]>
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> |
4b4e25f2 | 28 | #include "compat.h" |
0b86a832 CM |
29 | #include "ctree.h" |
30 | #include "extent_map.h" | |
31 | #include "disk-io.h" | |
32 | #include "transaction.h" | |
33 | #include "print-tree.h" | |
34 | #include "volumes.h" | |
8b712842 | 35 | #include "async-thread.h" |
21adbd5c | 36 | #include "check-integrity.h" |
606686ee | 37 | #include "rcu-string.h" |
3fed40cc | 38 | #include "math.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 | 230 | |
66657b31 | 231 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
232 | waitqueue_active(&fs_info->async_submit_wait)) |
233 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
234 | |
235 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 236 | |
2ab1ba68 CM |
237 | /* |
238 | * if we're doing the sync list, record that our | |
239 | * plug has some sync requests on it | |
240 | * | |
241 | * If we're doing the regular list and there are | |
242 | * sync requests sitting around, unplug before | |
243 | * we add more | |
244 | */ | |
245 | if (pending_bios == &device->pending_sync_bios) { | |
246 | sync_pending = 1; | |
247 | } else if (sync_pending) { | |
248 | blk_finish_plug(&plug); | |
249 | blk_start_plug(&plug); | |
250 | sync_pending = 0; | |
251 | } | |
252 | ||
21adbd5c | 253 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
254 | num_run++; |
255 | batch_run++; | |
7eaceacc | 256 | if (need_resched()) |
ffbd517d | 257 | cond_resched(); |
8b712842 CM |
258 | |
259 | /* | |
260 | * we made progress, there is more work to do and the bdi | |
261 | * is now congested. Back off and let other work structs | |
262 | * run instead | |
263 | */ | |
57fd5a5f | 264 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 265 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 266 | struct io_context *ioc; |
8b712842 | 267 | |
b765ead5 CM |
268 | ioc = current->io_context; |
269 | ||
270 | /* | |
271 | * the main goal here is that we don't want to | |
272 | * block if we're going to be able to submit | |
273 | * more requests without blocking. | |
274 | * | |
275 | * This code does two great things, it pokes into | |
276 | * the elevator code from a filesystem _and_ | |
277 | * it makes assumptions about how batching works. | |
278 | */ | |
279 | if (ioc && ioc->nr_batch_requests > 0 && | |
280 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
281 | (last_waited == 0 || | |
282 | ioc->last_waited == last_waited)) { | |
283 | /* | |
284 | * we want to go through our batch of | |
285 | * requests and stop. So, we copy out | |
286 | * the ioc->last_waited time and test | |
287 | * against it before looping | |
288 | */ | |
289 | last_waited = ioc->last_waited; | |
7eaceacc | 290 | if (need_resched()) |
ffbd517d | 291 | cond_resched(); |
b765ead5 CM |
292 | continue; |
293 | } | |
8b712842 | 294 | spin_lock(&device->io_lock); |
ffbd517d | 295 | requeue_list(pending_bios, pending, tail); |
a6837051 | 296 | device->running_pending = 1; |
8b712842 CM |
297 | |
298 | spin_unlock(&device->io_lock); | |
299 | btrfs_requeue_work(&device->work); | |
300 | goto done; | |
301 | } | |
d85c8a6f CM |
302 | /* unplug every 64 requests just for good measure */ |
303 | if (batch_run % 64 == 0) { | |
304 | blk_finish_plug(&plug); | |
305 | blk_start_plug(&plug); | |
306 | sync_pending = 0; | |
307 | } | |
8b712842 | 308 | } |
ffbd517d | 309 | |
51684082 CM |
310 | cond_resched(); |
311 | if (again) | |
312 | goto loop; | |
313 | ||
314 | spin_lock(&device->io_lock); | |
315 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
316 | goto loop_lock; | |
317 | spin_unlock(&device->io_lock); | |
318 | ||
8b712842 | 319 | done: |
211588ad | 320 | blk_finish_plug(&plug); |
8b712842 CM |
321 | } |
322 | ||
b2950863 | 323 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
324 | { |
325 | struct btrfs_device *device; | |
326 | ||
327 | device = container_of(work, struct btrfs_device, work); | |
328 | run_scheduled_bios(device); | |
329 | } | |
330 | ||
a1b32a59 | 331 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
332 | struct btrfs_super_block *disk_super, |
333 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
334 | { | |
335 | struct btrfs_device *device; | |
336 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 337 | struct rcu_string *name; |
8a4b83cc CM |
338 | u64 found_transid = btrfs_super_generation(disk_super); |
339 | ||
340 | fs_devices = find_fsid(disk_super->fsid); | |
341 | if (!fs_devices) { | |
515dc322 | 342 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
343 | if (!fs_devices) |
344 | return -ENOMEM; | |
345 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 346 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
347 | list_add(&fs_devices->list, &fs_uuids); |
348 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
349 | fs_devices->latest_devid = devid; | |
350 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 351 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
352 | device = NULL; |
353 | } else { | |
a443755f CM |
354 | device = __find_device(&fs_devices->devices, devid, |
355 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
356 | } |
357 | if (!device) { | |
2b82032c YZ |
358 | if (fs_devices->opened) |
359 | return -EBUSY; | |
360 | ||
8a4b83cc CM |
361 | device = kzalloc(sizeof(*device), GFP_NOFS); |
362 | if (!device) { | |
363 | /* we can safely leave the fs_devices entry around */ | |
364 | return -ENOMEM; | |
365 | } | |
366 | device->devid = devid; | |
733f4fbb | 367 | device->dev_stats_valid = 0; |
8b712842 | 368 | device->work.func = pending_bios_fn; |
a443755f CM |
369 | memcpy(device->uuid, disk_super->dev_item.uuid, |
370 | BTRFS_UUID_SIZE); | |
b248a415 | 371 | spin_lock_init(&device->io_lock); |
606686ee JB |
372 | |
373 | name = rcu_string_strdup(path, GFP_NOFS); | |
374 | if (!name) { | |
8a4b83cc CM |
375 | kfree(device); |
376 | return -ENOMEM; | |
377 | } | |
606686ee | 378 | rcu_assign_pointer(device->name, name); |
2b82032c | 379 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 380 | |
90519d66 AJ |
381 | /* init readahead state */ |
382 | spin_lock_init(&device->reada_lock); | |
383 | device->reada_curr_zone = NULL; | |
384 | atomic_set(&device->reada_in_flight, 0); | |
385 | device->reada_next = 0; | |
386 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
387 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
388 | ||
e5e9a520 | 389 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 390 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
391 | mutex_unlock(&fs_devices->device_list_mutex); |
392 | ||
2b82032c | 393 | device->fs_devices = fs_devices; |
8a4b83cc | 394 | fs_devices->num_devices++; |
606686ee JB |
395 | } else if (!device->name || strcmp(device->name->str, path)) { |
396 | name = rcu_string_strdup(path, GFP_NOFS); | |
3a0524dc TH |
397 | if (!name) |
398 | return -ENOMEM; | |
606686ee JB |
399 | rcu_string_free(device->name); |
400 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
401 | if (device->missing) { |
402 | fs_devices->missing_devices--; | |
403 | device->missing = 0; | |
404 | } | |
8a4b83cc CM |
405 | } |
406 | ||
407 | if (found_transid > fs_devices->latest_trans) { | |
408 | fs_devices->latest_devid = devid; | |
409 | fs_devices->latest_trans = found_transid; | |
410 | } | |
8a4b83cc CM |
411 | *fs_devices_ret = fs_devices; |
412 | return 0; | |
413 | } | |
414 | ||
e4404d6e YZ |
415 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
416 | { | |
417 | struct btrfs_fs_devices *fs_devices; | |
418 | struct btrfs_device *device; | |
419 | struct btrfs_device *orig_dev; | |
420 | ||
421 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
422 | if (!fs_devices) | |
423 | return ERR_PTR(-ENOMEM); | |
424 | ||
425 | INIT_LIST_HEAD(&fs_devices->devices); | |
426 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
427 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 428 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
429 | fs_devices->latest_devid = orig->latest_devid; |
430 | fs_devices->latest_trans = orig->latest_trans; | |
02db0844 | 431 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
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 */ | |
99f5944b JB |
571 | if (device->name) { |
572 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
573 | BUG_ON(device->name && !name); /* -ENOMEM */ | |
574 | rcu_assign_pointer(new_device->name, name); | |
575 | } | |
1f78160c XG |
576 | new_device->bdev = NULL; |
577 | new_device->writeable = 0; | |
578 | new_device->in_fs_metadata = 0; | |
d5e2003c | 579 | new_device->can_discard = 0; |
1f78160c XG |
580 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
581 | ||
582 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 583 | } |
c9513edb XG |
584 | mutex_unlock(&fs_devices->device_list_mutex); |
585 | ||
e4404d6e YZ |
586 | WARN_ON(fs_devices->open_devices); |
587 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
588 | fs_devices->opened = 0; |
589 | fs_devices->seeding = 0; | |
2b82032c | 590 | |
8a4b83cc CM |
591 | return 0; |
592 | } | |
593 | ||
2b82032c YZ |
594 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
595 | { | |
e4404d6e | 596 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
597 | int ret; |
598 | ||
599 | mutex_lock(&uuid_mutex); | |
600 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
601 | if (!fs_devices->opened) { |
602 | seed_devices = fs_devices->seed; | |
603 | fs_devices->seed = NULL; | |
604 | } | |
2b82032c | 605 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
606 | |
607 | while (seed_devices) { | |
608 | fs_devices = seed_devices; | |
609 | seed_devices = fs_devices->seed; | |
610 | __btrfs_close_devices(fs_devices); | |
611 | free_fs_devices(fs_devices); | |
612 | } | |
2b82032c YZ |
613 | return ret; |
614 | } | |
615 | ||
e4404d6e YZ |
616 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
617 | fmode_t flags, void *holder) | |
8a4b83cc | 618 | { |
d5e2003c | 619 | struct request_queue *q; |
8a4b83cc CM |
620 | struct block_device *bdev; |
621 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 622 | struct btrfs_device *device; |
a0af469b CM |
623 | struct block_device *latest_bdev = NULL; |
624 | struct buffer_head *bh; | |
625 | struct btrfs_super_block *disk_super; | |
626 | u64 latest_devid = 0; | |
627 | u64 latest_transid = 0; | |
a0af469b | 628 | u64 devid; |
2b82032c | 629 | int seeding = 1; |
a0af469b | 630 | int ret = 0; |
8a4b83cc | 631 | |
d4d77629 TH |
632 | flags |= FMODE_EXCL; |
633 | ||
c6e30871 | 634 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
635 | if (device->bdev) |
636 | continue; | |
dfe25020 CM |
637 | if (!device->name) |
638 | continue; | |
639 | ||
606686ee | 640 | bdev = blkdev_get_by_path(device->name->str, flags, holder); |
8a4b83cc | 641 | if (IS_ERR(bdev)) { |
48940662 | 642 | printk(KERN_INFO "btrfs: open %s failed\n", device->name->str); |
a0af469b | 643 | goto error; |
8a4b83cc | 644 | } |
3c4bb26b CM |
645 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
646 | invalidate_bdev(bdev); | |
a061fc8d | 647 | set_blocksize(bdev, 4096); |
a0af469b | 648 | |
a512bbf8 | 649 | bh = btrfs_read_dev_super(bdev); |
20bcd649 | 650 | if (!bh) |
a0af469b CM |
651 | goto error_close; |
652 | ||
653 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 654 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
655 | if (devid != device->devid) |
656 | goto error_brelse; | |
657 | ||
2b82032c YZ |
658 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
659 | BTRFS_UUID_SIZE)) | |
660 | goto error_brelse; | |
661 | ||
662 | device->generation = btrfs_super_generation(disk_super); | |
663 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 664 | latest_devid = devid; |
2b82032c | 665 | latest_transid = device->generation; |
a0af469b CM |
666 | latest_bdev = bdev; |
667 | } | |
668 | ||
2b82032c YZ |
669 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
670 | device->writeable = 0; | |
671 | } else { | |
672 | device->writeable = !bdev_read_only(bdev); | |
673 | seeding = 0; | |
674 | } | |
675 | ||
d5e2003c JB |
676 | q = bdev_get_queue(bdev); |
677 | if (blk_queue_discard(q)) { | |
678 | device->can_discard = 1; | |
679 | fs_devices->num_can_discard++; | |
680 | } | |
681 | ||
8a4b83cc | 682 | device->bdev = bdev; |
dfe25020 | 683 | device->in_fs_metadata = 0; |
15916de8 CM |
684 | device->mode = flags; |
685 | ||
c289811c CM |
686 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
687 | fs_devices->rotating = 1; | |
688 | ||
a0af469b | 689 | fs_devices->open_devices++; |
2b82032c YZ |
690 | if (device->writeable) { |
691 | fs_devices->rw_devices++; | |
692 | list_add(&device->dev_alloc_list, | |
693 | &fs_devices->alloc_list); | |
694 | } | |
4f6c9328 | 695 | brelse(bh); |
a0af469b | 696 | continue; |
a061fc8d | 697 | |
a0af469b CM |
698 | error_brelse: |
699 | brelse(bh); | |
700 | error_close: | |
d4d77629 | 701 | blkdev_put(bdev, flags); |
a0af469b CM |
702 | error: |
703 | continue; | |
8a4b83cc | 704 | } |
a0af469b | 705 | if (fs_devices->open_devices == 0) { |
20bcd649 | 706 | ret = -EINVAL; |
a0af469b CM |
707 | goto out; |
708 | } | |
2b82032c YZ |
709 | fs_devices->seeding = seeding; |
710 | fs_devices->opened = 1; | |
a0af469b CM |
711 | fs_devices->latest_bdev = latest_bdev; |
712 | fs_devices->latest_devid = latest_devid; | |
713 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 714 | fs_devices->total_rw_bytes = 0; |
a0af469b | 715 | out: |
2b82032c YZ |
716 | return ret; |
717 | } | |
718 | ||
719 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 720 | fmode_t flags, void *holder) |
2b82032c YZ |
721 | { |
722 | int ret; | |
723 | ||
724 | mutex_lock(&uuid_mutex); | |
725 | if (fs_devices->opened) { | |
e4404d6e YZ |
726 | fs_devices->opened++; |
727 | ret = 0; | |
2b82032c | 728 | } else { |
15916de8 | 729 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 730 | } |
8a4b83cc | 731 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
732 | return ret; |
733 | } | |
734 | ||
97288f2c | 735 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
736 | struct btrfs_fs_devices **fs_devices_ret) |
737 | { | |
738 | struct btrfs_super_block *disk_super; | |
739 | struct block_device *bdev; | |
740 | struct buffer_head *bh; | |
741 | int ret; | |
742 | u64 devid; | |
f2984462 | 743 | u64 transid; |
02db0844 | 744 | u64 total_devices; |
8a4b83cc | 745 | |
d4d77629 TH |
746 | flags |= FMODE_EXCL; |
747 | bdev = blkdev_get_by_path(path, flags, holder); | |
8a4b83cc CM |
748 | |
749 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
750 | ret = PTR_ERR(bdev); |
751 | goto error; | |
752 | } | |
753 | ||
10f6327b | 754 | mutex_lock(&uuid_mutex); |
8a4b83cc CM |
755 | ret = set_blocksize(bdev, 4096); |
756 | if (ret) | |
757 | goto error_close; | |
a512bbf8 | 758 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 759 | if (!bh) { |
20b45077 | 760 | ret = -EINVAL; |
8a4b83cc CM |
761 | goto error_close; |
762 | } | |
763 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 764 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 765 | transid = btrfs_super_generation(disk_super); |
02db0844 | 766 | total_devices = btrfs_super_num_devices(disk_super); |
d03f918a SB |
767 | if (disk_super->label[0]) { |
768 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
769 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
d397712b | 770 | printk(KERN_INFO "device label %s ", disk_super->label); |
d03f918a | 771 | } else { |
22b63a29 | 772 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); |
d03f918a | 773 | } |
119e10cf | 774 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 775 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc | 776 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
02db0844 JB |
777 | if (!ret && fs_devices_ret) |
778 | (*fs_devices_ret)->total_devices = total_devices; | |
8a4b83cc CM |
779 | brelse(bh); |
780 | error_close: | |
10f6327b | 781 | mutex_unlock(&uuid_mutex); |
d4d77629 | 782 | blkdev_put(bdev, flags); |
8a4b83cc | 783 | error: |
8a4b83cc CM |
784 | return ret; |
785 | } | |
0b86a832 | 786 | |
6d07bcec MX |
787 | /* helper to account the used device space in the range */ |
788 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
789 | u64 end, u64 *length) | |
790 | { | |
791 | struct btrfs_key key; | |
792 | struct btrfs_root *root = device->dev_root; | |
793 | struct btrfs_dev_extent *dev_extent; | |
794 | struct btrfs_path *path; | |
795 | u64 extent_end; | |
796 | int ret; | |
797 | int slot; | |
798 | struct extent_buffer *l; | |
799 | ||
800 | *length = 0; | |
801 | ||
802 | if (start >= device->total_bytes) | |
803 | return 0; | |
804 | ||
805 | path = btrfs_alloc_path(); | |
806 | if (!path) | |
807 | return -ENOMEM; | |
808 | path->reada = 2; | |
809 | ||
810 | key.objectid = device->devid; | |
811 | key.offset = start; | |
812 | key.type = BTRFS_DEV_EXTENT_KEY; | |
813 | ||
814 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
815 | if (ret < 0) | |
816 | goto out; | |
817 | if (ret > 0) { | |
818 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
819 | if (ret < 0) | |
820 | goto out; | |
821 | } | |
822 | ||
823 | while (1) { | |
824 | l = path->nodes[0]; | |
825 | slot = path->slots[0]; | |
826 | if (slot >= btrfs_header_nritems(l)) { | |
827 | ret = btrfs_next_leaf(root, path); | |
828 | if (ret == 0) | |
829 | continue; | |
830 | if (ret < 0) | |
831 | goto out; | |
832 | ||
833 | break; | |
834 | } | |
835 | btrfs_item_key_to_cpu(l, &key, slot); | |
836 | ||
837 | if (key.objectid < device->devid) | |
838 | goto next; | |
839 | ||
840 | if (key.objectid > device->devid) | |
841 | break; | |
842 | ||
843 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
844 | goto next; | |
845 | ||
846 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
847 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
848 | dev_extent); | |
849 | if (key.offset <= start && extent_end > end) { | |
850 | *length = end - start + 1; | |
851 | break; | |
852 | } else if (key.offset <= start && extent_end > start) | |
853 | *length += extent_end - start; | |
854 | else if (key.offset > start && extent_end <= end) | |
855 | *length += extent_end - key.offset; | |
856 | else if (key.offset > start && key.offset <= end) { | |
857 | *length += end - key.offset + 1; | |
858 | break; | |
859 | } else if (key.offset > end) | |
860 | break; | |
861 | ||
862 | next: | |
863 | path->slots[0]++; | |
864 | } | |
865 | ret = 0; | |
866 | out: | |
867 | btrfs_free_path(path); | |
868 | return ret; | |
869 | } | |
870 | ||
0b86a832 | 871 | /* |
7bfc837d | 872 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
873 | * @device: the device which we search the free space in |
874 | * @num_bytes: the size of the free space that we need | |
875 | * @start: store the start of the free space. | |
876 | * @len: the size of the free space. that we find, or the size of the max | |
877 | * free space if we don't find suitable free space | |
878 | * | |
0b86a832 CM |
879 | * this uses a pretty simple search, the expectation is that it is |
880 | * called very infrequently and that a given device has a small number | |
881 | * of extents | |
7bfc837d MX |
882 | * |
883 | * @start is used to store the start of the free space if we find. But if we | |
884 | * don't find suitable free space, it will be used to store the start position | |
885 | * of the max free space. | |
886 | * | |
887 | * @len is used to store the size of the free space that we find. | |
888 | * But if we don't find suitable free space, it is used to store the size of | |
889 | * the max free space. | |
0b86a832 | 890 | */ |
125ccb0a | 891 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 892 | u64 *start, u64 *len) |
0b86a832 CM |
893 | { |
894 | struct btrfs_key key; | |
895 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 896 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 897 | struct btrfs_path *path; |
7bfc837d MX |
898 | u64 hole_size; |
899 | u64 max_hole_start; | |
900 | u64 max_hole_size; | |
901 | u64 extent_end; | |
902 | u64 search_start; | |
0b86a832 CM |
903 | u64 search_end = device->total_bytes; |
904 | int ret; | |
7bfc837d | 905 | int slot; |
0b86a832 CM |
906 | struct extent_buffer *l; |
907 | ||
0b86a832 CM |
908 | /* FIXME use last free of some kind */ |
909 | ||
8a4b83cc CM |
910 | /* we don't want to overwrite the superblock on the drive, |
911 | * so we make sure to start at an offset of at least 1MB | |
912 | */ | |
a9c9bf68 | 913 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 914 | |
7bfc837d MX |
915 | max_hole_start = search_start; |
916 | max_hole_size = 0; | |
38c01b96 | 917 | hole_size = 0; |
7bfc837d MX |
918 | |
919 | if (search_start >= search_end) { | |
920 | ret = -ENOSPC; | |
921 | goto error; | |
922 | } | |
923 | ||
924 | path = btrfs_alloc_path(); | |
925 | if (!path) { | |
926 | ret = -ENOMEM; | |
927 | goto error; | |
928 | } | |
929 | path->reada = 2; | |
930 | ||
0b86a832 CM |
931 | key.objectid = device->devid; |
932 | key.offset = search_start; | |
933 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 934 | |
125ccb0a | 935 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 936 | if (ret < 0) |
7bfc837d | 937 | goto out; |
1fcbac58 YZ |
938 | if (ret > 0) { |
939 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
940 | if (ret < 0) | |
7bfc837d | 941 | goto out; |
1fcbac58 | 942 | } |
7bfc837d | 943 | |
0b86a832 CM |
944 | while (1) { |
945 | l = path->nodes[0]; | |
946 | slot = path->slots[0]; | |
947 | if (slot >= btrfs_header_nritems(l)) { | |
948 | ret = btrfs_next_leaf(root, path); | |
949 | if (ret == 0) | |
950 | continue; | |
951 | if (ret < 0) | |
7bfc837d MX |
952 | goto out; |
953 | ||
954 | break; | |
0b86a832 CM |
955 | } |
956 | btrfs_item_key_to_cpu(l, &key, slot); | |
957 | ||
958 | if (key.objectid < device->devid) | |
959 | goto next; | |
960 | ||
961 | if (key.objectid > device->devid) | |
7bfc837d | 962 | break; |
0b86a832 | 963 | |
7bfc837d MX |
964 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
965 | goto next; | |
9779b72f | 966 | |
7bfc837d MX |
967 | if (key.offset > search_start) { |
968 | hole_size = key.offset - search_start; | |
9779b72f | 969 | |
7bfc837d MX |
970 | if (hole_size > max_hole_size) { |
971 | max_hole_start = search_start; | |
972 | max_hole_size = hole_size; | |
973 | } | |
9779b72f | 974 | |
7bfc837d MX |
975 | /* |
976 | * If this free space is greater than which we need, | |
977 | * it must be the max free space that we have found | |
978 | * until now, so max_hole_start must point to the start | |
979 | * of this free space and the length of this free space | |
980 | * is stored in max_hole_size. Thus, we return | |
981 | * max_hole_start and max_hole_size and go back to the | |
982 | * caller. | |
983 | */ | |
984 | if (hole_size >= num_bytes) { | |
985 | ret = 0; | |
986 | goto out; | |
0b86a832 CM |
987 | } |
988 | } | |
0b86a832 | 989 | |
0b86a832 | 990 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
991 | extent_end = key.offset + btrfs_dev_extent_length(l, |
992 | dev_extent); | |
993 | if (extent_end > search_start) | |
994 | search_start = extent_end; | |
0b86a832 CM |
995 | next: |
996 | path->slots[0]++; | |
997 | cond_resched(); | |
998 | } | |
0b86a832 | 999 | |
38c01b96 | 1000 | /* |
1001 | * At this point, search_start should be the end of | |
1002 | * allocated dev extents, and when shrinking the device, | |
1003 | * search_end may be smaller than search_start. | |
1004 | */ | |
1005 | if (search_end > search_start) | |
1006 | hole_size = search_end - search_start; | |
1007 | ||
7bfc837d MX |
1008 | if (hole_size > max_hole_size) { |
1009 | max_hole_start = search_start; | |
1010 | max_hole_size = hole_size; | |
0b86a832 | 1011 | } |
0b86a832 | 1012 | |
7bfc837d MX |
1013 | /* See above. */ |
1014 | if (hole_size < num_bytes) | |
1015 | ret = -ENOSPC; | |
1016 | else | |
1017 | ret = 0; | |
1018 | ||
1019 | out: | |
2b82032c | 1020 | btrfs_free_path(path); |
7bfc837d MX |
1021 | error: |
1022 | *start = max_hole_start; | |
b2117a39 | 1023 | if (len) |
7bfc837d | 1024 | *len = max_hole_size; |
0b86a832 CM |
1025 | return ret; |
1026 | } | |
1027 | ||
b2950863 | 1028 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1029 | struct btrfs_device *device, |
1030 | u64 start) | |
1031 | { | |
1032 | int ret; | |
1033 | struct btrfs_path *path; | |
1034 | struct btrfs_root *root = device->dev_root; | |
1035 | struct btrfs_key key; | |
a061fc8d CM |
1036 | struct btrfs_key found_key; |
1037 | struct extent_buffer *leaf = NULL; | |
1038 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1039 | |
1040 | path = btrfs_alloc_path(); | |
1041 | if (!path) | |
1042 | return -ENOMEM; | |
1043 | ||
1044 | key.objectid = device->devid; | |
1045 | key.offset = start; | |
1046 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1047 | again: |
8f18cf13 | 1048 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1049 | if (ret > 0) { |
1050 | ret = btrfs_previous_item(root, path, key.objectid, | |
1051 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1052 | if (ret) |
1053 | goto out; | |
a061fc8d CM |
1054 | leaf = path->nodes[0]; |
1055 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1056 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1057 | struct btrfs_dev_extent); | |
1058 | BUG_ON(found_key.offset > start || found_key.offset + | |
1059 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1060 | key = found_key; |
1061 | btrfs_release_path(path); | |
1062 | goto again; | |
a061fc8d CM |
1063 | } else if (ret == 0) { |
1064 | leaf = path->nodes[0]; | |
1065 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1066 | struct btrfs_dev_extent); | |
79787eaa JM |
1067 | } else { |
1068 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1069 | goto out; | |
a061fc8d | 1070 | } |
8f18cf13 | 1071 | |
2bf64758 JB |
1072 | if (device->bytes_used > 0) { |
1073 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1074 | device->bytes_used -= len; | |
1075 | spin_lock(&root->fs_info->free_chunk_lock); | |
1076 | root->fs_info->free_chunk_space += len; | |
1077 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1078 | } | |
8f18cf13 | 1079 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1080 | if (ret) { |
1081 | btrfs_error(root->fs_info, ret, | |
1082 | "Failed to remove dev extent item"); | |
1083 | } | |
b0b802d7 | 1084 | out: |
8f18cf13 CM |
1085 | btrfs_free_path(path); |
1086 | return ret; | |
1087 | } | |
1088 | ||
2b82032c | 1089 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1090 | struct btrfs_device *device, |
e17cade2 | 1091 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1092 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1093 | { |
1094 | int ret; | |
1095 | struct btrfs_path *path; | |
1096 | struct btrfs_root *root = device->dev_root; | |
1097 | struct btrfs_dev_extent *extent; | |
1098 | struct extent_buffer *leaf; | |
1099 | struct btrfs_key key; | |
1100 | ||
dfe25020 | 1101 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1102 | path = btrfs_alloc_path(); |
1103 | if (!path) | |
1104 | return -ENOMEM; | |
1105 | ||
0b86a832 | 1106 | key.objectid = device->devid; |
2b82032c | 1107 | key.offset = start; |
0b86a832 CM |
1108 | key.type = BTRFS_DEV_EXTENT_KEY; |
1109 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1110 | sizeof(*extent)); | |
2cdcecbc MF |
1111 | if (ret) |
1112 | goto out; | |
0b86a832 CM |
1113 | |
1114 | leaf = path->nodes[0]; | |
1115 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1116 | struct btrfs_dev_extent); | |
e17cade2 CM |
1117 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1118 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1119 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1120 | ||
1121 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1122 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1123 | BTRFS_UUID_SIZE); | |
1124 | ||
0b86a832 CM |
1125 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1126 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1127 | out: |
0b86a832 CM |
1128 | btrfs_free_path(path); |
1129 | return ret; | |
1130 | } | |
1131 | ||
a1b32a59 CM |
1132 | static noinline int find_next_chunk(struct btrfs_root *root, |
1133 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1134 | { |
1135 | struct btrfs_path *path; | |
1136 | int ret; | |
1137 | struct btrfs_key key; | |
e17cade2 | 1138 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1139 | struct btrfs_key found_key; |
1140 | ||
1141 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1142 | if (!path) |
1143 | return -ENOMEM; | |
0b86a832 | 1144 | |
e17cade2 | 1145 | key.objectid = objectid; |
0b86a832 CM |
1146 | key.offset = (u64)-1; |
1147 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1148 | ||
1149 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1150 | if (ret < 0) | |
1151 | goto error; | |
1152 | ||
79787eaa | 1153 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1154 | |
1155 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1156 | if (ret) { | |
e17cade2 | 1157 | *offset = 0; |
0b86a832 CM |
1158 | } else { |
1159 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1160 | path->slots[0]); | |
e17cade2 CM |
1161 | if (found_key.objectid != objectid) |
1162 | *offset = 0; | |
1163 | else { | |
1164 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1165 | struct btrfs_chunk); | |
1166 | *offset = found_key.offset + | |
1167 | btrfs_chunk_length(path->nodes[0], chunk); | |
1168 | } | |
0b86a832 CM |
1169 | } |
1170 | ret = 0; | |
1171 | error: | |
1172 | btrfs_free_path(path); | |
1173 | return ret; | |
1174 | } | |
1175 | ||
2b82032c | 1176 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1177 | { |
1178 | int ret; | |
1179 | struct btrfs_key key; | |
1180 | struct btrfs_key found_key; | |
2b82032c YZ |
1181 | struct btrfs_path *path; |
1182 | ||
1183 | root = root->fs_info->chunk_root; | |
1184 | ||
1185 | path = btrfs_alloc_path(); | |
1186 | if (!path) | |
1187 | return -ENOMEM; | |
0b86a832 CM |
1188 | |
1189 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1190 | key.type = BTRFS_DEV_ITEM_KEY; | |
1191 | key.offset = (u64)-1; | |
1192 | ||
1193 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1194 | if (ret < 0) | |
1195 | goto error; | |
1196 | ||
79787eaa | 1197 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1198 | |
1199 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1200 | BTRFS_DEV_ITEM_KEY); | |
1201 | if (ret) { | |
1202 | *objectid = 1; | |
1203 | } else { | |
1204 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1205 | path->slots[0]); | |
1206 | *objectid = found_key.offset + 1; | |
1207 | } | |
1208 | ret = 0; | |
1209 | error: | |
2b82032c | 1210 | btrfs_free_path(path); |
0b86a832 CM |
1211 | return ret; |
1212 | } | |
1213 | ||
1214 | /* | |
1215 | * the device information is stored in the chunk root | |
1216 | * the btrfs_device struct should be fully filled in | |
1217 | */ | |
1218 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1219 | struct btrfs_root *root, | |
1220 | struct btrfs_device *device) | |
1221 | { | |
1222 | int ret; | |
1223 | struct btrfs_path *path; | |
1224 | struct btrfs_dev_item *dev_item; | |
1225 | struct extent_buffer *leaf; | |
1226 | struct btrfs_key key; | |
1227 | unsigned long ptr; | |
0b86a832 CM |
1228 | |
1229 | root = root->fs_info->chunk_root; | |
1230 | ||
1231 | path = btrfs_alloc_path(); | |
1232 | if (!path) | |
1233 | return -ENOMEM; | |
1234 | ||
0b86a832 CM |
1235 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1236 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1237 | key.offset = device->devid; |
0b86a832 CM |
1238 | |
1239 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1240 | sizeof(*dev_item)); |
0b86a832 CM |
1241 | if (ret) |
1242 | goto out; | |
1243 | ||
1244 | leaf = path->nodes[0]; | |
1245 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1246 | ||
1247 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1248 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1249 | btrfs_set_device_type(leaf, dev_item, device->type); |
1250 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1251 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1252 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1253 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1254 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1255 | btrfs_set_device_group(leaf, dev_item, 0); |
1256 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1257 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1258 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1259 | |
0b86a832 | 1260 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1261 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1262 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1263 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1264 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1265 | |
2b82032c | 1266 | ret = 0; |
0b86a832 CM |
1267 | out: |
1268 | btrfs_free_path(path); | |
1269 | return ret; | |
1270 | } | |
8f18cf13 | 1271 | |
a061fc8d CM |
1272 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1273 | struct btrfs_device *device) | |
1274 | { | |
1275 | int ret; | |
1276 | struct btrfs_path *path; | |
a061fc8d | 1277 | struct btrfs_key key; |
a061fc8d CM |
1278 | struct btrfs_trans_handle *trans; |
1279 | ||
1280 | root = root->fs_info->chunk_root; | |
1281 | ||
1282 | path = btrfs_alloc_path(); | |
1283 | if (!path) | |
1284 | return -ENOMEM; | |
1285 | ||
a22285a6 | 1286 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1287 | if (IS_ERR(trans)) { |
1288 | btrfs_free_path(path); | |
1289 | return PTR_ERR(trans); | |
1290 | } | |
a061fc8d CM |
1291 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1292 | key.type = BTRFS_DEV_ITEM_KEY; | |
1293 | key.offset = device->devid; | |
7d9eb12c | 1294 | lock_chunks(root); |
a061fc8d CM |
1295 | |
1296 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1297 | if (ret < 0) | |
1298 | goto out; | |
1299 | ||
1300 | if (ret > 0) { | |
1301 | ret = -ENOENT; | |
1302 | goto out; | |
1303 | } | |
1304 | ||
1305 | ret = btrfs_del_item(trans, root, path); | |
1306 | if (ret) | |
1307 | goto out; | |
a061fc8d CM |
1308 | out: |
1309 | btrfs_free_path(path); | |
7d9eb12c | 1310 | unlock_chunks(root); |
a061fc8d CM |
1311 | btrfs_commit_transaction(trans, root); |
1312 | return ret; | |
1313 | } | |
1314 | ||
1315 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1316 | { | |
1317 | struct btrfs_device *device; | |
2b82032c | 1318 | struct btrfs_device *next_device; |
a061fc8d | 1319 | struct block_device *bdev; |
dfe25020 | 1320 | struct buffer_head *bh = NULL; |
a061fc8d | 1321 | struct btrfs_super_block *disk_super; |
1f78160c | 1322 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1323 | u64 all_avail; |
1324 | u64 devid; | |
2b82032c YZ |
1325 | u64 num_devices; |
1326 | u8 *dev_uuid; | |
a061fc8d | 1327 | int ret = 0; |
1f78160c | 1328 | bool clear_super = false; |
a061fc8d | 1329 | |
a061fc8d CM |
1330 | mutex_lock(&uuid_mutex); |
1331 | ||
1332 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1333 | root->fs_info->avail_system_alloc_bits | | |
1334 | root->fs_info->avail_metadata_alloc_bits; | |
1335 | ||
1336 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1337 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1338 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1339 | "on raid10\n"); | |
a061fc8d CM |
1340 | ret = -EINVAL; |
1341 | goto out; | |
1342 | } | |
1343 | ||
1344 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1345 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1346 | printk(KERN_ERR "btrfs: unable to go below two " |
1347 | "devices on raid1\n"); | |
a061fc8d CM |
1348 | ret = -EINVAL; |
1349 | goto out; | |
1350 | } | |
1351 | ||
dfe25020 | 1352 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1353 | struct list_head *devices; |
1354 | struct btrfs_device *tmp; | |
a061fc8d | 1355 | |
dfe25020 CM |
1356 | device = NULL; |
1357 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1358 | /* |
1359 | * It is safe to read the devices since the volume_mutex | |
1360 | * is held. | |
1361 | */ | |
c6e30871 | 1362 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1363 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1364 | device = tmp; | |
1365 | break; | |
1366 | } | |
1367 | } | |
1368 | bdev = NULL; | |
1369 | bh = NULL; | |
1370 | disk_super = NULL; | |
1371 | if (!device) { | |
d397712b CM |
1372 | printk(KERN_ERR "btrfs: no missing devices found to " |
1373 | "remove\n"); | |
dfe25020 CM |
1374 | goto out; |
1375 | } | |
dfe25020 | 1376 | } else { |
d4d77629 TH |
1377 | bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL, |
1378 | root->fs_info->bdev_holder); | |
dfe25020 CM |
1379 | if (IS_ERR(bdev)) { |
1380 | ret = PTR_ERR(bdev); | |
1381 | goto out; | |
1382 | } | |
a061fc8d | 1383 | |
2b82032c | 1384 | set_blocksize(bdev, 4096); |
3c4bb26b | 1385 | invalidate_bdev(bdev); |
a512bbf8 | 1386 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1387 | if (!bh) { |
20b45077 | 1388 | ret = -EINVAL; |
dfe25020 CM |
1389 | goto error_close; |
1390 | } | |
1391 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1392 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1393 | dev_uuid = disk_super->dev_item.uuid; |
1394 | device = btrfs_find_device(root, devid, dev_uuid, | |
1395 | disk_super->fsid); | |
dfe25020 CM |
1396 | if (!device) { |
1397 | ret = -ENOENT; | |
1398 | goto error_brelse; | |
1399 | } | |
2b82032c | 1400 | } |
dfe25020 | 1401 | |
2b82032c | 1402 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1403 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1404 | "device\n"); | |
2b82032c YZ |
1405 | ret = -EINVAL; |
1406 | goto error_brelse; | |
1407 | } | |
1408 | ||
1409 | if (device->writeable) { | |
0c1daee0 | 1410 | lock_chunks(root); |
2b82032c | 1411 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1412 | unlock_chunks(root); |
2b82032c | 1413 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1414 | clear_super = true; |
dfe25020 | 1415 | } |
a061fc8d CM |
1416 | |
1417 | ret = btrfs_shrink_device(device, 0); | |
1418 | if (ret) | |
9b3517e9 | 1419 | goto error_undo; |
a061fc8d | 1420 | |
a061fc8d CM |
1421 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1422 | if (ret) | |
9b3517e9 | 1423 | goto error_undo; |
a061fc8d | 1424 | |
2bf64758 JB |
1425 | spin_lock(&root->fs_info->free_chunk_lock); |
1426 | root->fs_info->free_chunk_space = device->total_bytes - | |
1427 | device->bytes_used; | |
1428 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1429 | ||
2b82032c | 1430 | device->in_fs_metadata = 0; |
a2de733c | 1431 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1432 | |
1433 | /* | |
1434 | * the device list mutex makes sure that we don't change | |
1435 | * the device list while someone else is writing out all | |
1436 | * the device supers. | |
1437 | */ | |
1f78160c XG |
1438 | |
1439 | cur_devices = device->fs_devices; | |
e5e9a520 | 1440 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1441 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1442 | |
e4404d6e | 1443 | device->fs_devices->num_devices--; |
02db0844 | 1444 | device->fs_devices->total_devices--; |
2b82032c | 1445 | |
cd02dca5 CM |
1446 | if (device->missing) |
1447 | root->fs_info->fs_devices->missing_devices--; | |
1448 | ||
2b82032c YZ |
1449 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1450 | struct btrfs_device, dev_list); | |
1451 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1452 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1453 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1454 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1455 | ||
1f78160c | 1456 | if (device->bdev) |
e4404d6e | 1457 | device->fs_devices->open_devices--; |
1f78160c XG |
1458 | |
1459 | call_rcu(&device->rcu, free_device); | |
1460 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1461 | |
6c41761f DS |
1462 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1463 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1464 | |
1f78160c | 1465 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1466 | struct btrfs_fs_devices *fs_devices; |
1467 | fs_devices = root->fs_info->fs_devices; | |
1468 | while (fs_devices) { | |
1f78160c | 1469 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1470 | break; |
1471 | fs_devices = fs_devices->seed; | |
2b82032c | 1472 | } |
1f78160c XG |
1473 | fs_devices->seed = cur_devices->seed; |
1474 | cur_devices->seed = NULL; | |
0c1daee0 | 1475 | lock_chunks(root); |
1f78160c | 1476 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1477 | unlock_chunks(root); |
1f78160c | 1478 | free_fs_devices(cur_devices); |
2b82032c YZ |
1479 | } |
1480 | ||
5af3e8cc SB |
1481 | root->fs_info->num_tolerated_disk_barrier_failures = |
1482 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1483 | ||
2b82032c YZ |
1484 | /* |
1485 | * at this point, the device is zero sized. We want to | |
1486 | * remove it from the devices list and zero out the old super | |
1487 | */ | |
1f78160c | 1488 | if (clear_super) { |
dfe25020 CM |
1489 | /* make sure this device isn't detected as part of |
1490 | * the FS anymore | |
1491 | */ | |
1492 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1493 | set_buffer_dirty(bh); | |
1494 | sync_dirty_buffer(bh); | |
dfe25020 | 1495 | } |
a061fc8d | 1496 | |
a061fc8d | 1497 | ret = 0; |
a061fc8d CM |
1498 | |
1499 | error_brelse: | |
1500 | brelse(bh); | |
1501 | error_close: | |
dfe25020 | 1502 | if (bdev) |
e525fd89 | 1503 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1504 | out: |
1505 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1506 | return ret; |
9b3517e9 ID |
1507 | error_undo: |
1508 | if (device->writeable) { | |
0c1daee0 | 1509 | lock_chunks(root); |
9b3517e9 ID |
1510 | list_add(&device->dev_alloc_list, |
1511 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1512 | unlock_chunks(root); |
9b3517e9 ID |
1513 | root->fs_info->fs_devices->rw_devices++; |
1514 | } | |
1515 | goto error_brelse; | |
a061fc8d CM |
1516 | } |
1517 | ||
2b82032c YZ |
1518 | /* |
1519 | * does all the dirty work required for changing file system's UUID. | |
1520 | */ | |
125ccb0a | 1521 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1522 | { |
1523 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1524 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1525 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1526 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1527 | struct btrfs_device *device; |
1528 | u64 super_flags; | |
1529 | ||
1530 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1531 | if (!fs_devices->seeding) |
2b82032c YZ |
1532 | return -EINVAL; |
1533 | ||
e4404d6e YZ |
1534 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1535 | if (!seed_devices) | |
2b82032c YZ |
1536 | return -ENOMEM; |
1537 | ||
e4404d6e YZ |
1538 | old_devices = clone_fs_devices(fs_devices); |
1539 | if (IS_ERR(old_devices)) { | |
1540 | kfree(seed_devices); | |
1541 | return PTR_ERR(old_devices); | |
2b82032c | 1542 | } |
e4404d6e | 1543 | |
2b82032c YZ |
1544 | list_add(&old_devices->list, &fs_uuids); |
1545 | ||
e4404d6e YZ |
1546 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1547 | seed_devices->opened = 1; | |
1548 | INIT_LIST_HEAD(&seed_devices->devices); | |
1549 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1550 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1551 | |
1552 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1553 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1554 | synchronize_rcu); | |
c9513edb XG |
1555 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1556 | ||
e4404d6e YZ |
1557 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1558 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1559 | device->fs_devices = seed_devices; | |
1560 | } | |
1561 | ||
2b82032c YZ |
1562 | fs_devices->seeding = 0; |
1563 | fs_devices->num_devices = 0; | |
1564 | fs_devices->open_devices = 0; | |
02db0844 | 1565 | fs_devices->total_devices = 0; |
e4404d6e | 1566 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1567 | |
1568 | generate_random_uuid(fs_devices->fsid); | |
1569 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1570 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1571 | super_flags = btrfs_super_flags(disk_super) & | |
1572 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1573 | btrfs_set_super_flags(disk_super, super_flags); | |
1574 | ||
1575 | return 0; | |
1576 | } | |
1577 | ||
1578 | /* | |
1579 | * strore the expected generation for seed devices in device items. | |
1580 | */ | |
1581 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1582 | struct btrfs_root *root) | |
1583 | { | |
1584 | struct btrfs_path *path; | |
1585 | struct extent_buffer *leaf; | |
1586 | struct btrfs_dev_item *dev_item; | |
1587 | struct btrfs_device *device; | |
1588 | struct btrfs_key key; | |
1589 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1590 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1591 | u64 devid; | |
1592 | int ret; | |
1593 | ||
1594 | path = btrfs_alloc_path(); | |
1595 | if (!path) | |
1596 | return -ENOMEM; | |
1597 | ||
1598 | root = root->fs_info->chunk_root; | |
1599 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1600 | key.offset = 0; | |
1601 | key.type = BTRFS_DEV_ITEM_KEY; | |
1602 | ||
1603 | while (1) { | |
1604 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1605 | if (ret < 0) | |
1606 | goto error; | |
1607 | ||
1608 | leaf = path->nodes[0]; | |
1609 | next_slot: | |
1610 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1611 | ret = btrfs_next_leaf(root, path); | |
1612 | if (ret > 0) | |
1613 | break; | |
1614 | if (ret < 0) | |
1615 | goto error; | |
1616 | leaf = path->nodes[0]; | |
1617 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1618 | btrfs_release_path(path); |
2b82032c YZ |
1619 | continue; |
1620 | } | |
1621 | ||
1622 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1623 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1624 | key.type != BTRFS_DEV_ITEM_KEY) | |
1625 | break; | |
1626 | ||
1627 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1628 | struct btrfs_dev_item); | |
1629 | devid = btrfs_device_id(leaf, dev_item); | |
1630 | read_extent_buffer(leaf, dev_uuid, | |
1631 | (unsigned long)btrfs_device_uuid(dev_item), | |
1632 | BTRFS_UUID_SIZE); | |
1633 | read_extent_buffer(leaf, fs_uuid, | |
1634 | (unsigned long)btrfs_device_fsid(dev_item), | |
1635 | BTRFS_UUID_SIZE); | |
1636 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
79787eaa | 1637 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
1638 | |
1639 | if (device->fs_devices->seeding) { | |
1640 | btrfs_set_device_generation(leaf, dev_item, | |
1641 | device->generation); | |
1642 | btrfs_mark_buffer_dirty(leaf); | |
1643 | } | |
1644 | ||
1645 | path->slots[0]++; | |
1646 | goto next_slot; | |
1647 | } | |
1648 | ret = 0; | |
1649 | error: | |
1650 | btrfs_free_path(path); | |
1651 | return ret; | |
1652 | } | |
1653 | ||
788f20eb CM |
1654 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1655 | { | |
d5e2003c | 1656 | struct request_queue *q; |
788f20eb CM |
1657 | struct btrfs_trans_handle *trans; |
1658 | struct btrfs_device *device; | |
1659 | struct block_device *bdev; | |
788f20eb | 1660 | struct list_head *devices; |
2b82032c | 1661 | struct super_block *sb = root->fs_info->sb; |
606686ee | 1662 | struct rcu_string *name; |
788f20eb | 1663 | u64 total_bytes; |
2b82032c | 1664 | int seeding_dev = 0; |
788f20eb CM |
1665 | int ret = 0; |
1666 | ||
2b82032c | 1667 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 1668 | return -EROFS; |
788f20eb | 1669 | |
a5d16333 | 1670 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1671 | root->fs_info->bdev_holder); |
7f59203a JB |
1672 | if (IS_ERR(bdev)) |
1673 | return PTR_ERR(bdev); | |
a2135011 | 1674 | |
2b82032c YZ |
1675 | if (root->fs_info->fs_devices->seeding) { |
1676 | seeding_dev = 1; | |
1677 | down_write(&sb->s_umount); | |
1678 | mutex_lock(&uuid_mutex); | |
1679 | } | |
1680 | ||
8c8bee1d | 1681 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1682 | |
788f20eb | 1683 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
1684 | |
1685 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 1686 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1687 | if (device->bdev == bdev) { |
1688 | ret = -EEXIST; | |
d25628bd LB |
1689 | mutex_unlock( |
1690 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1691 | goto error; |
788f20eb CM |
1692 | } |
1693 | } | |
d25628bd | 1694 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb CM |
1695 | |
1696 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1697 | if (!device) { | |
1698 | /* we can safely leave the fs_devices entry around */ | |
1699 | ret = -ENOMEM; | |
2b82032c | 1700 | goto error; |
788f20eb CM |
1701 | } |
1702 | ||
606686ee JB |
1703 | name = rcu_string_strdup(device_path, GFP_NOFS); |
1704 | if (!name) { | |
788f20eb | 1705 | kfree(device); |
2b82032c YZ |
1706 | ret = -ENOMEM; |
1707 | goto error; | |
788f20eb | 1708 | } |
606686ee | 1709 | rcu_assign_pointer(device->name, name); |
2b82032c YZ |
1710 | |
1711 | ret = find_next_devid(root, &device->devid); | |
1712 | if (ret) { | |
606686ee | 1713 | rcu_string_free(device->name); |
2b82032c YZ |
1714 | kfree(device); |
1715 | goto error; | |
1716 | } | |
1717 | ||
a22285a6 | 1718 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1719 | if (IS_ERR(trans)) { |
606686ee | 1720 | rcu_string_free(device->name); |
98d5dc13 TI |
1721 | kfree(device); |
1722 | ret = PTR_ERR(trans); | |
1723 | goto error; | |
1724 | } | |
1725 | ||
2b82032c YZ |
1726 | lock_chunks(root); |
1727 | ||
d5e2003c JB |
1728 | q = bdev_get_queue(bdev); |
1729 | if (blk_queue_discard(q)) | |
1730 | device->can_discard = 1; | |
2b82032c YZ |
1731 | device->writeable = 1; |
1732 | device->work.func = pending_bios_fn; | |
1733 | generate_random_uuid(device->uuid); | |
1734 | spin_lock_init(&device->io_lock); | |
1735 | device->generation = trans->transid; | |
788f20eb CM |
1736 | device->io_width = root->sectorsize; |
1737 | device->io_align = root->sectorsize; | |
1738 | device->sector_size = root->sectorsize; | |
1739 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1740 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1741 | device->dev_root = root->fs_info->dev_root; |
1742 | device->bdev = bdev; | |
dfe25020 | 1743 | device->in_fs_metadata = 1; |
fb01aa85 | 1744 | device->mode = FMODE_EXCL; |
2b82032c | 1745 | set_blocksize(device->bdev, 4096); |
788f20eb | 1746 | |
2b82032c YZ |
1747 | if (seeding_dev) { |
1748 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 1749 | ret = btrfs_prepare_sprout(root); |
79787eaa | 1750 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 1751 | } |
788f20eb | 1752 | |
2b82032c | 1753 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 1754 | |
e5e9a520 | 1755 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1756 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1757 | list_add(&device->dev_alloc_list, |
1758 | &root->fs_info->fs_devices->alloc_list); | |
1759 | root->fs_info->fs_devices->num_devices++; | |
1760 | root->fs_info->fs_devices->open_devices++; | |
1761 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 1762 | root->fs_info->fs_devices->total_devices++; |
d5e2003c JB |
1763 | if (device->can_discard) |
1764 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1765 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1766 | |
2bf64758 JB |
1767 | spin_lock(&root->fs_info->free_chunk_lock); |
1768 | root->fs_info->free_chunk_space += device->total_bytes; | |
1769 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1770 | ||
c289811c CM |
1771 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1772 | root->fs_info->fs_devices->rotating = 1; | |
1773 | ||
6c41761f DS |
1774 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1775 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1776 | total_bytes + device->total_bytes); |
1777 | ||
6c41761f DS |
1778 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1779 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1780 | total_bytes + 1); |
e5e9a520 | 1781 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1782 | |
2b82032c YZ |
1783 | if (seeding_dev) { |
1784 | ret = init_first_rw_device(trans, root, device); | |
005d6427 DS |
1785 | if (ret) { |
1786 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1787 | goto error_trans; |
005d6427 | 1788 | } |
2b82032c | 1789 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
1790 | if (ret) { |
1791 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1792 | goto error_trans; |
005d6427 | 1793 | } |
2b82032c YZ |
1794 | } else { |
1795 | ret = btrfs_add_device(trans, root, device); | |
005d6427 DS |
1796 | if (ret) { |
1797 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1798 | goto error_trans; |
005d6427 | 1799 | } |
2b82032c YZ |
1800 | } |
1801 | ||
913d952e CM |
1802 | /* |
1803 | * we've got more storage, clear any full flags on the space | |
1804 | * infos | |
1805 | */ | |
1806 | btrfs_clear_space_info_full(root->fs_info); | |
1807 | ||
7d9eb12c | 1808 | unlock_chunks(root); |
5af3e8cc SB |
1809 | root->fs_info->num_tolerated_disk_barrier_failures = |
1810 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 1811 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 1812 | |
2b82032c YZ |
1813 | if (seeding_dev) { |
1814 | mutex_unlock(&uuid_mutex); | |
1815 | up_write(&sb->s_umount); | |
788f20eb | 1816 | |
79787eaa JM |
1817 | if (ret) /* transaction commit */ |
1818 | return ret; | |
1819 | ||
2b82032c | 1820 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
1821 | if (ret < 0) |
1822 | btrfs_error(root->fs_info, ret, | |
1823 | "Failed to relocate sys chunks after " | |
1824 | "device initialization. This can be fixed " | |
1825 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
1826 | trans = btrfs_attach_transaction(root); |
1827 | if (IS_ERR(trans)) { | |
1828 | if (PTR_ERR(trans) == -ENOENT) | |
1829 | return 0; | |
1830 | return PTR_ERR(trans); | |
1831 | } | |
1832 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 1833 | } |
c9e9f97b | 1834 | |
2b82032c | 1835 | return ret; |
79787eaa JM |
1836 | |
1837 | error_trans: | |
1838 | unlock_chunks(root); | |
79787eaa | 1839 | btrfs_end_transaction(trans, root); |
606686ee | 1840 | rcu_string_free(device->name); |
79787eaa | 1841 | kfree(device); |
2b82032c | 1842 | error: |
e525fd89 | 1843 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1844 | if (seeding_dev) { |
1845 | mutex_unlock(&uuid_mutex); | |
1846 | up_write(&sb->s_umount); | |
1847 | } | |
c9e9f97b | 1848 | return ret; |
788f20eb CM |
1849 | } |
1850 | ||
d397712b CM |
1851 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1852 | struct btrfs_device *device) | |
0b86a832 CM |
1853 | { |
1854 | int ret; | |
1855 | struct btrfs_path *path; | |
1856 | struct btrfs_root *root; | |
1857 | struct btrfs_dev_item *dev_item; | |
1858 | struct extent_buffer *leaf; | |
1859 | struct btrfs_key key; | |
1860 | ||
1861 | root = device->dev_root->fs_info->chunk_root; | |
1862 | ||
1863 | path = btrfs_alloc_path(); | |
1864 | if (!path) | |
1865 | return -ENOMEM; | |
1866 | ||
1867 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1868 | key.type = BTRFS_DEV_ITEM_KEY; | |
1869 | key.offset = device->devid; | |
1870 | ||
1871 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1872 | if (ret < 0) | |
1873 | goto out; | |
1874 | ||
1875 | if (ret > 0) { | |
1876 | ret = -ENOENT; | |
1877 | goto out; | |
1878 | } | |
1879 | ||
1880 | leaf = path->nodes[0]; | |
1881 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1882 | ||
1883 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1884 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1885 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1886 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1887 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1888 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1889 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1890 | btrfs_mark_buffer_dirty(leaf); | |
1891 | ||
1892 | out: | |
1893 | btrfs_free_path(path); | |
1894 | return ret; | |
1895 | } | |
1896 | ||
7d9eb12c | 1897 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1898 | struct btrfs_device *device, u64 new_size) |
1899 | { | |
1900 | struct btrfs_super_block *super_copy = | |
6c41761f | 1901 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1902 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1903 | u64 diff = new_size - device->total_bytes; | |
1904 | ||
2b82032c YZ |
1905 | if (!device->writeable) |
1906 | return -EACCES; | |
1907 | if (new_size <= device->total_bytes) | |
1908 | return -EINVAL; | |
1909 | ||
8f18cf13 | 1910 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1911 | device->fs_devices->total_rw_bytes += diff; |
1912 | ||
1913 | device->total_bytes = new_size; | |
9779b72f | 1914 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1915 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1916 | ||
8f18cf13 CM |
1917 | return btrfs_update_device(trans, device); |
1918 | } | |
1919 | ||
7d9eb12c CM |
1920 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1921 | struct btrfs_device *device, u64 new_size) | |
1922 | { | |
1923 | int ret; | |
1924 | lock_chunks(device->dev_root); | |
1925 | ret = __btrfs_grow_device(trans, device, new_size); | |
1926 | unlock_chunks(device->dev_root); | |
1927 | return ret; | |
1928 | } | |
1929 | ||
8f18cf13 CM |
1930 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1931 | struct btrfs_root *root, | |
1932 | u64 chunk_tree, u64 chunk_objectid, | |
1933 | u64 chunk_offset) | |
1934 | { | |
1935 | int ret; | |
1936 | struct btrfs_path *path; | |
1937 | struct btrfs_key key; | |
1938 | ||
1939 | root = root->fs_info->chunk_root; | |
1940 | path = btrfs_alloc_path(); | |
1941 | if (!path) | |
1942 | return -ENOMEM; | |
1943 | ||
1944 | key.objectid = chunk_objectid; | |
1945 | key.offset = chunk_offset; | |
1946 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1947 | ||
1948 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
1949 | if (ret < 0) |
1950 | goto out; | |
1951 | else if (ret > 0) { /* Logic error or corruption */ | |
1952 | btrfs_error(root->fs_info, -ENOENT, | |
1953 | "Failed lookup while freeing chunk."); | |
1954 | ret = -ENOENT; | |
1955 | goto out; | |
1956 | } | |
8f18cf13 CM |
1957 | |
1958 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
1959 | if (ret < 0) |
1960 | btrfs_error(root->fs_info, ret, | |
1961 | "Failed to delete chunk item."); | |
1962 | out: | |
8f18cf13 | 1963 | btrfs_free_path(path); |
65a246c5 | 1964 | return ret; |
8f18cf13 CM |
1965 | } |
1966 | ||
b2950863 | 1967 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1968 | chunk_offset) |
1969 | { | |
6c41761f | 1970 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
1971 | struct btrfs_disk_key *disk_key; |
1972 | struct btrfs_chunk *chunk; | |
1973 | u8 *ptr; | |
1974 | int ret = 0; | |
1975 | u32 num_stripes; | |
1976 | u32 array_size; | |
1977 | u32 len = 0; | |
1978 | u32 cur; | |
1979 | struct btrfs_key key; | |
1980 | ||
1981 | array_size = btrfs_super_sys_array_size(super_copy); | |
1982 | ||
1983 | ptr = super_copy->sys_chunk_array; | |
1984 | cur = 0; | |
1985 | ||
1986 | while (cur < array_size) { | |
1987 | disk_key = (struct btrfs_disk_key *)ptr; | |
1988 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1989 | ||
1990 | len = sizeof(*disk_key); | |
1991 | ||
1992 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1993 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1994 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1995 | len += btrfs_chunk_item_size(num_stripes); | |
1996 | } else { | |
1997 | ret = -EIO; | |
1998 | break; | |
1999 | } | |
2000 | if (key.objectid == chunk_objectid && | |
2001 | key.offset == chunk_offset) { | |
2002 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2003 | array_size -= len; | |
2004 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2005 | } else { | |
2006 | ptr += len; | |
2007 | cur += len; | |
2008 | } | |
2009 | } | |
2010 | return ret; | |
2011 | } | |
2012 | ||
b2950863 | 2013 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
2014 | u64 chunk_tree, u64 chunk_objectid, |
2015 | u64 chunk_offset) | |
2016 | { | |
2017 | struct extent_map_tree *em_tree; | |
2018 | struct btrfs_root *extent_root; | |
2019 | struct btrfs_trans_handle *trans; | |
2020 | struct extent_map *em; | |
2021 | struct map_lookup *map; | |
2022 | int ret; | |
2023 | int i; | |
2024 | ||
2025 | root = root->fs_info->chunk_root; | |
2026 | extent_root = root->fs_info->extent_root; | |
2027 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2028 | ||
ba1bf481 JB |
2029 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2030 | if (ret) | |
2031 | return -ENOSPC; | |
2032 | ||
8f18cf13 | 2033 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2034 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2035 | if (ret) |
2036 | return ret; | |
8f18cf13 | 2037 | |
a22285a6 | 2038 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2039 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 2040 | |
7d9eb12c CM |
2041 | lock_chunks(root); |
2042 | ||
8f18cf13 CM |
2043 | /* |
2044 | * step two, delete the device extents and the | |
2045 | * chunk tree entries | |
2046 | */ | |
890871be | 2047 | read_lock(&em_tree->lock); |
8f18cf13 | 2048 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2049 | read_unlock(&em_tree->lock); |
8f18cf13 | 2050 | |
285190d9 | 2051 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2052 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2053 | map = (struct map_lookup *)em->bdev; |
2054 | ||
2055 | for (i = 0; i < map->num_stripes; i++) { | |
2056 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
2057 | map->stripes[i].physical); | |
2058 | BUG_ON(ret); | |
a061fc8d | 2059 | |
dfe25020 CM |
2060 | if (map->stripes[i].dev) { |
2061 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2062 | BUG_ON(ret); | |
2063 | } | |
8f18cf13 CM |
2064 | } |
2065 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2066 | chunk_offset); | |
2067 | ||
2068 | BUG_ON(ret); | |
2069 | ||
1abe9b8a | 2070 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2071 | ||
8f18cf13 CM |
2072 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2073 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2074 | BUG_ON(ret); | |
8f18cf13 CM |
2075 | } |
2076 | ||
2b82032c YZ |
2077 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2078 | BUG_ON(ret); | |
2079 | ||
890871be | 2080 | write_lock(&em_tree->lock); |
2b82032c | 2081 | remove_extent_mapping(em_tree, em); |
890871be | 2082 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2083 | |
2084 | kfree(map); | |
2085 | em->bdev = NULL; | |
2086 | ||
2087 | /* once for the tree */ | |
2088 | free_extent_map(em); | |
2089 | /* once for us */ | |
2090 | free_extent_map(em); | |
2091 | ||
2092 | unlock_chunks(root); | |
2093 | btrfs_end_transaction(trans, root); | |
2094 | return 0; | |
2095 | } | |
2096 | ||
2097 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2098 | { | |
2099 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2100 | struct btrfs_path *path; | |
2101 | struct extent_buffer *leaf; | |
2102 | struct btrfs_chunk *chunk; | |
2103 | struct btrfs_key key; | |
2104 | struct btrfs_key found_key; | |
2105 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2106 | u64 chunk_type; | |
ba1bf481 JB |
2107 | bool retried = false; |
2108 | int failed = 0; | |
2b82032c YZ |
2109 | int ret; |
2110 | ||
2111 | path = btrfs_alloc_path(); | |
2112 | if (!path) | |
2113 | return -ENOMEM; | |
2114 | ||
ba1bf481 | 2115 | again: |
2b82032c YZ |
2116 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2117 | key.offset = (u64)-1; | |
2118 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2119 | ||
2120 | while (1) { | |
2121 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2122 | if (ret < 0) | |
2123 | goto error; | |
79787eaa | 2124 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2125 | |
2126 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2127 | key.type); | |
2128 | if (ret < 0) | |
2129 | goto error; | |
2130 | if (ret > 0) | |
2131 | break; | |
1a40e23b | 2132 | |
2b82032c YZ |
2133 | leaf = path->nodes[0]; |
2134 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2135 | |
2b82032c YZ |
2136 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2137 | struct btrfs_chunk); | |
2138 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2139 | btrfs_release_path(path); |
8f18cf13 | 2140 | |
2b82032c YZ |
2141 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2142 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2143 | found_key.objectid, | |
2144 | found_key.offset); | |
ba1bf481 JB |
2145 | if (ret == -ENOSPC) |
2146 | failed++; | |
2147 | else if (ret) | |
2148 | BUG(); | |
2b82032c | 2149 | } |
8f18cf13 | 2150 | |
2b82032c YZ |
2151 | if (found_key.offset == 0) |
2152 | break; | |
2153 | key.offset = found_key.offset - 1; | |
2154 | } | |
2155 | ret = 0; | |
ba1bf481 JB |
2156 | if (failed && !retried) { |
2157 | failed = 0; | |
2158 | retried = true; | |
2159 | goto again; | |
2160 | } else if (failed && retried) { | |
2161 | WARN_ON(1); | |
2162 | ret = -ENOSPC; | |
2163 | } | |
2b82032c YZ |
2164 | error: |
2165 | btrfs_free_path(path); | |
2166 | return ret; | |
8f18cf13 CM |
2167 | } |
2168 | ||
0940ebf6 ID |
2169 | static int insert_balance_item(struct btrfs_root *root, |
2170 | struct btrfs_balance_control *bctl) | |
2171 | { | |
2172 | struct btrfs_trans_handle *trans; | |
2173 | struct btrfs_balance_item *item; | |
2174 | struct btrfs_disk_balance_args disk_bargs; | |
2175 | struct btrfs_path *path; | |
2176 | struct extent_buffer *leaf; | |
2177 | struct btrfs_key key; | |
2178 | int ret, err; | |
2179 | ||
2180 | path = btrfs_alloc_path(); | |
2181 | if (!path) | |
2182 | return -ENOMEM; | |
2183 | ||
2184 | trans = btrfs_start_transaction(root, 0); | |
2185 | if (IS_ERR(trans)) { | |
2186 | btrfs_free_path(path); | |
2187 | return PTR_ERR(trans); | |
2188 | } | |
2189 | ||
2190 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2191 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2192 | key.offset = 0; | |
2193 | ||
2194 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2195 | sizeof(*item)); | |
2196 | if (ret) | |
2197 | goto out; | |
2198 | ||
2199 | leaf = path->nodes[0]; | |
2200 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2201 | ||
2202 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2203 | ||
2204 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2205 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2206 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2207 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2208 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2209 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2210 | ||
2211 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2212 | ||
2213 | btrfs_mark_buffer_dirty(leaf); | |
2214 | out: | |
2215 | btrfs_free_path(path); | |
2216 | err = btrfs_commit_transaction(trans, root); | |
2217 | if (err && !ret) | |
2218 | ret = err; | |
2219 | return ret; | |
2220 | } | |
2221 | ||
2222 | static int del_balance_item(struct btrfs_root *root) | |
2223 | { | |
2224 | struct btrfs_trans_handle *trans; | |
2225 | struct btrfs_path *path; | |
2226 | struct btrfs_key key; | |
2227 | int ret, err; | |
2228 | ||
2229 | path = btrfs_alloc_path(); | |
2230 | if (!path) | |
2231 | return -ENOMEM; | |
2232 | ||
2233 | trans = btrfs_start_transaction(root, 0); | |
2234 | if (IS_ERR(trans)) { | |
2235 | btrfs_free_path(path); | |
2236 | return PTR_ERR(trans); | |
2237 | } | |
2238 | ||
2239 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2240 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2241 | key.offset = 0; | |
2242 | ||
2243 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2244 | if (ret < 0) | |
2245 | goto out; | |
2246 | if (ret > 0) { | |
2247 | ret = -ENOENT; | |
2248 | goto out; | |
2249 | } | |
2250 | ||
2251 | ret = btrfs_del_item(trans, root, path); | |
2252 | out: | |
2253 | btrfs_free_path(path); | |
2254 | err = btrfs_commit_transaction(trans, root); | |
2255 | if (err && !ret) | |
2256 | ret = err; | |
2257 | return ret; | |
2258 | } | |
2259 | ||
59641015 ID |
2260 | /* |
2261 | * This is a heuristic used to reduce the number of chunks balanced on | |
2262 | * resume after balance was interrupted. | |
2263 | */ | |
2264 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2265 | { | |
2266 | /* | |
2267 | * Turn on soft mode for chunk types that were being converted. | |
2268 | */ | |
2269 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2270 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2271 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2272 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2273 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2274 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2275 | ||
2276 | /* | |
2277 | * Turn on usage filter if is not already used. The idea is | |
2278 | * that chunks that we have already balanced should be | |
2279 | * reasonably full. Don't do it for chunks that are being | |
2280 | * converted - that will keep us from relocating unconverted | |
2281 | * (albeit full) chunks. | |
2282 | */ | |
2283 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2284 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2285 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2286 | bctl->data.usage = 90; | |
2287 | } | |
2288 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2289 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2290 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2291 | bctl->sys.usage = 90; | |
2292 | } | |
2293 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2294 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2295 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2296 | bctl->meta.usage = 90; | |
2297 | } | |
2298 | } | |
2299 | ||
c9e9f97b ID |
2300 | /* |
2301 | * Should be called with both balance and volume mutexes held to | |
2302 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2303 | * restriper. Same goes for unset_balance_control. | |
2304 | */ | |
2305 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2306 | { | |
2307 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2308 | ||
2309 | BUG_ON(fs_info->balance_ctl); | |
2310 | ||
2311 | spin_lock(&fs_info->balance_lock); | |
2312 | fs_info->balance_ctl = bctl; | |
2313 | spin_unlock(&fs_info->balance_lock); | |
2314 | } | |
2315 | ||
2316 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2317 | { | |
2318 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2319 | ||
2320 | BUG_ON(!fs_info->balance_ctl); | |
2321 | ||
2322 | spin_lock(&fs_info->balance_lock); | |
2323 | fs_info->balance_ctl = NULL; | |
2324 | spin_unlock(&fs_info->balance_lock); | |
2325 | ||
2326 | kfree(bctl); | |
2327 | } | |
2328 | ||
ed25e9b2 ID |
2329 | /* |
2330 | * Balance filters. Return 1 if chunk should be filtered out | |
2331 | * (should not be balanced). | |
2332 | */ | |
899c81ea | 2333 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2334 | struct btrfs_balance_args *bargs) |
2335 | { | |
899c81ea ID |
2336 | chunk_type = chunk_to_extended(chunk_type) & |
2337 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2338 | |
899c81ea | 2339 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2340 | return 0; |
2341 | ||
2342 | return 1; | |
2343 | } | |
2344 | ||
5ce5b3c0 ID |
2345 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2346 | struct btrfs_balance_args *bargs) | |
2347 | { | |
2348 | struct btrfs_block_group_cache *cache; | |
2349 | u64 chunk_used, user_thresh; | |
2350 | int ret = 1; | |
2351 | ||
2352 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2353 | chunk_used = btrfs_block_group_used(&cache->item); | |
2354 | ||
2355 | user_thresh = div_factor_fine(cache->key.offset, bargs->usage); | |
2356 | if (chunk_used < user_thresh) | |
2357 | ret = 0; | |
2358 | ||
2359 | btrfs_put_block_group(cache); | |
2360 | return ret; | |
2361 | } | |
2362 | ||
409d404b ID |
2363 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2364 | struct btrfs_chunk *chunk, | |
2365 | struct btrfs_balance_args *bargs) | |
2366 | { | |
2367 | struct btrfs_stripe *stripe; | |
2368 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2369 | int i; | |
2370 | ||
2371 | for (i = 0; i < num_stripes; i++) { | |
2372 | stripe = btrfs_stripe_nr(chunk, i); | |
2373 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2374 | return 0; | |
2375 | } | |
2376 | ||
2377 | return 1; | |
2378 | } | |
2379 | ||
94e60d5a ID |
2380 | /* [pstart, pend) */ |
2381 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2382 | struct btrfs_chunk *chunk, | |
2383 | u64 chunk_offset, | |
2384 | struct btrfs_balance_args *bargs) | |
2385 | { | |
2386 | struct btrfs_stripe *stripe; | |
2387 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2388 | u64 stripe_offset; | |
2389 | u64 stripe_length; | |
2390 | int factor; | |
2391 | int i; | |
2392 | ||
2393 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2394 | return 0; | |
2395 | ||
2396 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
2397 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) | |
2398 | factor = 2; | |
2399 | else | |
2400 | factor = 1; | |
2401 | factor = num_stripes / factor; | |
2402 | ||
2403 | for (i = 0; i < num_stripes; i++) { | |
2404 | stripe = btrfs_stripe_nr(chunk, i); | |
2405 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2406 | continue; | |
2407 | ||
2408 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2409 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2410 | do_div(stripe_length, factor); | |
2411 | ||
2412 | if (stripe_offset < bargs->pend && | |
2413 | stripe_offset + stripe_length > bargs->pstart) | |
2414 | return 0; | |
2415 | } | |
2416 | ||
2417 | return 1; | |
2418 | } | |
2419 | ||
ea67176a ID |
2420 | /* [vstart, vend) */ |
2421 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2422 | struct btrfs_chunk *chunk, | |
2423 | u64 chunk_offset, | |
2424 | struct btrfs_balance_args *bargs) | |
2425 | { | |
2426 | if (chunk_offset < bargs->vend && | |
2427 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2428 | /* at least part of the chunk is inside this vrange */ | |
2429 | return 0; | |
2430 | ||
2431 | return 1; | |
2432 | } | |
2433 | ||
899c81ea | 2434 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2435 | struct btrfs_balance_args *bargs) |
2436 | { | |
2437 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2438 | return 0; | |
2439 | ||
899c81ea ID |
2440 | chunk_type = chunk_to_extended(chunk_type) & |
2441 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2442 | |
899c81ea | 2443 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2444 | return 1; |
2445 | ||
2446 | return 0; | |
2447 | } | |
2448 | ||
f43ffb60 ID |
2449 | static int should_balance_chunk(struct btrfs_root *root, |
2450 | struct extent_buffer *leaf, | |
2451 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2452 | { | |
2453 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2454 | struct btrfs_balance_args *bargs = NULL; | |
2455 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2456 | ||
2457 | /* type filter */ | |
2458 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2459 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2460 | return 0; | |
2461 | } | |
2462 | ||
2463 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2464 | bargs = &bctl->data; | |
2465 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2466 | bargs = &bctl->sys; | |
2467 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2468 | bargs = &bctl->meta; | |
2469 | ||
ed25e9b2 ID |
2470 | /* profiles filter */ |
2471 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2472 | chunk_profiles_filter(chunk_type, bargs)) { | |
2473 | return 0; | |
5ce5b3c0 ID |
2474 | } |
2475 | ||
2476 | /* usage filter */ | |
2477 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2478 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2479 | return 0; | |
409d404b ID |
2480 | } |
2481 | ||
2482 | /* devid filter */ | |
2483 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2484 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2485 | return 0; | |
94e60d5a ID |
2486 | } |
2487 | ||
2488 | /* drange filter, makes sense only with devid filter */ | |
2489 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2490 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2491 | return 0; | |
ea67176a ID |
2492 | } |
2493 | ||
2494 | /* vrange filter */ | |
2495 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2496 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2497 | return 0; | |
ed25e9b2 ID |
2498 | } |
2499 | ||
cfa4c961 ID |
2500 | /* soft profile changing mode */ |
2501 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2502 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2503 | return 0; | |
2504 | } | |
2505 | ||
f43ffb60 ID |
2506 | return 1; |
2507 | } | |
2508 | ||
c9e9f97b | 2509 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2510 | { |
19a39dce | 2511 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2512 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2513 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2514 | struct list_head *devices; | |
ec44a35c CM |
2515 | struct btrfs_device *device; |
2516 | u64 old_size; | |
2517 | u64 size_to_free; | |
f43ffb60 | 2518 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2519 | struct btrfs_path *path; |
2520 | struct btrfs_key key; | |
ec44a35c | 2521 | struct btrfs_key found_key; |
c9e9f97b | 2522 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2523 | struct extent_buffer *leaf; |
2524 | int slot; | |
c9e9f97b ID |
2525 | int ret; |
2526 | int enospc_errors = 0; | |
19a39dce | 2527 | bool counting = true; |
ec44a35c | 2528 | |
ec44a35c | 2529 | /* step one make some room on all the devices */ |
c9e9f97b | 2530 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2531 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2532 | old_size = device->total_bytes; |
2533 | size_to_free = div_factor(old_size, 1); | |
2534 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2535 | if (!device->writeable || |
2536 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2537 | continue; |
2538 | ||
2539 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2540 | if (ret == -ENOSPC) |
2541 | break; | |
ec44a35c CM |
2542 | BUG_ON(ret); |
2543 | ||
a22285a6 | 2544 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2545 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2546 | |
2547 | ret = btrfs_grow_device(trans, device, old_size); | |
2548 | BUG_ON(ret); | |
2549 | ||
2550 | btrfs_end_transaction(trans, dev_root); | |
2551 | } | |
2552 | ||
2553 | /* step two, relocate all the chunks */ | |
2554 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2555 | if (!path) { |
2556 | ret = -ENOMEM; | |
2557 | goto error; | |
2558 | } | |
19a39dce ID |
2559 | |
2560 | /* zero out stat counters */ | |
2561 | spin_lock(&fs_info->balance_lock); | |
2562 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2563 | spin_unlock(&fs_info->balance_lock); | |
2564 | again: | |
ec44a35c CM |
2565 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2566 | key.offset = (u64)-1; | |
2567 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2568 | ||
d397712b | 2569 | while (1) { |
19a39dce | 2570 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2571 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2572 | ret = -ECANCELED; |
2573 | goto error; | |
2574 | } | |
2575 | ||
ec44a35c CM |
2576 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2577 | if (ret < 0) | |
2578 | goto error; | |
2579 | ||
2580 | /* | |
2581 | * this shouldn't happen, it means the last relocate | |
2582 | * failed | |
2583 | */ | |
2584 | if (ret == 0) | |
c9e9f97b | 2585 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2586 | |
2587 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2588 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2589 | if (ret) { |
2590 | ret = 0; | |
ec44a35c | 2591 | break; |
c9e9f97b | 2592 | } |
7d9eb12c | 2593 | |
f43ffb60 ID |
2594 | leaf = path->nodes[0]; |
2595 | slot = path->slots[0]; | |
2596 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2597 | |
ec44a35c CM |
2598 | if (found_key.objectid != key.objectid) |
2599 | break; | |
7d9eb12c | 2600 | |
ec44a35c | 2601 | /* chunk zero is special */ |
ba1bf481 | 2602 | if (found_key.offset == 0) |
ec44a35c CM |
2603 | break; |
2604 | ||
f43ffb60 ID |
2605 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2606 | ||
19a39dce ID |
2607 | if (!counting) { |
2608 | spin_lock(&fs_info->balance_lock); | |
2609 | bctl->stat.considered++; | |
2610 | spin_unlock(&fs_info->balance_lock); | |
2611 | } | |
2612 | ||
f43ffb60 ID |
2613 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2614 | found_key.offset); | |
b3b4aa74 | 2615 | btrfs_release_path(path); |
f43ffb60 ID |
2616 | if (!ret) |
2617 | goto loop; | |
2618 | ||
19a39dce ID |
2619 | if (counting) { |
2620 | spin_lock(&fs_info->balance_lock); | |
2621 | bctl->stat.expected++; | |
2622 | spin_unlock(&fs_info->balance_lock); | |
2623 | goto loop; | |
2624 | } | |
2625 | ||
ec44a35c CM |
2626 | ret = btrfs_relocate_chunk(chunk_root, |
2627 | chunk_root->root_key.objectid, | |
2628 | found_key.objectid, | |
2629 | found_key.offset); | |
508794eb JB |
2630 | if (ret && ret != -ENOSPC) |
2631 | goto error; | |
19a39dce | 2632 | if (ret == -ENOSPC) { |
c9e9f97b | 2633 | enospc_errors++; |
19a39dce ID |
2634 | } else { |
2635 | spin_lock(&fs_info->balance_lock); | |
2636 | bctl->stat.completed++; | |
2637 | spin_unlock(&fs_info->balance_lock); | |
2638 | } | |
f43ffb60 | 2639 | loop: |
ba1bf481 | 2640 | key.offset = found_key.offset - 1; |
ec44a35c | 2641 | } |
c9e9f97b | 2642 | |
19a39dce ID |
2643 | if (counting) { |
2644 | btrfs_release_path(path); | |
2645 | counting = false; | |
2646 | goto again; | |
2647 | } | |
ec44a35c CM |
2648 | error: |
2649 | btrfs_free_path(path); | |
c9e9f97b ID |
2650 | if (enospc_errors) { |
2651 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
2652 | enospc_errors); | |
2653 | if (!ret) | |
2654 | ret = -ENOSPC; | |
2655 | } | |
2656 | ||
ec44a35c CM |
2657 | return ret; |
2658 | } | |
2659 | ||
0c460c0d ID |
2660 | /** |
2661 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
2662 | * @flags: profile to validate | |
2663 | * @extended: if true @flags is treated as an extended profile | |
2664 | */ | |
2665 | static int alloc_profile_is_valid(u64 flags, int extended) | |
2666 | { | |
2667 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
2668 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
2669 | ||
2670 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
2671 | ||
2672 | /* 1) check that all other bits are zeroed */ | |
2673 | if (flags & ~mask) | |
2674 | return 0; | |
2675 | ||
2676 | /* 2) see if profile is reduced */ | |
2677 | if (flags == 0) | |
2678 | return !extended; /* "0" is valid for usual profiles */ | |
2679 | ||
2680 | /* true if exactly one bit set */ | |
2681 | return (flags & (flags - 1)) == 0; | |
2682 | } | |
2683 | ||
837d5b6e ID |
2684 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
2685 | { | |
a7e99c69 ID |
2686 | /* cancel requested || normal exit path */ |
2687 | return atomic_read(&fs_info->balance_cancel_req) || | |
2688 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
2689 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
2690 | } |
2691 | ||
c9e9f97b ID |
2692 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
2693 | { | |
0940ebf6 ID |
2694 | int ret; |
2695 | ||
c9e9f97b | 2696 | unset_balance_control(fs_info); |
0940ebf6 ID |
2697 | ret = del_balance_item(fs_info->tree_root); |
2698 | BUG_ON(ret); | |
c9e9f97b ID |
2699 | } |
2700 | ||
19a39dce | 2701 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
2702 | struct btrfs_ioctl_balance_args *bargs); |
2703 | ||
2704 | /* | |
2705 | * Should be called with both balance and volume mutexes held | |
2706 | */ | |
2707 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
2708 | struct btrfs_ioctl_balance_args *bargs) | |
2709 | { | |
2710 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 2711 | u64 allowed; |
e4837f8f | 2712 | int mixed = 0; |
c9e9f97b ID |
2713 | int ret; |
2714 | ||
837d5b6e | 2715 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
2716 | atomic_read(&fs_info->balance_pause_req) || |
2717 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
2718 | ret = -EINVAL; |
2719 | goto out; | |
2720 | } | |
2721 | ||
e4837f8f ID |
2722 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
2723 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
2724 | mixed = 1; | |
2725 | ||
f43ffb60 ID |
2726 | /* |
2727 | * In case of mixed groups both data and meta should be picked, | |
2728 | * and identical options should be given for both of them. | |
2729 | */ | |
e4837f8f ID |
2730 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
2731 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
2732 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
2733 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
2734 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
2735 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
2736 | "metadata balance options must be the same\n"); | |
2737 | ret = -EINVAL; | |
2738 | goto out; | |
2739 | } | |
2740 | } | |
2741 | ||
e4d8ec0f ID |
2742 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
2743 | if (fs_info->fs_devices->num_devices == 1) | |
2744 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
2745 | else if (fs_info->fs_devices->num_devices < 4) | |
2746 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
2747 | else | |
2748 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
2749 | BTRFS_BLOCK_GROUP_RAID10); | |
2750 | ||
6728b198 ID |
2751 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2752 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
2753 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
2754 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2755 | "data profile %llu\n", | |
2756 | (unsigned long long)bctl->data.target); | |
2757 | ret = -EINVAL; | |
2758 | goto out; | |
2759 | } | |
6728b198 ID |
2760 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2761 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
2762 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
2763 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2764 | "metadata profile %llu\n", | |
2765 | (unsigned long long)bctl->meta.target); | |
2766 | ret = -EINVAL; | |
2767 | goto out; | |
2768 | } | |
6728b198 ID |
2769 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2770 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
2771 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
2772 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2773 | "system profile %llu\n", | |
2774 | (unsigned long long)bctl->sys.target); | |
2775 | ret = -EINVAL; | |
2776 | goto out; | |
2777 | } | |
2778 | ||
e4837f8f ID |
2779 | /* allow dup'ed data chunks only in mixed mode */ |
2780 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 2781 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
2782 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
2783 | ret = -EINVAL; | |
2784 | goto out; | |
2785 | } | |
2786 | ||
2787 | /* allow to reduce meta or sys integrity only if force set */ | |
2788 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2789 | BTRFS_BLOCK_GROUP_RAID10; | |
2790 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2791 | (fs_info->avail_system_alloc_bits & allowed) && | |
2792 | !(bctl->sys.target & allowed)) || | |
2793 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2794 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
2795 | !(bctl->meta.target & allowed))) { | |
2796 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
2797 | printk(KERN_INFO "btrfs: force reducing metadata " | |
2798 | "integrity\n"); | |
2799 | } else { | |
2800 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
2801 | "integrity, use force if you want this\n"); | |
2802 | ret = -EINVAL; | |
2803 | goto out; | |
2804 | } | |
2805 | } | |
2806 | ||
5af3e8cc SB |
2807 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
2808 | int num_tolerated_disk_barrier_failures; | |
2809 | u64 target = bctl->sys.target; | |
2810 | ||
2811 | num_tolerated_disk_barrier_failures = | |
2812 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
2813 | if (num_tolerated_disk_barrier_failures > 0 && | |
2814 | (target & | |
2815 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
2816 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
2817 | num_tolerated_disk_barrier_failures = 0; | |
2818 | else if (num_tolerated_disk_barrier_failures > 1 && | |
2819 | (target & | |
2820 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
2821 | num_tolerated_disk_barrier_failures = 1; | |
2822 | ||
2823 | fs_info->num_tolerated_disk_barrier_failures = | |
2824 | num_tolerated_disk_barrier_failures; | |
2825 | } | |
2826 | ||
0940ebf6 | 2827 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 2828 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
2829 | goto out; |
2830 | ||
59641015 ID |
2831 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
2832 | BUG_ON(ret == -EEXIST); | |
2833 | set_balance_control(bctl); | |
2834 | } else { | |
2835 | BUG_ON(ret != -EEXIST); | |
2836 | spin_lock(&fs_info->balance_lock); | |
2837 | update_balance_args(bctl); | |
2838 | spin_unlock(&fs_info->balance_lock); | |
2839 | } | |
c9e9f97b | 2840 | |
837d5b6e | 2841 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
2842 | mutex_unlock(&fs_info->balance_mutex); |
2843 | ||
2844 | ret = __btrfs_balance(fs_info); | |
2845 | ||
2846 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 2847 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
2848 | |
2849 | if (bargs) { | |
2850 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 2851 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
2852 | } |
2853 | ||
837d5b6e ID |
2854 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
2855 | balance_need_close(fs_info)) { | |
2856 | __cancel_balance(fs_info); | |
2857 | } | |
2858 | ||
5af3e8cc SB |
2859 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
2860 | fs_info->num_tolerated_disk_barrier_failures = | |
2861 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
2862 | } | |
2863 | ||
837d5b6e | 2864 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
2865 | |
2866 | return ret; | |
2867 | out: | |
59641015 ID |
2868 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
2869 | __cancel_balance(fs_info); | |
2870 | else | |
2871 | kfree(bctl); | |
2872 | return ret; | |
2873 | } | |
2874 | ||
2875 | static int balance_kthread(void *data) | |
2876 | { | |
2b6ba629 | 2877 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 2878 | int ret = 0; |
59641015 ID |
2879 | |
2880 | mutex_lock(&fs_info->volume_mutex); | |
2881 | mutex_lock(&fs_info->balance_mutex); | |
2882 | ||
2b6ba629 | 2883 | if (fs_info->balance_ctl) { |
9555c6c1 | 2884 | printk(KERN_INFO "btrfs: continuing balance\n"); |
2b6ba629 | 2885 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 2886 | } |
59641015 ID |
2887 | |
2888 | mutex_unlock(&fs_info->balance_mutex); | |
2889 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 2890 | |
59641015 ID |
2891 | return ret; |
2892 | } | |
2893 | ||
2b6ba629 ID |
2894 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
2895 | { | |
2896 | struct task_struct *tsk; | |
2897 | ||
2898 | spin_lock(&fs_info->balance_lock); | |
2899 | if (!fs_info->balance_ctl) { | |
2900 | spin_unlock(&fs_info->balance_lock); | |
2901 | return 0; | |
2902 | } | |
2903 | spin_unlock(&fs_info->balance_lock); | |
2904 | ||
2905 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
2906 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
2907 | return 0; | |
2908 | } | |
2909 | ||
2910 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
2911 | if (IS_ERR(tsk)) | |
2912 | return PTR_ERR(tsk); | |
2913 | ||
2914 | return 0; | |
2915 | } | |
2916 | ||
68310a5e | 2917 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 2918 | { |
59641015 ID |
2919 | struct btrfs_balance_control *bctl; |
2920 | struct btrfs_balance_item *item; | |
2921 | struct btrfs_disk_balance_args disk_bargs; | |
2922 | struct btrfs_path *path; | |
2923 | struct extent_buffer *leaf; | |
2924 | struct btrfs_key key; | |
2925 | int ret; | |
2926 | ||
2927 | path = btrfs_alloc_path(); | |
2928 | if (!path) | |
2929 | return -ENOMEM; | |
2930 | ||
59641015 ID |
2931 | key.objectid = BTRFS_BALANCE_OBJECTID; |
2932 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2933 | key.offset = 0; | |
2934 | ||
68310a5e | 2935 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 2936 | if (ret < 0) |
68310a5e | 2937 | goto out; |
59641015 ID |
2938 | if (ret > 0) { /* ret = -ENOENT; */ |
2939 | ret = 0; | |
68310a5e ID |
2940 | goto out; |
2941 | } | |
2942 | ||
2943 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
2944 | if (!bctl) { | |
2945 | ret = -ENOMEM; | |
2946 | goto out; | |
59641015 ID |
2947 | } |
2948 | ||
2949 | leaf = path->nodes[0]; | |
2950 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2951 | ||
68310a5e ID |
2952 | bctl->fs_info = fs_info; |
2953 | bctl->flags = btrfs_balance_flags(leaf, item); | |
2954 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
2955 | |
2956 | btrfs_balance_data(leaf, item, &disk_bargs); | |
2957 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
2958 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
2959 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
2960 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
2961 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
2962 | ||
68310a5e ID |
2963 | mutex_lock(&fs_info->volume_mutex); |
2964 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 2965 | |
68310a5e ID |
2966 | set_balance_control(bctl); |
2967 | ||
2968 | mutex_unlock(&fs_info->balance_mutex); | |
2969 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
2970 | out: |
2971 | btrfs_free_path(path); | |
ec44a35c CM |
2972 | return ret; |
2973 | } | |
2974 | ||
837d5b6e ID |
2975 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
2976 | { | |
2977 | int ret = 0; | |
2978 | ||
2979 | mutex_lock(&fs_info->balance_mutex); | |
2980 | if (!fs_info->balance_ctl) { | |
2981 | mutex_unlock(&fs_info->balance_mutex); | |
2982 | return -ENOTCONN; | |
2983 | } | |
2984 | ||
2985 | if (atomic_read(&fs_info->balance_running)) { | |
2986 | atomic_inc(&fs_info->balance_pause_req); | |
2987 | mutex_unlock(&fs_info->balance_mutex); | |
2988 | ||
2989 | wait_event(fs_info->balance_wait_q, | |
2990 | atomic_read(&fs_info->balance_running) == 0); | |
2991 | ||
2992 | mutex_lock(&fs_info->balance_mutex); | |
2993 | /* we are good with balance_ctl ripped off from under us */ | |
2994 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
2995 | atomic_dec(&fs_info->balance_pause_req); | |
2996 | } else { | |
2997 | ret = -ENOTCONN; | |
2998 | } | |
2999 | ||
3000 | mutex_unlock(&fs_info->balance_mutex); | |
3001 | return ret; | |
3002 | } | |
3003 | ||
a7e99c69 ID |
3004 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3005 | { | |
3006 | mutex_lock(&fs_info->balance_mutex); | |
3007 | if (!fs_info->balance_ctl) { | |
3008 | mutex_unlock(&fs_info->balance_mutex); | |
3009 | return -ENOTCONN; | |
3010 | } | |
3011 | ||
3012 | atomic_inc(&fs_info->balance_cancel_req); | |
3013 | /* | |
3014 | * if we are running just wait and return, balance item is | |
3015 | * deleted in btrfs_balance in this case | |
3016 | */ | |
3017 | if (atomic_read(&fs_info->balance_running)) { | |
3018 | mutex_unlock(&fs_info->balance_mutex); | |
3019 | wait_event(fs_info->balance_wait_q, | |
3020 | atomic_read(&fs_info->balance_running) == 0); | |
3021 | mutex_lock(&fs_info->balance_mutex); | |
3022 | } else { | |
3023 | /* __cancel_balance needs volume_mutex */ | |
3024 | mutex_unlock(&fs_info->balance_mutex); | |
3025 | mutex_lock(&fs_info->volume_mutex); | |
3026 | mutex_lock(&fs_info->balance_mutex); | |
3027 | ||
3028 | if (fs_info->balance_ctl) | |
3029 | __cancel_balance(fs_info); | |
3030 | ||
3031 | mutex_unlock(&fs_info->volume_mutex); | |
3032 | } | |
3033 | ||
3034 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3035 | atomic_dec(&fs_info->balance_cancel_req); | |
3036 | mutex_unlock(&fs_info->balance_mutex); | |
3037 | return 0; | |
3038 | } | |
3039 | ||
8f18cf13 CM |
3040 | /* |
3041 | * shrinking a device means finding all of the device extents past | |
3042 | * the new size, and then following the back refs to the chunks. | |
3043 | * The chunk relocation code actually frees the device extent | |
3044 | */ | |
3045 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3046 | { | |
3047 | struct btrfs_trans_handle *trans; | |
3048 | struct btrfs_root *root = device->dev_root; | |
3049 | struct btrfs_dev_extent *dev_extent = NULL; | |
3050 | struct btrfs_path *path; | |
3051 | u64 length; | |
3052 | u64 chunk_tree; | |
3053 | u64 chunk_objectid; | |
3054 | u64 chunk_offset; | |
3055 | int ret; | |
3056 | int slot; | |
ba1bf481 JB |
3057 | int failed = 0; |
3058 | bool retried = false; | |
8f18cf13 CM |
3059 | struct extent_buffer *l; |
3060 | struct btrfs_key key; | |
6c41761f | 3061 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3062 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 3063 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
3064 | u64 diff = device->total_bytes - new_size; |
3065 | ||
8f18cf13 CM |
3066 | path = btrfs_alloc_path(); |
3067 | if (!path) | |
3068 | return -ENOMEM; | |
3069 | ||
8f18cf13 CM |
3070 | path->reada = 2; |
3071 | ||
7d9eb12c CM |
3072 | lock_chunks(root); |
3073 | ||
8f18cf13 | 3074 | device->total_bytes = new_size; |
2bf64758 | 3075 | if (device->writeable) { |
2b82032c | 3076 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
3077 | spin_lock(&root->fs_info->free_chunk_lock); |
3078 | root->fs_info->free_chunk_space -= diff; | |
3079 | spin_unlock(&root->fs_info->free_chunk_lock); | |
3080 | } | |
7d9eb12c | 3081 | unlock_chunks(root); |
8f18cf13 | 3082 | |
ba1bf481 | 3083 | again: |
8f18cf13 CM |
3084 | key.objectid = device->devid; |
3085 | key.offset = (u64)-1; | |
3086 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3087 | ||
213e64da | 3088 | do { |
8f18cf13 CM |
3089 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3090 | if (ret < 0) | |
3091 | goto done; | |
3092 | ||
3093 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3094 | if (ret < 0) | |
3095 | goto done; | |
3096 | if (ret) { | |
3097 | ret = 0; | |
b3b4aa74 | 3098 | btrfs_release_path(path); |
bf1fb512 | 3099 | break; |
8f18cf13 CM |
3100 | } |
3101 | ||
3102 | l = path->nodes[0]; | |
3103 | slot = path->slots[0]; | |
3104 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3105 | ||
ba1bf481 | 3106 | if (key.objectid != device->devid) { |
b3b4aa74 | 3107 | btrfs_release_path(path); |
bf1fb512 | 3108 | break; |
ba1bf481 | 3109 | } |
8f18cf13 CM |
3110 | |
3111 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3112 | length = btrfs_dev_extent_length(l, dev_extent); | |
3113 | ||
ba1bf481 | 3114 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3115 | btrfs_release_path(path); |
d6397bae | 3116 | break; |
ba1bf481 | 3117 | } |
8f18cf13 CM |
3118 | |
3119 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3120 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3121 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3122 | btrfs_release_path(path); |
8f18cf13 CM |
3123 | |
3124 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3125 | chunk_offset); | |
ba1bf481 | 3126 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3127 | goto done; |
ba1bf481 JB |
3128 | if (ret == -ENOSPC) |
3129 | failed++; | |
213e64da | 3130 | } while (key.offset-- > 0); |
ba1bf481 JB |
3131 | |
3132 | if (failed && !retried) { | |
3133 | failed = 0; | |
3134 | retried = true; | |
3135 | goto again; | |
3136 | } else if (failed && retried) { | |
3137 | ret = -ENOSPC; | |
3138 | lock_chunks(root); | |
3139 | ||
3140 | device->total_bytes = old_size; | |
3141 | if (device->writeable) | |
3142 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3143 | spin_lock(&root->fs_info->free_chunk_lock); |
3144 | root->fs_info->free_chunk_space += diff; | |
3145 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3146 | unlock_chunks(root); |
3147 | goto done; | |
8f18cf13 CM |
3148 | } |
3149 | ||
d6397bae | 3150 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3151 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3152 | if (IS_ERR(trans)) { |
3153 | ret = PTR_ERR(trans); | |
3154 | goto done; | |
3155 | } | |
3156 | ||
d6397bae CB |
3157 | lock_chunks(root); |
3158 | ||
3159 | device->disk_total_bytes = new_size; | |
3160 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3161 | ret = btrfs_update_device(trans, device); | |
3162 | if (ret) { | |
3163 | unlock_chunks(root); | |
3164 | btrfs_end_transaction(trans, root); | |
3165 | goto done; | |
3166 | } | |
3167 | WARN_ON(diff > old_total); | |
3168 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3169 | unlock_chunks(root); | |
3170 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3171 | done: |
3172 | btrfs_free_path(path); | |
3173 | return ret; | |
3174 | } | |
3175 | ||
125ccb0a | 3176 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3177 | struct btrfs_key *key, |
3178 | struct btrfs_chunk *chunk, int item_size) | |
3179 | { | |
6c41761f | 3180 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3181 | struct btrfs_disk_key disk_key; |
3182 | u32 array_size; | |
3183 | u8 *ptr; | |
3184 | ||
3185 | array_size = btrfs_super_sys_array_size(super_copy); | |
3186 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3187 | return -EFBIG; | |
3188 | ||
3189 | ptr = super_copy->sys_chunk_array + array_size; | |
3190 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3191 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3192 | ptr += sizeof(disk_key); | |
3193 | memcpy(ptr, chunk, item_size); | |
3194 | item_size += sizeof(disk_key); | |
3195 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3196 | return 0; | |
3197 | } | |
3198 | ||
73c5de00 AJ |
3199 | /* |
3200 | * sort the devices in descending order by max_avail, total_avail | |
3201 | */ | |
3202 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3203 | { |
73c5de00 AJ |
3204 | const struct btrfs_device_info *di_a = a; |
3205 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3206 | |
73c5de00 | 3207 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3208 | return -1; |
73c5de00 | 3209 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3210 | return 1; |
73c5de00 AJ |
3211 | if (di_a->total_avail > di_b->total_avail) |
3212 | return -1; | |
3213 | if (di_a->total_avail < di_b->total_avail) | |
3214 | return 1; | |
3215 | return 0; | |
b2117a39 | 3216 | } |
0b86a832 | 3217 | |
73c5de00 AJ |
3218 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3219 | struct btrfs_root *extent_root, | |
3220 | struct map_lookup **map_ret, | |
3221 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3222 | u64 start, u64 type) | |
b2117a39 | 3223 | { |
73c5de00 AJ |
3224 | struct btrfs_fs_info *info = extent_root->fs_info; |
3225 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3226 | struct list_head *cur; | |
3227 | struct map_lookup *map = NULL; | |
3228 | struct extent_map_tree *em_tree; | |
3229 | struct extent_map *em; | |
3230 | struct btrfs_device_info *devices_info = NULL; | |
3231 | u64 total_avail; | |
3232 | int num_stripes; /* total number of stripes to allocate */ | |
3233 | int sub_stripes; /* sub_stripes info for map */ | |
3234 | int dev_stripes; /* stripes per dev */ | |
3235 | int devs_max; /* max devs to use */ | |
3236 | int devs_min; /* min devs needed */ | |
3237 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3238 | int ncopies; /* how many copies to data has */ | |
3239 | int ret; | |
3240 | u64 max_stripe_size; | |
3241 | u64 max_chunk_size; | |
3242 | u64 stripe_size; | |
3243 | u64 num_bytes; | |
3244 | int ndevs; | |
3245 | int i; | |
3246 | int j; | |
593060d7 | 3247 | |
0c460c0d | 3248 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3249 | |
73c5de00 AJ |
3250 | if (list_empty(&fs_devices->alloc_list)) |
3251 | return -ENOSPC; | |
b2117a39 | 3252 | |
73c5de00 AJ |
3253 | sub_stripes = 1; |
3254 | dev_stripes = 1; | |
3255 | devs_increment = 1; | |
3256 | ncopies = 1; | |
3257 | devs_max = 0; /* 0 == as many as possible */ | |
3258 | devs_min = 1; | |
3259 | ||
3260 | /* | |
3261 | * define the properties of each RAID type. | |
3262 | * FIXME: move this to a global table and use it in all RAID | |
3263 | * calculation code | |
3264 | */ | |
3265 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
3266 | dev_stripes = 2; | |
b2117a39 | 3267 | ncopies = 2; |
73c5de00 AJ |
3268 | devs_max = 1; |
3269 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
3270 | devs_min = 2; | |
3271 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
3272 | devs_increment = 2; | |
b2117a39 | 3273 | ncopies = 2; |
73c5de00 AJ |
3274 | devs_max = 2; |
3275 | devs_min = 2; | |
3276 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
3277 | sub_stripes = 2; | |
3278 | devs_increment = 2; | |
3279 | ncopies = 2; | |
3280 | devs_min = 4; | |
3281 | } else { | |
3282 | devs_max = 1; | |
3283 | } | |
b2117a39 | 3284 | |
9b3f68b9 | 3285 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3286 | max_stripe_size = 1024 * 1024 * 1024; |
3287 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3288 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3289 | /* for larger filesystems, use larger metadata chunks */ |
3290 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3291 | max_stripe_size = 1024 * 1024 * 1024; | |
3292 | else | |
3293 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3294 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3295 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3296 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3297 | max_chunk_size = 2 * max_stripe_size; |
3298 | } else { | |
3299 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3300 | type); | |
3301 | BUG_ON(1); | |
9b3f68b9 CM |
3302 | } |
3303 | ||
2b82032c YZ |
3304 | /* we don't want a chunk larger than 10% of writeable space */ |
3305 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3306 | max_chunk_size); | |
9b3f68b9 | 3307 | |
73c5de00 AJ |
3308 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3309 | GFP_NOFS); | |
3310 | if (!devices_info) | |
3311 | return -ENOMEM; | |
0cad8a11 | 3312 | |
73c5de00 | 3313 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3314 | |
9f680ce0 | 3315 | /* |
73c5de00 AJ |
3316 | * in the first pass through the devices list, we gather information |
3317 | * about the available holes on each device. | |
9f680ce0 | 3318 | */ |
73c5de00 AJ |
3319 | ndevs = 0; |
3320 | while (cur != &fs_devices->alloc_list) { | |
3321 | struct btrfs_device *device; | |
3322 | u64 max_avail; | |
3323 | u64 dev_offset; | |
b2117a39 | 3324 | |
73c5de00 | 3325 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3326 | |
73c5de00 | 3327 | cur = cur->next; |
b2117a39 | 3328 | |
73c5de00 | 3329 | if (!device->writeable) { |
31b1a2bd | 3330 | WARN(1, KERN_ERR |
73c5de00 | 3331 | "btrfs: read-only device in alloc_list\n"); |
73c5de00 AJ |
3332 | continue; |
3333 | } | |
b2117a39 | 3334 | |
73c5de00 AJ |
3335 | if (!device->in_fs_metadata) |
3336 | continue; | |
b2117a39 | 3337 | |
73c5de00 AJ |
3338 | if (device->total_bytes > device->bytes_used) |
3339 | total_avail = device->total_bytes - device->bytes_used; | |
3340 | else | |
3341 | total_avail = 0; | |
38c01b96 | 3342 | |
3343 | /* If there is no space on this device, skip it. */ | |
3344 | if (total_avail == 0) | |
3345 | continue; | |
b2117a39 | 3346 | |
125ccb0a | 3347 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3348 | max_stripe_size * dev_stripes, |
3349 | &dev_offset, &max_avail); | |
3350 | if (ret && ret != -ENOSPC) | |
3351 | goto error; | |
b2117a39 | 3352 | |
73c5de00 AJ |
3353 | if (ret == 0) |
3354 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3355 | |
73c5de00 AJ |
3356 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3357 | continue; | |
b2117a39 | 3358 | |
73c5de00 AJ |
3359 | devices_info[ndevs].dev_offset = dev_offset; |
3360 | devices_info[ndevs].max_avail = max_avail; | |
3361 | devices_info[ndevs].total_avail = total_avail; | |
3362 | devices_info[ndevs].dev = device; | |
3363 | ++ndevs; | |
3364 | } | |
b2117a39 | 3365 | |
73c5de00 AJ |
3366 | /* |
3367 | * now sort the devices by hole size / available space | |
3368 | */ | |
3369 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3370 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3371 | |
73c5de00 AJ |
3372 | /* round down to number of usable stripes */ |
3373 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3374 | |
73c5de00 AJ |
3375 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3376 | ret = -ENOSPC; | |
3377 | goto error; | |
b2117a39 | 3378 | } |
9f680ce0 | 3379 | |
73c5de00 AJ |
3380 | if (devs_max && ndevs > devs_max) |
3381 | ndevs = devs_max; | |
3382 | /* | |
3383 | * the primary goal is to maximize the number of stripes, so use as many | |
3384 | * devices as possible, even if the stripes are not maximum sized. | |
3385 | */ | |
3386 | stripe_size = devices_info[ndevs-1].max_avail; | |
3387 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3388 | |
37db63a4 | 3389 | if (stripe_size * ndevs > max_chunk_size * ncopies) { |
73c5de00 | 3390 | stripe_size = max_chunk_size * ncopies; |
37db63a4 | 3391 | do_div(stripe_size, ndevs); |
b2117a39 | 3392 | } |
b2117a39 | 3393 | |
73c5de00 | 3394 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
3395 | |
3396 | /* align to BTRFS_STRIPE_LEN */ | |
73c5de00 AJ |
3397 | do_div(stripe_size, BTRFS_STRIPE_LEN); |
3398 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
3399 | |
3400 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3401 | if (!map) { | |
3402 | ret = -ENOMEM; | |
3403 | goto error; | |
3404 | } | |
3405 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3406 | |
73c5de00 AJ |
3407 | for (i = 0; i < ndevs; ++i) { |
3408 | for (j = 0; j < dev_stripes; ++j) { | |
3409 | int s = i * dev_stripes + j; | |
3410 | map->stripes[s].dev = devices_info[i].dev; | |
3411 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3412 | j * stripe_size; | |
6324fbf3 | 3413 | } |
6324fbf3 | 3414 | } |
2b82032c | 3415 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
3416 | map->stripe_len = BTRFS_STRIPE_LEN; |
3417 | map->io_align = BTRFS_STRIPE_LEN; | |
3418 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 3419 | map->type = type; |
2b82032c | 3420 | map->sub_stripes = sub_stripes; |
0b86a832 | 3421 | |
2b82032c | 3422 | *map_ret = map; |
73c5de00 | 3423 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 3424 | |
73c5de00 AJ |
3425 | *stripe_size_out = stripe_size; |
3426 | *num_bytes_out = num_bytes; | |
0b86a832 | 3427 | |
73c5de00 | 3428 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3429 | |
172ddd60 | 3430 | em = alloc_extent_map(); |
2b82032c | 3431 | if (!em) { |
b2117a39 MX |
3432 | ret = -ENOMEM; |
3433 | goto error; | |
593060d7 | 3434 | } |
2b82032c YZ |
3435 | em->bdev = (struct block_device *)map; |
3436 | em->start = start; | |
73c5de00 | 3437 | em->len = num_bytes; |
2b82032c YZ |
3438 | em->block_start = 0; |
3439 | em->block_len = em->len; | |
593060d7 | 3440 | |
2b82032c | 3441 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3442 | write_lock(&em_tree->lock); |
2b82032c | 3443 | ret = add_extent_mapping(em_tree, em); |
890871be | 3444 | write_unlock(&em_tree->lock); |
2b82032c | 3445 | free_extent_map(em); |
1dd4602f MF |
3446 | if (ret) |
3447 | goto error; | |
0b86a832 | 3448 | |
2b82032c YZ |
3449 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
3450 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3451 | start, num_bytes); |
79787eaa JM |
3452 | if (ret) |
3453 | goto error; | |
611f0e00 | 3454 | |
73c5de00 AJ |
3455 | for (i = 0; i < map->num_stripes; ++i) { |
3456 | struct btrfs_device *device; | |
3457 | u64 dev_offset; | |
3458 | ||
3459 | device = map->stripes[i].dev; | |
3460 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3461 | |
3462 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3463 | info->chunk_root->root_key.objectid, |
3464 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3465 | start, dev_offset, stripe_size); |
79787eaa JM |
3466 | if (ret) { |
3467 | btrfs_abort_transaction(trans, extent_root, ret); | |
3468 | goto error; | |
3469 | } | |
2b82032c YZ |
3470 | } |
3471 | ||
b2117a39 | 3472 | kfree(devices_info); |
2b82032c | 3473 | return 0; |
b2117a39 MX |
3474 | |
3475 | error: | |
3476 | kfree(map); | |
3477 | kfree(devices_info); | |
3478 | return ret; | |
2b82032c YZ |
3479 | } |
3480 | ||
3481 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3482 | struct btrfs_root *extent_root, | |
3483 | struct map_lookup *map, u64 chunk_offset, | |
3484 | u64 chunk_size, u64 stripe_size) | |
3485 | { | |
3486 | u64 dev_offset; | |
3487 | struct btrfs_key key; | |
3488 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3489 | struct btrfs_device *device; | |
3490 | struct btrfs_chunk *chunk; | |
3491 | struct btrfs_stripe *stripe; | |
3492 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
3493 | int index = 0; | |
3494 | int ret; | |
3495 | ||
3496 | chunk = kzalloc(item_size, GFP_NOFS); | |
3497 | if (!chunk) | |
3498 | return -ENOMEM; | |
3499 | ||
3500 | index = 0; | |
3501 | while (index < map->num_stripes) { | |
3502 | device = map->stripes[index].dev; | |
3503 | device->bytes_used += stripe_size; | |
0b86a832 | 3504 | ret = btrfs_update_device(trans, device); |
3acd3953 MF |
3505 | if (ret) |
3506 | goto out_free; | |
2b82032c YZ |
3507 | index++; |
3508 | } | |
3509 | ||
2bf64758 JB |
3510 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
3511 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
3512 | map->num_stripes); | |
3513 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
3514 | ||
2b82032c YZ |
3515 | index = 0; |
3516 | stripe = &chunk->stripe; | |
3517 | while (index < map->num_stripes) { | |
3518 | device = map->stripes[index].dev; | |
3519 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 3520 | |
e17cade2 CM |
3521 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
3522 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
3523 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 3524 | stripe++; |
0b86a832 CM |
3525 | index++; |
3526 | } | |
3527 | ||
2b82032c | 3528 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 3529 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
3530 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
3531 | btrfs_set_stack_chunk_type(chunk, map->type); | |
3532 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
3533 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
3534 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 3535 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 3536 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 3537 | |
2b82032c YZ |
3538 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3539 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3540 | key.offset = chunk_offset; | |
0b86a832 | 3541 | |
2b82032c | 3542 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
0b86a832 | 3543 | |
4ed1d16e MF |
3544 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
3545 | /* | |
3546 | * TODO: Cleanup of inserted chunk root in case of | |
3547 | * failure. | |
3548 | */ | |
125ccb0a | 3549 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 3550 | item_size); |
8f18cf13 | 3551 | } |
1abe9b8a | 3552 | |
3acd3953 | 3553 | out_free: |
0b86a832 | 3554 | kfree(chunk); |
4ed1d16e | 3555 | return ret; |
2b82032c | 3556 | } |
0b86a832 | 3557 | |
2b82032c YZ |
3558 | /* |
3559 | * Chunk allocation falls into two parts. The first part does works | |
3560 | * that make the new allocated chunk useable, but not do any operation | |
3561 | * that modifies the chunk tree. The second part does the works that | |
3562 | * require modifying the chunk tree. This division is important for the | |
3563 | * bootstrap process of adding storage to a seed btrfs. | |
3564 | */ | |
3565 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
3566 | struct btrfs_root *extent_root, u64 type) | |
3567 | { | |
3568 | u64 chunk_offset; | |
3569 | u64 chunk_size; | |
3570 | u64 stripe_size; | |
3571 | struct map_lookup *map; | |
3572 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3573 | int ret; | |
3574 | ||
3575 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3576 | &chunk_offset); | |
3577 | if (ret) | |
3578 | return ret; | |
3579 | ||
3580 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3581 | &stripe_size, chunk_offset, type); | |
3582 | if (ret) | |
3583 | return ret; | |
3584 | ||
3585 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3586 | chunk_size, stripe_size); | |
79787eaa JM |
3587 | if (ret) |
3588 | return ret; | |
2b82032c YZ |
3589 | return 0; |
3590 | } | |
3591 | ||
d397712b | 3592 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
3593 | struct btrfs_root *root, |
3594 | struct btrfs_device *device) | |
3595 | { | |
3596 | u64 chunk_offset; | |
3597 | u64 sys_chunk_offset; | |
3598 | u64 chunk_size; | |
3599 | u64 sys_chunk_size; | |
3600 | u64 stripe_size; | |
3601 | u64 sys_stripe_size; | |
3602 | u64 alloc_profile; | |
3603 | struct map_lookup *map; | |
3604 | struct map_lookup *sys_map; | |
3605 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3606 | struct btrfs_root *extent_root = fs_info->extent_root; | |
3607 | int ret; | |
3608 | ||
3609 | ret = find_next_chunk(fs_info->chunk_root, | |
3610 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
3611 | if (ret) |
3612 | return ret; | |
2b82032c YZ |
3613 | |
3614 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
6fef8df1 | 3615 | fs_info->avail_metadata_alloc_bits; |
2b82032c YZ |
3616 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3617 | ||
3618 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3619 | &stripe_size, chunk_offset, alloc_profile); | |
79787eaa JM |
3620 | if (ret) |
3621 | return ret; | |
2b82032c YZ |
3622 | |
3623 | sys_chunk_offset = chunk_offset + chunk_size; | |
3624 | ||
3625 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
6fef8df1 | 3626 | fs_info->avail_system_alloc_bits; |
2b82032c YZ |
3627 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3628 | ||
3629 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
3630 | &sys_chunk_size, &sys_stripe_size, | |
3631 | sys_chunk_offset, alloc_profile); | |
005d6427 DS |
3632 | if (ret) { |
3633 | btrfs_abort_transaction(trans, root, ret); | |
3634 | goto out; | |
3635 | } | |
2b82032c YZ |
3636 | |
3637 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
005d6427 DS |
3638 | if (ret) { |
3639 | btrfs_abort_transaction(trans, root, ret); | |
3640 | goto out; | |
3641 | } | |
2b82032c YZ |
3642 | |
3643 | /* | |
3644 | * Modifying chunk tree needs allocating new blocks from both | |
3645 | * system block group and metadata block group. So we only can | |
3646 | * do operations require modifying the chunk tree after both | |
3647 | * block groups were created. | |
3648 | */ | |
3649 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3650 | chunk_size, stripe_size); | |
005d6427 DS |
3651 | if (ret) { |
3652 | btrfs_abort_transaction(trans, root, ret); | |
3653 | goto out; | |
3654 | } | |
2b82032c YZ |
3655 | |
3656 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
3657 | sys_chunk_offset, sys_chunk_size, | |
3658 | sys_stripe_size); | |
79787eaa | 3659 | if (ret) |
005d6427 | 3660 | btrfs_abort_transaction(trans, root, ret); |
79787eaa | 3661 | |
005d6427 | 3662 | out: |
79787eaa | 3663 | |
79787eaa | 3664 | return ret; |
2b82032c YZ |
3665 | } |
3666 | ||
3667 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
3668 | { | |
3669 | struct extent_map *em; | |
3670 | struct map_lookup *map; | |
3671 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3672 | int readonly = 0; | |
3673 | int i; | |
3674 | ||
890871be | 3675 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 3676 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 3677 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
3678 | if (!em) |
3679 | return 1; | |
3680 | ||
f48b9075 JB |
3681 | if (btrfs_test_opt(root, DEGRADED)) { |
3682 | free_extent_map(em); | |
3683 | return 0; | |
3684 | } | |
3685 | ||
2b82032c YZ |
3686 | map = (struct map_lookup *)em->bdev; |
3687 | for (i = 0; i < map->num_stripes; i++) { | |
3688 | if (!map->stripes[i].dev->writeable) { | |
3689 | readonly = 1; | |
3690 | break; | |
3691 | } | |
3692 | } | |
0b86a832 | 3693 | free_extent_map(em); |
2b82032c | 3694 | return readonly; |
0b86a832 CM |
3695 | } |
3696 | ||
3697 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
3698 | { | |
a8067e02 | 3699 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
3700 | } |
3701 | ||
3702 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
3703 | { | |
3704 | struct extent_map *em; | |
3705 | ||
d397712b | 3706 | while (1) { |
890871be | 3707 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
3708 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
3709 | if (em) | |
3710 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 3711 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
3712 | if (!em) |
3713 | break; | |
3714 | kfree(em->bdev); | |
3715 | /* once for us */ | |
3716 | free_extent_map(em); | |
3717 | /* once for the tree */ | |
3718 | free_extent_map(em); | |
3719 | } | |
3720 | } | |
3721 | ||
f188591e CM |
3722 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
3723 | { | |
3724 | struct extent_map *em; | |
3725 | struct map_lookup *map; | |
3726 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3727 | int ret; | |
3728 | ||
890871be | 3729 | read_lock(&em_tree->lock); |
f188591e | 3730 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 3731 | read_unlock(&em_tree->lock); |
f188591e CM |
3732 | BUG_ON(!em); |
3733 | ||
3734 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3735 | map = (struct map_lookup *)em->bdev; | |
3736 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
3737 | ret = map->num_stripes; | |
321aecc6 CM |
3738 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
3739 | ret = map->sub_stripes; | |
f188591e CM |
3740 | else |
3741 | ret = 1; | |
3742 | free_extent_map(em); | |
f188591e CM |
3743 | return ret; |
3744 | } | |
3745 | ||
dfe25020 CM |
3746 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
3747 | int optimal) | |
3748 | { | |
3749 | int i; | |
3750 | if (map->stripes[optimal].dev->bdev) | |
3751 | return optimal; | |
3752 | for (i = first; i < first + num; i++) { | |
3753 | if (map->stripes[i].dev->bdev) | |
3754 | return i; | |
3755 | } | |
3756 | /* we couldn't find one that doesn't fail. Just return something | |
3757 | * and the io error handling code will clean up eventually | |
3758 | */ | |
3759 | return optimal; | |
3760 | } | |
3761 | ||
f2d8d74d CM |
3762 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3763 | u64 logical, u64 *length, | |
a1d3c478 | 3764 | struct btrfs_bio **bbio_ret, |
7eaceacc | 3765 | int mirror_num) |
0b86a832 CM |
3766 | { |
3767 | struct extent_map *em; | |
3768 | struct map_lookup *map; | |
3769 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3770 | u64 offset; | |
593060d7 | 3771 | u64 stripe_offset; |
fce3bb9a | 3772 | u64 stripe_end_offset; |
593060d7 | 3773 | u64 stripe_nr; |
fce3bb9a LD |
3774 | u64 stripe_nr_orig; |
3775 | u64 stripe_nr_end; | |
593060d7 | 3776 | int stripe_index; |
cea9e445 | 3777 | int i; |
de11cc12 | 3778 | int ret = 0; |
f2d8d74d | 3779 | int num_stripes; |
a236aed1 | 3780 | int max_errors = 0; |
a1d3c478 | 3781 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3782 | |
890871be | 3783 | read_lock(&em_tree->lock); |
0b86a832 | 3784 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 3785 | read_unlock(&em_tree->lock); |
f2d8d74d | 3786 | |
3b951516 | 3787 | if (!em) { |
48940662 | 3788 | printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n", |
d397712b CM |
3789 | (unsigned long long)logical, |
3790 | (unsigned long long)*length); | |
f2d8d74d | 3791 | BUG(); |
3b951516 | 3792 | } |
0b86a832 CM |
3793 | |
3794 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3795 | map = (struct map_lookup *)em->bdev; | |
3796 | offset = logical - em->start; | |
593060d7 | 3797 | |
f188591e CM |
3798 | if (mirror_num > map->num_stripes) |
3799 | mirror_num = 0; | |
3800 | ||
593060d7 CM |
3801 | stripe_nr = offset; |
3802 | /* | |
3803 | * stripe_nr counts the total number of stripes we have to stride | |
3804 | * to get to this block | |
3805 | */ | |
3806 | do_div(stripe_nr, map->stripe_len); | |
3807 | ||
3808 | stripe_offset = stripe_nr * map->stripe_len; | |
3809 | BUG_ON(offset < stripe_offset); | |
3810 | ||
3811 | /* stripe_offset is the offset of this block in its stripe*/ | |
3812 | stripe_offset = offset - stripe_offset; | |
3813 | ||
fce3bb9a LD |
3814 | if (rw & REQ_DISCARD) |
3815 | *length = min_t(u64, em->len - offset, *length); | |
52ba6929 | 3816 | else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
cea9e445 CM |
3817 | /* we limit the length of each bio to what fits in a stripe */ |
3818 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 3819 | map->stripe_len - stripe_offset); |
cea9e445 CM |
3820 | } else { |
3821 | *length = em->len - offset; | |
3822 | } | |
f2d8d74d | 3823 | |
a1d3c478 | 3824 | if (!bbio_ret) |
cea9e445 CM |
3825 | goto out; |
3826 | ||
f2d8d74d | 3827 | num_stripes = 1; |
cea9e445 | 3828 | stripe_index = 0; |
fce3bb9a LD |
3829 | stripe_nr_orig = stripe_nr; |
3830 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
3831 | (~(map->stripe_len - 1)); | |
3832 | do_div(stripe_nr_end, map->stripe_len); | |
3833 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
3834 | (offset + *length); | |
3835 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3836 | if (rw & REQ_DISCARD) | |
3837 | num_stripes = min_t(u64, map->num_stripes, | |
3838 | stripe_nr_end - stripe_nr_orig); | |
3839 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3840 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3841 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3842 | num_stripes = map->num_stripes; |
2fff734f | 3843 | else if (mirror_num) |
f188591e | 3844 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3845 | else { |
3846 | stripe_index = find_live_mirror(map, 0, | |
3847 | map->num_stripes, | |
3848 | current->pid % map->num_stripes); | |
a1d3c478 | 3849 | mirror_num = stripe_index + 1; |
dfe25020 | 3850 | } |
2fff734f | 3851 | |
611f0e00 | 3852 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3853 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3854 | num_stripes = map->num_stripes; |
a1d3c478 | 3855 | } else if (mirror_num) { |
f188591e | 3856 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3857 | } else { |
3858 | mirror_num = 1; | |
3859 | } | |
2fff734f | 3860 | |
321aecc6 CM |
3861 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3862 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3863 | |
3864 | stripe_index = do_div(stripe_nr, factor); | |
3865 | stripe_index *= map->sub_stripes; | |
3866 | ||
7eaceacc | 3867 | if (rw & REQ_WRITE) |
f2d8d74d | 3868 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3869 | else if (rw & REQ_DISCARD) |
3870 | num_stripes = min_t(u64, map->sub_stripes * | |
3871 | (stripe_nr_end - stripe_nr_orig), | |
3872 | map->num_stripes); | |
321aecc6 CM |
3873 | else if (mirror_num) |
3874 | stripe_index += mirror_num - 1; | |
dfe25020 | 3875 | else { |
3e74317a | 3876 | int old_stripe_index = stripe_index; |
dfe25020 CM |
3877 | stripe_index = find_live_mirror(map, stripe_index, |
3878 | map->sub_stripes, stripe_index + | |
3879 | current->pid % map->sub_stripes); | |
3e74317a | 3880 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 3881 | } |
8790d502 CM |
3882 | } else { |
3883 | /* | |
3884 | * after this do_div call, stripe_nr is the number of stripes | |
3885 | * on this device we have to walk to find the data, and | |
3886 | * stripe_index is the number of our device in the stripe array | |
3887 | */ | |
3888 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3889 | mirror_num = stripe_index + 1; |
8790d502 | 3890 | } |
593060d7 | 3891 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3892 | |
de11cc12 LZ |
3893 | bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS); |
3894 | if (!bbio) { | |
3895 | ret = -ENOMEM; | |
3896 | goto out; | |
3897 | } | |
3898 | atomic_set(&bbio->error, 0); | |
3899 | ||
fce3bb9a | 3900 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
3901 | int factor = 0; |
3902 | int sub_stripes = 0; | |
3903 | u64 stripes_per_dev = 0; | |
3904 | u32 remaining_stripes = 0; | |
b89203f7 | 3905 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
3906 | |
3907 | if (map->type & | |
3908 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3909 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3910 | sub_stripes = 1; | |
3911 | else | |
3912 | sub_stripes = map->sub_stripes; | |
3913 | ||
3914 | factor = map->num_stripes / sub_stripes; | |
3915 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
3916 | stripe_nr_orig, | |
3917 | factor, | |
3918 | &remaining_stripes); | |
b89203f7 LB |
3919 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
3920 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
3921 | } |
3922 | ||
fce3bb9a | 3923 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 3924 | bbio->stripes[i].physical = |
f2d8d74d CM |
3925 | map->stripes[stripe_index].physical + |
3926 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3927 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 3928 | |
ec9ef7a1 LZ |
3929 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
3930 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3931 | bbio->stripes[i].length = stripes_per_dev * | |
3932 | map->stripe_len; | |
b89203f7 | 3933 | |
ec9ef7a1 LZ |
3934 | if (i / sub_stripes < remaining_stripes) |
3935 | bbio->stripes[i].length += | |
3936 | map->stripe_len; | |
b89203f7 LB |
3937 | |
3938 | /* | |
3939 | * Special for the first stripe and | |
3940 | * the last stripe: | |
3941 | * | |
3942 | * |-------|...|-------| | |
3943 | * |----------| | |
3944 | * off end_off | |
3945 | */ | |
ec9ef7a1 | 3946 | if (i < sub_stripes) |
a1d3c478 | 3947 | bbio->stripes[i].length -= |
fce3bb9a | 3948 | stripe_offset; |
b89203f7 LB |
3949 | |
3950 | if (stripe_index >= last_stripe && | |
3951 | stripe_index <= (last_stripe + | |
3952 | sub_stripes - 1)) | |
a1d3c478 | 3953 | bbio->stripes[i].length -= |
fce3bb9a | 3954 | stripe_end_offset; |
b89203f7 | 3955 | |
ec9ef7a1 LZ |
3956 | if (i == sub_stripes - 1) |
3957 | stripe_offset = 0; | |
fce3bb9a | 3958 | } else |
a1d3c478 | 3959 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
3960 | |
3961 | stripe_index++; | |
3962 | if (stripe_index == map->num_stripes) { | |
3963 | /* This could only happen for RAID0/10 */ | |
3964 | stripe_index = 0; | |
3965 | stripe_nr++; | |
3966 | } | |
3967 | } | |
3968 | } else { | |
3969 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3970 | bbio->stripes[i].physical = |
212a17ab LT |
3971 | map->stripes[stripe_index].physical + |
3972 | stripe_offset + | |
3973 | stripe_nr * map->stripe_len; | |
a1d3c478 | 3974 | bbio->stripes[i].dev = |
212a17ab | 3975 | map->stripes[stripe_index].dev; |
fce3bb9a | 3976 | stripe_index++; |
f2d8d74d | 3977 | } |
593060d7 | 3978 | } |
de11cc12 LZ |
3979 | |
3980 | if (rw & REQ_WRITE) { | |
3981 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
3982 | BTRFS_BLOCK_GROUP_RAID10 | | |
3983 | BTRFS_BLOCK_GROUP_DUP)) { | |
3984 | max_errors = 1; | |
3985 | } | |
f2d8d74d | 3986 | } |
de11cc12 LZ |
3987 | |
3988 | *bbio_ret = bbio; | |
3989 | bbio->num_stripes = num_stripes; | |
3990 | bbio->max_errors = max_errors; | |
3991 | bbio->mirror_num = mirror_num; | |
cea9e445 | 3992 | out: |
0b86a832 | 3993 | free_extent_map(em); |
de11cc12 | 3994 | return ret; |
0b86a832 CM |
3995 | } |
3996 | ||
f2d8d74d CM |
3997 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3998 | u64 logical, u64 *length, | |
a1d3c478 | 3999 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 4000 | { |
a1d3c478 | 4001 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 4002 | mirror_num); |
f2d8d74d CM |
4003 | } |
4004 | ||
a512bbf8 YZ |
4005 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
4006 | u64 chunk_start, u64 physical, u64 devid, | |
4007 | u64 **logical, int *naddrs, int *stripe_len) | |
4008 | { | |
4009 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4010 | struct extent_map *em; | |
4011 | struct map_lookup *map; | |
4012 | u64 *buf; | |
4013 | u64 bytenr; | |
4014 | u64 length; | |
4015 | u64 stripe_nr; | |
4016 | int i, j, nr = 0; | |
4017 | ||
890871be | 4018 | read_lock(&em_tree->lock); |
a512bbf8 | 4019 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 4020 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
4021 | |
4022 | BUG_ON(!em || em->start != chunk_start); | |
4023 | map = (struct map_lookup *)em->bdev; | |
4024 | ||
4025 | length = em->len; | |
4026 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
4027 | do_div(length, map->num_stripes / map->sub_stripes); | |
4028 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4029 | do_div(length, map->num_stripes); | |
4030 | ||
4031 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 4032 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
4033 | |
4034 | for (i = 0; i < map->num_stripes; i++) { | |
4035 | if (devid && map->stripes[i].dev->devid != devid) | |
4036 | continue; | |
4037 | if (map->stripes[i].physical > physical || | |
4038 | map->stripes[i].physical + length <= physical) | |
4039 | continue; | |
4040 | ||
4041 | stripe_nr = physical - map->stripes[i].physical; | |
4042 | do_div(stripe_nr, map->stripe_len); | |
4043 | ||
4044 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
4045 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4046 | do_div(stripe_nr, map->sub_stripes); | |
4047 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
4048 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4049 | } | |
4050 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 4051 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
4052 | for (j = 0; j < nr; j++) { |
4053 | if (buf[j] == bytenr) | |
4054 | break; | |
4055 | } | |
934d375b CM |
4056 | if (j == nr) { |
4057 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 4058 | buf[nr++] = bytenr; |
934d375b | 4059 | } |
a512bbf8 YZ |
4060 | } |
4061 | ||
a512bbf8 YZ |
4062 | *logical = buf; |
4063 | *naddrs = nr; | |
4064 | *stripe_len = map->stripe_len; | |
4065 | ||
4066 | free_extent_map(em); | |
4067 | return 0; | |
f2d8d74d CM |
4068 | } |
4069 | ||
442a4f63 SB |
4070 | static void *merge_stripe_index_into_bio_private(void *bi_private, |
4071 | unsigned int stripe_index) | |
4072 | { | |
4073 | /* | |
4074 | * with single, dup, RAID0, RAID1 and RAID10, stripe_index is | |
4075 | * at most 1. | |
4076 | * The alternative solution (instead of stealing bits from the | |
4077 | * pointer) would be to allocate an intermediate structure | |
4078 | * that contains the old private pointer plus the stripe_index. | |
4079 | */ | |
4080 | BUG_ON((((uintptr_t)bi_private) & 3) != 0); | |
4081 | BUG_ON(stripe_index > 3); | |
4082 | return (void *)(((uintptr_t)bi_private) | stripe_index); | |
4083 | } | |
4084 | ||
4085 | static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private) | |
4086 | { | |
4087 | return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3)); | |
4088 | } | |
4089 | ||
4090 | static unsigned int extract_stripe_index_from_bio_private(void *bi_private) | |
4091 | { | |
4092 | return (unsigned int)((uintptr_t)bi_private) & 3; | |
4093 | } | |
4094 | ||
a1d3c478 | 4095 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 4096 | { |
442a4f63 | 4097 | struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private); |
7d2b4daa | 4098 | int is_orig_bio = 0; |
8790d502 | 4099 | |
442a4f63 | 4100 | if (err) { |
a1d3c478 | 4101 | atomic_inc(&bbio->error); |
442a4f63 SB |
4102 | if (err == -EIO || err == -EREMOTEIO) { |
4103 | unsigned int stripe_index = | |
4104 | extract_stripe_index_from_bio_private( | |
4105 | bio->bi_private); | |
4106 | struct btrfs_device *dev; | |
4107 | ||
4108 | BUG_ON(stripe_index >= bbio->num_stripes); | |
4109 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
4110 | if (dev->bdev) { |
4111 | if (bio->bi_rw & WRITE) | |
4112 | btrfs_dev_stat_inc(dev, | |
4113 | BTRFS_DEV_STAT_WRITE_ERRS); | |
4114 | else | |
4115 | btrfs_dev_stat_inc(dev, | |
4116 | BTRFS_DEV_STAT_READ_ERRS); | |
4117 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
4118 | btrfs_dev_stat_inc(dev, | |
4119 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
4120 | btrfs_dev_stat_print_on_error(dev); | |
4121 | } | |
442a4f63 SB |
4122 | } |
4123 | } | |
8790d502 | 4124 | |
a1d3c478 | 4125 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
4126 | is_orig_bio = 1; |
4127 | ||
a1d3c478 | 4128 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
4129 | if (!is_orig_bio) { |
4130 | bio_put(bio); | |
a1d3c478 | 4131 | bio = bbio->orig_bio; |
7d2b4daa | 4132 | } |
a1d3c478 JS |
4133 | bio->bi_private = bbio->private; |
4134 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
4135 | bio->bi_bdev = (struct block_device *) |
4136 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
4137 | /* only send an error to the higher layers if it is |
4138 | * beyond the tolerance of the multi-bio | |
4139 | */ | |
a1d3c478 | 4140 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 4141 | err = -EIO; |
5dbc8fca | 4142 | } else { |
1259ab75 CM |
4143 | /* |
4144 | * this bio is actually up to date, we didn't | |
4145 | * go over the max number of errors | |
4146 | */ | |
4147 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 4148 | err = 0; |
1259ab75 | 4149 | } |
a1d3c478 | 4150 | kfree(bbio); |
8790d502 CM |
4151 | |
4152 | bio_endio(bio, err); | |
7d2b4daa | 4153 | } else if (!is_orig_bio) { |
8790d502 CM |
4154 | bio_put(bio); |
4155 | } | |
8790d502 CM |
4156 | } |
4157 | ||
8b712842 CM |
4158 | struct async_sched { |
4159 | struct bio *bio; | |
4160 | int rw; | |
4161 | struct btrfs_fs_info *info; | |
4162 | struct btrfs_work work; | |
4163 | }; | |
4164 | ||
4165 | /* | |
4166 | * see run_scheduled_bios for a description of why bios are collected for | |
4167 | * async submit. | |
4168 | * | |
4169 | * This will add one bio to the pending list for a device and make sure | |
4170 | * the work struct is scheduled. | |
4171 | */ | |
143bede5 | 4172 | static noinline void schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
4173 | struct btrfs_device *device, |
4174 | int rw, struct bio *bio) | |
8b712842 CM |
4175 | { |
4176 | int should_queue = 1; | |
ffbd517d | 4177 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
4178 | |
4179 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 4180 | if (!(rw & REQ_WRITE)) { |
492bb6de | 4181 | bio_get(bio); |
21adbd5c | 4182 | btrfsic_submit_bio(rw, bio); |
492bb6de | 4183 | bio_put(bio); |
143bede5 | 4184 | return; |
8b712842 CM |
4185 | } |
4186 | ||
4187 | /* | |
0986fe9e | 4188 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
4189 | * higher layers. Otherwise, the async bio makes it appear we have |
4190 | * made progress against dirty pages when we've really just put it | |
4191 | * on a queue for later | |
4192 | */ | |
0986fe9e | 4193 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 4194 | WARN_ON(bio->bi_next); |
8b712842 CM |
4195 | bio->bi_next = NULL; |
4196 | bio->bi_rw |= rw; | |
4197 | ||
4198 | spin_lock(&device->io_lock); | |
7b6d91da | 4199 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
4200 | pending_bios = &device->pending_sync_bios; |
4201 | else | |
4202 | pending_bios = &device->pending_bios; | |
8b712842 | 4203 | |
ffbd517d CM |
4204 | if (pending_bios->tail) |
4205 | pending_bios->tail->bi_next = bio; | |
8b712842 | 4206 | |
ffbd517d CM |
4207 | pending_bios->tail = bio; |
4208 | if (!pending_bios->head) | |
4209 | pending_bios->head = bio; | |
8b712842 CM |
4210 | if (device->running_pending) |
4211 | should_queue = 0; | |
4212 | ||
4213 | spin_unlock(&device->io_lock); | |
4214 | ||
4215 | if (should_queue) | |
1cc127b5 CM |
4216 | btrfs_queue_worker(&root->fs_info->submit_workers, |
4217 | &device->work); | |
8b712842 CM |
4218 | } |
4219 | ||
de1ee92a JB |
4220 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
4221 | sector_t sector) | |
4222 | { | |
4223 | struct bio_vec *prev; | |
4224 | struct request_queue *q = bdev_get_queue(bdev); | |
4225 | unsigned short max_sectors = queue_max_sectors(q); | |
4226 | struct bvec_merge_data bvm = { | |
4227 | .bi_bdev = bdev, | |
4228 | .bi_sector = sector, | |
4229 | .bi_rw = bio->bi_rw, | |
4230 | }; | |
4231 | ||
4232 | if (bio->bi_vcnt == 0) { | |
4233 | WARN_ON(1); | |
4234 | return 1; | |
4235 | } | |
4236 | ||
4237 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
4238 | if ((bio->bi_size >> 9) > max_sectors) | |
4239 | return 0; | |
4240 | ||
4241 | if (!q->merge_bvec_fn) | |
4242 | return 1; | |
4243 | ||
4244 | bvm.bi_size = bio->bi_size - prev->bv_len; | |
4245 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) | |
4246 | return 0; | |
4247 | return 1; | |
4248 | } | |
4249 | ||
4250 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
4251 | struct bio *bio, u64 physical, int dev_nr, | |
4252 | int rw, int async) | |
4253 | { | |
4254 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
4255 | ||
4256 | bio->bi_private = bbio; | |
4257 | bio->bi_private = merge_stripe_index_into_bio_private( | |
4258 | bio->bi_private, (unsigned int)dev_nr); | |
4259 | bio->bi_end_io = btrfs_end_bio; | |
4260 | bio->bi_sector = physical >> 9; | |
4261 | #ifdef DEBUG | |
4262 | { | |
4263 | struct rcu_string *name; | |
4264 | ||
4265 | rcu_read_lock(); | |
4266 | name = rcu_dereference(dev->name); | |
d1423248 | 4267 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a JB |
4268 | "(%s id %llu), size=%u\n", rw, |
4269 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
4270 | name->str, dev->devid, bio->bi_size); | |
4271 | rcu_read_unlock(); | |
4272 | } | |
4273 | #endif | |
4274 | bio->bi_bdev = dev->bdev; | |
4275 | if (async) | |
4276 | schedule_bio(root, dev, rw, bio); | |
4277 | else | |
4278 | btrfsic_submit_bio(rw, bio); | |
4279 | } | |
4280 | ||
4281 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
4282 | struct bio *first_bio, struct btrfs_device *dev, | |
4283 | int dev_nr, int rw, int async) | |
4284 | { | |
4285 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
4286 | struct bio *bio; | |
4287 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
4288 | u64 physical = bbio->stripes[dev_nr].physical; | |
4289 | ||
4290 | again: | |
4291 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
4292 | if (!bio) | |
4293 | return -ENOMEM; | |
4294 | ||
4295 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
4296 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
4297 | bvec->bv_offset) < bvec->bv_len) { | |
4298 | u64 len = bio->bi_size; | |
4299 | ||
4300 | atomic_inc(&bbio->stripes_pending); | |
4301 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
4302 | rw, async); | |
4303 | physical += len; | |
4304 | goto again; | |
4305 | } | |
4306 | bvec++; | |
4307 | } | |
4308 | ||
4309 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
4310 | return 0; | |
4311 | } | |
4312 | ||
4313 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
4314 | { | |
4315 | atomic_inc(&bbio->error); | |
4316 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
4317 | bio->bi_private = bbio->private; | |
4318 | bio->bi_end_io = bbio->end_io; | |
4319 | bio->bi_bdev = (struct block_device *) | |
4320 | (unsigned long)bbio->mirror_num; | |
4321 | bio->bi_sector = logical >> 9; | |
4322 | kfree(bbio); | |
4323 | bio_endio(bio, -EIO); | |
4324 | } | |
4325 | } | |
4326 | ||
f188591e | 4327 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 4328 | int mirror_num, int async_submit) |
0b86a832 CM |
4329 | { |
4330 | struct btrfs_mapping_tree *map_tree; | |
4331 | struct btrfs_device *dev; | |
8790d502 | 4332 | struct bio *first_bio = bio; |
a62b9401 | 4333 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
4334 | u64 length = 0; |
4335 | u64 map_length; | |
0b86a832 | 4336 | int ret; |
8790d502 CM |
4337 | int dev_nr = 0; |
4338 | int total_devs = 1; | |
a1d3c478 | 4339 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 4340 | |
f2d8d74d | 4341 | length = bio->bi_size; |
0b86a832 CM |
4342 | map_tree = &root->fs_info->mapping_tree; |
4343 | map_length = length; | |
cea9e445 | 4344 | |
a1d3c478 | 4345 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 4346 | mirror_num); |
79787eaa JM |
4347 | if (ret) /* -ENOMEM */ |
4348 | return ret; | |
cea9e445 | 4349 | |
a1d3c478 | 4350 | total_devs = bbio->num_stripes; |
cea9e445 | 4351 | if (map_length < length) { |
48940662 | 4352 | printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu " |
d397712b CM |
4353 | "len %llu\n", (unsigned long long)logical, |
4354 | (unsigned long long)length, | |
4355 | (unsigned long long)map_length); | |
cea9e445 CM |
4356 | BUG(); |
4357 | } | |
a1d3c478 JS |
4358 | |
4359 | bbio->orig_bio = first_bio; | |
4360 | bbio->private = first_bio->bi_private; | |
4361 | bbio->end_io = first_bio->bi_end_io; | |
4362 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 4363 | |
d397712b | 4364 | while (dev_nr < total_devs) { |
de1ee92a JB |
4365 | dev = bbio->stripes[dev_nr].dev; |
4366 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
4367 | bbio_error(bbio, first_bio, logical); | |
4368 | dev_nr++; | |
4369 | continue; | |
4370 | } | |
4371 | ||
4372 | /* | |
4373 | * Check and see if we're ok with this bio based on it's size | |
4374 | * and offset with the given device. | |
4375 | */ | |
4376 | if (!bio_size_ok(dev->bdev, first_bio, | |
4377 | bbio->stripes[dev_nr].physical >> 9)) { | |
4378 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
4379 | dev_nr, rw, async_submit); | |
4380 | BUG_ON(ret); | |
4381 | dev_nr++; | |
4382 | continue; | |
4383 | } | |
4384 | ||
a1d3c478 JS |
4385 | if (dev_nr < total_devs - 1) { |
4386 | bio = bio_clone(first_bio, GFP_NOFS); | |
79787eaa | 4387 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
4388 | } else { |
4389 | bio = first_bio; | |
8790d502 | 4390 | } |
de1ee92a JB |
4391 | |
4392 | submit_stripe_bio(root, bbio, bio, | |
4393 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
4394 | async_submit); | |
8790d502 CM |
4395 | dev_nr++; |
4396 | } | |
0b86a832 CM |
4397 | return 0; |
4398 | } | |
4399 | ||
a443755f | 4400 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 4401 | u8 *uuid, u8 *fsid) |
0b86a832 | 4402 | { |
2b82032c YZ |
4403 | struct btrfs_device *device; |
4404 | struct btrfs_fs_devices *cur_devices; | |
4405 | ||
4406 | cur_devices = root->fs_info->fs_devices; | |
4407 | while (cur_devices) { | |
4408 | if (!fsid || | |
4409 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4410 | device = __find_device(&cur_devices->devices, | |
4411 | devid, uuid); | |
4412 | if (device) | |
4413 | return device; | |
4414 | } | |
4415 | cur_devices = cur_devices->seed; | |
4416 | } | |
4417 | return NULL; | |
0b86a832 CM |
4418 | } |
4419 | ||
dfe25020 CM |
4420 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
4421 | u64 devid, u8 *dev_uuid) | |
4422 | { | |
4423 | struct btrfs_device *device; | |
4424 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4425 | ||
4426 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 4427 | if (!device) |
4428 | return NULL; | |
dfe25020 CM |
4429 | list_add(&device->dev_list, |
4430 | &fs_devices->devices); | |
dfe25020 CM |
4431 | device->dev_root = root->fs_info->dev_root; |
4432 | device->devid = devid; | |
8b712842 | 4433 | device->work.func = pending_bios_fn; |
e4404d6e | 4434 | device->fs_devices = fs_devices; |
cd02dca5 | 4435 | device->missing = 1; |
dfe25020 | 4436 | fs_devices->num_devices++; |
cd02dca5 | 4437 | fs_devices->missing_devices++; |
dfe25020 | 4438 | spin_lock_init(&device->io_lock); |
d20f7043 | 4439 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
4440 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
4441 | return device; | |
4442 | } | |
4443 | ||
0b86a832 CM |
4444 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
4445 | struct extent_buffer *leaf, | |
4446 | struct btrfs_chunk *chunk) | |
4447 | { | |
4448 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4449 | struct map_lookup *map; | |
4450 | struct extent_map *em; | |
4451 | u64 logical; | |
4452 | u64 length; | |
4453 | u64 devid; | |
a443755f | 4454 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 4455 | int num_stripes; |
0b86a832 | 4456 | int ret; |
593060d7 | 4457 | int i; |
0b86a832 | 4458 | |
e17cade2 CM |
4459 | logical = key->offset; |
4460 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 4461 | |
890871be | 4462 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 4463 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 4464 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
4465 | |
4466 | /* already mapped? */ | |
4467 | if (em && em->start <= logical && em->start + em->len > logical) { | |
4468 | free_extent_map(em); | |
0b86a832 CM |
4469 | return 0; |
4470 | } else if (em) { | |
4471 | free_extent_map(em); | |
4472 | } | |
0b86a832 | 4473 | |
172ddd60 | 4474 | em = alloc_extent_map(); |
0b86a832 CM |
4475 | if (!em) |
4476 | return -ENOMEM; | |
593060d7 CM |
4477 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
4478 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
4479 | if (!map) { |
4480 | free_extent_map(em); | |
4481 | return -ENOMEM; | |
4482 | } | |
4483 | ||
4484 | em->bdev = (struct block_device *)map; | |
4485 | em->start = logical; | |
4486 | em->len = length; | |
4487 | em->block_start = 0; | |
c8b97818 | 4488 | em->block_len = em->len; |
0b86a832 | 4489 | |
593060d7 CM |
4490 | map->num_stripes = num_stripes; |
4491 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
4492 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
4493 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
4494 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
4495 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 4496 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
4497 | for (i = 0; i < num_stripes; i++) { |
4498 | map->stripes[i].physical = | |
4499 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
4500 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
4501 | read_extent_buffer(leaf, uuid, (unsigned long) |
4502 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
4503 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4504 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
4505 | NULL); | |
dfe25020 | 4506 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
4507 | kfree(map); |
4508 | free_extent_map(em); | |
4509 | return -EIO; | |
4510 | } | |
dfe25020 CM |
4511 | if (!map->stripes[i].dev) { |
4512 | map->stripes[i].dev = | |
4513 | add_missing_dev(root, devid, uuid); | |
4514 | if (!map->stripes[i].dev) { | |
4515 | kfree(map); | |
4516 | free_extent_map(em); | |
4517 | return -EIO; | |
4518 | } | |
4519 | } | |
4520 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
4521 | } |
4522 | ||
890871be | 4523 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 4524 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 4525 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 4526 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
4527 | free_extent_map(em); |
4528 | ||
4529 | return 0; | |
4530 | } | |
4531 | ||
143bede5 | 4532 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
4533 | struct btrfs_dev_item *dev_item, |
4534 | struct btrfs_device *device) | |
4535 | { | |
4536 | unsigned long ptr; | |
0b86a832 CM |
4537 | |
4538 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
4539 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
4540 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
4541 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
4542 | device->type = btrfs_device_type(leaf, dev_item); | |
4543 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
4544 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
4545 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
4546 | |
4547 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 4548 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
4549 | } |
4550 | ||
2b82032c YZ |
4551 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
4552 | { | |
4553 | struct btrfs_fs_devices *fs_devices; | |
4554 | int ret; | |
4555 | ||
b367e47f | 4556 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
4557 | |
4558 | fs_devices = root->fs_info->fs_devices->seed; | |
4559 | while (fs_devices) { | |
4560 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4561 | ret = 0; | |
4562 | goto out; | |
4563 | } | |
4564 | fs_devices = fs_devices->seed; | |
4565 | } | |
4566 | ||
4567 | fs_devices = find_fsid(fsid); | |
4568 | if (!fs_devices) { | |
4569 | ret = -ENOENT; | |
4570 | goto out; | |
4571 | } | |
e4404d6e YZ |
4572 | |
4573 | fs_devices = clone_fs_devices(fs_devices); | |
4574 | if (IS_ERR(fs_devices)) { | |
4575 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
4576 | goto out; |
4577 | } | |
4578 | ||
97288f2c | 4579 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 4580 | root->fs_info->bdev_holder); |
48d28232 JL |
4581 | if (ret) { |
4582 | free_fs_devices(fs_devices); | |
2b82032c | 4583 | goto out; |
48d28232 | 4584 | } |
2b82032c YZ |
4585 | |
4586 | if (!fs_devices->seeding) { | |
4587 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 4588 | free_fs_devices(fs_devices); |
2b82032c YZ |
4589 | ret = -EINVAL; |
4590 | goto out; | |
4591 | } | |
4592 | ||
4593 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
4594 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 4595 | out: |
2b82032c YZ |
4596 | return ret; |
4597 | } | |
4598 | ||
0d81ba5d | 4599 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
4600 | struct extent_buffer *leaf, |
4601 | struct btrfs_dev_item *dev_item) | |
4602 | { | |
4603 | struct btrfs_device *device; | |
4604 | u64 devid; | |
4605 | int ret; | |
2b82032c | 4606 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
4607 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
4608 | ||
0b86a832 | 4609 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
4610 | read_extent_buffer(leaf, dev_uuid, |
4611 | (unsigned long)btrfs_device_uuid(dev_item), | |
4612 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4613 | read_extent_buffer(leaf, fs_uuid, |
4614 | (unsigned long)btrfs_device_fsid(dev_item), | |
4615 | BTRFS_UUID_SIZE); | |
4616 | ||
4617 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
4618 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 4619 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 4620 | return ret; |
2b82032c YZ |
4621 | } |
4622 | ||
4623 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
4624 | if (!device || !device->bdev) { | |
e4404d6e | 4625 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
4626 | return -EIO; |
4627 | ||
4628 | if (!device) { | |
d397712b CM |
4629 | printk(KERN_WARNING "warning devid %llu missing\n", |
4630 | (unsigned long long)devid); | |
2b82032c YZ |
4631 | device = add_missing_dev(root, devid, dev_uuid); |
4632 | if (!device) | |
4633 | return -ENOMEM; | |
cd02dca5 CM |
4634 | } else if (!device->missing) { |
4635 | /* | |
4636 | * this happens when a device that was properly setup | |
4637 | * in the device info lists suddenly goes bad. | |
4638 | * device->bdev is NULL, and so we have to set | |
4639 | * device->missing to one here | |
4640 | */ | |
4641 | root->fs_info->fs_devices->missing_devices++; | |
4642 | device->missing = 1; | |
2b82032c YZ |
4643 | } |
4644 | } | |
4645 | ||
4646 | if (device->fs_devices != root->fs_info->fs_devices) { | |
4647 | BUG_ON(device->writeable); | |
4648 | if (device->generation != | |
4649 | btrfs_device_generation(leaf, dev_item)) | |
4650 | return -EINVAL; | |
6324fbf3 | 4651 | } |
0b86a832 CM |
4652 | |
4653 | fill_device_from_item(leaf, dev_item, device); | |
4654 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 4655 | device->in_fs_metadata = 1; |
2bf64758 | 4656 | if (device->writeable) { |
2b82032c | 4657 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
4658 | spin_lock(&root->fs_info->free_chunk_lock); |
4659 | root->fs_info->free_chunk_space += device->total_bytes - | |
4660 | device->bytes_used; | |
4661 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4662 | } | |
0b86a832 | 4663 | ret = 0; |
0b86a832 CM |
4664 | return ret; |
4665 | } | |
4666 | ||
e4404d6e | 4667 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 4668 | { |
6c41761f | 4669 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 4670 | struct extent_buffer *sb; |
0b86a832 | 4671 | struct btrfs_disk_key *disk_key; |
0b86a832 | 4672 | struct btrfs_chunk *chunk; |
84eed90f CM |
4673 | u8 *ptr; |
4674 | unsigned long sb_ptr; | |
4675 | int ret = 0; | |
0b86a832 CM |
4676 | u32 num_stripes; |
4677 | u32 array_size; | |
4678 | u32 len = 0; | |
0b86a832 | 4679 | u32 cur; |
84eed90f | 4680 | struct btrfs_key key; |
0b86a832 | 4681 | |
e4404d6e | 4682 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
4683 | BTRFS_SUPER_INFO_SIZE); |
4684 | if (!sb) | |
4685 | return -ENOMEM; | |
4686 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 4687 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
4688 | /* |
4689 | * The sb extent buffer is artifical and just used to read the system array. | |
4690 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
4691 | * pages up-to-date when the page is larger: extent does not cover the | |
4692 | * whole page and consequently check_page_uptodate does not find all | |
4693 | * the page's extents up-to-date (the hole beyond sb), | |
4694 | * write_extent_buffer then triggers a WARN_ON. | |
4695 | * | |
4696 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
4697 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
4698 | * to silence the warning eg. on PowerPC 64. | |
4699 | */ | |
4700 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 4701 | SetPageUptodate(sb->pages[0]); |
4008c04a | 4702 | |
a061fc8d | 4703 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
4704 | array_size = btrfs_super_sys_array_size(super_copy); |
4705 | ||
0b86a832 CM |
4706 | ptr = super_copy->sys_chunk_array; |
4707 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
4708 | cur = 0; | |
4709 | ||
4710 | while (cur < array_size) { | |
4711 | disk_key = (struct btrfs_disk_key *)ptr; | |
4712 | btrfs_disk_key_to_cpu(&key, disk_key); | |
4713 | ||
a061fc8d | 4714 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
4715 | sb_ptr += len; |
4716 | cur += len; | |
4717 | ||
0d81ba5d | 4718 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 4719 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 4720 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
4721 | if (ret) |
4722 | break; | |
0b86a832 CM |
4723 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
4724 | len = btrfs_chunk_item_size(num_stripes); | |
4725 | } else { | |
84eed90f CM |
4726 | ret = -EIO; |
4727 | break; | |
0b86a832 CM |
4728 | } |
4729 | ptr += len; | |
4730 | sb_ptr += len; | |
4731 | cur += len; | |
4732 | } | |
a061fc8d | 4733 | free_extent_buffer(sb); |
84eed90f | 4734 | return ret; |
0b86a832 CM |
4735 | } |
4736 | ||
4737 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
4738 | { | |
4739 | struct btrfs_path *path; | |
4740 | struct extent_buffer *leaf; | |
4741 | struct btrfs_key key; | |
4742 | struct btrfs_key found_key; | |
4743 | int ret; | |
4744 | int slot; | |
4745 | ||
4746 | root = root->fs_info->chunk_root; | |
4747 | ||
4748 | path = btrfs_alloc_path(); | |
4749 | if (!path) | |
4750 | return -ENOMEM; | |
4751 | ||
b367e47f LZ |
4752 | mutex_lock(&uuid_mutex); |
4753 | lock_chunks(root); | |
4754 | ||
0b86a832 CM |
4755 | /* first we search for all of the device items, and then we |
4756 | * read in all of the chunk items. This way we can create chunk | |
4757 | * mappings that reference all of the devices that are afound | |
4758 | */ | |
4759 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
4760 | key.offset = 0; | |
4761 | key.type = 0; | |
4762 | again: | |
4763 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
4764 | if (ret < 0) |
4765 | goto error; | |
d397712b | 4766 | while (1) { |
0b86a832 CM |
4767 | leaf = path->nodes[0]; |
4768 | slot = path->slots[0]; | |
4769 | if (slot >= btrfs_header_nritems(leaf)) { | |
4770 | ret = btrfs_next_leaf(root, path); | |
4771 | if (ret == 0) | |
4772 | continue; | |
4773 | if (ret < 0) | |
4774 | goto error; | |
4775 | break; | |
4776 | } | |
4777 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4778 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4779 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
4780 | break; | |
4781 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
4782 | struct btrfs_dev_item *dev_item; | |
4783 | dev_item = btrfs_item_ptr(leaf, slot, | |
4784 | struct btrfs_dev_item); | |
0d81ba5d | 4785 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
4786 | if (ret) |
4787 | goto error; | |
0b86a832 CM |
4788 | } |
4789 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
4790 | struct btrfs_chunk *chunk; | |
4791 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
4792 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
4793 | if (ret) |
4794 | goto error; | |
0b86a832 CM |
4795 | } |
4796 | path->slots[0]++; | |
4797 | } | |
4798 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4799 | key.objectid = 0; | |
b3b4aa74 | 4800 | btrfs_release_path(path); |
0b86a832 CM |
4801 | goto again; |
4802 | } | |
0b86a832 CM |
4803 | ret = 0; |
4804 | error: | |
b367e47f LZ |
4805 | unlock_chunks(root); |
4806 | mutex_unlock(&uuid_mutex); | |
4807 | ||
2b82032c | 4808 | btrfs_free_path(path); |
0b86a832 CM |
4809 | return ret; |
4810 | } | |
442a4f63 | 4811 | |
733f4fbb SB |
4812 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
4813 | { | |
4814 | int i; | |
4815 | ||
4816 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4817 | btrfs_dev_stat_reset(dev, i); | |
4818 | } | |
4819 | ||
4820 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
4821 | { | |
4822 | struct btrfs_key key; | |
4823 | struct btrfs_key found_key; | |
4824 | struct btrfs_root *dev_root = fs_info->dev_root; | |
4825 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
4826 | struct extent_buffer *eb; | |
4827 | int slot; | |
4828 | int ret = 0; | |
4829 | struct btrfs_device *device; | |
4830 | struct btrfs_path *path = NULL; | |
4831 | int i; | |
4832 | ||
4833 | path = btrfs_alloc_path(); | |
4834 | if (!path) { | |
4835 | ret = -ENOMEM; | |
4836 | goto out; | |
4837 | } | |
4838 | ||
4839 | mutex_lock(&fs_devices->device_list_mutex); | |
4840 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
4841 | int item_size; | |
4842 | struct btrfs_dev_stats_item *ptr; | |
4843 | ||
4844 | key.objectid = 0; | |
4845 | key.type = BTRFS_DEV_STATS_KEY; | |
4846 | key.offset = device->devid; | |
4847 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
4848 | if (ret) { | |
733f4fbb SB |
4849 | __btrfs_reset_dev_stats(device); |
4850 | device->dev_stats_valid = 1; | |
4851 | btrfs_release_path(path); | |
4852 | continue; | |
4853 | } | |
4854 | slot = path->slots[0]; | |
4855 | eb = path->nodes[0]; | |
4856 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
4857 | item_size = btrfs_item_size_nr(eb, slot); | |
4858 | ||
4859 | ptr = btrfs_item_ptr(eb, slot, | |
4860 | struct btrfs_dev_stats_item); | |
4861 | ||
4862 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
4863 | if (item_size >= (1 + i) * sizeof(__le64)) | |
4864 | btrfs_dev_stat_set(device, i, | |
4865 | btrfs_dev_stats_value(eb, ptr, i)); | |
4866 | else | |
4867 | btrfs_dev_stat_reset(device, i); | |
4868 | } | |
4869 | ||
4870 | device->dev_stats_valid = 1; | |
4871 | btrfs_dev_stat_print_on_load(device); | |
4872 | btrfs_release_path(path); | |
4873 | } | |
4874 | mutex_unlock(&fs_devices->device_list_mutex); | |
4875 | ||
4876 | out: | |
4877 | btrfs_free_path(path); | |
4878 | return ret < 0 ? ret : 0; | |
4879 | } | |
4880 | ||
4881 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
4882 | struct btrfs_root *dev_root, | |
4883 | struct btrfs_device *device) | |
4884 | { | |
4885 | struct btrfs_path *path; | |
4886 | struct btrfs_key key; | |
4887 | struct extent_buffer *eb; | |
4888 | struct btrfs_dev_stats_item *ptr; | |
4889 | int ret; | |
4890 | int i; | |
4891 | ||
4892 | key.objectid = 0; | |
4893 | key.type = BTRFS_DEV_STATS_KEY; | |
4894 | key.offset = device->devid; | |
4895 | ||
4896 | path = btrfs_alloc_path(); | |
4897 | BUG_ON(!path); | |
4898 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
4899 | if (ret < 0) { | |
606686ee JB |
4900 | printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n", |
4901 | ret, rcu_str_deref(device->name)); | |
733f4fbb SB |
4902 | goto out; |
4903 | } | |
4904 | ||
4905 | if (ret == 0 && | |
4906 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
4907 | /* need to delete old one and insert a new one */ | |
4908 | ret = btrfs_del_item(trans, dev_root, path); | |
4909 | if (ret != 0) { | |
606686ee JB |
4910 | printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n", |
4911 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4912 | goto out; |
4913 | } | |
4914 | ret = 1; | |
4915 | } | |
4916 | ||
4917 | if (ret == 1) { | |
4918 | /* need to insert a new item */ | |
4919 | btrfs_release_path(path); | |
4920 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
4921 | &key, sizeof(*ptr)); | |
4922 | if (ret < 0) { | |
606686ee JB |
4923 | printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n", |
4924 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4925 | goto out; |
4926 | } | |
4927 | } | |
4928 | ||
4929 | eb = path->nodes[0]; | |
4930 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
4931 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4932 | btrfs_set_dev_stats_value(eb, ptr, i, | |
4933 | btrfs_dev_stat_read(device, i)); | |
4934 | btrfs_mark_buffer_dirty(eb); | |
4935 | ||
4936 | out: | |
4937 | btrfs_free_path(path); | |
4938 | return ret; | |
4939 | } | |
4940 | ||
4941 | /* | |
4942 | * called from commit_transaction. Writes all changed device stats to disk. | |
4943 | */ | |
4944 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
4945 | struct btrfs_fs_info *fs_info) | |
4946 | { | |
4947 | struct btrfs_root *dev_root = fs_info->dev_root; | |
4948 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
4949 | struct btrfs_device *device; | |
4950 | int ret = 0; | |
4951 | ||
4952 | mutex_lock(&fs_devices->device_list_mutex); | |
4953 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
4954 | if (!device->dev_stats_valid || !device->dev_stats_dirty) | |
4955 | continue; | |
4956 | ||
4957 | ret = update_dev_stat_item(trans, dev_root, device); | |
4958 | if (!ret) | |
4959 | device->dev_stats_dirty = 0; | |
4960 | } | |
4961 | mutex_unlock(&fs_devices->device_list_mutex); | |
4962 | ||
4963 | return ret; | |
4964 | } | |
4965 | ||
442a4f63 SB |
4966 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
4967 | { | |
4968 | btrfs_dev_stat_inc(dev, index); | |
4969 | btrfs_dev_stat_print_on_error(dev); | |
4970 | } | |
4971 | ||
4972 | void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) | |
4973 | { | |
733f4fbb SB |
4974 | if (!dev->dev_stats_valid) |
4975 | return; | |
606686ee | 4976 | printk_ratelimited_in_rcu(KERN_ERR |
442a4f63 | 4977 | "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
606686ee | 4978 | rcu_str_deref(dev->name), |
442a4f63 SB |
4979 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
4980 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
4981 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
4982 | btrfs_dev_stat_read(dev, | |
4983 | BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
4984 | btrfs_dev_stat_read(dev, | |
4985 | BTRFS_DEV_STAT_GENERATION_ERRS)); | |
4986 | } | |
c11d2c23 | 4987 | |
733f4fbb SB |
4988 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
4989 | { | |
a98cdb85 SB |
4990 | int i; |
4991 | ||
4992 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4993 | if (btrfs_dev_stat_read(dev, i) != 0) | |
4994 | break; | |
4995 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
4996 | return; /* all values == 0, suppress message */ | |
4997 | ||
606686ee JB |
4998 | printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
4999 | rcu_str_deref(dev->name), | |
733f4fbb SB |
5000 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
5001 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
5002 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
5003 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
5004 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
5005 | } | |
5006 | ||
c11d2c23 | 5007 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 5008 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
5009 | { |
5010 | struct btrfs_device *dev; | |
5011 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
5012 | int i; | |
5013 | ||
5014 | mutex_lock(&fs_devices->device_list_mutex); | |
5015 | dev = btrfs_find_device(root, stats->devid, NULL, NULL); | |
5016 | mutex_unlock(&fs_devices->device_list_mutex); | |
5017 | ||
5018 | if (!dev) { | |
5019 | printk(KERN_WARNING | |
5020 | "btrfs: get dev_stats failed, device not found\n"); | |
5021 | return -ENODEV; | |
733f4fbb SB |
5022 | } else if (!dev->dev_stats_valid) { |
5023 | printk(KERN_WARNING | |
5024 | "btrfs: get dev_stats failed, not yet valid\n"); | |
5025 | return -ENODEV; | |
b27f7c0c | 5026 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
5027 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
5028 | if (stats->nr_items > i) | |
5029 | stats->values[i] = | |
5030 | btrfs_dev_stat_read_and_reset(dev, i); | |
5031 | else | |
5032 | btrfs_dev_stat_reset(dev, i); | |
5033 | } | |
5034 | } else { | |
5035 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5036 | if (stats->nr_items > i) | |
5037 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
5038 | } | |
5039 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
5040 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
5041 | return 0; | |
5042 | } |