | |
2268 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2269 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2270 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2271 | bctl->meta.usage = 90; | |
2272 | } | |
2273 | } | |
2274 | ||
c9e9f97b ID |
2275 | /* |
2276 | * Should be called with both balance and volume mutexes held to | |
2277 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2278 | * restriper. Same goes for unset_balance_control. | |
2279 | */ | |
2280 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2281 | { | |
2282 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2283 | ||
2284 | BUG_ON(fs_info->balance_ctl); | |
2285 | ||
2286 | spin_lock(&fs_info->balance_lock); | |
2287 | fs_info->balance_ctl = bctl; | |
2288 | spin_unlock(&fs_info->balance_lock); | |
2289 | } | |
2290 | ||
2291 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2292 | { | |
2293 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2294 | ||
2295 | BUG_ON(!fs_info->balance_ctl); | |
2296 | ||
2297 | spin_lock(&fs_info->balance_lock); | |
2298 | fs_info->balance_ctl = NULL; | |
2299 | spin_unlock(&fs_info->balance_lock); | |
2300 | ||
2301 | kfree(bctl); | |
2302 | } | |
2303 | ||
ed25e9b2 ID |
2304 | /* |
2305 | * Balance filters. Return 1 if chunk should be filtered out | |
2306 | * (should not be balanced). | |
2307 | */ | |
899c81ea | 2308 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2309 | struct btrfs_balance_args *bargs) |
2310 | { | |
899c81ea ID |
2311 | chunk_type = chunk_to_extended(chunk_type) & |
2312 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2313 | |
899c81ea | 2314 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2315 | return 0; |
2316 | ||
2317 | return 1; | |
2318 | } | |
2319 | ||
5ce5b3c0 ID |
2320 | static u64 div_factor_fine(u64 num, int factor) |
2321 | { | |
2322 | if (factor <= 0) | |
2323 | return 0; | |
2324 | if (factor >= 100) | |
2325 | return num; | |
2326 | ||
2327 | num *= factor; | |
2328 | do_div(num, 100); | |
2329 | return num; | |
2330 | } | |
2331 | ||
2332 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, | |
2333 | struct btrfs_balance_args *bargs) | |
2334 | { | |
2335 | struct btrfs_block_group_cache *cache; | |
2336 | u64 chunk_used, user_thresh; | |
2337 | int ret = 1; | |
2338 | ||
2339 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2340 | chunk_used = btrfs_block_group_used(&cache->item); | |
2341 | ||
2342 | user_thresh = div_factor_fine(cache->key.offset, bargs->usage); | |
2343 | if (chunk_used < user_thresh) | |
2344 | ret = 0; | |
2345 | ||
2346 | btrfs_put_block_group(cache); | |
2347 | return ret; | |
2348 | } | |
2349 | ||
409d404b ID |
2350 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2351 | struct btrfs_chunk *chunk, | |
2352 | struct btrfs_balance_args *bargs) | |
2353 | { | |
2354 | struct btrfs_stripe *stripe; | |
2355 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2356 | int i; | |
2357 | ||
2358 | for (i = 0; i < num_stripes; i++) { | |
2359 | stripe = btrfs_stripe_nr(chunk, i); | |
2360 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2361 | return 0; | |
2362 | } | |
2363 | ||
2364 | return 1; | |
2365 | } | |
2366 | ||
94e60d5a ID |
2367 | /* [pstart, pend) */ |
2368 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2369 | struct btrfs_chunk *chunk, | |
2370 | u64 chunk_offset, | |
2371 | struct btrfs_balance_args *bargs) | |
2372 | { | |
2373 | struct btrfs_stripe *stripe; | |
2374 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2375 | u64 stripe_offset; | |
2376 | u64 stripe_length; | |
2377 | int factor; | |
2378 | int i; | |
2379 | ||
2380 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2381 | return 0; | |
2382 | ||
2383 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
2384 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) | |
2385 | factor = 2; | |
2386 | else | |
2387 | factor = 1; | |
2388 | factor = num_stripes / factor; | |
2389 | ||
2390 | for (i = 0; i < num_stripes; i++) { | |
2391 | stripe = btrfs_stripe_nr(chunk, i); | |
2392 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2393 | continue; | |
2394 | ||
2395 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2396 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2397 | do_div(stripe_length, factor); | |
2398 | ||
2399 | if (stripe_offset < bargs->pend && | |
2400 | stripe_offset + stripe_length > bargs->pstart) | |
2401 | return 0; | |
2402 | } | |
2403 | ||
2404 | return 1; | |
2405 | } | |
2406 | ||
ea67176a ID |
2407 | /* [vstart, vend) */ |
2408 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2409 | struct btrfs_chunk *chunk, | |
2410 | u64 chunk_offset, | |
2411 | struct btrfs_balance_args *bargs) | |
2412 | { | |
2413 | if (chunk_offset < bargs->vend && | |
2414 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2415 | /* at least part of the chunk is inside this vrange */ | |
2416 | return 0; | |
2417 | ||
2418 | return 1; | |
2419 | } | |
2420 | ||
899c81ea | 2421 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2422 | struct btrfs_balance_args *bargs) |
2423 | { | |
2424 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2425 | return 0; | |
2426 | ||
899c81ea ID |
2427 | chunk_type = chunk_to_extended(chunk_type) & |
2428 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2429 | |
899c81ea | 2430 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2431 | return 1; |
2432 | ||
2433 | return 0; | |
2434 | } | |
2435 | ||
f43ffb60 ID |
2436 | static int should_balance_chunk(struct btrfs_root *root, |
2437 | struct extent_buffer *leaf, | |
2438 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2439 | { | |
2440 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2441 | struct btrfs_balance_args *bargs = NULL; | |
2442 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2443 | ||
2444 | /* type filter */ | |
2445 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2446 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2447 | return 0; | |
2448 | } | |
2449 | ||
2450 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2451 | bargs = &bctl->data; | |
2452 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2453 | bargs = &bctl->sys; | |
2454 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2455 | bargs = &bctl->meta; | |
2456 | ||
ed25e9b2 ID |
2457 | /* profiles filter */ |
2458 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2459 | chunk_profiles_filter(chunk_type, bargs)) { | |
2460 | return 0; | |
5ce5b3c0 ID |
2461 | } |
2462 | ||
2463 | /* usage filter */ | |
2464 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2465 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2466 | return 0; | |
409d404b ID |
2467 | } |
2468 | ||
2469 | /* devid filter */ | |
2470 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2471 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2472 | return 0; | |
94e60d5a ID |
2473 | } |
2474 | ||
2475 | /* drange filter, makes sense only with devid filter */ | |
2476 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2477 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2478 | return 0; | |
ea67176a ID |
2479 | } |
2480 | ||
2481 | /* vrange filter */ | |
2482 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2483 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2484 | return 0; | |
ed25e9b2 ID |
2485 | } |
2486 | ||
cfa4c961 ID |
2487 | /* soft profile changing mode */ |
2488 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2489 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2490 | return 0; | |
2491 | } | |
2492 | ||
f43ffb60 ID |
2493 | return 1; |
2494 | } | |
2495 | ||
ec44a35c CM |
2496 | static u64 div_factor(u64 num, int factor) |
2497 | { | |
2498 | if (factor == 10) | |
2499 | return num; | |
2500 | num *= factor; | |
2501 | do_div(num, 10); | |
2502 | return num; | |
2503 | } | |
2504 | ||
c9e9f97b | 2505 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2506 | { |
19a39dce | 2507 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2508 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2509 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2510 | struct list_head *devices; | |
ec44a35c CM |
2511 | struct btrfs_device *device; |
2512 | u64 old_size; | |
2513 | u64 size_to_free; | |
f43ffb60 | 2514 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2515 | struct btrfs_path *path; |
2516 | struct btrfs_key key; | |
ec44a35c | 2517 | struct btrfs_key found_key; |
c9e9f97b | 2518 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2519 | struct extent_buffer *leaf; |
2520 | int slot; | |
c9e9f97b ID |
2521 | int ret; |
2522 | int enospc_errors = 0; | |
19a39dce | 2523 | bool counting = true; |
ec44a35c | 2524 | |
ec44a35c | 2525 | /* step one make some room on all the devices */ |
c9e9f97b | 2526 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2527 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2528 | old_size = device->total_bytes; |
2529 | size_to_free = div_factor(old_size, 1); | |
2530 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2531 | if (!device->writeable || |
2532 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2533 | continue; |
2534 | ||
2535 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2536 | if (ret == -ENOSPC) |
2537 | break; | |
ec44a35c CM |
2538 | BUG_ON(ret); |
2539 | ||
a22285a6 | 2540 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2541 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2542 | |
2543 | ret = btrfs_grow_device(trans, device, old_size); | |
2544 | BUG_ON(ret); | |
2545 | ||
2546 | btrfs_end_transaction(trans, dev_root); | |
2547 | } | |
2548 | ||
2549 | /* step two, relocate all the chunks */ | |
2550 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2551 | if (!path) { |
2552 | ret = -ENOMEM; | |
2553 | goto error; | |
2554 | } | |
19a39dce ID |
2555 | |
2556 | /* zero out stat counters */ | |
2557 | spin_lock(&fs_info->balance_lock); | |
2558 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2559 | spin_unlock(&fs_info->balance_lock); | |
2560 | again: | |
ec44a35c CM |
2561 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2562 | key.offset = (u64)-1; | |
2563 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2564 | ||
d397712b | 2565 | while (1) { |
19a39dce | 2566 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2567 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2568 | ret = -ECANCELED; |
2569 | goto error; | |
2570 | } | |
2571 | ||
ec44a35c CM |
2572 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2573 | if (ret < 0) | |
2574 | goto error; | |
2575 | ||
2576 | /* | |
2577 | * this shouldn't happen, it means the last relocate | |
2578 | * failed | |
2579 | */ | |
2580 | if (ret == 0) | |
c9e9f97b | 2581 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2582 | |
2583 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2584 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2585 | if (ret) { |
2586 | ret = 0; | |
ec44a35c | 2587 | break; |
c9e9f97b | 2588 | } |
7d9eb12c | 2589 | |
f43ffb60 ID |
2590 | leaf = path->nodes[0]; |
2591 | slot = path->slots[0]; | |
2592 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2593 | |
ec44a35c CM |
2594 | if (found_key.objectid != key.objectid) |
2595 | break; | |
7d9eb12c | 2596 | |
ec44a35c | 2597 | /* chunk zero is special */ |
ba1bf481 | 2598 | if (found_key.offset == 0) |
ec44a35c CM |
2599 | break; |
2600 | ||
f43ffb60 ID |
2601 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2602 | ||
19a39dce ID |
2603 | if (!counting) { |
2604 | spin_lock(&fs_info->balance_lock); | |
2605 | bctl->stat.considered++; | |
2606 | spin_unlock(&fs_info->balance_lock); | |
2607 | } | |
2608 | ||
f43ffb60 ID |
2609 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2610 | found_key.offset); | |
b3b4aa74 | 2611 | btrfs_release_path(path); |
f43ffb60 ID |
2612 | if (!ret) |
2613 | goto loop; | |
2614 | ||
19a39dce ID |
2615 | if (counting) { |
2616 | spin_lock(&fs_info->balance_lock); | |
2617 | bctl->stat.expected++; | |
2618 | spin_unlock(&fs_info->balance_lock); | |
2619 | goto loop; | |
2620 | } | |
2621 | ||
ec44a35c CM |
2622 | ret = btrfs_relocate_chunk(chunk_root, |
2623 | chunk_root->root_key.objectid, | |
2624 | found_key.objectid, | |
2625 | found_key.offset); | |
508794eb JB |
2626 | if (ret && ret != -ENOSPC) |
2627 | goto error; | |
19a39dce | 2628 | if (ret == -ENOSPC) { |
c9e9f97b | 2629 | enospc_errors++; |
19a39dce ID |
2630 | } else { |
2631 | spin_lock(&fs_info->balance_lock); | |
2632 | bctl->stat.completed++; | |
2633 | spin_unlock(&fs_info->balance_lock); | |
2634 | } | |
f43ffb60 | 2635 | loop: |
ba1bf481 | 2636 | key.offset = found_key.offset - 1; |
ec44a35c | 2637 | } |
c9e9f97b | 2638 | |
19a39dce ID |
2639 | if (counting) { |
2640 | btrfs_release_path(path); | |
2641 | counting = false; | |
2642 | goto again; | |
2643 | } | |
ec44a35c CM |
2644 | error: |
2645 | btrfs_free_path(path); | |
c9e9f97b ID |
2646 | if (enospc_errors) { |
2647 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
2648 | enospc_errors); | |
2649 | if (!ret) | |
2650 | ret = -ENOSPC; | |
2651 | } | |
2652 | ||
ec44a35c CM |
2653 | return ret; |
2654 | } | |
2655 | ||
0c460c0d ID |
2656 | /** |
2657 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
2658 | * @flags: profile to validate | |
2659 | * @extended: if true @flags is treated as an extended profile | |
2660 | */ | |
2661 | static int alloc_profile_is_valid(u64 flags, int extended) | |
2662 | { | |
2663 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
2664 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
2665 | ||
2666 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
2667 | ||
2668 | /* 1) check that all other bits are zeroed */ | |
2669 | if (flags & ~mask) | |
2670 | return 0; | |
2671 | ||
2672 | /* 2) see if profile is reduced */ | |
2673 | if (flags == 0) | |
2674 | return !extended; /* "0" is valid for usual profiles */ | |
2675 | ||
2676 | /* true if exactly one bit set */ | |
2677 | return (flags & (flags - 1)) == 0; | |
2678 | } | |
2679 | ||
837d5b6e ID |
2680 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
2681 | { | |
a7e99c69 ID |
2682 | /* cancel requested || normal exit path */ |
2683 | return atomic_read(&fs_info->balance_cancel_req) || | |
2684 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
2685 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
2686 | } |
2687 | ||
c9e9f97b ID |
2688 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
2689 | { | |
0940ebf6 ID |
2690 | int ret; |
2691 | ||
c9e9f97b | 2692 | unset_balance_control(fs_info); |
0940ebf6 ID |
2693 | ret = del_balance_item(fs_info->tree_root); |
2694 | BUG_ON(ret); | |
c9e9f97b ID |
2695 | } |
2696 | ||
19a39dce | 2697 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
2698 | struct btrfs_ioctl_balance_args *bargs); |
2699 | ||
2700 | /* | |
2701 | * Should be called with both balance and volume mutexes held | |
2702 | */ | |
2703 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
2704 | struct btrfs_ioctl_balance_args *bargs) | |
2705 | { | |
2706 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 2707 | u64 allowed; |
e4837f8f | 2708 | int mixed = 0; |
c9e9f97b ID |
2709 | int ret; |
2710 | ||
837d5b6e | 2711 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
2712 | atomic_read(&fs_info->balance_pause_req) || |
2713 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
2714 | ret = -EINVAL; |
2715 | goto out; | |
2716 | } | |
2717 | ||
e4837f8f ID |
2718 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
2719 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
2720 | mixed = 1; | |
2721 | ||
f43ffb60 ID |
2722 | /* |
2723 | * In case of mixed groups both data and meta should be picked, | |
2724 | * and identical options should be given for both of them. | |
2725 | */ | |
e4837f8f ID |
2726 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
2727 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
2728 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
2729 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
2730 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
2731 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
2732 | "metadata balance options must be the same\n"); | |
2733 | ret = -EINVAL; | |
2734 | goto out; | |
2735 | } | |
2736 | } | |
2737 | ||
e4d8ec0f ID |
2738 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
2739 | if (fs_info->fs_devices->num_devices == 1) | |
2740 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
2741 | else if (fs_info->fs_devices->num_devices < 4) | |
2742 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
2743 | else | |
2744 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
2745 | BTRFS_BLOCK_GROUP_RAID10); | |
2746 | ||
6728b198 ID |
2747 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2748 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
2749 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
2750 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2751 | "data profile %llu\n", | |
2752 | (unsigned long long)bctl->data.target); | |
2753 | ret = -EINVAL; | |
2754 | goto out; | |
2755 | } | |
6728b198 ID |
2756 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2757 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
2758 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
2759 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2760 | "metadata profile %llu\n", | |
2761 | (unsigned long long)bctl->meta.target); | |
2762 | ret = -EINVAL; | |
2763 | goto out; | |
2764 | } | |
6728b198 ID |
2765 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2766 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
2767 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
2768 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2769 | "system profile %llu\n", | |
2770 | (unsigned long long)bctl->sys.target); | |
2771 | ret = -EINVAL; | |
2772 | goto out; | |
2773 | } | |
2774 | ||
e4837f8f ID |
2775 | /* allow dup'ed data chunks only in mixed mode */ |
2776 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 2777 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
2778 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
2779 | ret = -EINVAL; | |
2780 | goto out; | |
2781 | } | |
2782 | ||
2783 | /* allow to reduce meta or sys integrity only if force set */ | |
2784 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2785 | BTRFS_BLOCK_GROUP_RAID10; | |
2786 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2787 | (fs_info->avail_system_alloc_bits & allowed) && | |
2788 | !(bctl->sys.target & allowed)) || | |
2789 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2790 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
2791 | !(bctl->meta.target & allowed))) { | |
2792 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
2793 | printk(KERN_INFO "btrfs: force reducing metadata " | |
2794 | "integrity\n"); | |
2795 | } else { | |
2796 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
2797 | "integrity, use force if you want this\n"); | |
2798 | ret = -EINVAL; | |
2799 | goto out; | |
2800 | } | |
2801 | } | |
2802 | ||
0940ebf6 | 2803 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 2804 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
2805 | goto out; |
2806 | ||
59641015 ID |
2807 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
2808 | BUG_ON(ret == -EEXIST); | |
2809 | set_balance_control(bctl); | |
2810 | } else { | |
2811 | BUG_ON(ret != -EEXIST); | |
2812 | spin_lock(&fs_info->balance_lock); | |
2813 | update_balance_args(bctl); | |
2814 | spin_unlock(&fs_info->balance_lock); | |
2815 | } | |
c9e9f97b | 2816 | |
837d5b6e | 2817 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
2818 | mutex_unlock(&fs_info->balance_mutex); |
2819 | ||
2820 | ret = __btrfs_balance(fs_info); | |
2821 | ||
2822 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 2823 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
2824 | |
2825 | if (bargs) { | |
2826 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 2827 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
2828 | } |
2829 | ||
837d5b6e ID |
2830 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
2831 | balance_need_close(fs_info)) { | |
2832 | __cancel_balance(fs_info); | |
2833 | } | |
2834 | ||
2835 | wake_up(&fs_info->balance_wait_q); | |
c9e9f97b ID |
2836 | |
2837 | return ret; | |
2838 | out: | |
59641015 ID |
2839 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
2840 | __cancel_balance(fs_info); | |
2841 | else | |
2842 | kfree(bctl); | |
2843 | return ret; | |
2844 | } | |
2845 | ||
2846 | static int balance_kthread(void *data) | |
2847 | { | |
2b6ba629 | 2848 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 2849 | int ret = 0; |
59641015 ID |
2850 | |
2851 | mutex_lock(&fs_info->volume_mutex); | |
2852 | mutex_lock(&fs_info->balance_mutex); | |
2853 | ||
2b6ba629 | 2854 | if (fs_info->balance_ctl) { |
9555c6c1 | 2855 | printk(KERN_INFO "btrfs: continuing balance\n"); |
2b6ba629 | 2856 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 2857 | } |
59641015 ID |
2858 | |
2859 | mutex_unlock(&fs_info->balance_mutex); | |
2860 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 2861 | |
59641015 ID |
2862 | return ret; |
2863 | } | |
2864 | ||
2b6ba629 ID |
2865 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
2866 | { | |
2867 | struct task_struct *tsk; | |
2868 | ||
2869 | spin_lock(&fs_info->balance_lock); | |
2870 | if (!fs_info->balance_ctl) { | |
2871 | spin_unlock(&fs_info->balance_lock); | |
2872 | return 0; | |
2873 | } | |
2874 | spin_unlock(&fs_info->balance_lock); | |
2875 | ||
2876 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
2877 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
2878 | return 0; | |
2879 | } | |
2880 | ||
2881 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
2882 | if (IS_ERR(tsk)) | |
2883 | return PTR_ERR(tsk); | |
2884 | ||
2885 | return 0; | |
2886 | } | |
2887 | ||
68310a5e | 2888 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 2889 | { |
59641015 ID |
2890 | struct btrfs_balance_control *bctl; |
2891 | struct btrfs_balance_item *item; | |
2892 | struct btrfs_disk_balance_args disk_bargs; | |
2893 | struct btrfs_path *path; | |
2894 | struct extent_buffer *leaf; | |
2895 | struct btrfs_key key; | |
2896 | int ret; | |
2897 | ||
2898 | path = btrfs_alloc_path(); | |
2899 | if (!path) | |
2900 | return -ENOMEM; | |
2901 | ||
59641015 ID |
2902 | key.objectid = BTRFS_BALANCE_OBJECTID; |
2903 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2904 | key.offset = 0; | |
2905 | ||
68310a5e | 2906 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 2907 | if (ret < 0) |
68310a5e | 2908 | goto out; |
59641015 ID |
2909 | if (ret > 0) { /* ret = -ENOENT; */ |
2910 | ret = 0; | |
68310a5e ID |
2911 | goto out; |
2912 | } | |
2913 | ||
2914 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
2915 | if (!bctl) { | |
2916 | ret = -ENOMEM; | |
2917 | goto out; | |
59641015 ID |
2918 | } |
2919 | ||
2920 | leaf = path->nodes[0]; | |
2921 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2922 | ||
68310a5e ID |
2923 | bctl->fs_info = fs_info; |
2924 | bctl->flags = btrfs_balance_flags(leaf, item); | |
2925 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
2926 | |
2927 | btrfs_balance_data(leaf, item, &disk_bargs); | |
2928 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
2929 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
2930 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
2931 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
2932 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
2933 | ||
68310a5e ID |
2934 | mutex_lock(&fs_info->volume_mutex); |
2935 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 2936 | |
68310a5e ID |
2937 | set_balance_control(bctl); |
2938 | ||
2939 | mutex_unlock(&fs_info->balance_mutex); | |
2940 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
2941 | out: |
2942 | btrfs_free_path(path); | |
ec44a35c CM |
2943 | return ret; |
2944 | } | |
2945 | ||
837d5b6e ID |
2946 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
2947 | { | |
2948 | int ret = 0; | |
2949 | ||
2950 | mutex_lock(&fs_info->balance_mutex); | |
2951 | if (!fs_info->balance_ctl) { | |
2952 | mutex_unlock(&fs_info->balance_mutex); | |
2953 | return -ENOTCONN; | |
2954 | } | |
2955 | ||
2956 | if (atomic_read(&fs_info->balance_running)) { | |
2957 | atomic_inc(&fs_info->balance_pause_req); | |
2958 | mutex_unlock(&fs_info->balance_mutex); | |
2959 | ||
2960 | wait_event(fs_info->balance_wait_q, | |
2961 | atomic_read(&fs_info->balance_running) == 0); | |
2962 | ||
2963 | mutex_lock(&fs_info->balance_mutex); | |
2964 | /* we are good with balance_ctl ripped off from under us */ | |
2965 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
2966 | atomic_dec(&fs_info->balance_pause_req); | |
2967 | } else { | |
2968 | ret = -ENOTCONN; | |
2969 | } | |
2970 | ||
2971 | mutex_unlock(&fs_info->balance_mutex); | |
2972 | return ret; | |
2973 | } | |
2974 | ||
a7e99c69 ID |
2975 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
2976 | { | |
2977 | mutex_lock(&fs_info->balance_mutex); | |
2978 | if (!fs_info->balance_ctl) { | |
2979 | mutex_unlock(&fs_info->balance_mutex); | |
2980 | return -ENOTCONN; | |
2981 | } | |
2982 | ||
2983 | atomic_inc(&fs_info->balance_cancel_req); | |
2984 | /* | |
2985 | * if we are running just wait and return, balance item is | |
2986 | * deleted in btrfs_balance in this case | |
2987 | */ | |
2988 | if (atomic_read(&fs_info->balance_running)) { | |
2989 | mutex_unlock(&fs_info->balance_mutex); | |
2990 | wait_event(fs_info->balance_wait_q, | |
2991 | atomic_read(&fs_info->balance_running) == 0); | |
2992 | mutex_lock(&fs_info->balance_mutex); | |
2993 | } else { | |
2994 | /* __cancel_balance needs volume_mutex */ | |
2995 | mutex_unlock(&fs_info->balance_mutex); | |
2996 | mutex_lock(&fs_info->volume_mutex); | |
2997 | mutex_lock(&fs_info->balance_mutex); | |
2998 | ||
2999 | if (fs_info->balance_ctl) | |
3000 | __cancel_balance(fs_info); | |
3001 | ||
3002 | mutex_unlock(&fs_info->volume_mutex); | |
3003 | } | |
3004 | ||
3005 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3006 | atomic_dec(&fs_info->balance_cancel_req); | |
3007 | mutex_unlock(&fs_info->balance_mutex); | |
3008 | return 0; | |
3009 | } | |
3010 | ||
8f18cf13 CM |
3011 | /* |
3012 | * shrinking a device means finding all of the device extents past | |
3013 | * the new size, and then following the back refs to the chunks. | |
3014 | * The chunk relocation code actually frees the device extent | |
3015 | */ | |
3016 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3017 | { | |
3018 | struct btrfs_trans_handle *trans; | |
3019 | struct btrfs_root *root = device->dev_root; | |
3020 | struct btrfs_dev_extent *dev_extent = NULL; | |
3021 | struct btrfs_path *path; | |
3022 | u64 length; | |
3023 | u64 chunk_tree; | |
3024 | u64 chunk_objectid; | |
3025 | u64 chunk_offset; | |
3026 | int ret; | |
3027 | int slot; | |
ba1bf481 JB |
3028 | int failed = 0; |
3029 | bool retried = false; | |
8f18cf13 CM |
3030 | struct extent_buffer *l; |
3031 | struct btrfs_key key; | |
6c41761f | 3032 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3033 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 3034 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
3035 | u64 diff = device->total_bytes - new_size; |
3036 | ||
2b82032c YZ |
3037 | if (new_size >= device->total_bytes) |
3038 | return -EINVAL; | |
8f18cf13 CM |
3039 | |
3040 | path = btrfs_alloc_path(); | |
3041 | if (!path) | |
3042 | return -ENOMEM; | |
3043 | ||
8f18cf13 CM |
3044 | path->reada = 2; |
3045 | ||
7d9eb12c CM |
3046 | lock_chunks(root); |
3047 | ||
8f18cf13 | 3048 | device->total_bytes = new_size; |
2bf64758 | 3049 | if (device->writeable) { |
2b82032c | 3050 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3051 | spin_lock(&root->fs_info->free_chunk_lock); |
3052 | root->fs_info->free_chunk_space -= diff; | |
3053 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3054 | } | |
7d9eb12c | 3055 | unlock_chunks(root); |
8f18cf13 | 3056 | |
ba1bf481 | 3057 | again: |
8f18cf13 CM |
3058 | key.objectid = device->devid; |
3059 | key.offset = (u64)-1; | |
3060 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3061 | ||
213e64da | 3062 | do { |
8f18cf13 CM |
3063 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3064 | if (ret < 0) | |
3065 | goto done; | |
3066 | ||
3067 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3068 | if (ret < 0) | |
3069 | goto done; | |
3070 | if (ret) { | |
3071 | ret = 0; | |
b3b4aa74 | 3072 | btrfs_release_path(path); |
bf1fb512 | 3073 | break; |
8f18cf13 CM |
3074 | } |
3075 | ||
3076 | l = path->nodes[0]; | |
3077 | slot = path->slots[0]; | |
3078 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3079 | ||
ba1bf481 | 3080 | if (key.objectid != device->devid) { |
b3b4aa74 | 3081 | btrfs_release_path(path); |
bf1fb512 | 3082 | break; |
ba1bf481 | 3083 | } |
8f18cf13 CM |
3084 | |
3085 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3086 | length = btrfs_dev_extent_length(l, dev_extent); | |
3087 | ||
ba1bf481 | 3088 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3089 | btrfs_release_path(path); |
d6397bae | 3090 | break; |
ba1bf481 | 3091 | } |
8f18cf13 CM |
3092 | |
3093 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3094 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3095 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3096 | btrfs_release_path(path); |
8f18cf13 CM |
3097 | |
3098 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3099 | chunk_offset); | |
ba1bf481 | 3100 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3101 | goto done; |
ba1bf481 JB |
3102 | if (ret == -ENOSPC) |
3103 | failed++; | |
213e64da | 3104 | } while (key.offset-- > 0); |
ba1bf481 JB |
3105 | |
3106 | if (failed && !retried) { | |
3107 | failed = 0; | |
3108 | retried = true; | |
3109 | goto again; | |
3110 | } else if (failed && retried) { | |
3111 | ret = -ENOSPC; | |
3112 | lock_chunks(root); | |
3113 | ||
3114 | device->total_bytes = old_size; | |
3115 | if (device->writeable) | |
3116 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3117 | spin_lock(&root->fs_info->free_chunk_lock); |
3118 | root->fs_info->free_chunk_space += diff; | |
3119 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3120 | unlock_chunks(root); |
3121 | goto done; | |
8f18cf13 CM |
3122 | } |
3123 | ||
d6397bae | 3124 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3125 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3126 | if (IS_ERR(trans)) { |
3127 | ret = PTR_ERR(trans); | |
3128 | goto done; | |
3129 | } | |
3130 | ||
d6397bae CB |
3131 | lock_chunks(root); |
3132 | ||
3133 | device->disk_total_bytes = new_size; | |
3134 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3135 | ret = btrfs_update_device(trans, device); | |
3136 | if (ret) { | |
3137 | unlock_chunks(root); | |
3138 | btrfs_end_transaction(trans, root); | |
3139 | goto done; | |
3140 | } | |
3141 | WARN_ON(diff > old_total); | |
3142 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3143 | unlock_chunks(root); | |
3144 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3145 | done: |
3146 | btrfs_free_path(path); | |
3147 | return ret; | |
3148 | } | |
3149 | ||
125ccb0a | 3150 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3151 | struct btrfs_key *key, |
3152 | struct btrfs_chunk *chunk, int item_size) | |
3153 | { | |
6c41761f | 3154 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3155 | struct btrfs_disk_key disk_key; |
3156 | u32 array_size; | |
3157 | u8 *ptr; | |
3158 | ||
3159 | array_size = btrfs_super_sys_array_size(super_copy); | |
3160 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3161 | return -EFBIG; | |
3162 | ||
3163 | ptr = super_copy->sys_chunk_array + array_size; | |
3164 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3165 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3166 | ptr += sizeof(disk_key); | |
3167 | memcpy(ptr, chunk, item_size); | |
3168 | item_size += sizeof(disk_key); | |
3169 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3170 | return 0; | |
3171 | } | |
3172 | ||
73c5de00 AJ |
3173 | /* |
3174 | * sort the devices in descending order by max_avail, total_avail | |
3175 | */ | |
3176 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3177 | { |
73c5de00 AJ |
3178 | const struct btrfs_device_info *di_a = a; |
3179 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3180 | |
73c5de00 | 3181 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3182 | return -1; |
73c5de00 | 3183 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3184 | return 1; |
73c5de00 AJ |
3185 | if (di_a->total_avail > di_b->total_avail) |
3186 | return -1; | |
3187 | if (di_a->total_avail < di_b->total_avail) | |
3188 | return 1; | |
3189 | return 0; | |
b2117a39 | 3190 | } |
0b86a832 | 3191 | |
73c5de00 AJ |
3192 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3193 | struct btrfs_root *extent_root, | |
3194 | struct map_lookup **map_ret, | |
3195 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3196 | u64 start, u64 type) | |
b2117a39 | 3197 | { |
73c5de00 AJ |
3198 | struct btrfs_fs_info *info = extent_root->fs_info; |
3199 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3200 | struct list_head *cur; | |
3201 | struct map_lookup *map = NULL; | |
3202 | struct extent_map_tree *em_tree; | |
3203 | struct extent_map *em; | |
3204 | struct btrfs_device_info *devices_info = NULL; | |
3205 | u64 total_avail; | |
3206 | int num_stripes; /* total number of stripes to allocate */ | |
3207 | int sub_stripes; /* sub_stripes info for map */ | |
3208 | int dev_stripes; /* stripes per dev */ | |
3209 | int devs_max; /* max devs to use */ | |
3210 | int devs_min; /* min devs needed */ | |
3211 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3212 | int ncopies; /* how many copies to data has */ | |
3213 | int ret; | |
3214 | u64 max_stripe_size; | |
3215 | u64 max_chunk_size; | |
3216 | u64 stripe_size; | |
3217 | u64 num_bytes; | |
3218 | int ndevs; | |
3219 | int i; | |
3220 | int j; | |
593060d7 | 3221 | |
0c460c0d | 3222 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3223 | |
73c5de00 AJ |
3224 | if (list_empty(&fs_devices->alloc_list)) |
3225 | return -ENOSPC; | |
b2117a39 | 3226 | |
73c5de00 AJ |
3227 | sub_stripes = 1; |
3228 | dev_stripes = 1; | |
3229 | devs_increment = 1; | |
3230 | ncopies = 1; | |
3231 | devs_max = 0; /* 0 == as many as possible */ | |
3232 | devs_min = 1; | |
3233 | ||
3234 | /* | |
3235 | * define the properties of each RAID type. | |
3236 | * FIXME: move this to a global table and use it in all RAID | |
3237 | * calculation code | |
3238 | */ | |
3239 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
3240 | dev_stripes = 2; | |
b2117a39 | 3241 | ncopies = 2; |
73c5de00 AJ |
3242 | devs_max = 1; |
3243 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
3244 | devs_min = 2; | |
3245 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
3246 | devs_increment = 2; | |
b2117a39 | 3247 | ncopies = 2; |
73c5de00 AJ |
3248 | devs_max = 2; |
3249 | devs_min = 2; | |
3250 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
3251 | sub_stripes = 2; | |
3252 | devs_increment = 2; | |
3253 | ncopies = 2; | |
3254 | devs_min = 4; | |
3255 | } else { | |
3256 | devs_max = 1; | |
3257 | } | |
b2117a39 | 3258 | |
9b3f68b9 | 3259 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3260 | max_stripe_size = 1024 * 1024 * 1024; |
3261 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3262 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3263 | /* for larger filesystems, use larger metadata chunks */ |
3264 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3265 | max_stripe_size = 1024 * 1024 * 1024; | |
3266 | else | |
3267 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3268 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3269 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3270 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3271 | max_chunk_size = 2 * max_stripe_size; |
3272 | } else { | |
3273 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3274 | type); | |
3275 | BUG_ON(1); | |
9b3f68b9 CM |
3276 | } |
3277 | ||
2b82032c YZ |
3278 | /* we don't want a chunk larger than 10% of writeable space */ |
3279 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3280 | max_chunk_size); | |
9b3f68b9 | 3281 | |
73c5de00 AJ |
3282 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3283 | GFP_NOFS); | |
3284 | if (!devices_info) | |
3285 | return -ENOMEM; | |
0cad8a11 | 3286 | |
73c5de00 | 3287 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3288 | |
9f680ce0 | 3289 | /* |
73c5de00 AJ |
3290 | * in the first pass through the devices list, we gather information |
3291 | * about the available holes on each device. | |
9f680ce0 | 3292 | */ |
73c5de00 AJ |
3293 | ndevs = 0; |
3294 | while (cur != &fs_devices->alloc_list) { | |
3295 | struct btrfs_device *device; | |
3296 | u64 max_avail; | |
3297 | u64 dev_offset; | |
b2117a39 | 3298 | |
73c5de00 | 3299 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3300 | |
73c5de00 | 3301 | cur = cur->next; |
b2117a39 | 3302 | |
73c5de00 AJ |
3303 | if (!device->writeable) { |
3304 | printk(KERN_ERR | |
3305 | "btrfs: read-only device in alloc_list\n"); | |
3306 | WARN_ON(1); | |
3307 | continue; | |
3308 | } | |
b2117a39 | 3309 | |
73c5de00 AJ |
3310 | if (!device->in_fs_metadata) |
3311 | continue; | |
b2117a39 | 3312 | |
73c5de00 AJ |
3313 | if (device->total_bytes > device->bytes_used) |
3314 | total_avail = device->total_bytes - device->bytes_used; | |
3315 | else | |
3316 | total_avail = 0; | |
38c01b96 | 3317 | |
3318 | /* If there is no space on this device, skip it. */ | |
3319 | if (total_avail == 0) | |
3320 | continue; | |
b2117a39 | 3321 | |
125ccb0a | 3322 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3323 | max_stripe_size * dev_stripes, |
3324 | &dev_offset, &max_avail); | |
3325 | if (ret && ret != -ENOSPC) | |
3326 | goto error; | |
b2117a39 | 3327 | |
73c5de00 AJ |
3328 | if (ret == 0) |
3329 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3330 | |
73c5de00 AJ |
3331 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3332 | continue; | |
b2117a39 | 3333 | |
73c5de00 AJ |
3334 | devices_info[ndevs].dev_offset = dev_offset; |
3335 | devices_info[ndevs].max_avail = max_avail; | |
3336 | devices_info[ndevs].total_avail = total_avail; | |
3337 | devices_info[ndevs].dev = device; | |
3338 | ++ndevs; | |
3339 | } | |
b2117a39 | 3340 | |
73c5de00 AJ |
3341 | /* |
3342 | * now sort the devices by hole size / available space | |
3343 | */ | |
3344 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3345 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3346 | |
73c5de00 AJ |
3347 | /* round down to number of usable stripes */ |
3348 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3349 | |
73c5de00 AJ |
3350 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3351 | ret = -ENOSPC; | |
3352 | goto error; | |
b2117a39 | 3353 | } |
9f680ce0 | 3354 | |
73c5de00 AJ |
3355 | if (devs_max && ndevs > devs_max) |
3356 | ndevs = devs_max; | |
3357 | /* | |
3358 | * the primary goal is to maximize the number of stripes, so use as many | |
3359 | * devices as possible, even if the stripes are not maximum sized. | |
3360 | */ | |
3361 | stripe_size = devices_info[ndevs-1].max_avail; | |
3362 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3363 | |
37db63a4 | 3364 | if (stripe_size * ndevs > max_chunk_size * ncopies) { |
73c5de00 | 3365 | stripe_size = max_chunk_size * ncopies; |
37db63a4 | 3366 | do_div(stripe_size, ndevs); |
b2117a39 | 3367 | } |
b2117a39 | 3368 | |
73c5de00 | 3369 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
3370 | |
3371 | /* align to BTRFS_STRIPE_LEN */ | |
73c5de00 AJ |
3372 | do_div(stripe_size, BTRFS_STRIPE_LEN); |
3373 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
3374 | |
3375 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3376 | if (!map) { | |
3377 | ret = -ENOMEM; | |
3378 | goto error; | |
3379 | } | |
3380 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3381 | |
73c5de00 AJ |
3382 | for (i = 0; i < ndevs; ++i) { |
3383 | for (j = 0; j < dev_stripes; ++j) { | |
3384 | int s = i * dev_stripes + j; | |
3385 | map->stripes[s].dev = devices_info[i].dev; | |
3386 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3387 | j * stripe_size; | |
6324fbf3 | 3388 | } |
6324fbf3 | 3389 | } |
2b82032c | 3390 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
3391 | map->stripe_len = BTRFS_STRIPE_LEN; |
3392 | map->io_align = BTRFS_STRIPE_LEN; | |
3393 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 3394 | map->type = type; |
2b82032c | 3395 | map->sub_stripes = sub_stripes; |
0b86a832 | 3396 | |
2b82032c | 3397 | *map_ret = map; |
73c5de00 | 3398 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 3399 | |
73c5de00 AJ |
3400 | *stripe_size_out = stripe_size; |
3401 | *num_bytes_out = num_bytes; | |
0b86a832 | 3402 | |
73c5de00 | 3403 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3404 | |
172ddd60 | 3405 | em = alloc_extent_map(); |
2b82032c | 3406 | if (!em) { |
b2117a39 MX |
3407 | ret = -ENOMEM; |
3408 | goto error; | |
593060d7 | 3409 | } |
2b82032c YZ |
3410 | em->bdev = (struct block_device *)map; |
3411 | em->start = start; | |
73c5de00 | 3412 | em->len = num_bytes; |
2b82032c YZ |
3413 | em->block_start = 0; |
3414 | em->block_len = em->len; | |
593060d7 | 3415 | |
2b82032c | 3416 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3417 | write_lock(&em_tree->lock); |
2b82032c | 3418 | ret = add_extent_mapping(em_tree, em); |
890871be | 3419 | write_unlock(&em_tree->lock); |
2b82032c | 3420 | free_extent_map(em); |
1dd4602f MF |
3421 | if (ret) |
3422 | goto error; | |
0b86a832 | 3423 | |
2b82032c YZ |
3424 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
3425 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3426 | start, num_bytes); |
79787eaa JM |
3427 | if (ret) |
3428 | goto error; | |
611f0e00 | 3429 | |
73c5de00 AJ |
3430 | for (i = 0; i < map->num_stripes; ++i) { |
3431 | struct btrfs_device *device; | |
3432 | u64 dev_offset; | |
3433 | ||
3434 | device = map->stripes[i].dev; | |
3435 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3436 | |
3437 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3438 | info->chunk_root->root_key.objectid, |
3439 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3440 | start, dev_offset, stripe_size); |
79787eaa JM |
3441 | if (ret) { |
3442 | btrfs_abort_transaction(trans, extent_root, ret); | |
3443 | goto error; | |
3444 | } | |
2b82032c YZ |
3445 | } |
3446 | ||
b2117a39 | 3447 | kfree(devices_info); |
2b82032c | 3448 | return 0; |
b2117a39 MX |
3449 | |
3450 | error: | |
3451 | kfree(map); | |
3452 | kfree(devices_info); | |
3453 | return ret; | |
2b82032c YZ |
3454 | } |
3455 | ||
3456 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3457 | struct btrfs_root *extent_root, | |
3458 | struct map_lookup *map, u64 chunk_offset, | |
3459 | u64 chunk_size, u64 stripe_size) | |
3460 | { | |
3461 | u64 dev_offset; | |
3462 | struct btrfs_key key; | |
3463 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3464 | struct btrfs_device *device; | |
3465 | struct btrfs_chunk *chunk; | |
3466 | struct btrfs_stripe *stripe; | |
3467 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
3468 | int index = 0; | |
3469 | int ret; | |
3470 | ||
3471 | chunk = kzalloc(item_size, GFP_NOFS); | |
3472 | if (!chunk) | |
3473 | return -ENOMEM; | |
3474 | ||
3475 | index = 0; | |
3476 | while (index < map->num_stripes) { | |
3477 | device = map->stripes[index].dev; | |
3478 | device->bytes_used += stripe_size; | |
0b86a832 | 3479 | ret = btrfs_update_device(trans, device); |
3acd3953 MF |
3480 | if (ret) |
3481 | goto out_free; | |
2b82032c YZ |
3482 | index++; |
3483 | } | |
3484 | ||
2bf64758 JB |
3485 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
3486 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
3487 | map->num_stripes); | |
3488 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
3489 | ||
2b82032c YZ |
3490 | index = 0; |
3491 | stripe = &chunk->stripe; | |
3492 | while (index < map->num_stripes) { | |
3493 | device = map->stripes[index].dev; | |
3494 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 3495 | |
e17cade2 CM |
3496 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
3497 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
3498 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 3499 | stripe++; |
0b86a832 CM |
3500 | index++; |
3501 | } | |
3502 | ||
2b82032c | 3503 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 3504 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
3505 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
3506 | btrfs_set_stack_chunk_type(chunk, map->type); | |
3507 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
3508 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
3509 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 3510 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 3511 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 3512 | |
2b82032c YZ |
3513 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3514 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3515 | key.offset = chunk_offset; | |
0b86a832 | 3516 | |
2b82032c | 3517 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
0b86a832 | 3518 | |
4ed1d16e MF |
3519 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
3520 | /* | |
3521 | * TODO: Cleanup of inserted chunk root in case of | |
3522 | * failure. | |
3523 | */ | |
125ccb0a | 3524 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 3525 | item_size); |
8f18cf13 | 3526 | } |
1abe9b8a | 3527 | |
3acd3953 | 3528 | out_free: |
0b86a832 | 3529 | kfree(chunk); |
4ed1d16e | 3530 | return ret; |
2b82032c | 3531 | } |
0b86a832 | 3532 | |
2b82032c YZ |
3533 | /* |
3534 | * Chunk allocation falls into two parts. The first part does works | |
3535 | * that make the new allocated chunk useable, but not do any operation | |
3536 | * that modifies the chunk tree. The second part does the works that | |
3537 | * require modifying the chunk tree. This division is important for the | |
3538 | * bootstrap process of adding storage to a seed btrfs. | |
3539 | */ | |
3540 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
3541 | struct btrfs_root *extent_root, u64 type) | |
3542 | { | |
3543 | u64 chunk_offset; | |
3544 | u64 chunk_size; | |
3545 | u64 stripe_size; | |
3546 | struct map_lookup *map; | |
3547 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3548 | int ret; | |
3549 | ||
3550 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3551 | &chunk_offset); | |
3552 | if (ret) | |
3553 | return ret; | |
3554 | ||
3555 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3556 | &stripe_size, chunk_offset, type); | |
3557 | if (ret) | |
3558 | return ret; | |
3559 | ||
3560 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3561 | chunk_size, stripe_size); | |
79787eaa JM |
3562 | if (ret) |
3563 | return ret; | |
2b82032c YZ |
3564 | return 0; |
3565 | } | |
3566 | ||
d397712b | 3567 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
3568 | struct btrfs_root *root, |
3569 | struct btrfs_device *device) | |
3570 | { | |
3571 | u64 chunk_offset; | |
3572 | u64 sys_chunk_offset; | |
3573 | u64 chunk_size; | |
3574 | u64 sys_chunk_size; | |
3575 | u64 stripe_size; | |
3576 | u64 sys_stripe_size; | |
3577 | u64 alloc_profile; | |
3578 | struct map_lookup *map; | |
3579 | struct map_lookup *sys_map; | |
3580 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3581 | struct btrfs_root *extent_root = fs_info->extent_root; | |
3582 | int ret; | |
3583 | ||
3584 | ret = find_next_chunk(fs_info->chunk_root, | |
3585 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
3586 | if (ret) |
3587 | return ret; | |
2b82032c YZ |
3588 | |
3589 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
6fef8df1 | 3590 | fs_info->avail_metadata_alloc_bits; |
2b82032c YZ |
3591 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3592 | ||
3593 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3594 | &stripe_size, chunk_offset, alloc_profile); | |
79787eaa JM |
3595 | if (ret) |
3596 | return ret; | |
2b82032c YZ |
3597 | |
3598 | sys_chunk_offset = chunk_offset + chunk_size; | |
3599 | ||
3600 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
6fef8df1 | 3601 | fs_info->avail_system_alloc_bits; |
2b82032c YZ |
3602 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3603 | ||
3604 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
3605 | &sys_chunk_size, &sys_stripe_size, | |
3606 | sys_chunk_offset, alloc_profile); | |
79787eaa JM |
3607 | if (ret) |
3608 | goto abort; | |
2b82032c YZ |
3609 | |
3610 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
79787eaa JM |
3611 | if (ret) |
3612 | goto abort; | |
2b82032c YZ |
3613 | |
3614 | /* | |
3615 | * Modifying chunk tree needs allocating new blocks from both | |
3616 | * system block group and metadata block group. So we only can | |
3617 | * do operations require modifying the chunk tree after both | |
3618 | * block groups were created. | |
3619 | */ | |
3620 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3621 | chunk_size, stripe_size); | |
79787eaa JM |
3622 | if (ret) |
3623 | goto abort; | |
2b82032c YZ |
3624 | |
3625 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
3626 | sys_chunk_offset, sys_chunk_size, | |
3627 | sys_stripe_size); | |
79787eaa JM |
3628 | if (ret) |
3629 | goto abort; | |
3630 | ||
2b82032c | 3631 | return 0; |
79787eaa JM |
3632 | |
3633 | abort: | |
3634 | btrfs_abort_transaction(trans, root, ret); | |
3635 | return ret; | |
2b82032c YZ |
3636 | } |
3637 | ||
3638 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
3639 | { | |
3640 | struct extent_map *em; | |
3641 | struct map_lookup *map; | |
3642 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3643 | int readonly = 0; | |
3644 | int i; | |
3645 | ||
890871be | 3646 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 3647 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 3648 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
3649 | if (!em) |
3650 | return 1; | |
3651 | ||
f48b9075 JB |
3652 | if (btrfs_test_opt(root, DEGRADED)) { |
3653 | free_extent_map(em); | |
3654 | return 0; | |
3655 | } | |
3656 | ||
2b82032c YZ |
3657 | map = (struct map_lookup *)em->bdev; |
3658 | for (i = 0; i < map->num_stripes; i++) { | |
3659 | if (!map->stripes[i].dev->writeable) { | |
3660 | readonly = 1; | |
3661 | break; | |
3662 | } | |
3663 | } | |
0b86a832 | 3664 | free_extent_map(em); |
2b82032c | 3665 | return readonly; |
0b86a832 CM |
3666 | } |
3667 | ||
3668 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
3669 | { | |
a8067e02 | 3670 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
3671 | } |
3672 | ||
3673 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
3674 | { | |
3675 | struct extent_map *em; | |
3676 | ||
d397712b | 3677 | while (1) { |
890871be | 3678 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
3679 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
3680 | if (em) | |
3681 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 3682 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
3683 | if (!em) |
3684 | break; | |
3685 | kfree(em->bdev); | |
3686 | /* once for us */ | |
3687 | free_extent_map(em); | |
3688 | /* once for the tree */ | |
3689 | free_extent_map(em); | |
3690 | } | |
3691 | } | |
3692 | ||
f188591e CM |
3693 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
3694 | { | |
3695 | struct extent_map *em; | |
3696 | struct map_lookup *map; | |
3697 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3698 | int ret; | |
3699 | ||
890871be | 3700 | read_lock(&em_tree->lock); |
f188591e | 3701 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 3702 | read_unlock(&em_tree->lock); |
f188591e CM |
3703 | BUG_ON(!em); |
3704 | ||
3705 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3706 | map = (struct map_lookup *)em->bdev; | |
3707 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
3708 | ret = map->num_stripes; | |
321aecc6 CM |
3709 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
3710 | ret = map->sub_stripes; | |
f188591e CM |
3711 | else |
3712 | ret = 1; | |
3713 | free_extent_map(em); | |
f188591e CM |
3714 | return ret; |
3715 | } | |
3716 | ||
dfe25020 CM |
3717 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
3718 | int optimal) | |
3719 | { | |
3720 | int i; | |
3721 | if (map->stripes[optimal].dev->bdev) | |
3722 | return optimal; | |
3723 | for (i = first; i < first + num; i++) { | |
3724 | if (map->stripes[i].dev->bdev) | |
3725 | return i; | |
3726 | } | |
3727 | /* we couldn't find one that doesn't fail. Just return something | |
3728 | * and the io error handling code will clean up eventually | |
3729 | */ | |
3730 | return optimal; | |
3731 | } | |
3732 | ||
f2d8d74d CM |
3733 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3734 | u64 logical, u64 *length, | |
a1d3c478 | 3735 | struct btrfs_bio **bbio_ret, |
7eaceacc | 3736 | int mirror_num) |
0b86a832 CM |
3737 | { |
3738 | struct extent_map *em; | |
3739 | struct map_lookup *map; | |
3740 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3741 | u64 offset; | |
593060d7 | 3742 | u64 stripe_offset; |
fce3bb9a | 3743 | u64 stripe_end_offset; |
593060d7 | 3744 | u64 stripe_nr; |
fce3bb9a LD |
3745 | u64 stripe_nr_orig; |
3746 | u64 stripe_nr_end; | |
593060d7 | 3747 | int stripe_index; |
cea9e445 | 3748 | int i; |
de11cc12 | 3749 | int ret = 0; |
f2d8d74d | 3750 | int num_stripes; |
a236aed1 | 3751 | int max_errors = 0; |
a1d3c478 | 3752 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3753 | |
890871be | 3754 | read_lock(&em_tree->lock); |
0b86a832 | 3755 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 3756 | read_unlock(&em_tree->lock); |
f2d8d74d | 3757 | |
3b951516 | 3758 | if (!em) { |
d397712b CM |
3759 | printk(KERN_CRIT "unable to find logical %llu len %llu\n", |
3760 | (unsigned long long)logical, | |
3761 | (unsigned long long)*length); | |
f2d8d74d | 3762 | BUG(); |
3b951516 | 3763 | } |
0b86a832 CM |
3764 | |
3765 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3766 | map = (struct map_lookup *)em->bdev; | |
3767 | offset = logical - em->start; | |
593060d7 | 3768 | |
f188591e CM |
3769 | if (mirror_num > map->num_stripes) |
3770 | mirror_num = 0; | |
3771 | ||
593060d7 CM |
3772 | stripe_nr = offset; |
3773 | /* | |
3774 | * stripe_nr counts the total number of stripes we have to stride | |
3775 | * to get to this block | |
3776 | */ | |
3777 | do_div(stripe_nr, map->stripe_len); | |
3778 | ||
3779 | stripe_offset = stripe_nr * map->stripe_len; | |
3780 | BUG_ON(offset < stripe_offset); | |
3781 | ||
3782 | /* stripe_offset is the offset of this block in its stripe*/ | |
3783 | stripe_offset = offset - stripe_offset; | |
3784 | ||
fce3bb9a LD |
3785 | if (rw & REQ_DISCARD) |
3786 | *length = min_t(u64, em->len - offset, *length); | |
52ba6929 | 3787 | else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
cea9e445 CM |
3788 | /* we limit the length of each bio to what fits in a stripe */ |
3789 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 3790 | map->stripe_len - stripe_offset); |
cea9e445 CM |
3791 | } else { |
3792 | *length = em->len - offset; | |
3793 | } | |
f2d8d74d | 3794 | |
a1d3c478 | 3795 | if (!bbio_ret) |
cea9e445 CM |
3796 | goto out; |
3797 | ||
f2d8d74d | 3798 | num_stripes = 1; |
cea9e445 | 3799 | stripe_index = 0; |
fce3bb9a LD |
3800 | stripe_nr_orig = stripe_nr; |
3801 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
3802 | (~(map->stripe_len - 1)); | |
3803 | do_div(stripe_nr_end, map->stripe_len); | |
3804 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
3805 | (offset + *length); | |
3806 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3807 | if (rw & REQ_DISCARD) | |
3808 | num_stripes = min_t(u64, map->num_stripes, | |
3809 | stripe_nr_end - stripe_nr_orig); | |
3810 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3811 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3812 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3813 | num_stripes = map->num_stripes; |
2fff734f | 3814 | else if (mirror_num) |
f188591e | 3815 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3816 | else { |
3817 | stripe_index = find_live_mirror(map, 0, | |
3818 | map->num_stripes, | |
3819 | current->pid % map->num_stripes); | |
a1d3c478 | 3820 | mirror_num = stripe_index + 1; |
dfe25020 | 3821 | } |
2fff734f | 3822 | |
611f0e00 | 3823 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3824 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3825 | num_stripes = map->num_stripes; |
a1d3c478 | 3826 | } else if (mirror_num) { |
f188591e | 3827 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3828 | } else { |
3829 | mirror_num = 1; | |
3830 | } | |
2fff734f | 3831 | |
321aecc6 CM |
3832 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3833 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3834 | |
3835 | stripe_index = do_div(stripe_nr, factor); | |
3836 | stripe_index *= map->sub_stripes; | |
3837 | ||
7eaceacc | 3838 | if (rw & REQ_WRITE) |
f2d8d74d | 3839 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3840 | else if (rw & REQ_DISCARD) |
3841 | num_stripes = min_t(u64, map->sub_stripes * | |
3842 | (stripe_nr_end - stripe_nr_orig), | |
3843 | map->num_stripes); | |
321aecc6 CM |
3844 | else if (mirror_num) |
3845 | stripe_index += mirror_num - 1; | |
dfe25020 | 3846 | else { |
3e74317a | 3847 | int old_stripe_index = stripe_index; |
dfe25020 CM |
3848 | stripe_index = find_live_mirror(map, stripe_index, |
3849 | map->sub_stripes, stripe_index + | |
3850 | current->pid % map->sub_stripes); | |
3e74317a | 3851 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 3852 | } |
8790d502 CM |
3853 | } else { |
3854 | /* | |
3855 | * after this do_div call, stripe_nr is the number of stripes | |
3856 | * on this device we have to walk to find the data, and | |
3857 | * stripe_index is the number of our device in the stripe array | |
3858 | */ | |
3859 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3860 | mirror_num = stripe_index + 1; |
8790d502 | 3861 | } |
593060d7 | 3862 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3863 | |
de11cc12 LZ |
3864 | bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS); |
3865 | if (!bbio) { | |
3866 | ret = -ENOMEM; | |
3867 | goto out; | |
3868 | } | |
3869 | atomic_set(&bbio->error, 0); | |
3870 | ||
fce3bb9a | 3871 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
3872 | int factor = 0; |
3873 | int sub_stripes = 0; | |
3874 | u64 stripes_per_dev = 0; | |
3875 | u32 remaining_stripes = 0; | |
b89203f7 | 3876 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
3877 | |
3878 | if (map->type & | |
3879 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3880 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3881 | sub_stripes = 1; | |
3882 | else | |
3883 | sub_stripes = map->sub_stripes; | |
3884 | ||
3885 | factor = map->num_stripes / sub_stripes; | |
3886 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
3887 | stripe_nr_orig, | |
3888 | factor, | |
3889 | &remaining_stripes); | |
b89203f7 LB |
3890 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
3891 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
3892 | } |
3893 | ||
fce3bb9a | 3894 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 3895 | bbio->stripes[i].physical = |
f2d8d74d CM |
3896 | map->stripes[stripe_index].physical + |
3897 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3898 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 3899 | |
ec9ef7a1 LZ |
3900 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
3901 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3902 | bbio->stripes[i].length = stripes_per_dev * | |
3903 | map->stripe_len; | |
b89203f7 | 3904 | |
ec9ef7a1 LZ |
3905 | if (i / sub_stripes < remaining_stripes) |
3906 | bbio->stripes[i].length += | |
3907 | map->stripe_len; | |
b89203f7 LB |
3908 | |
3909 | /* | |
3910 | * Special for the first stripe and | |
3911 | * the last stripe: | |
3912 | * | |
3913 | * |-------|...|-------| | |
3914 | * |----------| | |
3915 | * off end_off | |
3916 | */ | |
ec9ef7a1 | 3917 | if (i < sub_stripes) |
a1d3c478 | 3918 | bbio->stripes[i].length -= |
fce3bb9a | 3919 | stripe_offset; |
b89203f7 LB |
3920 | |
3921 | if (stripe_index >= last_stripe && | |
3922 | stripe_index <= (last_stripe + | |
3923 | sub_stripes - 1)) | |
a1d3c478 | 3924 | bbio->stripes[i].length -= |
fce3bb9a | 3925 | stripe_end_offset; |
b89203f7 | 3926 | |
ec9ef7a1 LZ |
3927 | if (i == sub_stripes - 1) |
3928 | stripe_offset = 0; | |
fce3bb9a | 3929 | } else |
a1d3c478 | 3930 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
3931 | |
3932 | stripe_index++; | |
3933 | if (stripe_index == map->num_stripes) { | |
3934 | /* This could only happen for RAID0/10 */ | |
3935 | stripe_index = 0; | |
3936 | stripe_nr++; | |
3937 | } | |
3938 | } | |
3939 | } else { | |
3940 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3941 | bbio->stripes[i].physical = |
212a17ab LT |
3942 | map->stripes[stripe_index].physical + |
3943 | stripe_offset + | |
3944 | stripe_nr * map->stripe_len; | |
a1d3c478 | 3945 | bbio->stripes[i].dev = |
212a17ab | 3946 | map->stripes[stripe_index].dev; |
fce3bb9a | 3947 | stripe_index++; |
f2d8d74d | 3948 | } |
593060d7 | 3949 | } |
de11cc12 LZ |
3950 | |
3951 | if (rw & REQ_WRITE) { | |
3952 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
3953 | BTRFS_BLOCK_GROUP_RAID10 | | |
3954 | BTRFS_BLOCK_GROUP_DUP)) { | |
3955 | max_errors = 1; | |
3956 | } | |
f2d8d74d | 3957 | } |
de11cc12 LZ |
3958 | |
3959 | *bbio_ret = bbio; | |
3960 | bbio->num_stripes = num_stripes; | |
3961 | bbio->max_errors = max_errors; | |
3962 | bbio->mirror_num = mirror_num; | |
cea9e445 | 3963 | out: |
0b86a832 | 3964 | free_extent_map(em); |
de11cc12 | 3965 | return ret; |
0b86a832 CM |
3966 | } |
3967 | ||
f2d8d74d CM |
3968 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3969 | u64 logical, u64 *length, | |
a1d3c478 | 3970 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 3971 | { |
a1d3c478 | 3972 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 3973 | mirror_num); |
f2d8d74d CM |
3974 | } |
3975 | ||
a512bbf8 YZ |
3976 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
3977 | u64 chunk_start, u64 physical, u64 devid, | |
3978 | u64 **logical, int *naddrs, int *stripe_len) | |
3979 | { | |
3980 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3981 | struct extent_map *em; | |
3982 | struct map_lookup *map; | |
3983 | u64 *buf; | |
3984 | u64 bytenr; | |
3985 | u64 length; | |
3986 | u64 stripe_nr; | |
3987 | int i, j, nr = 0; | |
3988 | ||
890871be | 3989 | read_lock(&em_tree->lock); |
a512bbf8 | 3990 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 3991 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
3992 | |
3993 | BUG_ON(!em || em->start != chunk_start); | |
3994 | map = (struct map_lookup *)em->bdev; | |
3995 | ||
3996 | length = em->len; | |
3997 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
3998 | do_div(length, map->num_stripes / map->sub_stripes); | |
3999 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4000 | do_div(length, map->num_stripes); | |
4001 | ||
4002 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 4003 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
4004 | |
4005 | for (i = 0; i < map->num_stripes; i++) { | |
4006 | if (devid && map->stripes[i].dev->devid != devid) | |
4007 | continue; | |
4008 | if (map->stripes[i].physical > physical || | |
4009 | map->stripes[i].physical + length <= physical) | |
4010 | continue; | |
4011 | ||
4012 | stripe_nr = physical - map->stripes[i].physical; | |
4013 | do_div(stripe_nr, map->stripe_len); | |
4014 | ||
4015 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
4016 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4017 | do_div(stripe_nr, map->sub_stripes); | |
4018 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
4019 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4020 | } | |
4021 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 4022 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
4023 | for (j = 0; j < nr; j++) { |
4024 | if (buf[j] == bytenr) | |
4025 | break; | |
4026 | } | |
934d375b CM |
4027 | if (j == nr) { |
4028 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 4029 | buf[nr++] = bytenr; |
934d375b | 4030 | } |
a512bbf8 YZ |
4031 | } |
4032 | ||
a512bbf8 YZ |
4033 | *logical = buf; |
4034 | *naddrs = nr; | |
4035 | *stripe_len = map->stripe_len; | |
4036 | ||
4037 | free_extent_map(em); | |
4038 | return 0; | |
f2d8d74d CM |
4039 | } |
4040 | ||
442a4f63 SB |
4041 | static void *merge_stripe_index_into_bio_private(void *bi_private, |
4042 | unsigned int stripe_index) | |
4043 | { | |
4044 | /* | |
4045 | * with single, dup, RAID0, RAID1 and RAID10, stripe_index is | |
4046 | * at most 1. | |
4047 | * The alternative solution (instead of stealing bits from the | |
4048 | * pointer) would be to allocate an intermediate structure | |
4049 | * that contains the old private pointer plus the stripe_index. | |
4050 | */ | |
4051 | BUG_ON((((uintptr_t)bi_private) & 3) != 0); | |
4052 | BUG_ON(stripe_index > 3); | |
4053 | return (void *)(((uintptr_t)bi_private) | stripe_index); | |
4054 | } | |
4055 | ||
4056 | static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private) | |
4057 | { | |
4058 | return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3)); | |
4059 | } | |
4060 | ||
4061 | static unsigned int extract_stripe_index_from_bio_private(void *bi_private) | |
4062 | { | |
4063 | return (unsigned int)((uintptr_t)bi_private) & 3; | |
4064 | } | |
4065 | ||
a1d3c478 | 4066 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 4067 | { |
442a4f63 | 4068 | struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private); |
7d2b4daa | 4069 | int is_orig_bio = 0; |
8790d502 | 4070 | |
442a4f63 | 4071 | if (err) { |
a1d3c478 | 4072 | atomic_inc(&bbio->error); |
442a4f63 SB |
4073 | if (err == -EIO || err == -EREMOTEIO) { |
4074 | unsigned int stripe_index = | |
4075 | extract_stripe_index_from_bio_private( | |
4076 | bio->bi_private); | |
4077 | struct btrfs_device *dev; | |
4078 | ||
4079 | BUG_ON(stripe_index >= bbio->num_stripes); | |
4080 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
4081 | if (dev->bdev) { |
4082 | if (bio->bi_rw & WRITE) | |
4083 | btrfs_dev_stat_inc(dev, | |
4084 | BTRFS_DEV_STAT_WRITE_ERRS); | |
4085 | else | |
4086 | btrfs_dev_stat_inc(dev, | |
4087 | BTRFS_DEV_STAT_READ_ERRS); | |
4088 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
4089 | btrfs_dev_stat_inc(dev, | |
4090 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
4091 | btrfs_dev_stat_print_on_error(dev); | |
4092 | } | |
442a4f63 SB |
4093 | } |
4094 | } | |
8790d502 | 4095 | |
a1d3c478 | 4096 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
4097 | is_orig_bio = 1; |
4098 | ||
a1d3c478 | 4099 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
4100 | if (!is_orig_bio) { |
4101 | bio_put(bio); | |
a1d3c478 | 4102 | bio = bbio->orig_bio; |
7d2b4daa | 4103 | } |
a1d3c478 JS |
4104 | bio->bi_private = bbio->private; |
4105 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
4106 | bio->bi_bdev = (struct block_device *) |
4107 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
4108 | /* only send an error to the higher layers if it is |
4109 | * beyond the tolerance of the multi-bio | |
4110 | */ | |
a1d3c478 | 4111 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 4112 | err = -EIO; |
5dbc8fca | 4113 | } else { |
1259ab75 CM |
4114 | /* |
4115 | * this bio is actually up to date, we didn't | |
4116 | * go over the max number of errors | |
4117 | */ | |
4118 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 4119 | err = 0; |
1259ab75 | 4120 | } |
a1d3c478 | 4121 | kfree(bbio); |
8790d502 CM |
4122 | |
4123 | bio_endio(bio, err); | |
7d2b4daa | 4124 | } else if (!is_orig_bio) { |
8790d502 CM |
4125 | bio_put(bio); |
4126 | } | |
8790d502 CM |
4127 | } |
4128 | ||
8b712842 CM |
4129 | struct async_sched { |
4130 | struct bio *bio; | |
4131 | int rw; | |
4132 | struct btrfs_fs_info *info; | |
4133 | struct btrfs_work work; | |
4134 | }; | |
4135 | ||
4136 | /* | |
4137 | * see run_scheduled_bios for a description of why bios are collected for | |
4138 | * async submit. | |
4139 | * | |
4140 | * This will add one bio to the pending list for a device and make sure | |
4141 | * the work struct is scheduled. | |
4142 | */ | |
143bede5 | 4143 | static noinline void schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
4144 | struct btrfs_device *device, |
4145 | int rw, struct bio *bio) | |
8b712842 CM |
4146 | { |
4147 | int should_queue = 1; | |
ffbd517d | 4148 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
4149 | |
4150 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 4151 | if (!(rw & REQ_WRITE)) { |
492bb6de | 4152 | bio_get(bio); |
21adbd5c | 4153 | btrfsic_submit_bio(rw, bio); |
492bb6de | 4154 | bio_put(bio); |
143bede5 | 4155 | return; |
8b712842 CM |
4156 | } |
4157 | ||
4158 | /* | |
0986fe9e | 4159 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
4160 | * higher layers. Otherwise, the async bio makes it appear we have |
4161 | * made progress against dirty pages when we've really just put it | |
4162 | * on a queue for later | |
4163 | */ | |
0986fe9e | 4164 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 4165 | WARN_ON(bio->bi_next); |
8b712842 CM |
4166 | bio->bi_next = NULL; |
4167 | bio->bi_rw |= rw; | |
4168 | ||
4169 | spin_lock(&device->io_lock); | |
7b6d91da | 4170 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
4171 | pending_bios = &device->pending_sync_bios; |
4172 | else | |
4173 | pending_bios = &device->pending_bios; | |
8b712842 | 4174 | |
ffbd517d CM |
4175 | if (pending_bios->tail) |
4176 | pending_bios->tail->bi_next = bio; | |
8b712842 | 4177 | |
ffbd517d CM |
4178 | pending_bios->tail = bio; |
4179 | if (!pending_bios->head) | |
4180 | pending_bios->head = bio; | |
8b712842 CM |
4181 | if (device->running_pending) |
4182 | should_queue = 0; | |
4183 | ||
4184 | spin_unlock(&device->io_lock); | |
4185 | ||
4186 | if (should_queue) | |
1cc127b5 CM |
4187 | btrfs_queue_worker(&root->fs_info->submit_workers, |
4188 | &device->work); | |
8b712842 CM |
4189 | } |
4190 | ||
f188591e | 4191 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 4192 | int mirror_num, int async_submit) |
0b86a832 CM |
4193 | { |
4194 | struct btrfs_mapping_tree *map_tree; | |
4195 | struct btrfs_device *dev; | |
8790d502 | 4196 | struct bio *first_bio = bio; |
a62b9401 | 4197 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
4198 | u64 length = 0; |
4199 | u64 map_length; | |
0b86a832 | 4200 | int ret; |
8790d502 CM |
4201 | int dev_nr = 0; |
4202 | int total_devs = 1; | |
a1d3c478 | 4203 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 4204 | |
f2d8d74d | 4205 | length = bio->bi_size; |
0b86a832 CM |
4206 | map_tree = &root->fs_info->mapping_tree; |
4207 | map_length = length; | |
cea9e445 | 4208 | |
a1d3c478 | 4209 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 4210 | mirror_num); |
79787eaa JM |
4211 | if (ret) /* -ENOMEM */ |
4212 | return ret; | |
cea9e445 | 4213 | |
a1d3c478 | 4214 | total_devs = bbio->num_stripes; |
cea9e445 | 4215 | if (map_length < length) { |
d397712b CM |
4216 | printk(KERN_CRIT "mapping failed logical %llu bio len %llu " |
4217 | "len %llu\n", (unsigned long long)logical, | |
4218 | (unsigned long long)length, | |
4219 | (unsigned long long)map_length); | |
cea9e445 CM |
4220 | BUG(); |
4221 | } | |
a1d3c478 JS |
4222 | |
4223 | bbio->orig_bio = first_bio; | |
4224 | bbio->private = first_bio->bi_private; | |
4225 | bbio->end_io = first_bio->bi_end_io; | |
4226 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 4227 | |
d397712b | 4228 | while (dev_nr < total_devs) { |
a1d3c478 JS |
4229 | if (dev_nr < total_devs - 1) { |
4230 | bio = bio_clone(first_bio, GFP_NOFS); | |
79787eaa | 4231 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
4232 | } else { |
4233 | bio = first_bio; | |
8790d502 | 4234 | } |
a1d3c478 | 4235 | bio->bi_private = bbio; |
442a4f63 SB |
4236 | bio->bi_private = merge_stripe_index_into_bio_private( |
4237 | bio->bi_private, (unsigned int)dev_nr); | |
a1d3c478 JS |
4238 | bio->bi_end_io = btrfs_end_bio; |
4239 | bio->bi_sector = bbio->stripes[dev_nr].physical >> 9; | |
4240 | dev = bbio->stripes[dev_nr].dev; | |
18e503d6 | 4241 | if (dev && dev->bdev && (rw != WRITE || dev->writeable)) { |
606686ee JB |
4242 | #ifdef DEBUG |
4243 | struct rcu_string *name; | |
4244 | ||
4245 | rcu_read_lock(); | |
4246 | name = rcu_dereference(dev->name); | |
a1d3c478 JS |
4247 | pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu " |
4248 | "(%s id %llu), size=%u\n", rw, | |
4249 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
606686ee JB |
4250 | name->str, dev->devid, bio->bi_size); |
4251 | rcu_read_unlock(); | |
4252 | #endif | |
dfe25020 | 4253 | bio->bi_bdev = dev->bdev; |
8b712842 CM |
4254 | if (async_submit) |
4255 | schedule_bio(root, dev, rw, bio); | |
4256 | else | |
21adbd5c | 4257 | btrfsic_submit_bio(rw, bio); |
dfe25020 CM |
4258 | } else { |
4259 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
4260 | bio->bi_sector = logical >> 9; | |
dfe25020 | 4261 | bio_endio(bio, -EIO); |
dfe25020 | 4262 | } |
8790d502 CM |
4263 | dev_nr++; |
4264 | } | |
0b86a832 CM |
4265 | return 0; |
4266 | } | |
4267 | ||
a443755f | 4268 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 4269 | u8 *uuid, u8 *fsid) |
0b86a832 | 4270 | { |
2b82032c YZ |
4271 | struct btrfs_device *device; |
4272 | struct btrfs_fs_devices *cur_devices; | |
4273 | ||
4274 | cur_devices = root->fs_info->fs_devices; | |
4275 | while (cur_devices) { | |
4276 | if (!fsid || | |
4277 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4278 | device = __find_device(&cur_devices->devices, | |
4279 | devid, uuid); | |
4280 | if (device) | |
4281 | return device; | |
4282 | } | |
4283 | cur_devices = cur_devices->seed; | |
4284 | } | |
4285 | return NULL; | |
0b86a832 CM |
4286 | } |
4287 | ||
dfe25020 CM |
4288 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
4289 | u64 devid, u8 *dev_uuid) | |
4290 | { | |
4291 | struct btrfs_device *device; | |
4292 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4293 | ||
4294 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 4295 | if (!device) |
4296 | return NULL; | |
dfe25020 CM |
4297 | list_add(&device->dev_list, |
4298 | &fs_devices->devices); | |
dfe25020 CM |
4299 | device->dev_root = root->fs_info->dev_root; |
4300 | device->devid = devid; | |
8b712842 | 4301 | device->work.func = pending_bios_fn; |
e4404d6e | 4302 | device->fs_devices = fs_devices; |
cd02dca5 | 4303 | device->missing = 1; |
dfe25020 | 4304 | fs_devices->num_devices++; |
cd02dca5 | 4305 | fs_devices->missing_devices++; |
dfe25020 | 4306 | spin_lock_init(&device->io_lock); |
d20f7043 | 4307 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
4308 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
4309 | return device; | |
4310 | } | |
4311 | ||
0b86a832 CM |
4312 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
4313 | struct extent_buffer *leaf, | |
4314 | struct btrfs_chunk *chunk) | |
4315 | { | |
4316 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4317 | struct map_lookup *map; | |
4318 | struct extent_map *em; | |
4319 | u64 logical; | |
4320 | u64 length; | |
4321 | u64 devid; | |
a443755f | 4322 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 4323 | int num_stripes; |
0b86a832 | 4324 | int ret; |
593060d7 | 4325 | int i; |
0b86a832 | 4326 | |
e17cade2 CM |
4327 | logical = key->offset; |
4328 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 4329 | |
890871be | 4330 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 4331 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 4332 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
4333 | |
4334 | /* already mapped? */ | |
4335 | if (em && em->start <= logical && em->start + em->len > logical) { | |
4336 | free_extent_map(em); | |
0b86a832 CM |
4337 | return 0; |
4338 | } else if (em) { | |
4339 | free_extent_map(em); | |
4340 | } | |
0b86a832 | 4341 | |
172ddd60 | 4342 | em = alloc_extent_map(); |
0b86a832 CM |
4343 | if (!em) |
4344 | return -ENOMEM; | |
593060d7 CM |
4345 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
4346 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
4347 | if (!map) { |
4348 | free_extent_map(em); | |
4349 | return -ENOMEM; | |
4350 | } | |
4351 | ||
4352 | em->bdev = (struct block_device *)map; | |
4353 | em->start = logical; | |
4354 | em->len = length; | |
4355 | em->block_start = 0; | |
c8b97818 | 4356 | em->block_len = em->len; |
0b86a832 | 4357 | |
593060d7 CM |
4358 | map->num_stripes = num_stripes; |
4359 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
4360 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
4361 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
4362 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
4363 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 4364 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
4365 | for (i = 0; i < num_stripes; i++) { |
4366 | map->stripes[i].physical = | |
4367 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
4368 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
4369 | read_extent_buffer(leaf, uuid, (unsigned long) |
4370 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
4371 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4372 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
4373 | NULL); | |
dfe25020 | 4374 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
4375 | kfree(map); |
4376 | free_extent_map(em); | |
4377 | return -EIO; | |
4378 | } | |
dfe25020 CM |
4379 | if (!map->stripes[i].dev) { |
4380 | map->stripes[i].dev = | |
4381 | add_missing_dev(root, devid, uuid); | |
4382 | if (!map->stripes[i].dev) { | |
4383 | kfree(map); | |
4384 | free_extent_map(em); | |
4385 | return -EIO; | |
4386 | } | |
4387 | } | |
4388 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
4389 | } |
4390 | ||
890871be | 4391 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 4392 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 4393 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 4394 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
4395 | free_extent_map(em); |
4396 | ||
4397 | return 0; | |
4398 | } | |
4399 | ||
143bede5 | 4400 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
4401 | struct btrfs_dev_item *dev_item, |
4402 | struct btrfs_device *device) | |
4403 | { | |
4404 | unsigned long ptr; | |
0b86a832 CM |
4405 | |
4406 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
4407 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
4408 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
4409 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
4410 | device->type = btrfs_device_type(leaf, dev_item); | |
4411 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
4412 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
4413 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
4414 | |
4415 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 4416 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
4417 | } |
4418 | ||
2b82032c YZ |
4419 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
4420 | { | |
4421 | struct btrfs_fs_devices *fs_devices; | |
4422 | int ret; | |
4423 | ||
b367e47f | 4424 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
4425 | |
4426 | fs_devices = root->fs_info->fs_devices->seed; | |
4427 | while (fs_devices) { | |
4428 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4429 | ret = 0; | |
4430 | goto out; | |
4431 | } | |
4432 | fs_devices = fs_devices->seed; | |
4433 | } | |
4434 | ||
4435 | fs_devices = find_fsid(fsid); | |
4436 | if (!fs_devices) { | |
4437 | ret = -ENOENT; | |
4438 | goto out; | |
4439 | } | |
e4404d6e YZ |
4440 | |
4441 | fs_devices = clone_fs_devices(fs_devices); | |
4442 | if (IS_ERR(fs_devices)) { | |
4443 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
4444 | goto out; |
4445 | } | |
4446 | ||
97288f2c | 4447 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 4448 | root->fs_info->bdev_holder); |
48d28232 JL |
4449 | if (ret) { |
4450 | free_fs_devices(fs_devices); | |
2b82032c | 4451 | goto out; |
48d28232 | 4452 | } |
2b82032c YZ |
4453 | |
4454 | if (!fs_devices->seeding) { | |
4455 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 4456 | free_fs_devices(fs_devices); |
2b82032c YZ |
4457 | ret = -EINVAL; |
4458 | goto out; | |
4459 | } | |
4460 | ||
4461 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
4462 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 4463 | out: |
2b82032c YZ |
4464 | return ret; |
4465 | } | |
4466 | ||
0d81ba5d | 4467 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
4468 | struct extent_buffer *leaf, |
4469 | struct btrfs_dev_item *dev_item) | |
4470 | { | |
4471 | struct btrfs_device *device; | |
4472 | u64 devid; | |
4473 | int ret; | |
2b82032c | 4474 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
4475 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
4476 | ||
0b86a832 | 4477 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
4478 | read_extent_buffer(leaf, dev_uuid, |
4479 | (unsigned long)btrfs_device_uuid(dev_item), | |
4480 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4481 | read_extent_buffer(leaf, fs_uuid, |
4482 | (unsigned long)btrfs_device_fsid(dev_item), | |
4483 | BTRFS_UUID_SIZE); | |
4484 | ||
4485 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
4486 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 4487 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 4488 | return ret; |
2b82032c YZ |
4489 | } |
4490 | ||
4491 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
4492 | if (!device || !device->bdev) { | |
e4404d6e | 4493 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
4494 | return -EIO; |
4495 | ||
4496 | if (!device) { | |
d397712b CM |
4497 | printk(KERN_WARNING "warning devid %llu missing\n", |
4498 | (unsigned long long)devid); | |
2b82032c YZ |
4499 | device = add_missing_dev(root, devid, dev_uuid); |
4500 | if (!device) | |
4501 | return -ENOMEM; | |
cd02dca5 CM |
4502 | } else if (!device->missing) { |
4503 | /* | |
4504 | * this happens when a device that was properly setup | |
4505 | * in the device info lists suddenly goes bad. | |
4506 | * device->bdev is NULL, and so we have to set | |
4507 | * device->missing to one here | |
4508 | */ | |
4509 | root->fs_info->fs_devices->missing_devices++; | |
4510 | device->missing = 1; | |
2b82032c YZ |
4511 | } |
4512 | } | |
4513 | ||
4514 | if (device->fs_devices != root->fs_info->fs_devices) { | |
4515 | BUG_ON(device->writeable); | |
4516 | if (device->generation != | |
4517 | btrfs_device_generation(leaf, dev_item)) | |
4518 | return -EINVAL; | |
6324fbf3 | 4519 | } |
0b86a832 CM |
4520 | |
4521 | fill_device_from_item(leaf, dev_item, device); | |
4522 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 4523 | device->in_fs_metadata = 1; |
2bf64758 | 4524 | if (device->writeable) { |
2b82032c | 4525 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
4526 | spin_lock(&root->fs_info->free_chunk_lock); |
4527 | root->fs_info->free_chunk_space += device->total_bytes - | |
4528 | device->bytes_used; | |
4529 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4530 | } | |
0b86a832 | 4531 | ret = 0; |
0b86a832 CM |
4532 | return ret; |
4533 | } | |
4534 | ||
e4404d6e | 4535 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 4536 | { |
6c41761f | 4537 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 4538 | struct extent_buffer *sb; |
0b86a832 | 4539 | struct btrfs_disk_key *disk_key; |
0b86a832 | 4540 | struct btrfs_chunk *chunk; |
84eed90f CM |
4541 | u8 *ptr; |
4542 | unsigned long sb_ptr; | |
4543 | int ret = 0; | |
0b86a832 CM |
4544 | u32 num_stripes; |
4545 | u32 array_size; | |
4546 | u32 len = 0; | |
0b86a832 | 4547 | u32 cur; |
84eed90f | 4548 | struct btrfs_key key; |
0b86a832 | 4549 | |
e4404d6e | 4550 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
4551 | BTRFS_SUPER_INFO_SIZE); |
4552 | if (!sb) | |
4553 | return -ENOMEM; | |
4554 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 4555 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
4556 | /* |
4557 | * The sb extent buffer is artifical and just used to read the system array. | |
4558 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
4559 | * pages up-to-date when the page is larger: extent does not cover the | |
4560 | * whole page and consequently check_page_uptodate does not find all | |
4561 | * the page's extents up-to-date (the hole beyond sb), | |
4562 | * write_extent_buffer then triggers a WARN_ON. | |
4563 | * | |
4564 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
4565 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
4566 | * to silence the warning eg. on PowerPC 64. | |
4567 | */ | |
4568 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 4569 | SetPageUptodate(sb->pages[0]); |
4008c04a | 4570 | |
a061fc8d | 4571 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
4572 | array_size = btrfs_super_sys_array_size(super_copy); |
4573 | ||
0b86a832 CM |
4574 | ptr = super_copy->sys_chunk_array; |
4575 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
4576 | cur = 0; | |
4577 | ||
4578 | while (cur < array_size) { | |
4579 | disk_key = (struct btrfs_disk_key *)ptr; | |
4580 | btrfs_disk_key_to_cpu(&key, disk_key); | |
4581 | ||
a061fc8d | 4582 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
4583 | sb_ptr += len; |
4584 | cur += len; | |
4585 | ||
0d81ba5d | 4586 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 4587 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 4588 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
4589 | if (ret) |
4590 | break; | |
0b86a832 CM |
4591 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
4592 | len = btrfs_chunk_item_size(num_stripes); | |
4593 | } else { | |
84eed90f CM |
4594 | ret = -EIO; |
4595 | break; | |
0b86a832 CM |
4596 | } |
4597 | ptr += len; | |
4598 | sb_ptr += len; | |
4599 | cur += len; | |
4600 | } | |
a061fc8d | 4601 | free_extent_buffer(sb); |
84eed90f | 4602 | return ret; |
0b86a832 CM |
4603 | } |
4604 | ||
442a4f63 SB |
4605 | struct btrfs_device *btrfs_find_device_for_logical(struct btrfs_root *root, |
4606 | u64 logical, int mirror_num) | |
4607 | { | |
4608 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4609 | int ret; | |
4610 | u64 map_length = 0; | |
4611 | struct btrfs_bio *bbio = NULL; | |
4612 | struct btrfs_device *device; | |
4613 | ||
4614 | BUG_ON(mirror_num == 0); | |
4615 | ret = btrfs_map_block(map_tree, WRITE, logical, &map_length, &bbio, | |
4616 | mirror_num); | |
4617 | if (ret) { | |
4618 | BUG_ON(bbio != NULL); | |
4619 | return NULL; | |
4620 | } | |
4621 | BUG_ON(mirror_num != bbio->mirror_num); | |
4622 | device = bbio->stripes[mirror_num - 1].dev; | |
4623 | kfree(bbio); | |
4624 | return device; | |
4625 | } | |
4626 | ||
0b86a832 CM |
4627 | int btrfs_read_chunk_tree(struct btrfs_root *root) |
4628 | { | |
4629 | struct btrfs_path *path; | |
4630 | struct extent_buffer *leaf; | |
4631 | struct btrfs_key key; | |
4632 | struct btrfs_key found_key; | |
4633 | int ret; | |
4634 | int slot; | |
4635 | ||
4636 | root = root->fs_info->chunk_root; | |
4637 | ||
4638 | path = btrfs_alloc_path(); | |
4639 | if (!path) | |
4640 | return -ENOMEM; | |
4641 | ||
b367e47f LZ |
4642 | mutex_lock(&uuid_mutex); |
4643 | lock_chunks(root); | |
4644 | ||
0b86a832 CM |
4645 | /* first we search for all of the device items, and then we |
4646 | * read in all of the chunk items. This way we can create chunk | |
4647 | * mappings that reference all of the devices that are afound | |
4648 | */ | |
4649 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
4650 | key.offset = 0; | |
4651 | key.type = 0; | |
4652 | again: | |
4653 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
4654 | if (ret < 0) |
4655 | goto error; | |
d397712b | 4656 | while (1) { |
0b86a832 CM |
4657 | leaf = path->nodes[0]; |
4658 | slot = path->slots[0]; | |
4659 | if (slot >= btrfs_header_nritems(leaf)) { | |
4660 | ret = btrfs_next_leaf(root, path); | |
4661 | if (ret == 0) | |
4662 | continue; | |
4663 | if (ret < 0) | |
4664 | goto error; | |
4665 | break; | |
4666 | } | |
4667 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4668 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4669 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
4670 | break; | |
4671 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
4672 | struct btrfs_dev_item *dev_item; | |
4673 | dev_item = btrfs_item_ptr(leaf, slot, | |
4674 | struct btrfs_dev_item); | |
0d81ba5d | 4675 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
4676 | if (ret) |
4677 | goto error; | |
0b86a832 CM |
4678 | } |
4679 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
4680 | struct btrfs_chunk *chunk; | |
4681 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
4682 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
4683 | if (ret) |
4684 | goto error; | |
0b86a832 CM |
4685 | } |
4686 | path->slots[0]++; | |
4687 | } | |
4688 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4689 | key.objectid = 0; | |
b3b4aa74 | 4690 | btrfs_release_path(path); |
0b86a832 CM |
4691 | goto again; |
4692 | } | |
0b86a832 CM |
4693 | ret = 0; |
4694 | error: | |
b367e47f LZ |
4695 | unlock_chunks(root); |
4696 | mutex_unlock(&uuid_mutex); | |
4697 | ||
2b82032c | 4698 | btrfs_free_path(path); |
0b86a832 CM |
4699 | return ret; |
4700 | } | |
442a4f63 | 4701 | |
733f4fbb SB |
4702 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
4703 | { | |
4704 | int i; | |
4705 | ||
4706 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4707 | btrfs_dev_stat_reset(dev, i); | |
4708 | } | |
4709 | ||
4710 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
4711 | { | |
4712 | struct btrfs_key key; | |
4713 | struct btrfs_key found_key; | |
4714 | struct btrfs_root *dev_root = fs_info->dev_root; | |
4715 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
4716 | struct extent_buffer *eb; | |
4717 | int slot; | |
4718 | int ret = 0; | |
4719 | struct btrfs_device *device; | |
4720 | struct btrfs_path *path = NULL; | |
4721 | int i; | |
4722 | ||
4723 | path = btrfs_alloc_path(); | |
4724 | if (!path) { | |
4725 | ret = -ENOMEM; | |
4726 | goto out; | |
4727 | } | |
4728 | ||
4729 | mutex_lock(&fs_devices->device_list_mutex); | |
4730 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
4731 | int item_size; | |
4732 | struct btrfs_dev_stats_item *ptr; | |
4733 | ||
4734 | key.objectid = 0; | |
4735 | key.type = BTRFS_DEV_STATS_KEY; | |
4736 | key.offset = device->devid; | |
4737 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
4738 | if (ret) { | |
606686ee JB |
4739 | printk_in_rcu(KERN_WARNING "btrfs: no dev_stats entry found for device %s (devid %llu) (OK on first mount after mkfs)\n", |
4740 | rcu_str_deref(device->name), | |
4741 | (unsigned long long)device->devid); | |
733f4fbb SB |
4742 | __btrfs_reset_dev_stats(device); |
4743 | device->dev_stats_valid = 1; | |
4744 | btrfs_release_path(path); | |
4745 | continue; | |
4746 | } | |
4747 | slot = path->slots[0]; | |
4748 | eb = path->nodes[0]; | |
4749 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
4750 | item_size = btrfs_item_size_nr(eb, slot); | |
4751 | ||
4752 | ptr = btrfs_item_ptr(eb, slot, | |
4753 | struct btrfs_dev_stats_item); | |
4754 | ||
4755 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
4756 | if (item_size >= (1 + i) * sizeof(__le64)) | |
4757 | btrfs_dev_stat_set(device, i, | |
4758 | btrfs_dev_stats_value(eb, ptr, i)); | |
4759 | else | |
4760 | btrfs_dev_stat_reset(device, i); | |
4761 | } | |
4762 | ||
4763 | device->dev_stats_valid = 1; | |
4764 | btrfs_dev_stat_print_on_load(device); | |
4765 | btrfs_release_path(path); | |
4766 | } | |
4767 | mutex_unlock(&fs_devices->device_list_mutex); | |
4768 | ||
4769 | out: | |
4770 | btrfs_free_path(path); | |
4771 | return ret < 0 ? ret : 0; | |
4772 | } | |
4773 | ||
4774 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
4775 | struct btrfs_root *dev_root, | |
4776 | struct btrfs_device *device) | |
4777 | { | |
4778 | struct btrfs_path *path; | |
4779 | struct btrfs_key key; | |
4780 | struct extent_buffer *eb; | |
4781 | struct btrfs_dev_stats_item *ptr; | |
4782 | int ret; | |
4783 | int i; | |
4784 | ||
4785 | key.objectid = 0; | |
4786 | key.type = BTRFS_DEV_STATS_KEY; | |
4787 | key.offset = device->devid; | |
4788 | ||
4789 | path = btrfs_alloc_path(); | |
4790 | BUG_ON(!path); | |
4791 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
4792 | if (ret < 0) { | |
606686ee JB |
4793 | printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n", |
4794 | ret, rcu_str_deref(device->name)); | |
733f4fbb SB |
4795 | goto out; |
4796 | } | |
4797 | ||
4798 | if (ret == 0 && | |
4799 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
4800 | /* need to delete old one and insert a new one */ | |
4801 | ret = btrfs_del_item(trans, dev_root, path); | |
4802 | if (ret != 0) { | |
606686ee JB |
4803 | printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n", |
4804 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4805 | goto out; |
4806 | } | |
4807 | ret = 1; | |
4808 | } | |
4809 | ||
4810 | if (ret == 1) { | |
4811 | /* need to insert a new item */ | |
4812 | btrfs_release_path(path); | |
4813 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
4814 | &key, sizeof(*ptr)); | |
4815 | if (ret < 0) { | |
606686ee JB |
4816 | printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n", |
4817 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4818 | goto out; |
4819 | } | |
4820 | } | |
4821 | ||
4822 | eb = path->nodes[0]; | |
4823 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
4824 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4825 | btrfs_set_dev_stats_value(eb, ptr, i, | |
4826 | btrfs_dev_stat_read(device, i)); | |
4827 | btrfs_mark_buffer_dirty(eb); | |
4828 | ||
4829 | out: | |
4830 | btrfs_free_path(path); | |
4831 | return ret; | |
4832 | } | |
4833 | ||
4834 | /* | |
4835 | * called from commit_transaction. Writes all changed device stats to disk. | |
4836 | */ | |
4837 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
4838 | struct btrfs_fs_info *fs_info) | |
4839 | { | |
4840 | struct btrfs_root *dev_root = fs_info->dev_root; | |
4841 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
4842 | struct btrfs_device *device; | |
4843 | int ret = 0; | |
4844 | ||
4845 | mutex_lock(&fs_devices->device_list_mutex); | |
4846 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
4847 | if (!device->dev_stats_valid || !device->dev_stats_dirty) | |
4848 | continue; | |
4849 | ||
4850 | ret = update_dev_stat_item(trans, dev_root, device); | |
4851 | if (!ret) | |
4852 | device->dev_stats_dirty = 0; | |
4853 | } | |
4854 | mutex_unlock(&fs_devices->device_list_mutex); | |
4855 | ||
4856 | return ret; | |
4857 | } | |
4858 | ||
442a4f63 SB |
4859 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
4860 | { | |
4861 | btrfs_dev_stat_inc(dev, index); | |
4862 | btrfs_dev_stat_print_on_error(dev); | |
4863 | } | |
4864 | ||
4865 | void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) | |
4866 | { | |
733f4fbb SB |
4867 | if (!dev->dev_stats_valid) |
4868 | return; | |
606686ee | 4869 | printk_ratelimited_in_rcu(KERN_ERR |
442a4f63 | 4870 | "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
606686ee | 4871 | rcu_str_deref(dev->name), |
442a4f63 SB |
4872 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
4873 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
4874 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
4875 | btrfs_dev_stat_read(dev, | |
4876 | BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
4877 | btrfs_dev_stat_read(dev, | |
4878 | BTRFS_DEV_STAT_GENERATION_ERRS)); | |
4879 | } | |
c11d2c23 | 4880 | |
733f4fbb SB |
4881 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
4882 | { | |
606686ee JB |
4883 | printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
4884 | rcu_str_deref(dev->name), | |
733f4fbb SB |
4885 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
4886 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
4887 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
4888 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
4889 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
4890 | } | |
4891 | ||
c11d2c23 SB |
4892 | int btrfs_get_dev_stats(struct btrfs_root *root, |
4893 | struct btrfs_ioctl_get_dev_stats *stats, | |
4894 | int reset_after_read) | |
4895 | { | |
4896 | struct btrfs_device *dev; | |
4897 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4898 | int i; | |
4899 | ||
4900 | mutex_lock(&fs_devices->device_list_mutex); | |
4901 | dev = btrfs_find_device(root, stats->devid, NULL, NULL); | |
4902 | mutex_unlock(&fs_devices->device_list_mutex); | |
4903 | ||
4904 | if (!dev) { | |
4905 | printk(KERN_WARNING | |
4906 | "btrfs: get dev_stats failed, device not found\n"); | |
4907 | return -ENODEV; | |
733f4fbb SB |
4908 | } else if (!dev->dev_stats_valid) { |
4909 | printk(KERN_WARNING | |
4910 | "btrfs: get dev_stats failed, not yet valid\n"); | |
4911 | return -ENODEV; | |
c11d2c23 SB |
4912 | } else if (reset_after_read) { |
4913 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
4914 | if (stats->nr_items > i) | |
4915 | stats->values[i] = | |
4916 | btrfs_dev_stat_read_and_reset(dev, i); | |
4917 | else | |
4918 | btrfs_dev_stat_reset(dev, i); | |
4919 | } | |
4920 | } else { | |
4921 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4922 | if (stats->nr_items > i) | |
4923 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
4924 | } | |
4925 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
4926 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
4927 | return 0; | |
4928 | } |