]>
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 | ||
beaf8ab3 SB |
111 | static int |
112 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
113 | int flush, struct block_device **bdev, | |
114 | struct buffer_head **bh) | |
115 | { | |
116 | int ret; | |
117 | ||
118 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
119 | ||
120 | if (IS_ERR(*bdev)) { | |
121 | ret = PTR_ERR(*bdev); | |
122 | printk(KERN_INFO "btrfs: open %s failed\n", device_path); | |
123 | goto error; | |
124 | } | |
125 | ||
126 | if (flush) | |
127 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
128 | ret = set_blocksize(*bdev, 4096); | |
129 | if (ret) { | |
130 | blkdev_put(*bdev, flags); | |
131 | goto error; | |
132 | } | |
133 | invalidate_bdev(*bdev); | |
134 | *bh = btrfs_read_dev_super(*bdev); | |
135 | if (!*bh) { | |
136 | ret = -EINVAL; | |
137 | blkdev_put(*bdev, flags); | |
138 | goto error; | |
139 | } | |
140 | ||
141 | return 0; | |
142 | ||
143 | error: | |
144 | *bdev = NULL; | |
145 | *bh = NULL; | |
146 | return ret; | |
147 | } | |
148 | ||
ffbd517d CM |
149 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
150 | struct bio *head, struct bio *tail) | |
151 | { | |
152 | ||
153 | struct bio *old_head; | |
154 | ||
155 | old_head = pending_bios->head; | |
156 | pending_bios->head = head; | |
157 | if (pending_bios->tail) | |
158 | tail->bi_next = old_head; | |
159 | else | |
160 | pending_bios->tail = tail; | |
161 | } | |
162 | ||
8b712842 CM |
163 | /* |
164 | * we try to collect pending bios for a device so we don't get a large | |
165 | * number of procs sending bios down to the same device. This greatly | |
166 | * improves the schedulers ability to collect and merge the bios. | |
167 | * | |
168 | * But, it also turns into a long list of bios to process and that is sure | |
169 | * to eventually make the worker thread block. The solution here is to | |
170 | * make some progress and then put this work struct back at the end of | |
171 | * the list if the block device is congested. This way, multiple devices | |
172 | * can make progress from a single worker thread. | |
173 | */ | |
143bede5 | 174 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
175 | { |
176 | struct bio *pending; | |
177 | struct backing_dev_info *bdi; | |
b64a2851 | 178 | struct btrfs_fs_info *fs_info; |
ffbd517d | 179 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
180 | struct bio *tail; |
181 | struct bio *cur; | |
182 | int again = 0; | |
ffbd517d | 183 | unsigned long num_run; |
d644d8a1 | 184 | unsigned long batch_run = 0; |
b64a2851 | 185 | unsigned long limit; |
b765ead5 | 186 | unsigned long last_waited = 0; |
d84275c9 | 187 | int force_reg = 0; |
0e588859 | 188 | int sync_pending = 0; |
211588ad CM |
189 | struct blk_plug plug; |
190 | ||
191 | /* | |
192 | * this function runs all the bios we've collected for | |
193 | * a particular device. We don't want to wander off to | |
194 | * another device without first sending all of these down. | |
195 | * So, setup a plug here and finish it off before we return | |
196 | */ | |
197 | blk_start_plug(&plug); | |
8b712842 | 198 | |
bedf762b | 199 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
200 | fs_info = device->dev_root->fs_info; |
201 | limit = btrfs_async_submit_limit(fs_info); | |
202 | limit = limit * 2 / 3; | |
203 | ||
8b712842 CM |
204 | loop: |
205 | spin_lock(&device->io_lock); | |
206 | ||
a6837051 | 207 | loop_lock: |
d84275c9 | 208 | num_run = 0; |
ffbd517d | 209 | |
8b712842 CM |
210 | /* take all the bios off the list at once and process them |
211 | * later on (without the lock held). But, remember the | |
212 | * tail and other pointers so the bios can be properly reinserted | |
213 | * into the list if we hit congestion | |
214 | */ | |
d84275c9 | 215 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 216 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
217 | force_reg = 1; |
218 | } else { | |
ffbd517d | 219 | pending_bios = &device->pending_bios; |
d84275c9 CM |
220 | force_reg = 0; |
221 | } | |
ffbd517d CM |
222 | |
223 | pending = pending_bios->head; | |
224 | tail = pending_bios->tail; | |
8b712842 | 225 | WARN_ON(pending && !tail); |
8b712842 CM |
226 | |
227 | /* | |
228 | * if pending was null this time around, no bios need processing | |
229 | * at all and we can stop. Otherwise it'll loop back up again | |
230 | * and do an additional check so no bios are missed. | |
231 | * | |
232 | * device->running_pending is used to synchronize with the | |
233 | * schedule_bio code. | |
234 | */ | |
ffbd517d CM |
235 | if (device->pending_sync_bios.head == NULL && |
236 | device->pending_bios.head == NULL) { | |
8b712842 CM |
237 | again = 0; |
238 | device->running_pending = 0; | |
ffbd517d CM |
239 | } else { |
240 | again = 1; | |
241 | device->running_pending = 1; | |
8b712842 | 242 | } |
ffbd517d CM |
243 | |
244 | pending_bios->head = NULL; | |
245 | pending_bios->tail = NULL; | |
246 | ||
8b712842 CM |
247 | spin_unlock(&device->io_lock); |
248 | ||
d397712b | 249 | while (pending) { |
ffbd517d CM |
250 | |
251 | rmb(); | |
d84275c9 CM |
252 | /* we want to work on both lists, but do more bios on the |
253 | * sync list than the regular list | |
254 | */ | |
255 | if ((num_run > 32 && | |
256 | pending_bios != &device->pending_sync_bios && | |
257 | device->pending_sync_bios.head) || | |
258 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
259 | device->pending_bios.head)) { | |
ffbd517d CM |
260 | spin_lock(&device->io_lock); |
261 | requeue_list(pending_bios, pending, tail); | |
262 | goto loop_lock; | |
263 | } | |
264 | ||
8b712842 CM |
265 | cur = pending; |
266 | pending = pending->bi_next; | |
267 | cur->bi_next = NULL; | |
b64a2851 | 268 | |
66657b31 | 269 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
270 | waitqueue_active(&fs_info->async_submit_wait)) |
271 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
272 | |
273 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 274 | |
2ab1ba68 CM |
275 | /* |
276 | * if we're doing the sync list, record that our | |
277 | * plug has some sync requests on it | |
278 | * | |
279 | * If we're doing the regular list and there are | |
280 | * sync requests sitting around, unplug before | |
281 | * we add more | |
282 | */ | |
283 | if (pending_bios == &device->pending_sync_bios) { | |
284 | sync_pending = 1; | |
285 | } else if (sync_pending) { | |
286 | blk_finish_plug(&plug); | |
287 | blk_start_plug(&plug); | |
288 | sync_pending = 0; | |
289 | } | |
290 | ||
21adbd5c | 291 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
292 | num_run++; |
293 | batch_run++; | |
7eaceacc | 294 | if (need_resched()) |
ffbd517d | 295 | cond_resched(); |
8b712842 CM |
296 | |
297 | /* | |
298 | * we made progress, there is more work to do and the bdi | |
299 | * is now congested. Back off and let other work structs | |
300 | * run instead | |
301 | */ | |
57fd5a5f | 302 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 303 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 304 | struct io_context *ioc; |
8b712842 | 305 | |
b765ead5 CM |
306 | ioc = current->io_context; |
307 | ||
308 | /* | |
309 | * the main goal here is that we don't want to | |
310 | * block if we're going to be able to submit | |
311 | * more requests without blocking. | |
312 | * | |
313 | * This code does two great things, it pokes into | |
314 | * the elevator code from a filesystem _and_ | |
315 | * it makes assumptions about how batching works. | |
316 | */ | |
317 | if (ioc && ioc->nr_batch_requests > 0 && | |
318 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
319 | (last_waited == 0 || | |
320 | ioc->last_waited == last_waited)) { | |
321 | /* | |
322 | * we want to go through our batch of | |
323 | * requests and stop. So, we copy out | |
324 | * the ioc->last_waited time and test | |
325 | * against it before looping | |
326 | */ | |
327 | last_waited = ioc->last_waited; | |
7eaceacc | 328 | if (need_resched()) |
ffbd517d | 329 | cond_resched(); |
b765ead5 CM |
330 | continue; |
331 | } | |
8b712842 | 332 | spin_lock(&device->io_lock); |
ffbd517d | 333 | requeue_list(pending_bios, pending, tail); |
a6837051 | 334 | device->running_pending = 1; |
8b712842 CM |
335 | |
336 | spin_unlock(&device->io_lock); | |
337 | btrfs_requeue_work(&device->work); | |
338 | goto done; | |
339 | } | |
d85c8a6f CM |
340 | /* unplug every 64 requests just for good measure */ |
341 | if (batch_run % 64 == 0) { | |
342 | blk_finish_plug(&plug); | |
343 | blk_start_plug(&plug); | |
344 | sync_pending = 0; | |
345 | } | |
8b712842 | 346 | } |
ffbd517d | 347 | |
51684082 CM |
348 | cond_resched(); |
349 | if (again) | |
350 | goto loop; | |
351 | ||
352 | spin_lock(&device->io_lock); | |
353 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
354 | goto loop_lock; | |
355 | spin_unlock(&device->io_lock); | |
356 | ||
8b712842 | 357 | done: |
211588ad | 358 | blk_finish_plug(&plug); |
8b712842 CM |
359 | } |
360 | ||
b2950863 | 361 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
362 | { |
363 | struct btrfs_device *device; | |
364 | ||
365 | device = container_of(work, struct btrfs_device, work); | |
366 | run_scheduled_bios(device); | |
367 | } | |
368 | ||
a1b32a59 | 369 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
370 | struct btrfs_super_block *disk_super, |
371 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
372 | { | |
373 | struct btrfs_device *device; | |
374 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 375 | struct rcu_string *name; |
8a4b83cc CM |
376 | u64 found_transid = btrfs_super_generation(disk_super); |
377 | ||
378 | fs_devices = find_fsid(disk_super->fsid); | |
379 | if (!fs_devices) { | |
515dc322 | 380 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
381 | if (!fs_devices) |
382 | return -ENOMEM; | |
383 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 384 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
385 | list_add(&fs_devices->list, &fs_uuids); |
386 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
387 | fs_devices->latest_devid = devid; | |
388 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 389 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
390 | device = NULL; |
391 | } else { | |
a443755f CM |
392 | device = __find_device(&fs_devices->devices, devid, |
393 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
394 | } |
395 | if (!device) { | |
2b82032c YZ |
396 | if (fs_devices->opened) |
397 | return -EBUSY; | |
398 | ||
8a4b83cc CM |
399 | device = kzalloc(sizeof(*device), GFP_NOFS); |
400 | if (!device) { | |
401 | /* we can safely leave the fs_devices entry around */ | |
402 | return -ENOMEM; | |
403 | } | |
404 | device->devid = devid; | |
733f4fbb | 405 | device->dev_stats_valid = 0; |
8b712842 | 406 | device->work.func = pending_bios_fn; |
a443755f CM |
407 | memcpy(device->uuid, disk_super->dev_item.uuid, |
408 | BTRFS_UUID_SIZE); | |
b248a415 | 409 | spin_lock_init(&device->io_lock); |
606686ee JB |
410 | |
411 | name = rcu_string_strdup(path, GFP_NOFS); | |
412 | if (!name) { | |
8a4b83cc CM |
413 | kfree(device); |
414 | return -ENOMEM; | |
415 | } | |
606686ee | 416 | rcu_assign_pointer(device->name, name); |
2b82032c | 417 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 418 | |
90519d66 AJ |
419 | /* init readahead state */ |
420 | spin_lock_init(&device->reada_lock); | |
421 | device->reada_curr_zone = NULL; | |
422 | atomic_set(&device->reada_in_flight, 0); | |
423 | device->reada_next = 0; | |
424 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
425 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
426 | ||
e5e9a520 | 427 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 428 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
429 | mutex_unlock(&fs_devices->device_list_mutex); |
430 | ||
2b82032c | 431 | device->fs_devices = fs_devices; |
8a4b83cc | 432 | fs_devices->num_devices++; |
606686ee JB |
433 | } else if (!device->name || strcmp(device->name->str, path)) { |
434 | name = rcu_string_strdup(path, GFP_NOFS); | |
3a0524dc TH |
435 | if (!name) |
436 | return -ENOMEM; | |
606686ee JB |
437 | rcu_string_free(device->name); |
438 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
439 | if (device->missing) { |
440 | fs_devices->missing_devices--; | |
441 | device->missing = 0; | |
442 | } | |
8a4b83cc CM |
443 | } |
444 | ||
445 | if (found_transid > fs_devices->latest_trans) { | |
446 | fs_devices->latest_devid = devid; | |
447 | fs_devices->latest_trans = found_transid; | |
448 | } | |
8a4b83cc CM |
449 | *fs_devices_ret = fs_devices; |
450 | return 0; | |
451 | } | |
452 | ||
e4404d6e YZ |
453 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
454 | { | |
455 | struct btrfs_fs_devices *fs_devices; | |
456 | struct btrfs_device *device; | |
457 | struct btrfs_device *orig_dev; | |
458 | ||
459 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
460 | if (!fs_devices) | |
461 | return ERR_PTR(-ENOMEM); | |
462 | ||
463 | INIT_LIST_HEAD(&fs_devices->devices); | |
464 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
465 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 466 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
467 | fs_devices->latest_devid = orig->latest_devid; |
468 | fs_devices->latest_trans = orig->latest_trans; | |
02db0844 | 469 | fs_devices->total_devices = orig->total_devices; |
e4404d6e YZ |
470 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); |
471 | ||
46224705 | 472 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 473 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
474 | struct rcu_string *name; |
475 | ||
e4404d6e YZ |
476 | device = kzalloc(sizeof(*device), GFP_NOFS); |
477 | if (!device) | |
478 | goto error; | |
479 | ||
606686ee JB |
480 | /* |
481 | * This is ok to do without rcu read locked because we hold the | |
482 | * uuid mutex so nothing we touch in here is going to disappear. | |
483 | */ | |
484 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
485 | if (!name) { | |
fd2696f3 | 486 | kfree(device); |
e4404d6e | 487 | goto error; |
fd2696f3 | 488 | } |
606686ee | 489 | rcu_assign_pointer(device->name, name); |
e4404d6e YZ |
490 | |
491 | device->devid = orig_dev->devid; | |
492 | device->work.func = pending_bios_fn; | |
493 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
494 | spin_lock_init(&device->io_lock); |
495 | INIT_LIST_HEAD(&device->dev_list); | |
496 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
497 | ||
498 | list_add(&device->dev_list, &fs_devices->devices); | |
499 | device->fs_devices = fs_devices; | |
500 | fs_devices->num_devices++; | |
501 | } | |
502 | return fs_devices; | |
503 | error: | |
504 | free_fs_devices(fs_devices); | |
505 | return ERR_PTR(-ENOMEM); | |
506 | } | |
507 | ||
143bede5 | 508 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
dfe25020 | 509 | { |
c6e30871 | 510 | struct btrfs_device *device, *next; |
dfe25020 | 511 | |
a6b0d5c8 CM |
512 | struct block_device *latest_bdev = NULL; |
513 | u64 latest_devid = 0; | |
514 | u64 latest_transid = 0; | |
515 | ||
dfe25020 CM |
516 | mutex_lock(&uuid_mutex); |
517 | again: | |
46224705 | 518 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 519 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 CM |
520 | if (device->in_fs_metadata) { |
521 | if (!latest_transid || | |
522 | device->generation > latest_transid) { | |
523 | latest_devid = device->devid; | |
524 | latest_transid = device->generation; | |
525 | latest_bdev = device->bdev; | |
526 | } | |
2b82032c | 527 | continue; |
a6b0d5c8 | 528 | } |
2b82032c YZ |
529 | |
530 | if (device->bdev) { | |
d4d77629 | 531 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
532 | device->bdev = NULL; |
533 | fs_devices->open_devices--; | |
534 | } | |
535 | if (device->writeable) { | |
536 | list_del_init(&device->dev_alloc_list); | |
537 | device->writeable = 0; | |
538 | fs_devices->rw_devices--; | |
539 | } | |
e4404d6e YZ |
540 | list_del_init(&device->dev_list); |
541 | fs_devices->num_devices--; | |
606686ee | 542 | rcu_string_free(device->name); |
e4404d6e | 543 | kfree(device); |
dfe25020 | 544 | } |
2b82032c YZ |
545 | |
546 | if (fs_devices->seed) { | |
547 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
548 | goto again; |
549 | } | |
550 | ||
a6b0d5c8 CM |
551 | fs_devices->latest_bdev = latest_bdev; |
552 | fs_devices->latest_devid = latest_devid; | |
553 | fs_devices->latest_trans = latest_transid; | |
554 | ||
dfe25020 | 555 | mutex_unlock(&uuid_mutex); |
dfe25020 | 556 | } |
a0af469b | 557 | |
1f78160c XG |
558 | static void __free_device(struct work_struct *work) |
559 | { | |
560 | struct btrfs_device *device; | |
561 | ||
562 | device = container_of(work, struct btrfs_device, rcu_work); | |
563 | ||
564 | if (device->bdev) | |
565 | blkdev_put(device->bdev, device->mode); | |
566 | ||
606686ee | 567 | rcu_string_free(device->name); |
1f78160c XG |
568 | kfree(device); |
569 | } | |
570 | ||
571 | static void free_device(struct rcu_head *head) | |
572 | { | |
573 | struct btrfs_device *device; | |
574 | ||
575 | device = container_of(head, struct btrfs_device, rcu); | |
576 | ||
577 | INIT_WORK(&device->rcu_work, __free_device); | |
578 | schedule_work(&device->rcu_work); | |
579 | } | |
580 | ||
2b82032c | 581 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 582 | { |
8a4b83cc | 583 | struct btrfs_device *device; |
e4404d6e | 584 | |
2b82032c YZ |
585 | if (--fs_devices->opened > 0) |
586 | return 0; | |
8a4b83cc | 587 | |
c9513edb | 588 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 589 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c | 590 | struct btrfs_device *new_device; |
606686ee | 591 | struct rcu_string *name; |
1f78160c XG |
592 | |
593 | if (device->bdev) | |
a0af469b | 594 | fs_devices->open_devices--; |
1f78160c | 595 | |
2b82032c YZ |
596 | if (device->writeable) { |
597 | list_del_init(&device->dev_alloc_list); | |
598 | fs_devices->rw_devices--; | |
599 | } | |
600 | ||
d5e2003c JB |
601 | if (device->can_discard) |
602 | fs_devices->num_can_discard--; | |
603 | ||
1f78160c | 604 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
79787eaa | 605 | BUG_ON(!new_device); /* -ENOMEM */ |
1f78160c | 606 | memcpy(new_device, device, sizeof(*new_device)); |
606686ee JB |
607 | |
608 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
609 | if (device->name) { |
610 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
611 | BUG_ON(device->name && !name); /* -ENOMEM */ | |
612 | rcu_assign_pointer(new_device->name, name); | |
613 | } | |
1f78160c XG |
614 | new_device->bdev = NULL; |
615 | new_device->writeable = 0; | |
616 | new_device->in_fs_metadata = 0; | |
d5e2003c | 617 | new_device->can_discard = 0; |
1f78160c XG |
618 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
619 | ||
620 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 621 | } |
c9513edb XG |
622 | mutex_unlock(&fs_devices->device_list_mutex); |
623 | ||
e4404d6e YZ |
624 | WARN_ON(fs_devices->open_devices); |
625 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
626 | fs_devices->opened = 0; |
627 | fs_devices->seeding = 0; | |
2b82032c | 628 | |
8a4b83cc CM |
629 | return 0; |
630 | } | |
631 | ||
2b82032c YZ |
632 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
633 | { | |
e4404d6e | 634 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
635 | int ret; |
636 | ||
637 | mutex_lock(&uuid_mutex); | |
638 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
639 | if (!fs_devices->opened) { |
640 | seed_devices = fs_devices->seed; | |
641 | fs_devices->seed = NULL; | |
642 | } | |
2b82032c | 643 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
644 | |
645 | while (seed_devices) { | |
646 | fs_devices = seed_devices; | |
647 | seed_devices = fs_devices->seed; | |
648 | __btrfs_close_devices(fs_devices); | |
649 | free_fs_devices(fs_devices); | |
650 | } | |
2b82032c YZ |
651 | return ret; |
652 | } | |
653 | ||
e4404d6e YZ |
654 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
655 | fmode_t flags, void *holder) | |
8a4b83cc | 656 | { |
d5e2003c | 657 | struct request_queue *q; |
8a4b83cc CM |
658 | struct block_device *bdev; |
659 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 660 | struct btrfs_device *device; |
a0af469b CM |
661 | struct block_device *latest_bdev = NULL; |
662 | struct buffer_head *bh; | |
663 | struct btrfs_super_block *disk_super; | |
664 | u64 latest_devid = 0; | |
665 | u64 latest_transid = 0; | |
a0af469b | 666 | u64 devid; |
2b82032c | 667 | int seeding = 1; |
a0af469b | 668 | int ret = 0; |
8a4b83cc | 669 | |
d4d77629 TH |
670 | flags |= FMODE_EXCL; |
671 | ||
c6e30871 | 672 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
673 | if (device->bdev) |
674 | continue; | |
dfe25020 CM |
675 | if (!device->name) |
676 | continue; | |
677 | ||
beaf8ab3 SB |
678 | ret = btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, |
679 | &bdev, &bh); | |
680 | if (ret) | |
681 | continue; | |
a0af469b CM |
682 | |
683 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 684 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
685 | if (devid != device->devid) |
686 | goto error_brelse; | |
687 | ||
2b82032c YZ |
688 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
689 | BTRFS_UUID_SIZE)) | |
690 | goto error_brelse; | |
691 | ||
692 | device->generation = btrfs_super_generation(disk_super); | |
693 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 694 | latest_devid = devid; |
2b82032c | 695 | latest_transid = device->generation; |
a0af469b CM |
696 | latest_bdev = bdev; |
697 | } | |
698 | ||
2b82032c YZ |
699 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
700 | device->writeable = 0; | |
701 | } else { | |
702 | device->writeable = !bdev_read_only(bdev); | |
703 | seeding = 0; | |
704 | } | |
705 | ||
d5e2003c JB |
706 | q = bdev_get_queue(bdev); |
707 | if (blk_queue_discard(q)) { | |
708 | device->can_discard = 1; | |
709 | fs_devices->num_can_discard++; | |
710 | } | |
711 | ||
8a4b83cc | 712 | device->bdev = bdev; |
dfe25020 | 713 | device->in_fs_metadata = 0; |
15916de8 CM |
714 | device->mode = flags; |
715 | ||
c289811c CM |
716 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
717 | fs_devices->rotating = 1; | |
718 | ||
a0af469b | 719 | fs_devices->open_devices++; |
2b82032c YZ |
720 | if (device->writeable) { |
721 | fs_devices->rw_devices++; | |
722 | list_add(&device->dev_alloc_list, | |
723 | &fs_devices->alloc_list); | |
724 | } | |
4f6c9328 | 725 | brelse(bh); |
a0af469b | 726 | continue; |
a061fc8d | 727 | |
a0af469b CM |
728 | error_brelse: |
729 | brelse(bh); | |
d4d77629 | 730 | blkdev_put(bdev, flags); |
a0af469b | 731 | continue; |
8a4b83cc | 732 | } |
a0af469b | 733 | if (fs_devices->open_devices == 0) { |
20bcd649 | 734 | ret = -EINVAL; |
a0af469b CM |
735 | goto out; |
736 | } | |
2b82032c YZ |
737 | fs_devices->seeding = seeding; |
738 | fs_devices->opened = 1; | |
a0af469b CM |
739 | fs_devices->latest_bdev = latest_bdev; |
740 | fs_devices->latest_devid = latest_devid; | |
741 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 742 | fs_devices->total_rw_bytes = 0; |
a0af469b | 743 | out: |
2b82032c YZ |
744 | return ret; |
745 | } | |
746 | ||
747 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 748 | fmode_t flags, void *holder) |
2b82032c YZ |
749 | { |
750 | int ret; | |
751 | ||
752 | mutex_lock(&uuid_mutex); | |
753 | if (fs_devices->opened) { | |
e4404d6e YZ |
754 | fs_devices->opened++; |
755 | ret = 0; | |
2b82032c | 756 | } else { |
15916de8 | 757 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 758 | } |
8a4b83cc | 759 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
760 | return ret; |
761 | } | |
762 | ||
97288f2c | 763 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
764 | struct btrfs_fs_devices **fs_devices_ret) |
765 | { | |
766 | struct btrfs_super_block *disk_super; | |
767 | struct block_device *bdev; | |
768 | struct buffer_head *bh; | |
769 | int ret; | |
770 | u64 devid; | |
f2984462 | 771 | u64 transid; |
02db0844 | 772 | u64 total_devices; |
8a4b83cc | 773 | |
d4d77629 | 774 | flags |= FMODE_EXCL; |
10f6327b | 775 | mutex_lock(&uuid_mutex); |
beaf8ab3 | 776 | ret = btrfs_get_bdev_and_sb(path, flags, holder, 0, &bdev, &bh); |
8a4b83cc | 777 | if (ret) |
beaf8ab3 | 778 | goto error; |
8a4b83cc | 779 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 780 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 781 | transid = btrfs_super_generation(disk_super); |
02db0844 | 782 | total_devices = btrfs_super_num_devices(disk_super); |
d03f918a SB |
783 | if (disk_super->label[0]) { |
784 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
785 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
d397712b | 786 | printk(KERN_INFO "device label %s ", disk_super->label); |
d03f918a | 787 | } else { |
22b63a29 | 788 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); |
d03f918a | 789 | } |
119e10cf | 790 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 791 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc | 792 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
02db0844 JB |
793 | if (!ret && fs_devices_ret) |
794 | (*fs_devices_ret)->total_devices = total_devices; | |
8a4b83cc | 795 | brelse(bh); |
d4d77629 | 796 | blkdev_put(bdev, flags); |
8a4b83cc | 797 | error: |
beaf8ab3 | 798 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
799 | return ret; |
800 | } | |
0b86a832 | 801 | |
6d07bcec MX |
802 | /* helper to account the used device space in the range */ |
803 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
804 | u64 end, u64 *length) | |
805 | { | |
806 | struct btrfs_key key; | |
807 | struct btrfs_root *root = device->dev_root; | |
808 | struct btrfs_dev_extent *dev_extent; | |
809 | struct btrfs_path *path; | |
810 | u64 extent_end; | |
811 | int ret; | |
812 | int slot; | |
813 | struct extent_buffer *l; | |
814 | ||
815 | *length = 0; | |
816 | ||
817 | if (start >= device->total_bytes) | |
818 | return 0; | |
819 | ||
820 | path = btrfs_alloc_path(); | |
821 | if (!path) | |
822 | return -ENOMEM; | |
823 | path->reada = 2; | |
824 | ||
825 | key.objectid = device->devid; | |
826 | key.offset = start; | |
827 | key.type = BTRFS_DEV_EXTENT_KEY; | |
828 | ||
829 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
830 | if (ret < 0) | |
831 | goto out; | |
832 | if (ret > 0) { | |
833 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
834 | if (ret < 0) | |
835 | goto out; | |
836 | } | |
837 | ||
838 | while (1) { | |
839 | l = path->nodes[0]; | |
840 | slot = path->slots[0]; | |
841 | if (slot >= btrfs_header_nritems(l)) { | |
842 | ret = btrfs_next_leaf(root, path); | |
843 | if (ret == 0) | |
844 | continue; | |
845 | if (ret < 0) | |
846 | goto out; | |
847 | ||
848 | break; | |
849 | } | |
850 | btrfs_item_key_to_cpu(l, &key, slot); | |
851 | ||
852 | if (key.objectid < device->devid) | |
853 | goto next; | |
854 | ||
855 | if (key.objectid > device->devid) | |
856 | break; | |
857 | ||
858 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
859 | goto next; | |
860 | ||
861 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
862 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
863 | dev_extent); | |
864 | if (key.offset <= start && extent_end > end) { | |
865 | *length = end - start + 1; | |
866 | break; | |
867 | } else if (key.offset <= start && extent_end > start) | |
868 | *length += extent_end - start; | |
869 | else if (key.offset > start && extent_end <= end) | |
870 | *length += extent_end - key.offset; | |
871 | else if (key.offset > start && key.offset <= end) { | |
872 | *length += end - key.offset + 1; | |
873 | break; | |
874 | } else if (key.offset > end) | |
875 | break; | |
876 | ||
877 | next: | |
878 | path->slots[0]++; | |
879 | } | |
880 | ret = 0; | |
881 | out: | |
882 | btrfs_free_path(path); | |
883 | return ret; | |
884 | } | |
885 | ||
0b86a832 | 886 | /* |
7bfc837d | 887 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
888 | * @device: the device which we search the free space in |
889 | * @num_bytes: the size of the free space that we need | |
890 | * @start: store the start of the free space. | |
891 | * @len: the size of the free space. that we find, or the size of the max | |
892 | * free space if we don't find suitable free space | |
893 | * | |
0b86a832 CM |
894 | * this uses a pretty simple search, the expectation is that it is |
895 | * called very infrequently and that a given device has a small number | |
896 | * of extents | |
7bfc837d MX |
897 | * |
898 | * @start is used to store the start of the free space if we find. But if we | |
899 | * don't find suitable free space, it will be used to store the start position | |
900 | * of the max free space. | |
901 | * | |
902 | * @len is used to store the size of the free space that we find. | |
903 | * But if we don't find suitable free space, it is used to store the size of | |
904 | * the max free space. | |
0b86a832 | 905 | */ |
125ccb0a | 906 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 907 | u64 *start, u64 *len) |
0b86a832 CM |
908 | { |
909 | struct btrfs_key key; | |
910 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 911 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 912 | struct btrfs_path *path; |
7bfc837d MX |
913 | u64 hole_size; |
914 | u64 max_hole_start; | |
915 | u64 max_hole_size; | |
916 | u64 extent_end; | |
917 | u64 search_start; | |
0b86a832 CM |
918 | u64 search_end = device->total_bytes; |
919 | int ret; | |
7bfc837d | 920 | int slot; |
0b86a832 CM |
921 | struct extent_buffer *l; |
922 | ||
0b86a832 CM |
923 | /* FIXME use last free of some kind */ |
924 | ||
8a4b83cc CM |
925 | /* we don't want to overwrite the superblock on the drive, |
926 | * so we make sure to start at an offset of at least 1MB | |
927 | */ | |
a9c9bf68 | 928 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 929 | |
7bfc837d MX |
930 | max_hole_start = search_start; |
931 | max_hole_size = 0; | |
38c01b96 | 932 | hole_size = 0; |
7bfc837d MX |
933 | |
934 | if (search_start >= search_end) { | |
935 | ret = -ENOSPC; | |
936 | goto error; | |
937 | } | |
938 | ||
939 | path = btrfs_alloc_path(); | |
940 | if (!path) { | |
941 | ret = -ENOMEM; | |
942 | goto error; | |
943 | } | |
944 | path->reada = 2; | |
945 | ||
0b86a832 CM |
946 | key.objectid = device->devid; |
947 | key.offset = search_start; | |
948 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 949 | |
125ccb0a | 950 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 951 | if (ret < 0) |
7bfc837d | 952 | goto out; |
1fcbac58 YZ |
953 | if (ret > 0) { |
954 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
955 | if (ret < 0) | |
7bfc837d | 956 | goto out; |
1fcbac58 | 957 | } |
7bfc837d | 958 | |
0b86a832 CM |
959 | while (1) { |
960 | l = path->nodes[0]; | |
961 | slot = path->slots[0]; | |
962 | if (slot >= btrfs_header_nritems(l)) { | |
963 | ret = btrfs_next_leaf(root, path); | |
964 | if (ret == 0) | |
965 | continue; | |
966 | if (ret < 0) | |
7bfc837d MX |
967 | goto out; |
968 | ||
969 | break; | |
0b86a832 CM |
970 | } |
971 | btrfs_item_key_to_cpu(l, &key, slot); | |
972 | ||
973 | if (key.objectid < device->devid) | |
974 | goto next; | |
975 | ||
976 | if (key.objectid > device->devid) | |
7bfc837d | 977 | break; |
0b86a832 | 978 | |
7bfc837d MX |
979 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
980 | goto next; | |
9779b72f | 981 | |
7bfc837d MX |
982 | if (key.offset > search_start) { |
983 | hole_size = key.offset - search_start; | |
9779b72f | 984 | |
7bfc837d MX |
985 | if (hole_size > max_hole_size) { |
986 | max_hole_start = search_start; | |
987 | max_hole_size = hole_size; | |
988 | } | |
9779b72f | 989 | |
7bfc837d MX |
990 | /* |
991 | * If this free space is greater than which we need, | |
992 | * it must be the max free space that we have found | |
993 | * until now, so max_hole_start must point to the start | |
994 | * of this free space and the length of this free space | |
995 | * is stored in max_hole_size. Thus, we return | |
996 | * max_hole_start and max_hole_size and go back to the | |
997 | * caller. | |
998 | */ | |
999 | if (hole_size >= num_bytes) { | |
1000 | ret = 0; | |
1001 | goto out; | |
0b86a832 CM |
1002 | } |
1003 | } | |
0b86a832 | 1004 | |
0b86a832 | 1005 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1006 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1007 | dev_extent); | |
1008 | if (extent_end > search_start) | |
1009 | search_start = extent_end; | |
0b86a832 CM |
1010 | next: |
1011 | path->slots[0]++; | |
1012 | cond_resched(); | |
1013 | } | |
0b86a832 | 1014 | |
38c01b96 | 1015 | /* |
1016 | * At this point, search_start should be the end of | |
1017 | * allocated dev extents, and when shrinking the device, | |
1018 | * search_end may be smaller than search_start. | |
1019 | */ | |
1020 | if (search_end > search_start) | |
1021 | hole_size = search_end - search_start; | |
1022 | ||
7bfc837d MX |
1023 | if (hole_size > max_hole_size) { |
1024 | max_hole_start = search_start; | |
1025 | max_hole_size = hole_size; | |
0b86a832 | 1026 | } |
0b86a832 | 1027 | |
7bfc837d MX |
1028 | /* See above. */ |
1029 | if (hole_size < num_bytes) | |
1030 | ret = -ENOSPC; | |
1031 | else | |
1032 | ret = 0; | |
1033 | ||
1034 | out: | |
2b82032c | 1035 | btrfs_free_path(path); |
7bfc837d MX |
1036 | error: |
1037 | *start = max_hole_start; | |
b2117a39 | 1038 | if (len) |
7bfc837d | 1039 | *len = max_hole_size; |
0b86a832 CM |
1040 | return ret; |
1041 | } | |
1042 | ||
b2950863 | 1043 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1044 | struct btrfs_device *device, |
1045 | u64 start) | |
1046 | { | |
1047 | int ret; | |
1048 | struct btrfs_path *path; | |
1049 | struct btrfs_root *root = device->dev_root; | |
1050 | struct btrfs_key key; | |
a061fc8d CM |
1051 | struct btrfs_key found_key; |
1052 | struct extent_buffer *leaf = NULL; | |
1053 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1054 | |
1055 | path = btrfs_alloc_path(); | |
1056 | if (!path) | |
1057 | return -ENOMEM; | |
1058 | ||
1059 | key.objectid = device->devid; | |
1060 | key.offset = start; | |
1061 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1062 | again: |
8f18cf13 | 1063 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1064 | if (ret > 0) { |
1065 | ret = btrfs_previous_item(root, path, key.objectid, | |
1066 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1067 | if (ret) |
1068 | goto out; | |
a061fc8d CM |
1069 | leaf = path->nodes[0]; |
1070 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1071 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1072 | struct btrfs_dev_extent); | |
1073 | BUG_ON(found_key.offset > start || found_key.offset + | |
1074 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1075 | key = found_key; |
1076 | btrfs_release_path(path); | |
1077 | goto again; | |
a061fc8d CM |
1078 | } else if (ret == 0) { |
1079 | leaf = path->nodes[0]; | |
1080 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1081 | struct btrfs_dev_extent); | |
79787eaa JM |
1082 | } else { |
1083 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1084 | goto out; | |
a061fc8d | 1085 | } |
8f18cf13 | 1086 | |
2bf64758 JB |
1087 | if (device->bytes_used > 0) { |
1088 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1089 | device->bytes_used -= len; | |
1090 | spin_lock(&root->fs_info->free_chunk_lock); | |
1091 | root->fs_info->free_chunk_space += len; | |
1092 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1093 | } | |
8f18cf13 | 1094 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1095 | if (ret) { |
1096 | btrfs_error(root->fs_info, ret, | |
1097 | "Failed to remove dev extent item"); | |
1098 | } | |
b0b802d7 | 1099 | out: |
8f18cf13 CM |
1100 | btrfs_free_path(path); |
1101 | return ret; | |
1102 | } | |
1103 | ||
2b82032c | 1104 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1105 | struct btrfs_device *device, |
e17cade2 | 1106 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1107 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1108 | { |
1109 | int ret; | |
1110 | struct btrfs_path *path; | |
1111 | struct btrfs_root *root = device->dev_root; | |
1112 | struct btrfs_dev_extent *extent; | |
1113 | struct extent_buffer *leaf; | |
1114 | struct btrfs_key key; | |
1115 | ||
dfe25020 | 1116 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1117 | path = btrfs_alloc_path(); |
1118 | if (!path) | |
1119 | return -ENOMEM; | |
1120 | ||
0b86a832 | 1121 | key.objectid = device->devid; |
2b82032c | 1122 | key.offset = start; |
0b86a832 CM |
1123 | key.type = BTRFS_DEV_EXTENT_KEY; |
1124 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1125 | sizeof(*extent)); | |
2cdcecbc MF |
1126 | if (ret) |
1127 | goto out; | |
0b86a832 CM |
1128 | |
1129 | leaf = path->nodes[0]; | |
1130 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1131 | struct btrfs_dev_extent); | |
e17cade2 CM |
1132 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1133 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1134 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1135 | ||
1136 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1137 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1138 | BTRFS_UUID_SIZE); | |
1139 | ||
0b86a832 CM |
1140 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1141 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1142 | out: |
0b86a832 CM |
1143 | btrfs_free_path(path); |
1144 | return ret; | |
1145 | } | |
1146 | ||
a1b32a59 CM |
1147 | static noinline int find_next_chunk(struct btrfs_root *root, |
1148 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1149 | { |
1150 | struct btrfs_path *path; | |
1151 | int ret; | |
1152 | struct btrfs_key key; | |
e17cade2 | 1153 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1154 | struct btrfs_key found_key; |
1155 | ||
1156 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1157 | if (!path) |
1158 | return -ENOMEM; | |
0b86a832 | 1159 | |
e17cade2 | 1160 | key.objectid = objectid; |
0b86a832 CM |
1161 | key.offset = (u64)-1; |
1162 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1163 | ||
1164 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1165 | if (ret < 0) | |
1166 | goto error; | |
1167 | ||
79787eaa | 1168 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1169 | |
1170 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1171 | if (ret) { | |
e17cade2 | 1172 | *offset = 0; |
0b86a832 CM |
1173 | } else { |
1174 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1175 | path->slots[0]); | |
e17cade2 CM |
1176 | if (found_key.objectid != objectid) |
1177 | *offset = 0; | |
1178 | else { | |
1179 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1180 | struct btrfs_chunk); | |
1181 | *offset = found_key.offset + | |
1182 | btrfs_chunk_length(path->nodes[0], chunk); | |
1183 | } | |
0b86a832 CM |
1184 | } |
1185 | ret = 0; | |
1186 | error: | |
1187 | btrfs_free_path(path); | |
1188 | return ret; | |
1189 | } | |
1190 | ||
2b82032c | 1191 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1192 | { |
1193 | int ret; | |
1194 | struct btrfs_key key; | |
1195 | struct btrfs_key found_key; | |
2b82032c YZ |
1196 | struct btrfs_path *path; |
1197 | ||
1198 | root = root->fs_info->chunk_root; | |
1199 | ||
1200 | path = btrfs_alloc_path(); | |
1201 | if (!path) | |
1202 | return -ENOMEM; | |
0b86a832 CM |
1203 | |
1204 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1205 | key.type = BTRFS_DEV_ITEM_KEY; | |
1206 | key.offset = (u64)-1; | |
1207 | ||
1208 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1209 | if (ret < 0) | |
1210 | goto error; | |
1211 | ||
79787eaa | 1212 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 CM |
1213 | |
1214 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1215 | BTRFS_DEV_ITEM_KEY); | |
1216 | if (ret) { | |
1217 | *objectid = 1; | |
1218 | } else { | |
1219 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1220 | path->slots[0]); | |
1221 | *objectid = found_key.offset + 1; | |
1222 | } | |
1223 | ret = 0; | |
1224 | error: | |
2b82032c | 1225 | btrfs_free_path(path); |
0b86a832 CM |
1226 | return ret; |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * the device information is stored in the chunk root | |
1231 | * the btrfs_device struct should be fully filled in | |
1232 | */ | |
1233 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1234 | struct btrfs_root *root, | |
1235 | struct btrfs_device *device) | |
1236 | { | |
1237 | int ret; | |
1238 | struct btrfs_path *path; | |
1239 | struct btrfs_dev_item *dev_item; | |
1240 | struct extent_buffer *leaf; | |
1241 | struct btrfs_key key; | |
1242 | unsigned long ptr; | |
0b86a832 CM |
1243 | |
1244 | root = root->fs_info->chunk_root; | |
1245 | ||
1246 | path = btrfs_alloc_path(); | |
1247 | if (!path) | |
1248 | return -ENOMEM; | |
1249 | ||
0b86a832 CM |
1250 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1251 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1252 | key.offset = device->devid; |
0b86a832 CM |
1253 | |
1254 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1255 | sizeof(*dev_item)); |
0b86a832 CM |
1256 | if (ret) |
1257 | goto out; | |
1258 | ||
1259 | leaf = path->nodes[0]; | |
1260 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1261 | ||
1262 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1263 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1264 | btrfs_set_device_type(leaf, dev_item, device->type); |
1265 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1266 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1267 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1268 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1269 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1270 | btrfs_set_device_group(leaf, dev_item, 0); |
1271 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1272 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1273 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1274 | |
0b86a832 | 1275 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1276 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1277 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1278 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1279 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1280 | |
2b82032c | 1281 | ret = 0; |
0b86a832 CM |
1282 | out: |
1283 | btrfs_free_path(path); | |
1284 | return ret; | |
1285 | } | |
8f18cf13 | 1286 | |
a061fc8d CM |
1287 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1288 | struct btrfs_device *device) | |
1289 | { | |
1290 | int ret; | |
1291 | struct btrfs_path *path; | |
a061fc8d | 1292 | struct btrfs_key key; |
a061fc8d CM |
1293 | struct btrfs_trans_handle *trans; |
1294 | ||
1295 | root = root->fs_info->chunk_root; | |
1296 | ||
1297 | path = btrfs_alloc_path(); | |
1298 | if (!path) | |
1299 | return -ENOMEM; | |
1300 | ||
a22285a6 | 1301 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1302 | if (IS_ERR(trans)) { |
1303 | btrfs_free_path(path); | |
1304 | return PTR_ERR(trans); | |
1305 | } | |
a061fc8d CM |
1306 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1307 | key.type = BTRFS_DEV_ITEM_KEY; | |
1308 | key.offset = device->devid; | |
7d9eb12c | 1309 | lock_chunks(root); |
a061fc8d CM |
1310 | |
1311 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1312 | if (ret < 0) | |
1313 | goto out; | |
1314 | ||
1315 | if (ret > 0) { | |
1316 | ret = -ENOENT; | |
1317 | goto out; | |
1318 | } | |
1319 | ||
1320 | ret = btrfs_del_item(trans, root, path); | |
1321 | if (ret) | |
1322 | goto out; | |
a061fc8d CM |
1323 | out: |
1324 | btrfs_free_path(path); | |
7d9eb12c | 1325 | unlock_chunks(root); |
a061fc8d CM |
1326 | btrfs_commit_transaction(trans, root); |
1327 | return ret; | |
1328 | } | |
1329 | ||
1330 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1331 | { | |
1332 | struct btrfs_device *device; | |
2b82032c | 1333 | struct btrfs_device *next_device; |
a061fc8d | 1334 | struct block_device *bdev; |
dfe25020 | 1335 | struct buffer_head *bh = NULL; |
a061fc8d | 1336 | struct btrfs_super_block *disk_super; |
1f78160c | 1337 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1338 | u64 all_avail; |
1339 | u64 devid; | |
2b82032c YZ |
1340 | u64 num_devices; |
1341 | u8 *dev_uuid; | |
a061fc8d | 1342 | int ret = 0; |
1f78160c | 1343 | bool clear_super = false; |
a061fc8d | 1344 | |
a061fc8d CM |
1345 | mutex_lock(&uuid_mutex); |
1346 | ||
1347 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1348 | root->fs_info->avail_system_alloc_bits | | |
1349 | root->fs_info->avail_metadata_alloc_bits; | |
1350 | ||
1351 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1352 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1353 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1354 | "on raid10\n"); | |
a061fc8d CM |
1355 | ret = -EINVAL; |
1356 | goto out; | |
1357 | } | |
1358 | ||
1359 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1360 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1361 | printk(KERN_ERR "btrfs: unable to go below two " |
1362 | "devices on raid1\n"); | |
a061fc8d CM |
1363 | ret = -EINVAL; |
1364 | goto out; | |
1365 | } | |
1366 | ||
dfe25020 | 1367 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1368 | struct list_head *devices; |
1369 | struct btrfs_device *tmp; | |
a061fc8d | 1370 | |
dfe25020 CM |
1371 | device = NULL; |
1372 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1373 | /* |
1374 | * It is safe to read the devices since the volume_mutex | |
1375 | * is held. | |
1376 | */ | |
c6e30871 | 1377 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1378 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1379 | device = tmp; | |
1380 | break; | |
1381 | } | |
1382 | } | |
1383 | bdev = NULL; | |
1384 | bh = NULL; | |
1385 | disk_super = NULL; | |
1386 | if (!device) { | |
d397712b CM |
1387 | printk(KERN_ERR "btrfs: no missing devices found to " |
1388 | "remove\n"); | |
dfe25020 CM |
1389 | goto out; |
1390 | } | |
dfe25020 | 1391 | } else { |
beaf8ab3 SB |
1392 | ret = btrfs_get_bdev_and_sb(device_path, |
1393 | FMODE_READ | FMODE_EXCL, | |
1394 | root->fs_info->bdev_holder, 0, | |
1395 | &bdev, &bh); | |
1396 | if (ret) | |
dfe25020 | 1397 | goto out; |
dfe25020 | 1398 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1399 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1400 | dev_uuid = disk_super->dev_item.uuid; |
1401 | device = btrfs_find_device(root, devid, dev_uuid, | |
1402 | disk_super->fsid); | |
dfe25020 CM |
1403 | if (!device) { |
1404 | ret = -ENOENT; | |
1405 | goto error_brelse; | |
1406 | } | |
2b82032c | 1407 | } |
dfe25020 | 1408 | |
2b82032c | 1409 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1410 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1411 | "device\n"); | |
2b82032c YZ |
1412 | ret = -EINVAL; |
1413 | goto error_brelse; | |
1414 | } | |
1415 | ||
1416 | if (device->writeable) { | |
0c1daee0 | 1417 | lock_chunks(root); |
2b82032c | 1418 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1419 | unlock_chunks(root); |
2b82032c | 1420 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1421 | clear_super = true; |
dfe25020 | 1422 | } |
a061fc8d CM |
1423 | |
1424 | ret = btrfs_shrink_device(device, 0); | |
1425 | if (ret) | |
9b3517e9 | 1426 | goto error_undo; |
a061fc8d | 1427 | |
a061fc8d CM |
1428 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1429 | if (ret) | |
9b3517e9 | 1430 | goto error_undo; |
a061fc8d | 1431 | |
2bf64758 JB |
1432 | spin_lock(&root->fs_info->free_chunk_lock); |
1433 | root->fs_info->free_chunk_space = device->total_bytes - | |
1434 | device->bytes_used; | |
1435 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1436 | ||
2b82032c | 1437 | device->in_fs_metadata = 0; |
a2de733c | 1438 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1439 | |
1440 | /* | |
1441 | * the device list mutex makes sure that we don't change | |
1442 | * the device list while someone else is writing out all | |
1443 | * the device supers. | |
1444 | */ | |
1f78160c XG |
1445 | |
1446 | cur_devices = device->fs_devices; | |
e5e9a520 | 1447 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1448 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1449 | |
e4404d6e | 1450 | device->fs_devices->num_devices--; |
02db0844 | 1451 | device->fs_devices->total_devices--; |
2b82032c | 1452 | |
cd02dca5 CM |
1453 | if (device->missing) |
1454 | root->fs_info->fs_devices->missing_devices--; | |
1455 | ||
2b82032c YZ |
1456 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1457 | struct btrfs_device, dev_list); | |
1458 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1459 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1460 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1461 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1462 | ||
1f78160c | 1463 | if (device->bdev) |
e4404d6e | 1464 | device->fs_devices->open_devices--; |
1f78160c XG |
1465 | |
1466 | call_rcu(&device->rcu, free_device); | |
1467 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1468 | |
6c41761f DS |
1469 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1470 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1471 | |
1f78160c | 1472 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1473 | struct btrfs_fs_devices *fs_devices; |
1474 | fs_devices = root->fs_info->fs_devices; | |
1475 | while (fs_devices) { | |
1f78160c | 1476 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1477 | break; |
1478 | fs_devices = fs_devices->seed; | |
2b82032c | 1479 | } |
1f78160c XG |
1480 | fs_devices->seed = cur_devices->seed; |
1481 | cur_devices->seed = NULL; | |
0c1daee0 | 1482 | lock_chunks(root); |
1f78160c | 1483 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1484 | unlock_chunks(root); |
1f78160c | 1485 | free_fs_devices(cur_devices); |
2b82032c YZ |
1486 | } |
1487 | ||
5af3e8cc SB |
1488 | root->fs_info->num_tolerated_disk_barrier_failures = |
1489 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1490 | ||
2b82032c YZ |
1491 | /* |
1492 | * at this point, the device is zero sized. We want to | |
1493 | * remove it from the devices list and zero out the old super | |
1494 | */ | |
1f78160c | 1495 | if (clear_super) { |
dfe25020 CM |
1496 | /* make sure this device isn't detected as part of |
1497 | * the FS anymore | |
1498 | */ | |
1499 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1500 | set_buffer_dirty(bh); | |
1501 | sync_dirty_buffer(bh); | |
dfe25020 | 1502 | } |
a061fc8d | 1503 | |
a061fc8d | 1504 | ret = 0; |
a061fc8d CM |
1505 | |
1506 | error_brelse: | |
1507 | brelse(bh); | |
1508 | error_close: | |
dfe25020 | 1509 | if (bdev) |
e525fd89 | 1510 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1511 | out: |
1512 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1513 | return ret; |
9b3517e9 ID |
1514 | error_undo: |
1515 | if (device->writeable) { | |
0c1daee0 | 1516 | lock_chunks(root); |
9b3517e9 ID |
1517 | list_add(&device->dev_alloc_list, |
1518 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1519 | unlock_chunks(root); |
9b3517e9 ID |
1520 | root->fs_info->fs_devices->rw_devices++; |
1521 | } | |
1522 | goto error_brelse; | |
a061fc8d CM |
1523 | } |
1524 | ||
7ba15b7d SB |
1525 | int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1526 | struct btrfs_device **device) | |
1527 | { | |
1528 | int ret = 0; | |
1529 | struct btrfs_super_block *disk_super; | |
1530 | u64 devid; | |
1531 | u8 *dev_uuid; | |
1532 | struct block_device *bdev; | |
1533 | struct buffer_head *bh; | |
1534 | ||
1535 | *device = NULL; | |
1536 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1537 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1538 | if (ret) | |
1539 | return ret; | |
1540 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1541 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1542 | dev_uuid = disk_super->dev_item.uuid; | |
1543 | *device = btrfs_find_device(root, devid, dev_uuid, | |
1544 | disk_super->fsid); | |
1545 | brelse(bh); | |
1546 | if (!*device) | |
1547 | ret = -ENOENT; | |
1548 | blkdev_put(bdev, FMODE_READ); | |
1549 | return ret; | |
1550 | } | |
1551 | ||
1552 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
1553 | char *device_path, | |
1554 | struct btrfs_device **device) | |
1555 | { | |
1556 | *device = NULL; | |
1557 | if (strcmp(device_path, "missing") == 0) { | |
1558 | struct list_head *devices; | |
1559 | struct btrfs_device *tmp; | |
1560 | ||
1561 | devices = &root->fs_info->fs_devices->devices; | |
1562 | /* | |
1563 | * It is safe to read the devices since the volume_mutex | |
1564 | * is held by the caller. | |
1565 | */ | |
1566 | list_for_each_entry(tmp, devices, dev_list) { | |
1567 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
1568 | *device = tmp; | |
1569 | break; | |
1570 | } | |
1571 | } | |
1572 | ||
1573 | if (!*device) { | |
1574 | pr_err("btrfs: no missing device found\n"); | |
1575 | return -ENOENT; | |
1576 | } | |
1577 | ||
1578 | return 0; | |
1579 | } else { | |
1580 | return btrfs_find_device_by_path(root, device_path, device); | |
1581 | } | |
1582 | } | |
1583 | ||
2b82032c YZ |
1584 | /* |
1585 | * does all the dirty work required for changing file system's UUID. | |
1586 | */ | |
125ccb0a | 1587 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1588 | { |
1589 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1590 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1591 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1592 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1593 | struct btrfs_device *device; |
1594 | u64 super_flags; | |
1595 | ||
1596 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1597 | if (!fs_devices->seeding) |
2b82032c YZ |
1598 | return -EINVAL; |
1599 | ||
e4404d6e YZ |
1600 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1601 | if (!seed_devices) | |
2b82032c YZ |
1602 | return -ENOMEM; |
1603 | ||
e4404d6e YZ |
1604 | old_devices = clone_fs_devices(fs_devices); |
1605 | if (IS_ERR(old_devices)) { | |
1606 | kfree(seed_devices); | |
1607 | return PTR_ERR(old_devices); | |
2b82032c | 1608 | } |
e4404d6e | 1609 | |
2b82032c YZ |
1610 | list_add(&old_devices->list, &fs_uuids); |
1611 | ||
e4404d6e YZ |
1612 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1613 | seed_devices->opened = 1; | |
1614 | INIT_LIST_HEAD(&seed_devices->devices); | |
1615 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1616 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1617 | |
1618 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1619 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1620 | synchronize_rcu); | |
c9513edb XG |
1621 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1622 | ||
e4404d6e YZ |
1623 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1624 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1625 | device->fs_devices = seed_devices; | |
1626 | } | |
1627 | ||
2b82032c YZ |
1628 | fs_devices->seeding = 0; |
1629 | fs_devices->num_devices = 0; | |
1630 | fs_devices->open_devices = 0; | |
02db0844 | 1631 | fs_devices->total_devices = 0; |
e4404d6e | 1632 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1633 | |
1634 | generate_random_uuid(fs_devices->fsid); | |
1635 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1636 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1637 | super_flags = btrfs_super_flags(disk_super) & | |
1638 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1639 | btrfs_set_super_flags(disk_super, super_flags); | |
1640 | ||
1641 | return 0; | |
1642 | } | |
1643 | ||
1644 | /* | |
1645 | * strore the expected generation for seed devices in device items. | |
1646 | */ | |
1647 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1648 | struct btrfs_root *root) | |
1649 | { | |
1650 | struct btrfs_path *path; | |
1651 | struct extent_buffer *leaf; | |
1652 | struct btrfs_dev_item *dev_item; | |
1653 | struct btrfs_device *device; | |
1654 | struct btrfs_key key; | |
1655 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1656 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1657 | u64 devid; | |
1658 | int ret; | |
1659 | ||
1660 | path = btrfs_alloc_path(); | |
1661 | if (!path) | |
1662 | return -ENOMEM; | |
1663 | ||
1664 | root = root->fs_info->chunk_root; | |
1665 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1666 | key.offset = 0; | |
1667 | key.type = BTRFS_DEV_ITEM_KEY; | |
1668 | ||
1669 | while (1) { | |
1670 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1671 | if (ret < 0) | |
1672 | goto error; | |
1673 | ||
1674 | leaf = path->nodes[0]; | |
1675 | next_slot: | |
1676 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1677 | ret = btrfs_next_leaf(root, path); | |
1678 | if (ret > 0) | |
1679 | break; | |
1680 | if (ret < 0) | |
1681 | goto error; | |
1682 | leaf = path->nodes[0]; | |
1683 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1684 | btrfs_release_path(path); |
2b82032c YZ |
1685 | continue; |
1686 | } | |
1687 | ||
1688 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1689 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1690 | key.type != BTRFS_DEV_ITEM_KEY) | |
1691 | break; | |
1692 | ||
1693 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1694 | struct btrfs_dev_item); | |
1695 | devid = btrfs_device_id(leaf, dev_item); | |
1696 | read_extent_buffer(leaf, dev_uuid, | |
1697 | (unsigned long)btrfs_device_uuid(dev_item), | |
1698 | BTRFS_UUID_SIZE); | |
1699 | read_extent_buffer(leaf, fs_uuid, | |
1700 | (unsigned long)btrfs_device_fsid(dev_item), | |
1701 | BTRFS_UUID_SIZE); | |
1702 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
79787eaa | 1703 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
1704 | |
1705 | if (device->fs_devices->seeding) { | |
1706 | btrfs_set_device_generation(leaf, dev_item, | |
1707 | device->generation); | |
1708 | btrfs_mark_buffer_dirty(leaf); | |
1709 | } | |
1710 | ||
1711 | path->slots[0]++; | |
1712 | goto next_slot; | |
1713 | } | |
1714 | ret = 0; | |
1715 | error: | |
1716 | btrfs_free_path(path); | |
1717 | return ret; | |
1718 | } | |
1719 | ||
788f20eb CM |
1720 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1721 | { | |
d5e2003c | 1722 | struct request_queue *q; |
788f20eb CM |
1723 | struct btrfs_trans_handle *trans; |
1724 | struct btrfs_device *device; | |
1725 | struct block_device *bdev; | |
788f20eb | 1726 | struct list_head *devices; |
2b82032c | 1727 | struct super_block *sb = root->fs_info->sb; |
606686ee | 1728 | struct rcu_string *name; |
788f20eb | 1729 | u64 total_bytes; |
2b82032c | 1730 | int seeding_dev = 0; |
788f20eb CM |
1731 | int ret = 0; |
1732 | ||
2b82032c | 1733 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 1734 | return -EROFS; |
788f20eb | 1735 | |
a5d16333 | 1736 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1737 | root->fs_info->bdev_holder); |
7f59203a JB |
1738 | if (IS_ERR(bdev)) |
1739 | return PTR_ERR(bdev); | |
a2135011 | 1740 | |
2b82032c YZ |
1741 | if (root->fs_info->fs_devices->seeding) { |
1742 | seeding_dev = 1; | |
1743 | down_write(&sb->s_umount); | |
1744 | mutex_lock(&uuid_mutex); | |
1745 | } | |
1746 | ||
8c8bee1d | 1747 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1748 | |
788f20eb | 1749 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
1750 | |
1751 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 1752 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1753 | if (device->bdev == bdev) { |
1754 | ret = -EEXIST; | |
d25628bd LB |
1755 | mutex_unlock( |
1756 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1757 | goto error; |
788f20eb CM |
1758 | } |
1759 | } | |
d25628bd | 1760 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb CM |
1761 | |
1762 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1763 | if (!device) { | |
1764 | /* we can safely leave the fs_devices entry around */ | |
1765 | ret = -ENOMEM; | |
2b82032c | 1766 | goto error; |
788f20eb CM |
1767 | } |
1768 | ||
606686ee JB |
1769 | name = rcu_string_strdup(device_path, GFP_NOFS); |
1770 | if (!name) { | |
788f20eb | 1771 | kfree(device); |
2b82032c YZ |
1772 | ret = -ENOMEM; |
1773 | goto error; | |
788f20eb | 1774 | } |
606686ee | 1775 | rcu_assign_pointer(device->name, name); |
2b82032c YZ |
1776 | |
1777 | ret = find_next_devid(root, &device->devid); | |
1778 | if (ret) { | |
606686ee | 1779 | rcu_string_free(device->name); |
2b82032c YZ |
1780 | kfree(device); |
1781 | goto error; | |
1782 | } | |
1783 | ||
a22285a6 | 1784 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1785 | if (IS_ERR(trans)) { |
606686ee | 1786 | rcu_string_free(device->name); |
98d5dc13 TI |
1787 | kfree(device); |
1788 | ret = PTR_ERR(trans); | |
1789 | goto error; | |
1790 | } | |
1791 | ||
2b82032c YZ |
1792 | lock_chunks(root); |
1793 | ||
d5e2003c JB |
1794 | q = bdev_get_queue(bdev); |
1795 | if (blk_queue_discard(q)) | |
1796 | device->can_discard = 1; | |
2b82032c YZ |
1797 | device->writeable = 1; |
1798 | device->work.func = pending_bios_fn; | |
1799 | generate_random_uuid(device->uuid); | |
1800 | spin_lock_init(&device->io_lock); | |
1801 | device->generation = trans->transid; | |
788f20eb CM |
1802 | device->io_width = root->sectorsize; |
1803 | device->io_align = root->sectorsize; | |
1804 | device->sector_size = root->sectorsize; | |
1805 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1806 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1807 | device->dev_root = root->fs_info->dev_root; |
1808 | device->bdev = bdev; | |
dfe25020 | 1809 | device->in_fs_metadata = 1; |
fb01aa85 | 1810 | device->mode = FMODE_EXCL; |
2b82032c | 1811 | set_blocksize(device->bdev, 4096); |
788f20eb | 1812 | |
2b82032c YZ |
1813 | if (seeding_dev) { |
1814 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 1815 | ret = btrfs_prepare_sprout(root); |
79787eaa | 1816 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 1817 | } |
788f20eb | 1818 | |
2b82032c | 1819 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 1820 | |
e5e9a520 | 1821 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1822 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1823 | list_add(&device->dev_alloc_list, |
1824 | &root->fs_info->fs_devices->alloc_list); | |
1825 | root->fs_info->fs_devices->num_devices++; | |
1826 | root->fs_info->fs_devices->open_devices++; | |
1827 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 1828 | root->fs_info->fs_devices->total_devices++; |
d5e2003c JB |
1829 | if (device->can_discard) |
1830 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1831 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1832 | |
2bf64758 JB |
1833 | spin_lock(&root->fs_info->free_chunk_lock); |
1834 | root->fs_info->free_chunk_space += device->total_bytes; | |
1835 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1836 | ||
c289811c CM |
1837 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1838 | root->fs_info->fs_devices->rotating = 1; | |
1839 | ||
6c41761f DS |
1840 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1841 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1842 | total_bytes + device->total_bytes); |
1843 | ||
6c41761f DS |
1844 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1845 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1846 | total_bytes + 1); |
e5e9a520 | 1847 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1848 | |
2b82032c YZ |
1849 | if (seeding_dev) { |
1850 | ret = init_first_rw_device(trans, root, device); | |
005d6427 DS |
1851 | if (ret) { |
1852 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1853 | goto error_trans; |
005d6427 | 1854 | } |
2b82032c | 1855 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
1856 | if (ret) { |
1857 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1858 | goto error_trans; |
005d6427 | 1859 | } |
2b82032c YZ |
1860 | } else { |
1861 | ret = btrfs_add_device(trans, root, device); | |
005d6427 DS |
1862 | if (ret) { |
1863 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 1864 | goto error_trans; |
005d6427 | 1865 | } |
2b82032c YZ |
1866 | } |
1867 | ||
913d952e CM |
1868 | /* |
1869 | * we've got more storage, clear any full flags on the space | |
1870 | * infos | |
1871 | */ | |
1872 | btrfs_clear_space_info_full(root->fs_info); | |
1873 | ||
7d9eb12c | 1874 | unlock_chunks(root); |
5af3e8cc SB |
1875 | root->fs_info->num_tolerated_disk_barrier_failures = |
1876 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 1877 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 1878 | |
2b82032c YZ |
1879 | if (seeding_dev) { |
1880 | mutex_unlock(&uuid_mutex); | |
1881 | up_write(&sb->s_umount); | |
788f20eb | 1882 | |
79787eaa JM |
1883 | if (ret) /* transaction commit */ |
1884 | return ret; | |
1885 | ||
2b82032c | 1886 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
1887 | if (ret < 0) |
1888 | btrfs_error(root->fs_info, ret, | |
1889 | "Failed to relocate sys chunks after " | |
1890 | "device initialization. This can be fixed " | |
1891 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
1892 | trans = btrfs_attach_transaction(root); |
1893 | if (IS_ERR(trans)) { | |
1894 | if (PTR_ERR(trans) == -ENOENT) | |
1895 | return 0; | |
1896 | return PTR_ERR(trans); | |
1897 | } | |
1898 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 1899 | } |
c9e9f97b | 1900 | |
2b82032c | 1901 | return ret; |
79787eaa JM |
1902 | |
1903 | error_trans: | |
1904 | unlock_chunks(root); | |
79787eaa | 1905 | btrfs_end_transaction(trans, root); |
606686ee | 1906 | rcu_string_free(device->name); |
79787eaa | 1907 | kfree(device); |
2b82032c | 1908 | error: |
e525fd89 | 1909 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1910 | if (seeding_dev) { |
1911 | mutex_unlock(&uuid_mutex); | |
1912 | up_write(&sb->s_umount); | |
1913 | } | |
c9e9f97b | 1914 | return ret; |
788f20eb CM |
1915 | } |
1916 | ||
d397712b CM |
1917 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1918 | struct btrfs_device *device) | |
0b86a832 CM |
1919 | { |
1920 | int ret; | |
1921 | struct btrfs_path *path; | |
1922 | struct btrfs_root *root; | |
1923 | struct btrfs_dev_item *dev_item; | |
1924 | struct extent_buffer *leaf; | |
1925 | struct btrfs_key key; | |
1926 | ||
1927 | root = device->dev_root->fs_info->chunk_root; | |
1928 | ||
1929 | path = btrfs_alloc_path(); | |
1930 | if (!path) | |
1931 | return -ENOMEM; | |
1932 | ||
1933 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1934 | key.type = BTRFS_DEV_ITEM_KEY; | |
1935 | key.offset = device->devid; | |
1936 | ||
1937 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1938 | if (ret < 0) | |
1939 | goto out; | |
1940 | ||
1941 | if (ret > 0) { | |
1942 | ret = -ENOENT; | |
1943 | goto out; | |
1944 | } | |
1945 | ||
1946 | leaf = path->nodes[0]; | |
1947 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1948 | ||
1949 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1950 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1951 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1952 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1953 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1954 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1955 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1956 | btrfs_mark_buffer_dirty(leaf); | |
1957 | ||
1958 | out: | |
1959 | btrfs_free_path(path); | |
1960 | return ret; | |
1961 | } | |
1962 | ||
7d9eb12c | 1963 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1964 | struct btrfs_device *device, u64 new_size) |
1965 | { | |
1966 | struct btrfs_super_block *super_copy = | |
6c41761f | 1967 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1968 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1969 | u64 diff = new_size - device->total_bytes; | |
1970 | ||
2b82032c YZ |
1971 | if (!device->writeable) |
1972 | return -EACCES; | |
1973 | if (new_size <= device->total_bytes) | |
1974 | return -EINVAL; | |
1975 | ||
8f18cf13 | 1976 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1977 | device->fs_devices->total_rw_bytes += diff; |
1978 | ||
1979 | device->total_bytes = new_size; | |
9779b72f | 1980 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1981 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1982 | ||
8f18cf13 CM |
1983 | return btrfs_update_device(trans, device); |
1984 | } | |
1985 | ||
7d9eb12c CM |
1986 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1987 | struct btrfs_device *device, u64 new_size) | |
1988 | { | |
1989 | int ret; | |
1990 | lock_chunks(device->dev_root); | |
1991 | ret = __btrfs_grow_device(trans, device, new_size); | |
1992 | unlock_chunks(device->dev_root); | |
1993 | return ret; | |
1994 | } | |
1995 | ||
8f18cf13 CM |
1996 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1997 | struct btrfs_root *root, | |
1998 | u64 chunk_tree, u64 chunk_objectid, | |
1999 | u64 chunk_offset) | |
2000 | { | |
2001 | int ret; | |
2002 | struct btrfs_path *path; | |
2003 | struct btrfs_key key; | |
2004 | ||
2005 | root = root->fs_info->chunk_root; | |
2006 | path = btrfs_alloc_path(); | |
2007 | if (!path) | |
2008 | return -ENOMEM; | |
2009 | ||
2010 | key.objectid = chunk_objectid; | |
2011 | key.offset = chunk_offset; | |
2012 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2013 | ||
2014 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2015 | if (ret < 0) |
2016 | goto out; | |
2017 | else if (ret > 0) { /* Logic error or corruption */ | |
2018 | btrfs_error(root->fs_info, -ENOENT, | |
2019 | "Failed lookup while freeing chunk."); | |
2020 | ret = -ENOENT; | |
2021 | goto out; | |
2022 | } | |
8f18cf13 CM |
2023 | |
2024 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2025 | if (ret < 0) |
2026 | btrfs_error(root->fs_info, ret, | |
2027 | "Failed to delete chunk item."); | |
2028 | out: | |
8f18cf13 | 2029 | btrfs_free_path(path); |
65a246c5 | 2030 | return ret; |
8f18cf13 CM |
2031 | } |
2032 | ||
b2950863 | 2033 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2034 | chunk_offset) |
2035 | { | |
6c41761f | 2036 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2037 | struct btrfs_disk_key *disk_key; |
2038 | struct btrfs_chunk *chunk; | |
2039 | u8 *ptr; | |
2040 | int ret = 0; | |
2041 | u32 num_stripes; | |
2042 | u32 array_size; | |
2043 | u32 len = 0; | |
2044 | u32 cur; | |
2045 | struct btrfs_key key; | |
2046 | ||
2047 | array_size = btrfs_super_sys_array_size(super_copy); | |
2048 | ||
2049 | ptr = super_copy->sys_chunk_array; | |
2050 | cur = 0; | |
2051 | ||
2052 | while (cur < array_size) { | |
2053 | disk_key = (struct btrfs_disk_key *)ptr; | |
2054 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2055 | ||
2056 | len = sizeof(*disk_key); | |
2057 | ||
2058 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2059 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2060 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2061 | len += btrfs_chunk_item_size(num_stripes); | |
2062 | } else { | |
2063 | ret = -EIO; | |
2064 | break; | |
2065 | } | |
2066 | if (key.objectid == chunk_objectid && | |
2067 | key.offset == chunk_offset) { | |
2068 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2069 | array_size -= len; | |
2070 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2071 | } else { | |
2072 | ptr += len; | |
2073 | cur += len; | |
2074 | } | |
2075 | } | |
2076 | return ret; | |
2077 | } | |
2078 | ||
b2950863 | 2079 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
2080 | u64 chunk_tree, u64 chunk_objectid, |
2081 | u64 chunk_offset) | |
2082 | { | |
2083 | struct extent_map_tree *em_tree; | |
2084 | struct btrfs_root *extent_root; | |
2085 | struct btrfs_trans_handle *trans; | |
2086 | struct extent_map *em; | |
2087 | struct map_lookup *map; | |
2088 | int ret; | |
2089 | int i; | |
2090 | ||
2091 | root = root->fs_info->chunk_root; | |
2092 | extent_root = root->fs_info->extent_root; | |
2093 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2094 | ||
ba1bf481 JB |
2095 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
2096 | if (ret) | |
2097 | return -ENOSPC; | |
2098 | ||
8f18cf13 | 2099 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 2100 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
2101 | if (ret) |
2102 | return ret; | |
8f18cf13 | 2103 | |
a22285a6 | 2104 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2105 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 2106 | |
7d9eb12c CM |
2107 | lock_chunks(root); |
2108 | ||
8f18cf13 CM |
2109 | /* |
2110 | * step two, delete the device extents and the | |
2111 | * chunk tree entries | |
2112 | */ | |
890871be | 2113 | read_lock(&em_tree->lock); |
8f18cf13 | 2114 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2115 | read_unlock(&em_tree->lock); |
8f18cf13 | 2116 | |
285190d9 | 2117 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 2118 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
2119 | map = (struct map_lookup *)em->bdev; |
2120 | ||
2121 | for (i = 0; i < map->num_stripes; i++) { | |
2122 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
2123 | map->stripes[i].physical); | |
2124 | BUG_ON(ret); | |
a061fc8d | 2125 | |
dfe25020 CM |
2126 | if (map->stripes[i].dev) { |
2127 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2128 | BUG_ON(ret); | |
2129 | } | |
8f18cf13 CM |
2130 | } |
2131 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
2132 | chunk_offset); | |
2133 | ||
2134 | BUG_ON(ret); | |
2135 | ||
1abe9b8a | 2136 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2137 | ||
8f18cf13 CM |
2138 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2139 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2140 | BUG_ON(ret); | |
8f18cf13 CM |
2141 | } |
2142 | ||
2b82032c YZ |
2143 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
2144 | BUG_ON(ret); | |
2145 | ||
890871be | 2146 | write_lock(&em_tree->lock); |
2b82032c | 2147 | remove_extent_mapping(em_tree, em); |
890871be | 2148 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2149 | |
2150 | kfree(map); | |
2151 | em->bdev = NULL; | |
2152 | ||
2153 | /* once for the tree */ | |
2154 | free_extent_map(em); | |
2155 | /* once for us */ | |
2156 | free_extent_map(em); | |
2157 | ||
2158 | unlock_chunks(root); | |
2159 | btrfs_end_transaction(trans, root); | |
2160 | return 0; | |
2161 | } | |
2162 | ||
2163 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2164 | { | |
2165 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2166 | struct btrfs_path *path; | |
2167 | struct extent_buffer *leaf; | |
2168 | struct btrfs_chunk *chunk; | |
2169 | struct btrfs_key key; | |
2170 | struct btrfs_key found_key; | |
2171 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2172 | u64 chunk_type; | |
ba1bf481 JB |
2173 | bool retried = false; |
2174 | int failed = 0; | |
2b82032c YZ |
2175 | int ret; |
2176 | ||
2177 | path = btrfs_alloc_path(); | |
2178 | if (!path) | |
2179 | return -ENOMEM; | |
2180 | ||
ba1bf481 | 2181 | again: |
2b82032c YZ |
2182 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2183 | key.offset = (u64)-1; | |
2184 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2185 | ||
2186 | while (1) { | |
2187 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2188 | if (ret < 0) | |
2189 | goto error; | |
79787eaa | 2190 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2191 | |
2192 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2193 | key.type); | |
2194 | if (ret < 0) | |
2195 | goto error; | |
2196 | if (ret > 0) | |
2197 | break; | |
1a40e23b | 2198 | |
2b82032c YZ |
2199 | leaf = path->nodes[0]; |
2200 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2201 | |
2b82032c YZ |
2202 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2203 | struct btrfs_chunk); | |
2204 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2205 | btrfs_release_path(path); |
8f18cf13 | 2206 | |
2b82032c YZ |
2207 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2208 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2209 | found_key.objectid, | |
2210 | found_key.offset); | |
ba1bf481 JB |
2211 | if (ret == -ENOSPC) |
2212 | failed++; | |
2213 | else if (ret) | |
2214 | BUG(); | |
2b82032c | 2215 | } |
8f18cf13 | 2216 | |
2b82032c YZ |
2217 | if (found_key.offset == 0) |
2218 | break; | |
2219 | key.offset = found_key.offset - 1; | |
2220 | } | |
2221 | ret = 0; | |
ba1bf481 JB |
2222 | if (failed && !retried) { |
2223 | failed = 0; | |
2224 | retried = true; | |
2225 | goto again; | |
2226 | } else if (failed && retried) { | |
2227 | WARN_ON(1); | |
2228 | ret = -ENOSPC; | |
2229 | } | |
2b82032c YZ |
2230 | error: |
2231 | btrfs_free_path(path); | |
2232 | return ret; | |
8f18cf13 CM |
2233 | } |
2234 | ||
0940ebf6 ID |
2235 | static int insert_balance_item(struct btrfs_root *root, |
2236 | struct btrfs_balance_control *bctl) | |
2237 | { | |
2238 | struct btrfs_trans_handle *trans; | |
2239 | struct btrfs_balance_item *item; | |
2240 | struct btrfs_disk_balance_args disk_bargs; | |
2241 | struct btrfs_path *path; | |
2242 | struct extent_buffer *leaf; | |
2243 | struct btrfs_key key; | |
2244 | int ret, err; | |
2245 | ||
2246 | path = btrfs_alloc_path(); | |
2247 | if (!path) | |
2248 | return -ENOMEM; | |
2249 | ||
2250 | trans = btrfs_start_transaction(root, 0); | |
2251 | if (IS_ERR(trans)) { | |
2252 | btrfs_free_path(path); | |
2253 | return PTR_ERR(trans); | |
2254 | } | |
2255 | ||
2256 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2257 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2258 | key.offset = 0; | |
2259 | ||
2260 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2261 | sizeof(*item)); | |
2262 | if (ret) | |
2263 | goto out; | |
2264 | ||
2265 | leaf = path->nodes[0]; | |
2266 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2267 | ||
2268 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2269 | ||
2270 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2271 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2272 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2273 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2274 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2275 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2276 | ||
2277 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2278 | ||
2279 | btrfs_mark_buffer_dirty(leaf); | |
2280 | out: | |
2281 | btrfs_free_path(path); | |
2282 | err = btrfs_commit_transaction(trans, root); | |
2283 | if (err && !ret) | |
2284 | ret = err; | |
2285 | return ret; | |
2286 | } | |
2287 | ||
2288 | static int del_balance_item(struct btrfs_root *root) | |
2289 | { | |
2290 | struct btrfs_trans_handle *trans; | |
2291 | struct btrfs_path *path; | |
2292 | struct btrfs_key key; | |
2293 | int ret, err; | |
2294 | ||
2295 | path = btrfs_alloc_path(); | |
2296 | if (!path) | |
2297 | return -ENOMEM; | |
2298 | ||
2299 | trans = btrfs_start_transaction(root, 0); | |
2300 | if (IS_ERR(trans)) { | |
2301 | btrfs_free_path(path); | |
2302 | return PTR_ERR(trans); | |
2303 | } | |
2304 | ||
2305 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2306 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2307 | key.offset = 0; | |
2308 | ||
2309 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2310 | if (ret < 0) | |
2311 | goto out; | |
2312 | if (ret > 0) { | |
2313 | ret = -ENOENT; | |
2314 | goto out; | |
2315 | } | |
2316 | ||
2317 | ret = btrfs_del_item(trans, root, path); | |
2318 | out: | |
2319 | btrfs_free_path(path); | |
2320 | err = btrfs_commit_transaction(trans, root); | |
2321 | if (err && !ret) | |
2322 | ret = err; | |
2323 | return ret; | |
2324 | } | |
2325 | ||
59641015 ID |
2326 | /* |
2327 | * This is a heuristic used to reduce the number of chunks balanced on | |
2328 | * resume after balance was interrupted. | |
2329 | */ | |
2330 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2331 | { | |
2332 | /* | |
2333 | * Turn on soft mode for chunk types that were being converted. | |
2334 | */ | |
2335 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2336 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2337 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2338 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2339 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2340 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2341 | ||
2342 | /* | |
2343 | * Turn on usage filter if is not already used. The idea is | |
2344 | * that chunks that we have already balanced should be | |
2345 | * reasonably full. Don't do it for chunks that are being | |
2346 | * converted - that will keep us from relocating unconverted | |
2347 | * (albeit full) chunks. | |
2348 | */ | |
2349 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2350 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2351 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2352 | bctl->data.usage = 90; | |
2353 | } | |
2354 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2355 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2356 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2357 | bctl->sys.usage = 90; | |
2358 | } | |
2359 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2360 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2361 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2362 | bctl->meta.usage = 90; | |
2363 | } | |
2364 | } | |
2365 | ||
c9e9f97b ID |
2366 | /* |
2367 | * Should be called with both balance and volume mutexes held to | |
2368 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2369 | * restriper. Same goes for unset_balance_control. | |
2370 | */ | |
2371 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2372 | { | |
2373 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2374 | ||
2375 | BUG_ON(fs_info->balance_ctl); | |
2376 | ||
2377 | spin_lock(&fs_info->balance_lock); | |
2378 | fs_info->balance_ctl = bctl; | |
2379 | spin_unlock(&fs_info->balance_lock); | |
2380 | } | |
2381 | ||
2382 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2383 | { | |
2384 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2385 | ||
2386 | BUG_ON(!fs_info->balance_ctl); | |
2387 | ||
2388 | spin_lock(&fs_info->balance_lock); | |
2389 | fs_info->balance_ctl = NULL; | |
2390 | spin_unlock(&fs_info->balance_lock); | |
2391 | ||
2392 | kfree(bctl); | |
2393 | } | |
2394 | ||
ed25e9b2 ID |
2395 | /* |
2396 | * Balance filters. Return 1 if chunk should be filtered out | |
2397 | * (should not be balanced). | |
2398 | */ | |
899c81ea | 2399 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2400 | struct btrfs_balance_args *bargs) |
2401 | { | |
899c81ea ID |
2402 | chunk_type = chunk_to_extended(chunk_type) & |
2403 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2404 | |
899c81ea | 2405 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2406 | return 0; |
2407 | ||
2408 | return 1; | |
2409 | } | |
2410 | ||
5ce5b3c0 ID |
2411 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
2412 | struct btrfs_balance_args *bargs) | |
2413 | { | |
2414 | struct btrfs_block_group_cache *cache; | |
2415 | u64 chunk_used, user_thresh; | |
2416 | int ret = 1; | |
2417 | ||
2418 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2419 | chunk_used = btrfs_block_group_used(&cache->item); | |
2420 | ||
2421 | user_thresh = div_factor_fine(cache->key.offset, bargs->usage); | |
2422 | if (chunk_used < user_thresh) | |
2423 | ret = 0; | |
2424 | ||
2425 | btrfs_put_block_group(cache); | |
2426 | return ret; | |
2427 | } | |
2428 | ||
409d404b ID |
2429 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2430 | struct btrfs_chunk *chunk, | |
2431 | struct btrfs_balance_args *bargs) | |
2432 | { | |
2433 | struct btrfs_stripe *stripe; | |
2434 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2435 | int i; | |
2436 | ||
2437 | for (i = 0; i < num_stripes; i++) { | |
2438 | stripe = btrfs_stripe_nr(chunk, i); | |
2439 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2440 | return 0; | |
2441 | } | |
2442 | ||
2443 | return 1; | |
2444 | } | |
2445 | ||
94e60d5a ID |
2446 | /* [pstart, pend) */ |
2447 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2448 | struct btrfs_chunk *chunk, | |
2449 | u64 chunk_offset, | |
2450 | struct btrfs_balance_args *bargs) | |
2451 | { | |
2452 | struct btrfs_stripe *stripe; | |
2453 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2454 | u64 stripe_offset; | |
2455 | u64 stripe_length; | |
2456 | int factor; | |
2457 | int i; | |
2458 | ||
2459 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2460 | return 0; | |
2461 | ||
2462 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
2463 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) | |
2464 | factor = 2; | |
2465 | else | |
2466 | factor = 1; | |
2467 | factor = num_stripes / factor; | |
2468 | ||
2469 | for (i = 0; i < num_stripes; i++) { | |
2470 | stripe = btrfs_stripe_nr(chunk, i); | |
2471 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2472 | continue; | |
2473 | ||
2474 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2475 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2476 | do_div(stripe_length, factor); | |
2477 | ||
2478 | if (stripe_offset < bargs->pend && | |
2479 | stripe_offset + stripe_length > bargs->pstart) | |
2480 | return 0; | |
2481 | } | |
2482 | ||
2483 | return 1; | |
2484 | } | |
2485 | ||
ea67176a ID |
2486 | /* [vstart, vend) */ |
2487 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2488 | struct btrfs_chunk *chunk, | |
2489 | u64 chunk_offset, | |
2490 | struct btrfs_balance_args *bargs) | |
2491 | { | |
2492 | if (chunk_offset < bargs->vend && | |
2493 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2494 | /* at least part of the chunk is inside this vrange */ | |
2495 | return 0; | |
2496 | ||
2497 | return 1; | |
2498 | } | |
2499 | ||
899c81ea | 2500 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2501 | struct btrfs_balance_args *bargs) |
2502 | { | |
2503 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2504 | return 0; | |
2505 | ||
899c81ea ID |
2506 | chunk_type = chunk_to_extended(chunk_type) & |
2507 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2508 | |
899c81ea | 2509 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2510 | return 1; |
2511 | ||
2512 | return 0; | |
2513 | } | |
2514 | ||
f43ffb60 ID |
2515 | static int should_balance_chunk(struct btrfs_root *root, |
2516 | struct extent_buffer *leaf, | |
2517 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2518 | { | |
2519 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2520 | struct btrfs_balance_args *bargs = NULL; | |
2521 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2522 | ||
2523 | /* type filter */ | |
2524 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2525 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2526 | return 0; | |
2527 | } | |
2528 | ||
2529 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2530 | bargs = &bctl->data; | |
2531 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2532 | bargs = &bctl->sys; | |
2533 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2534 | bargs = &bctl->meta; | |
2535 | ||
ed25e9b2 ID |
2536 | /* profiles filter */ |
2537 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2538 | chunk_profiles_filter(chunk_type, bargs)) { | |
2539 | return 0; | |
5ce5b3c0 ID |
2540 | } |
2541 | ||
2542 | /* usage filter */ | |
2543 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2544 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2545 | return 0; | |
409d404b ID |
2546 | } |
2547 | ||
2548 | /* devid filter */ | |
2549 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2550 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2551 | return 0; | |
94e60d5a ID |
2552 | } |
2553 | ||
2554 | /* drange filter, makes sense only with devid filter */ | |
2555 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2556 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2557 | return 0; | |
ea67176a ID |
2558 | } |
2559 | ||
2560 | /* vrange filter */ | |
2561 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2562 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2563 | return 0; | |
ed25e9b2 ID |
2564 | } |
2565 | ||
cfa4c961 ID |
2566 | /* soft profile changing mode */ |
2567 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2568 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2569 | return 0; | |
2570 | } | |
2571 | ||
f43ffb60 ID |
2572 | return 1; |
2573 | } | |
2574 | ||
c9e9f97b | 2575 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2576 | { |
19a39dce | 2577 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2578 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2579 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2580 | struct list_head *devices; | |
ec44a35c CM |
2581 | struct btrfs_device *device; |
2582 | u64 old_size; | |
2583 | u64 size_to_free; | |
f43ffb60 | 2584 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2585 | struct btrfs_path *path; |
2586 | struct btrfs_key key; | |
ec44a35c | 2587 | struct btrfs_key found_key; |
c9e9f97b | 2588 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2589 | struct extent_buffer *leaf; |
2590 | int slot; | |
c9e9f97b ID |
2591 | int ret; |
2592 | int enospc_errors = 0; | |
19a39dce | 2593 | bool counting = true; |
ec44a35c | 2594 | |
ec44a35c | 2595 | /* step one make some room on all the devices */ |
c9e9f97b | 2596 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2597 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2598 | old_size = device->total_bytes; |
2599 | size_to_free = div_factor(old_size, 1); | |
2600 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2601 | if (!device->writeable || |
2602 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2603 | continue; |
2604 | ||
2605 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2606 | if (ret == -ENOSPC) |
2607 | break; | |
ec44a35c CM |
2608 | BUG_ON(ret); |
2609 | ||
a22285a6 | 2610 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2611 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2612 | |
2613 | ret = btrfs_grow_device(trans, device, old_size); | |
2614 | BUG_ON(ret); | |
2615 | ||
2616 | btrfs_end_transaction(trans, dev_root); | |
2617 | } | |
2618 | ||
2619 | /* step two, relocate all the chunks */ | |
2620 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2621 | if (!path) { |
2622 | ret = -ENOMEM; | |
2623 | goto error; | |
2624 | } | |
19a39dce ID |
2625 | |
2626 | /* zero out stat counters */ | |
2627 | spin_lock(&fs_info->balance_lock); | |
2628 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2629 | spin_unlock(&fs_info->balance_lock); | |
2630 | again: | |
ec44a35c CM |
2631 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2632 | key.offset = (u64)-1; | |
2633 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2634 | ||
d397712b | 2635 | while (1) { |
19a39dce | 2636 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2637 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2638 | ret = -ECANCELED; |
2639 | goto error; | |
2640 | } | |
2641 | ||
ec44a35c CM |
2642 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2643 | if (ret < 0) | |
2644 | goto error; | |
2645 | ||
2646 | /* | |
2647 | * this shouldn't happen, it means the last relocate | |
2648 | * failed | |
2649 | */ | |
2650 | if (ret == 0) | |
c9e9f97b | 2651 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2652 | |
2653 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2654 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2655 | if (ret) { |
2656 | ret = 0; | |
ec44a35c | 2657 | break; |
c9e9f97b | 2658 | } |
7d9eb12c | 2659 | |
f43ffb60 ID |
2660 | leaf = path->nodes[0]; |
2661 | slot = path->slots[0]; | |
2662 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2663 | |
ec44a35c CM |
2664 | if (found_key.objectid != key.objectid) |
2665 | break; | |
7d9eb12c | 2666 | |
ec44a35c | 2667 | /* chunk zero is special */ |
ba1bf481 | 2668 | if (found_key.offset == 0) |
ec44a35c CM |
2669 | break; |
2670 | ||
f43ffb60 ID |
2671 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2672 | ||
19a39dce ID |
2673 | if (!counting) { |
2674 | spin_lock(&fs_info->balance_lock); | |
2675 | bctl->stat.considered++; | |
2676 | spin_unlock(&fs_info->balance_lock); | |
2677 | } | |
2678 | ||
f43ffb60 ID |
2679 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2680 | found_key.offset); | |
b3b4aa74 | 2681 | btrfs_release_path(path); |
f43ffb60 ID |
2682 | if (!ret) |
2683 | goto loop; | |
2684 | ||
19a39dce ID |
2685 | if (counting) { |
2686 | spin_lock(&fs_info->balance_lock); | |
2687 | bctl->stat.expected++; | |
2688 | spin_unlock(&fs_info->balance_lock); | |
2689 | goto loop; | |
2690 | } | |
2691 | ||
ec44a35c CM |
2692 | ret = btrfs_relocate_chunk(chunk_root, |
2693 | chunk_root->root_key.objectid, | |
2694 | found_key.objectid, | |
2695 | found_key.offset); | |
508794eb JB |
2696 | if (ret && ret != -ENOSPC) |
2697 | goto error; | |
19a39dce | 2698 | if (ret == -ENOSPC) { |
c9e9f97b | 2699 | enospc_errors++; |
19a39dce ID |
2700 | } else { |
2701 | spin_lock(&fs_info->balance_lock); | |
2702 | bctl->stat.completed++; | |
2703 | spin_unlock(&fs_info->balance_lock); | |
2704 | } | |
f43ffb60 | 2705 | loop: |
ba1bf481 | 2706 | key.offset = found_key.offset - 1; |
ec44a35c | 2707 | } |
c9e9f97b | 2708 | |
19a39dce ID |
2709 | if (counting) { |
2710 | btrfs_release_path(path); | |
2711 | counting = false; | |
2712 | goto again; | |
2713 | } | |
ec44a35c CM |
2714 | error: |
2715 | btrfs_free_path(path); | |
c9e9f97b ID |
2716 | if (enospc_errors) { |
2717 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
2718 | enospc_errors); | |
2719 | if (!ret) | |
2720 | ret = -ENOSPC; | |
2721 | } | |
2722 | ||
ec44a35c CM |
2723 | return ret; |
2724 | } | |
2725 | ||
0c460c0d ID |
2726 | /** |
2727 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
2728 | * @flags: profile to validate | |
2729 | * @extended: if true @flags is treated as an extended profile | |
2730 | */ | |
2731 | static int alloc_profile_is_valid(u64 flags, int extended) | |
2732 | { | |
2733 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
2734 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
2735 | ||
2736 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
2737 | ||
2738 | /* 1) check that all other bits are zeroed */ | |
2739 | if (flags & ~mask) | |
2740 | return 0; | |
2741 | ||
2742 | /* 2) see if profile is reduced */ | |
2743 | if (flags == 0) | |
2744 | return !extended; /* "0" is valid for usual profiles */ | |
2745 | ||
2746 | /* true if exactly one bit set */ | |
2747 | return (flags & (flags - 1)) == 0; | |
2748 | } | |
2749 | ||
837d5b6e ID |
2750 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
2751 | { | |
a7e99c69 ID |
2752 | /* cancel requested || normal exit path */ |
2753 | return atomic_read(&fs_info->balance_cancel_req) || | |
2754 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
2755 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
2756 | } |
2757 | ||
c9e9f97b ID |
2758 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
2759 | { | |
0940ebf6 ID |
2760 | int ret; |
2761 | ||
c9e9f97b | 2762 | unset_balance_control(fs_info); |
0940ebf6 ID |
2763 | ret = del_balance_item(fs_info->tree_root); |
2764 | BUG_ON(ret); | |
c9e9f97b ID |
2765 | } |
2766 | ||
19a39dce | 2767 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
2768 | struct btrfs_ioctl_balance_args *bargs); |
2769 | ||
2770 | /* | |
2771 | * Should be called with both balance and volume mutexes held | |
2772 | */ | |
2773 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
2774 | struct btrfs_ioctl_balance_args *bargs) | |
2775 | { | |
2776 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 2777 | u64 allowed; |
e4837f8f | 2778 | int mixed = 0; |
c9e9f97b ID |
2779 | int ret; |
2780 | ||
837d5b6e | 2781 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
2782 | atomic_read(&fs_info->balance_pause_req) || |
2783 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
2784 | ret = -EINVAL; |
2785 | goto out; | |
2786 | } | |
2787 | ||
e4837f8f ID |
2788 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
2789 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
2790 | mixed = 1; | |
2791 | ||
f43ffb60 ID |
2792 | /* |
2793 | * In case of mixed groups both data and meta should be picked, | |
2794 | * and identical options should be given for both of them. | |
2795 | */ | |
e4837f8f ID |
2796 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
2797 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
2798 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
2799 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
2800 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
2801 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
2802 | "metadata balance options must be the same\n"); | |
2803 | ret = -EINVAL; | |
2804 | goto out; | |
2805 | } | |
2806 | } | |
2807 | ||
e4d8ec0f ID |
2808 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
2809 | if (fs_info->fs_devices->num_devices == 1) | |
2810 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
2811 | else if (fs_info->fs_devices->num_devices < 4) | |
2812 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
2813 | else | |
2814 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
2815 | BTRFS_BLOCK_GROUP_RAID10); | |
2816 | ||
6728b198 ID |
2817 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2818 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
2819 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
2820 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2821 | "data profile %llu\n", | |
2822 | (unsigned long long)bctl->data.target); | |
2823 | ret = -EINVAL; | |
2824 | goto out; | |
2825 | } | |
6728b198 ID |
2826 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2827 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
2828 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
2829 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2830 | "metadata profile %llu\n", | |
2831 | (unsigned long long)bctl->meta.target); | |
2832 | ret = -EINVAL; | |
2833 | goto out; | |
2834 | } | |
6728b198 ID |
2835 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2836 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
2837 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
2838 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2839 | "system profile %llu\n", | |
2840 | (unsigned long long)bctl->sys.target); | |
2841 | ret = -EINVAL; | |
2842 | goto out; | |
2843 | } | |
2844 | ||
e4837f8f ID |
2845 | /* allow dup'ed data chunks only in mixed mode */ |
2846 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 2847 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
2848 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
2849 | ret = -EINVAL; | |
2850 | goto out; | |
2851 | } | |
2852 | ||
2853 | /* allow to reduce meta or sys integrity only if force set */ | |
2854 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2855 | BTRFS_BLOCK_GROUP_RAID10; | |
2856 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2857 | (fs_info->avail_system_alloc_bits & allowed) && | |
2858 | !(bctl->sys.target & allowed)) || | |
2859 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2860 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
2861 | !(bctl->meta.target & allowed))) { | |
2862 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
2863 | printk(KERN_INFO "btrfs: force reducing metadata " | |
2864 | "integrity\n"); | |
2865 | } else { | |
2866 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
2867 | "integrity, use force if you want this\n"); | |
2868 | ret = -EINVAL; | |
2869 | goto out; | |
2870 | } | |
2871 | } | |
2872 | ||
5af3e8cc SB |
2873 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
2874 | int num_tolerated_disk_barrier_failures; | |
2875 | u64 target = bctl->sys.target; | |
2876 | ||
2877 | num_tolerated_disk_barrier_failures = | |
2878 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
2879 | if (num_tolerated_disk_barrier_failures > 0 && | |
2880 | (target & | |
2881 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
2882 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
2883 | num_tolerated_disk_barrier_failures = 0; | |
2884 | else if (num_tolerated_disk_barrier_failures > 1 && | |
2885 | (target & | |
2886 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
2887 | num_tolerated_disk_barrier_failures = 1; | |
2888 | ||
2889 | fs_info->num_tolerated_disk_barrier_failures = | |
2890 | num_tolerated_disk_barrier_failures; | |
2891 | } | |
2892 | ||
0940ebf6 | 2893 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 2894 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
2895 | goto out; |
2896 | ||
59641015 ID |
2897 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
2898 | BUG_ON(ret == -EEXIST); | |
2899 | set_balance_control(bctl); | |
2900 | } else { | |
2901 | BUG_ON(ret != -EEXIST); | |
2902 | spin_lock(&fs_info->balance_lock); | |
2903 | update_balance_args(bctl); | |
2904 | spin_unlock(&fs_info->balance_lock); | |
2905 | } | |
c9e9f97b | 2906 | |
837d5b6e | 2907 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
2908 | mutex_unlock(&fs_info->balance_mutex); |
2909 | ||
2910 | ret = __btrfs_balance(fs_info); | |
2911 | ||
2912 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 2913 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
2914 | |
2915 | if (bargs) { | |
2916 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 2917 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
2918 | } |
2919 | ||
837d5b6e ID |
2920 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
2921 | balance_need_close(fs_info)) { | |
2922 | __cancel_balance(fs_info); | |
2923 | } | |
2924 | ||
5af3e8cc SB |
2925 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
2926 | fs_info->num_tolerated_disk_barrier_failures = | |
2927 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
2928 | } | |
2929 | ||
837d5b6e | 2930 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
2931 | |
2932 | return ret; | |
2933 | out: | |
59641015 ID |
2934 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
2935 | __cancel_balance(fs_info); | |
2936 | else | |
2937 | kfree(bctl); | |
2938 | return ret; | |
2939 | } | |
2940 | ||
2941 | static int balance_kthread(void *data) | |
2942 | { | |
2b6ba629 | 2943 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 2944 | int ret = 0; |
59641015 ID |
2945 | |
2946 | mutex_lock(&fs_info->volume_mutex); | |
2947 | mutex_lock(&fs_info->balance_mutex); | |
2948 | ||
2b6ba629 | 2949 | if (fs_info->balance_ctl) { |
9555c6c1 | 2950 | printk(KERN_INFO "btrfs: continuing balance\n"); |
2b6ba629 | 2951 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 2952 | } |
59641015 ID |
2953 | |
2954 | mutex_unlock(&fs_info->balance_mutex); | |
2955 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 2956 | |
59641015 ID |
2957 | return ret; |
2958 | } | |
2959 | ||
2b6ba629 ID |
2960 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
2961 | { | |
2962 | struct task_struct *tsk; | |
2963 | ||
2964 | spin_lock(&fs_info->balance_lock); | |
2965 | if (!fs_info->balance_ctl) { | |
2966 | spin_unlock(&fs_info->balance_lock); | |
2967 | return 0; | |
2968 | } | |
2969 | spin_unlock(&fs_info->balance_lock); | |
2970 | ||
2971 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
2972 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
2973 | return 0; | |
2974 | } | |
2975 | ||
2976 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
2977 | if (IS_ERR(tsk)) | |
2978 | return PTR_ERR(tsk); | |
2979 | ||
2980 | return 0; | |
2981 | } | |
2982 | ||
68310a5e | 2983 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 2984 | { |
59641015 ID |
2985 | struct btrfs_balance_control *bctl; |
2986 | struct btrfs_balance_item *item; | |
2987 | struct btrfs_disk_balance_args disk_bargs; | |
2988 | struct btrfs_path *path; | |
2989 | struct extent_buffer *leaf; | |
2990 | struct btrfs_key key; | |
2991 | int ret; | |
2992 | ||
2993 | path = btrfs_alloc_path(); | |
2994 | if (!path) | |
2995 | return -ENOMEM; | |
2996 | ||
59641015 ID |
2997 | key.objectid = BTRFS_BALANCE_OBJECTID; |
2998 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2999 | key.offset = 0; | |
3000 | ||
68310a5e | 3001 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3002 | if (ret < 0) |
68310a5e | 3003 | goto out; |
59641015 ID |
3004 | if (ret > 0) { /* ret = -ENOENT; */ |
3005 | ret = 0; | |
68310a5e ID |
3006 | goto out; |
3007 | } | |
3008 | ||
3009 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3010 | if (!bctl) { | |
3011 | ret = -ENOMEM; | |
3012 | goto out; | |
59641015 ID |
3013 | } |
3014 | ||
3015 | leaf = path->nodes[0]; | |
3016 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3017 | ||
68310a5e ID |
3018 | bctl->fs_info = fs_info; |
3019 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3020 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3021 | |
3022 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3023 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3024 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3025 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3026 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3027 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3028 | ||
68310a5e ID |
3029 | mutex_lock(&fs_info->volume_mutex); |
3030 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3031 | |
68310a5e ID |
3032 | set_balance_control(bctl); |
3033 | ||
3034 | mutex_unlock(&fs_info->balance_mutex); | |
3035 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3036 | out: |
3037 | btrfs_free_path(path); | |
ec44a35c CM |
3038 | return ret; |
3039 | } | |
3040 | ||
837d5b6e ID |
3041 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3042 | { | |
3043 | int ret = 0; | |
3044 | ||
3045 | mutex_lock(&fs_info->balance_mutex); | |
3046 | if (!fs_info->balance_ctl) { | |
3047 | mutex_unlock(&fs_info->balance_mutex); | |
3048 | return -ENOTCONN; | |
3049 | } | |
3050 | ||
3051 | if (atomic_read(&fs_info->balance_running)) { | |
3052 | atomic_inc(&fs_info->balance_pause_req); | |
3053 | mutex_unlock(&fs_info->balance_mutex); | |
3054 | ||
3055 | wait_event(fs_info->balance_wait_q, | |
3056 | atomic_read(&fs_info->balance_running) == 0); | |
3057 | ||
3058 | mutex_lock(&fs_info->balance_mutex); | |
3059 | /* we are good with balance_ctl ripped off from under us */ | |
3060 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3061 | atomic_dec(&fs_info->balance_pause_req); | |
3062 | } else { | |
3063 | ret = -ENOTCONN; | |
3064 | } | |
3065 | ||
3066 | mutex_unlock(&fs_info->balance_mutex); | |
3067 | return ret; | |
3068 | } | |
3069 | ||
a7e99c69 ID |
3070 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3071 | { | |
3072 | mutex_lock(&fs_info->balance_mutex); | |
3073 | if (!fs_info->balance_ctl) { | |
3074 | mutex_unlock(&fs_info->balance_mutex); | |
3075 | return -ENOTCONN; | |
3076 | } | |
3077 | ||
3078 | atomic_inc(&fs_info->balance_cancel_req); | |
3079 | /* | |
3080 | * if we are running just wait and return, balance item is | |
3081 | * deleted in btrfs_balance in this case | |
3082 | */ | |
3083 | if (atomic_read(&fs_info->balance_running)) { | |
3084 | mutex_unlock(&fs_info->balance_mutex); | |
3085 | wait_event(fs_info->balance_wait_q, | |
3086 | atomic_read(&fs_info->balance_running) == 0); | |
3087 | mutex_lock(&fs_info->balance_mutex); | |
3088 | } else { | |
3089 | /* __cancel_balance needs volume_mutex */ | |
3090 | mutex_unlock(&fs_info->balance_mutex); | |
3091 | mutex_lock(&fs_info->volume_mutex); | |
3092 | mutex_lock(&fs_info->balance_mutex); | |
3093 | ||
3094 | if (fs_info->balance_ctl) | |
3095 | __cancel_balance(fs_info); | |
3096 | ||
3097 | mutex_unlock(&fs_info->volume_mutex); | |
3098 | } | |
3099 | ||
3100 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3101 | atomic_dec(&fs_info->balance_cancel_req); | |
3102 | mutex_unlock(&fs_info->balance_mutex); | |
3103 | return 0; | |
3104 | } | |
3105 | ||
8f18cf13 CM |
3106 | /* |
3107 | * shrinking a device means finding all of the device extents past | |
3108 | * the new size, and then following the back refs to the chunks. | |
3109 | * The chunk relocation code actually frees the device extent | |
3110 | */ | |
3111 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
3112 | { | |
3113 | struct btrfs_trans_handle *trans; | |
3114 | struct btrfs_root *root = device->dev_root; | |
3115 | struct btrfs_dev_extent *dev_extent = NULL; | |
3116 | struct btrfs_path *path; | |
3117 | u64 length; | |
3118 | u64 chunk_tree; | |
3119 | u64 chunk_objectid; | |
3120 | u64 chunk_offset; | |
3121 | int ret; | |
3122 | int slot; | |
ba1bf481 JB |
3123 | int failed = 0; |
3124 | bool retried = false; | |
8f18cf13 CM |
3125 | struct extent_buffer *l; |
3126 | struct btrfs_key key; | |
6c41761f | 3127 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 3128 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 3129 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
3130 | u64 diff = device->total_bytes - new_size; |
3131 | ||
8f18cf13 CM |
3132 | path = btrfs_alloc_path(); |
3133 | if (!path) | |
3134 | return -ENOMEM; | |
3135 | ||
8f18cf13 CM |
3136 | path->reada = 2; |
3137 | ||
7d9eb12c CM |
3138 | lock_chunks(root); |
3139 | ||
8f18cf13 | 3140 | device->total_bytes = new_size; |
2bf64758 | 3141 | if (device->writeable) { |
2b82032c | 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); | |
3146 | } | |
7d9eb12c | 3147 | unlock_chunks(root); |
8f18cf13 | 3148 | |
ba1bf481 | 3149 | again: |
8f18cf13 CM |
3150 | key.objectid = device->devid; |
3151 | key.offset = (u64)-1; | |
3152 | key.type = BTRFS_DEV_EXTENT_KEY; | |
3153 | ||
213e64da | 3154 | do { |
8f18cf13 CM |
3155 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
3156 | if (ret < 0) | |
3157 | goto done; | |
3158 | ||
3159 | ret = btrfs_previous_item(root, path, 0, key.type); | |
3160 | if (ret < 0) | |
3161 | goto done; | |
3162 | if (ret) { | |
3163 | ret = 0; | |
b3b4aa74 | 3164 | btrfs_release_path(path); |
bf1fb512 | 3165 | break; |
8f18cf13 CM |
3166 | } |
3167 | ||
3168 | l = path->nodes[0]; | |
3169 | slot = path->slots[0]; | |
3170 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3171 | ||
ba1bf481 | 3172 | if (key.objectid != device->devid) { |
b3b4aa74 | 3173 | btrfs_release_path(path); |
bf1fb512 | 3174 | break; |
ba1bf481 | 3175 | } |
8f18cf13 CM |
3176 | |
3177 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3178 | length = btrfs_dev_extent_length(l, dev_extent); | |
3179 | ||
ba1bf481 | 3180 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3181 | btrfs_release_path(path); |
d6397bae | 3182 | break; |
ba1bf481 | 3183 | } |
8f18cf13 CM |
3184 | |
3185 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3186 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3187 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3188 | btrfs_release_path(path); |
8f18cf13 CM |
3189 | |
3190 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3191 | chunk_offset); | |
ba1bf481 | 3192 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3193 | goto done; |
ba1bf481 JB |
3194 | if (ret == -ENOSPC) |
3195 | failed++; | |
213e64da | 3196 | } while (key.offset-- > 0); |
ba1bf481 JB |
3197 | |
3198 | if (failed && !retried) { | |
3199 | failed = 0; | |
3200 | retried = true; | |
3201 | goto again; | |
3202 | } else if (failed && retried) { | |
3203 | ret = -ENOSPC; | |
3204 | lock_chunks(root); | |
3205 | ||
3206 | device->total_bytes = old_size; | |
3207 | if (device->writeable) | |
3208 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3209 | spin_lock(&root->fs_info->free_chunk_lock); |
3210 | root->fs_info->free_chunk_space += diff; | |
3211 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3212 | unlock_chunks(root); |
3213 | goto done; | |
8f18cf13 CM |
3214 | } |
3215 | ||
d6397bae | 3216 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3217 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3218 | if (IS_ERR(trans)) { |
3219 | ret = PTR_ERR(trans); | |
3220 | goto done; | |
3221 | } | |
3222 | ||
d6397bae CB |
3223 | lock_chunks(root); |
3224 | ||
3225 | device->disk_total_bytes = new_size; | |
3226 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3227 | ret = btrfs_update_device(trans, device); | |
3228 | if (ret) { | |
3229 | unlock_chunks(root); | |
3230 | btrfs_end_transaction(trans, root); | |
3231 | goto done; | |
3232 | } | |
3233 | WARN_ON(diff > old_total); | |
3234 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3235 | unlock_chunks(root); | |
3236 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3237 | done: |
3238 | btrfs_free_path(path); | |
3239 | return ret; | |
3240 | } | |
3241 | ||
125ccb0a | 3242 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3243 | struct btrfs_key *key, |
3244 | struct btrfs_chunk *chunk, int item_size) | |
3245 | { | |
6c41761f | 3246 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3247 | struct btrfs_disk_key disk_key; |
3248 | u32 array_size; | |
3249 | u8 *ptr; | |
3250 | ||
3251 | array_size = btrfs_super_sys_array_size(super_copy); | |
3252 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3253 | return -EFBIG; | |
3254 | ||
3255 | ptr = super_copy->sys_chunk_array + array_size; | |
3256 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3257 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3258 | ptr += sizeof(disk_key); | |
3259 | memcpy(ptr, chunk, item_size); | |
3260 | item_size += sizeof(disk_key); | |
3261 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3262 | return 0; | |
3263 | } | |
3264 | ||
73c5de00 AJ |
3265 | /* |
3266 | * sort the devices in descending order by max_avail, total_avail | |
3267 | */ | |
3268 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3269 | { |
73c5de00 AJ |
3270 | const struct btrfs_device_info *di_a = a; |
3271 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3272 | |
73c5de00 | 3273 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3274 | return -1; |
73c5de00 | 3275 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3276 | return 1; |
73c5de00 AJ |
3277 | if (di_a->total_avail > di_b->total_avail) |
3278 | return -1; | |
3279 | if (di_a->total_avail < di_b->total_avail) | |
3280 | return 1; | |
3281 | return 0; | |
b2117a39 | 3282 | } |
0b86a832 | 3283 | |
73c5de00 AJ |
3284 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3285 | struct btrfs_root *extent_root, | |
3286 | struct map_lookup **map_ret, | |
3287 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3288 | u64 start, u64 type) | |
b2117a39 | 3289 | { |
73c5de00 AJ |
3290 | struct btrfs_fs_info *info = extent_root->fs_info; |
3291 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3292 | struct list_head *cur; | |
3293 | struct map_lookup *map = NULL; | |
3294 | struct extent_map_tree *em_tree; | |
3295 | struct extent_map *em; | |
3296 | struct btrfs_device_info *devices_info = NULL; | |
3297 | u64 total_avail; | |
3298 | int num_stripes; /* total number of stripes to allocate */ | |
3299 | int sub_stripes; /* sub_stripes info for map */ | |
3300 | int dev_stripes; /* stripes per dev */ | |
3301 | int devs_max; /* max devs to use */ | |
3302 | int devs_min; /* min devs needed */ | |
3303 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3304 | int ncopies; /* how many copies to data has */ | |
3305 | int ret; | |
3306 | u64 max_stripe_size; | |
3307 | u64 max_chunk_size; | |
3308 | u64 stripe_size; | |
3309 | u64 num_bytes; | |
3310 | int ndevs; | |
3311 | int i; | |
3312 | int j; | |
593060d7 | 3313 | |
0c460c0d | 3314 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3315 | |
73c5de00 AJ |
3316 | if (list_empty(&fs_devices->alloc_list)) |
3317 | return -ENOSPC; | |
b2117a39 | 3318 | |
73c5de00 AJ |
3319 | sub_stripes = 1; |
3320 | dev_stripes = 1; | |
3321 | devs_increment = 1; | |
3322 | ncopies = 1; | |
3323 | devs_max = 0; /* 0 == as many as possible */ | |
3324 | devs_min = 1; | |
3325 | ||
3326 | /* | |
3327 | * define the properties of each RAID type. | |
3328 | * FIXME: move this to a global table and use it in all RAID | |
3329 | * calculation code | |
3330 | */ | |
3331 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
3332 | dev_stripes = 2; | |
b2117a39 | 3333 | ncopies = 2; |
73c5de00 AJ |
3334 | devs_max = 1; |
3335 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
3336 | devs_min = 2; | |
3337 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
3338 | devs_increment = 2; | |
b2117a39 | 3339 | ncopies = 2; |
73c5de00 AJ |
3340 | devs_max = 2; |
3341 | devs_min = 2; | |
3342 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
3343 | sub_stripes = 2; | |
3344 | devs_increment = 2; | |
3345 | ncopies = 2; | |
3346 | devs_min = 4; | |
3347 | } else { | |
3348 | devs_max = 1; | |
3349 | } | |
b2117a39 | 3350 | |
9b3f68b9 | 3351 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3352 | max_stripe_size = 1024 * 1024 * 1024; |
3353 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3354 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3355 | /* for larger filesystems, use larger metadata chunks */ |
3356 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3357 | max_stripe_size = 1024 * 1024 * 1024; | |
3358 | else | |
3359 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3360 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3361 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3362 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3363 | max_chunk_size = 2 * max_stripe_size; |
3364 | } else { | |
3365 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3366 | type); | |
3367 | BUG_ON(1); | |
9b3f68b9 CM |
3368 | } |
3369 | ||
2b82032c YZ |
3370 | /* we don't want a chunk larger than 10% of writeable space */ |
3371 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3372 | max_chunk_size); | |
9b3f68b9 | 3373 | |
73c5de00 AJ |
3374 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3375 | GFP_NOFS); | |
3376 | if (!devices_info) | |
3377 | return -ENOMEM; | |
0cad8a11 | 3378 | |
73c5de00 | 3379 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3380 | |
9f680ce0 | 3381 | /* |
73c5de00 AJ |
3382 | * in the first pass through the devices list, we gather information |
3383 | * about the available holes on each device. | |
9f680ce0 | 3384 | */ |
73c5de00 AJ |
3385 | ndevs = 0; |
3386 | while (cur != &fs_devices->alloc_list) { | |
3387 | struct btrfs_device *device; | |
3388 | u64 max_avail; | |
3389 | u64 dev_offset; | |
b2117a39 | 3390 | |
73c5de00 | 3391 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3392 | |
73c5de00 | 3393 | cur = cur->next; |
b2117a39 | 3394 | |
73c5de00 | 3395 | if (!device->writeable) { |
31b1a2bd | 3396 | WARN(1, KERN_ERR |
73c5de00 | 3397 | "btrfs: read-only device in alloc_list\n"); |
73c5de00 AJ |
3398 | continue; |
3399 | } | |
b2117a39 | 3400 | |
73c5de00 AJ |
3401 | if (!device->in_fs_metadata) |
3402 | continue; | |
b2117a39 | 3403 | |
73c5de00 AJ |
3404 | if (device->total_bytes > device->bytes_used) |
3405 | total_avail = device->total_bytes - device->bytes_used; | |
3406 | else | |
3407 | total_avail = 0; | |
38c01b96 | 3408 | |
3409 | /* If there is no space on this device, skip it. */ | |
3410 | if (total_avail == 0) | |
3411 | continue; | |
b2117a39 | 3412 | |
125ccb0a | 3413 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3414 | max_stripe_size * dev_stripes, |
3415 | &dev_offset, &max_avail); | |
3416 | if (ret && ret != -ENOSPC) | |
3417 | goto error; | |
b2117a39 | 3418 | |
73c5de00 AJ |
3419 | if (ret == 0) |
3420 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3421 | |
73c5de00 AJ |
3422 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3423 | continue; | |
b2117a39 | 3424 | |
73c5de00 AJ |
3425 | devices_info[ndevs].dev_offset = dev_offset; |
3426 | devices_info[ndevs].max_avail = max_avail; | |
3427 | devices_info[ndevs].total_avail = total_avail; | |
3428 | devices_info[ndevs].dev = device; | |
3429 | ++ndevs; | |
3430 | } | |
b2117a39 | 3431 | |
73c5de00 AJ |
3432 | /* |
3433 | * now sort the devices by hole size / available space | |
3434 | */ | |
3435 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3436 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3437 | |
73c5de00 AJ |
3438 | /* round down to number of usable stripes */ |
3439 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3440 | |
73c5de00 AJ |
3441 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3442 | ret = -ENOSPC; | |
3443 | goto error; | |
b2117a39 | 3444 | } |
9f680ce0 | 3445 | |
73c5de00 AJ |
3446 | if (devs_max && ndevs > devs_max) |
3447 | ndevs = devs_max; | |
3448 | /* | |
3449 | * the primary goal is to maximize the number of stripes, so use as many | |
3450 | * devices as possible, even if the stripes are not maximum sized. | |
3451 | */ | |
3452 | stripe_size = devices_info[ndevs-1].max_avail; | |
3453 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3454 | |
37db63a4 | 3455 | if (stripe_size * ndevs > max_chunk_size * ncopies) { |
73c5de00 | 3456 | stripe_size = max_chunk_size * ncopies; |
37db63a4 | 3457 | do_div(stripe_size, ndevs); |
b2117a39 | 3458 | } |
b2117a39 | 3459 | |
73c5de00 | 3460 | do_div(stripe_size, dev_stripes); |
37db63a4 ID |
3461 | |
3462 | /* align to BTRFS_STRIPE_LEN */ | |
73c5de00 AJ |
3463 | do_div(stripe_size, BTRFS_STRIPE_LEN); |
3464 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
3465 | |
3466 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3467 | if (!map) { | |
3468 | ret = -ENOMEM; | |
3469 | goto error; | |
3470 | } | |
3471 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3472 | |
73c5de00 AJ |
3473 | for (i = 0; i < ndevs; ++i) { |
3474 | for (j = 0; j < dev_stripes; ++j) { | |
3475 | int s = i * dev_stripes + j; | |
3476 | map->stripes[s].dev = devices_info[i].dev; | |
3477 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3478 | j * stripe_size; | |
6324fbf3 | 3479 | } |
6324fbf3 | 3480 | } |
2b82032c | 3481 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
3482 | map->stripe_len = BTRFS_STRIPE_LEN; |
3483 | map->io_align = BTRFS_STRIPE_LEN; | |
3484 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 3485 | map->type = type; |
2b82032c | 3486 | map->sub_stripes = sub_stripes; |
0b86a832 | 3487 | |
2b82032c | 3488 | *map_ret = map; |
73c5de00 | 3489 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 3490 | |
73c5de00 AJ |
3491 | *stripe_size_out = stripe_size; |
3492 | *num_bytes_out = num_bytes; | |
0b86a832 | 3493 | |
73c5de00 | 3494 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3495 | |
172ddd60 | 3496 | em = alloc_extent_map(); |
2b82032c | 3497 | if (!em) { |
b2117a39 MX |
3498 | ret = -ENOMEM; |
3499 | goto error; | |
593060d7 | 3500 | } |
2b82032c YZ |
3501 | em->bdev = (struct block_device *)map; |
3502 | em->start = start; | |
73c5de00 | 3503 | em->len = num_bytes; |
2b82032c YZ |
3504 | em->block_start = 0; |
3505 | em->block_len = em->len; | |
593060d7 | 3506 | |
2b82032c | 3507 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3508 | write_lock(&em_tree->lock); |
2b82032c | 3509 | ret = add_extent_mapping(em_tree, em); |
890871be | 3510 | write_unlock(&em_tree->lock); |
2b82032c | 3511 | free_extent_map(em); |
1dd4602f MF |
3512 | if (ret) |
3513 | goto error; | |
0b86a832 | 3514 | |
2b82032c YZ |
3515 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
3516 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3517 | start, num_bytes); |
79787eaa JM |
3518 | if (ret) |
3519 | goto error; | |
611f0e00 | 3520 | |
73c5de00 AJ |
3521 | for (i = 0; i < map->num_stripes; ++i) { |
3522 | struct btrfs_device *device; | |
3523 | u64 dev_offset; | |
3524 | ||
3525 | device = map->stripes[i].dev; | |
3526 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3527 | |
3528 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3529 | info->chunk_root->root_key.objectid, |
3530 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3531 | start, dev_offset, stripe_size); |
79787eaa JM |
3532 | if (ret) { |
3533 | btrfs_abort_transaction(trans, extent_root, ret); | |
3534 | goto error; | |
3535 | } | |
2b82032c YZ |
3536 | } |
3537 | ||
b2117a39 | 3538 | kfree(devices_info); |
2b82032c | 3539 | return 0; |
b2117a39 MX |
3540 | |
3541 | error: | |
3542 | kfree(map); | |
3543 | kfree(devices_info); | |
3544 | return ret; | |
2b82032c YZ |
3545 | } |
3546 | ||
3547 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3548 | struct btrfs_root *extent_root, | |
3549 | struct map_lookup *map, u64 chunk_offset, | |
3550 | u64 chunk_size, u64 stripe_size) | |
3551 | { | |
3552 | u64 dev_offset; | |
3553 | struct btrfs_key key; | |
3554 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3555 | struct btrfs_device *device; | |
3556 | struct btrfs_chunk *chunk; | |
3557 | struct btrfs_stripe *stripe; | |
3558 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
3559 | int index = 0; | |
3560 | int ret; | |
3561 | ||
3562 | chunk = kzalloc(item_size, GFP_NOFS); | |
3563 | if (!chunk) | |
3564 | return -ENOMEM; | |
3565 | ||
3566 | index = 0; | |
3567 | while (index < map->num_stripes) { | |
3568 | device = map->stripes[index].dev; | |
3569 | device->bytes_used += stripe_size; | |
0b86a832 | 3570 | ret = btrfs_update_device(trans, device); |
3acd3953 MF |
3571 | if (ret) |
3572 | goto out_free; | |
2b82032c YZ |
3573 | index++; |
3574 | } | |
3575 | ||
2bf64758 JB |
3576 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
3577 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
3578 | map->num_stripes); | |
3579 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
3580 | ||
2b82032c YZ |
3581 | index = 0; |
3582 | stripe = &chunk->stripe; | |
3583 | while (index < map->num_stripes) { | |
3584 | device = map->stripes[index].dev; | |
3585 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 3586 | |
e17cade2 CM |
3587 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
3588 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
3589 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 3590 | stripe++; |
0b86a832 CM |
3591 | index++; |
3592 | } | |
3593 | ||
2b82032c | 3594 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 3595 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
3596 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
3597 | btrfs_set_stack_chunk_type(chunk, map->type); | |
3598 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
3599 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
3600 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 3601 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 3602 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 3603 | |
2b82032c YZ |
3604 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3605 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3606 | key.offset = chunk_offset; | |
0b86a832 | 3607 | |
2b82032c | 3608 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
0b86a832 | 3609 | |
4ed1d16e MF |
3610 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
3611 | /* | |
3612 | * TODO: Cleanup of inserted chunk root in case of | |
3613 | * failure. | |
3614 | */ | |
125ccb0a | 3615 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 3616 | item_size); |
8f18cf13 | 3617 | } |
1abe9b8a | 3618 | |
3acd3953 | 3619 | out_free: |
0b86a832 | 3620 | kfree(chunk); |
4ed1d16e | 3621 | return ret; |
2b82032c | 3622 | } |
0b86a832 | 3623 | |
2b82032c YZ |
3624 | /* |
3625 | * Chunk allocation falls into two parts. The first part does works | |
3626 | * that make the new allocated chunk useable, but not do any operation | |
3627 | * that modifies the chunk tree. The second part does the works that | |
3628 | * require modifying the chunk tree. This division is important for the | |
3629 | * bootstrap process of adding storage to a seed btrfs. | |
3630 | */ | |
3631 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
3632 | struct btrfs_root *extent_root, u64 type) | |
3633 | { | |
3634 | u64 chunk_offset; | |
3635 | u64 chunk_size; | |
3636 | u64 stripe_size; | |
3637 | struct map_lookup *map; | |
3638 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3639 | int ret; | |
3640 | ||
3641 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3642 | &chunk_offset); | |
3643 | if (ret) | |
3644 | return ret; | |
3645 | ||
3646 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3647 | &stripe_size, chunk_offset, type); | |
3648 | if (ret) | |
3649 | return ret; | |
3650 | ||
3651 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3652 | chunk_size, stripe_size); | |
79787eaa JM |
3653 | if (ret) |
3654 | return ret; | |
2b82032c YZ |
3655 | return 0; |
3656 | } | |
3657 | ||
d397712b | 3658 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
3659 | struct btrfs_root *root, |
3660 | struct btrfs_device *device) | |
3661 | { | |
3662 | u64 chunk_offset; | |
3663 | u64 sys_chunk_offset; | |
3664 | u64 chunk_size; | |
3665 | u64 sys_chunk_size; | |
3666 | u64 stripe_size; | |
3667 | u64 sys_stripe_size; | |
3668 | u64 alloc_profile; | |
3669 | struct map_lookup *map; | |
3670 | struct map_lookup *sys_map; | |
3671 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3672 | struct btrfs_root *extent_root = fs_info->extent_root; | |
3673 | int ret; | |
3674 | ||
3675 | ret = find_next_chunk(fs_info->chunk_root, | |
3676 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
3677 | if (ret) |
3678 | return ret; | |
2b82032c YZ |
3679 | |
3680 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
6fef8df1 | 3681 | fs_info->avail_metadata_alloc_bits; |
2b82032c YZ |
3682 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3683 | ||
3684 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3685 | &stripe_size, chunk_offset, alloc_profile); | |
79787eaa JM |
3686 | if (ret) |
3687 | return ret; | |
2b82032c YZ |
3688 | |
3689 | sys_chunk_offset = chunk_offset + chunk_size; | |
3690 | ||
3691 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
6fef8df1 | 3692 | fs_info->avail_system_alloc_bits; |
2b82032c YZ |
3693 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3694 | ||
3695 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
3696 | &sys_chunk_size, &sys_stripe_size, | |
3697 | sys_chunk_offset, alloc_profile); | |
005d6427 DS |
3698 | if (ret) { |
3699 | btrfs_abort_transaction(trans, root, ret); | |
3700 | goto out; | |
3701 | } | |
2b82032c YZ |
3702 | |
3703 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
005d6427 DS |
3704 | if (ret) { |
3705 | btrfs_abort_transaction(trans, root, ret); | |
3706 | goto out; | |
3707 | } | |
2b82032c YZ |
3708 | |
3709 | /* | |
3710 | * Modifying chunk tree needs allocating new blocks from both | |
3711 | * system block group and metadata block group. So we only can | |
3712 | * do operations require modifying the chunk tree after both | |
3713 | * block groups were created. | |
3714 | */ | |
3715 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3716 | chunk_size, stripe_size); | |
005d6427 DS |
3717 | if (ret) { |
3718 | btrfs_abort_transaction(trans, root, ret); | |
3719 | goto out; | |
3720 | } | |
2b82032c YZ |
3721 | |
3722 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
3723 | sys_chunk_offset, sys_chunk_size, | |
3724 | sys_stripe_size); | |
79787eaa | 3725 | if (ret) |
005d6427 | 3726 | btrfs_abort_transaction(trans, root, ret); |
79787eaa | 3727 | |
005d6427 | 3728 | out: |
79787eaa | 3729 | |
79787eaa | 3730 | return ret; |
2b82032c YZ |
3731 | } |
3732 | ||
3733 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
3734 | { | |
3735 | struct extent_map *em; | |
3736 | struct map_lookup *map; | |
3737 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3738 | int readonly = 0; | |
3739 | int i; | |
3740 | ||
890871be | 3741 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 3742 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 3743 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
3744 | if (!em) |
3745 | return 1; | |
3746 | ||
f48b9075 JB |
3747 | if (btrfs_test_opt(root, DEGRADED)) { |
3748 | free_extent_map(em); | |
3749 | return 0; | |
3750 | } | |
3751 | ||
2b82032c YZ |
3752 | map = (struct map_lookup *)em->bdev; |
3753 | for (i = 0; i < map->num_stripes; i++) { | |
3754 | if (!map->stripes[i].dev->writeable) { | |
3755 | readonly = 1; | |
3756 | break; | |
3757 | } | |
3758 | } | |
0b86a832 | 3759 | free_extent_map(em); |
2b82032c | 3760 | return readonly; |
0b86a832 CM |
3761 | } |
3762 | ||
3763 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
3764 | { | |
a8067e02 | 3765 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
3766 | } |
3767 | ||
3768 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
3769 | { | |
3770 | struct extent_map *em; | |
3771 | ||
d397712b | 3772 | while (1) { |
890871be | 3773 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
3774 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
3775 | if (em) | |
3776 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 3777 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
3778 | if (!em) |
3779 | break; | |
3780 | kfree(em->bdev); | |
3781 | /* once for us */ | |
3782 | free_extent_map(em); | |
3783 | /* once for the tree */ | |
3784 | free_extent_map(em); | |
3785 | } | |
3786 | } | |
3787 | ||
f188591e CM |
3788 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
3789 | { | |
3790 | struct extent_map *em; | |
3791 | struct map_lookup *map; | |
3792 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3793 | int ret; | |
3794 | ||
890871be | 3795 | read_lock(&em_tree->lock); |
f188591e | 3796 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 3797 | read_unlock(&em_tree->lock); |
f188591e CM |
3798 | BUG_ON(!em); |
3799 | ||
3800 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3801 | map = (struct map_lookup *)em->bdev; | |
3802 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
3803 | ret = map->num_stripes; | |
321aecc6 CM |
3804 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
3805 | ret = map->sub_stripes; | |
f188591e CM |
3806 | else |
3807 | ret = 1; | |
3808 | free_extent_map(em); | |
f188591e CM |
3809 | return ret; |
3810 | } | |
3811 | ||
dfe25020 CM |
3812 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
3813 | int optimal) | |
3814 | { | |
3815 | int i; | |
3816 | if (map->stripes[optimal].dev->bdev) | |
3817 | return optimal; | |
3818 | for (i = first; i < first + num; i++) { | |
3819 | if (map->stripes[i].dev->bdev) | |
3820 | return i; | |
3821 | } | |
3822 | /* we couldn't find one that doesn't fail. Just return something | |
3823 | * and the io error handling code will clean up eventually | |
3824 | */ | |
3825 | return optimal; | |
3826 | } | |
3827 | ||
f2d8d74d CM |
3828 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3829 | u64 logical, u64 *length, | |
a1d3c478 | 3830 | struct btrfs_bio **bbio_ret, |
7eaceacc | 3831 | int mirror_num) |
0b86a832 CM |
3832 | { |
3833 | struct extent_map *em; | |
3834 | struct map_lookup *map; | |
3835 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3836 | u64 offset; | |
593060d7 | 3837 | u64 stripe_offset; |
fce3bb9a | 3838 | u64 stripe_end_offset; |
593060d7 | 3839 | u64 stripe_nr; |
fce3bb9a LD |
3840 | u64 stripe_nr_orig; |
3841 | u64 stripe_nr_end; | |
593060d7 | 3842 | int stripe_index; |
cea9e445 | 3843 | int i; |
de11cc12 | 3844 | int ret = 0; |
f2d8d74d | 3845 | int num_stripes; |
a236aed1 | 3846 | int max_errors = 0; |
a1d3c478 | 3847 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3848 | |
890871be | 3849 | read_lock(&em_tree->lock); |
0b86a832 | 3850 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 3851 | read_unlock(&em_tree->lock); |
f2d8d74d | 3852 | |
3b951516 | 3853 | if (!em) { |
48940662 | 3854 | printk(KERN_CRIT "btrfs: unable to find logical %llu len %llu\n", |
d397712b CM |
3855 | (unsigned long long)logical, |
3856 | (unsigned long long)*length); | |
f2d8d74d | 3857 | BUG(); |
3b951516 | 3858 | } |
0b86a832 CM |
3859 | |
3860 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3861 | map = (struct map_lookup *)em->bdev; | |
3862 | offset = logical - em->start; | |
593060d7 | 3863 | |
f188591e CM |
3864 | if (mirror_num > map->num_stripes) |
3865 | mirror_num = 0; | |
3866 | ||
593060d7 CM |
3867 | stripe_nr = offset; |
3868 | /* | |
3869 | * stripe_nr counts the total number of stripes we have to stride | |
3870 | * to get to this block | |
3871 | */ | |
3872 | do_div(stripe_nr, map->stripe_len); | |
3873 | ||
3874 | stripe_offset = stripe_nr * map->stripe_len; | |
3875 | BUG_ON(offset < stripe_offset); | |
3876 | ||
3877 | /* stripe_offset is the offset of this block in its stripe*/ | |
3878 | stripe_offset = offset - stripe_offset; | |
3879 | ||
fce3bb9a LD |
3880 | if (rw & REQ_DISCARD) |
3881 | *length = min_t(u64, em->len - offset, *length); | |
52ba6929 | 3882 | else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
cea9e445 CM |
3883 | /* we limit the length of each bio to what fits in a stripe */ |
3884 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 3885 | map->stripe_len - stripe_offset); |
cea9e445 CM |
3886 | } else { |
3887 | *length = em->len - offset; | |
3888 | } | |
f2d8d74d | 3889 | |
a1d3c478 | 3890 | if (!bbio_ret) |
cea9e445 CM |
3891 | goto out; |
3892 | ||
f2d8d74d | 3893 | num_stripes = 1; |
cea9e445 | 3894 | stripe_index = 0; |
fce3bb9a LD |
3895 | stripe_nr_orig = stripe_nr; |
3896 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
3897 | (~(map->stripe_len - 1)); | |
3898 | do_div(stripe_nr_end, map->stripe_len); | |
3899 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
3900 | (offset + *length); | |
3901 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3902 | if (rw & REQ_DISCARD) | |
3903 | num_stripes = min_t(u64, map->num_stripes, | |
3904 | stripe_nr_end - stripe_nr_orig); | |
3905 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3906 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3907 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3908 | num_stripes = map->num_stripes; |
2fff734f | 3909 | else if (mirror_num) |
f188591e | 3910 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3911 | else { |
3912 | stripe_index = find_live_mirror(map, 0, | |
3913 | map->num_stripes, | |
3914 | current->pid % map->num_stripes); | |
a1d3c478 | 3915 | mirror_num = stripe_index + 1; |
dfe25020 | 3916 | } |
2fff734f | 3917 | |
611f0e00 | 3918 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3919 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3920 | num_stripes = map->num_stripes; |
a1d3c478 | 3921 | } else if (mirror_num) { |
f188591e | 3922 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3923 | } else { |
3924 | mirror_num = 1; | |
3925 | } | |
2fff734f | 3926 | |
321aecc6 CM |
3927 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3928 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3929 | |
3930 | stripe_index = do_div(stripe_nr, factor); | |
3931 | stripe_index *= map->sub_stripes; | |
3932 | ||
7eaceacc | 3933 | if (rw & REQ_WRITE) |
f2d8d74d | 3934 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3935 | else if (rw & REQ_DISCARD) |
3936 | num_stripes = min_t(u64, map->sub_stripes * | |
3937 | (stripe_nr_end - stripe_nr_orig), | |
3938 | map->num_stripes); | |
321aecc6 CM |
3939 | else if (mirror_num) |
3940 | stripe_index += mirror_num - 1; | |
dfe25020 | 3941 | else { |
3e74317a | 3942 | int old_stripe_index = stripe_index; |
dfe25020 CM |
3943 | stripe_index = find_live_mirror(map, stripe_index, |
3944 | map->sub_stripes, stripe_index + | |
3945 | current->pid % map->sub_stripes); | |
3e74317a | 3946 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 3947 | } |
8790d502 CM |
3948 | } else { |
3949 | /* | |
3950 | * after this do_div call, stripe_nr is the number of stripes | |
3951 | * on this device we have to walk to find the data, and | |
3952 | * stripe_index is the number of our device in the stripe array | |
3953 | */ | |
3954 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3955 | mirror_num = stripe_index + 1; |
8790d502 | 3956 | } |
593060d7 | 3957 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3958 | |
de11cc12 LZ |
3959 | bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS); |
3960 | if (!bbio) { | |
3961 | ret = -ENOMEM; | |
3962 | goto out; | |
3963 | } | |
3964 | atomic_set(&bbio->error, 0); | |
3965 | ||
fce3bb9a | 3966 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
3967 | int factor = 0; |
3968 | int sub_stripes = 0; | |
3969 | u64 stripes_per_dev = 0; | |
3970 | u32 remaining_stripes = 0; | |
b89203f7 | 3971 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
3972 | |
3973 | if (map->type & | |
3974 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3975 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3976 | sub_stripes = 1; | |
3977 | else | |
3978 | sub_stripes = map->sub_stripes; | |
3979 | ||
3980 | factor = map->num_stripes / sub_stripes; | |
3981 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
3982 | stripe_nr_orig, | |
3983 | factor, | |
3984 | &remaining_stripes); | |
b89203f7 LB |
3985 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
3986 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
3987 | } |
3988 | ||
fce3bb9a | 3989 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 3990 | bbio->stripes[i].physical = |
f2d8d74d CM |
3991 | map->stripes[stripe_index].physical + |
3992 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3993 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 3994 | |
ec9ef7a1 LZ |
3995 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
3996 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3997 | bbio->stripes[i].length = stripes_per_dev * | |
3998 | map->stripe_len; | |
b89203f7 | 3999 | |
ec9ef7a1 LZ |
4000 | if (i / sub_stripes < remaining_stripes) |
4001 | bbio->stripes[i].length += | |
4002 | map->stripe_len; | |
b89203f7 LB |
4003 | |
4004 | /* | |
4005 | * Special for the first stripe and | |
4006 | * the last stripe: | |
4007 | * | |
4008 | * |-------|...|-------| | |
4009 | * |----------| | |
4010 | * off end_off | |
4011 | */ | |
ec9ef7a1 | 4012 | if (i < sub_stripes) |
a1d3c478 | 4013 | bbio->stripes[i].length -= |
fce3bb9a | 4014 | stripe_offset; |
b89203f7 LB |
4015 | |
4016 | if (stripe_index >= last_stripe && | |
4017 | stripe_index <= (last_stripe + | |
4018 | sub_stripes - 1)) | |
a1d3c478 | 4019 | bbio->stripes[i].length -= |
fce3bb9a | 4020 | stripe_end_offset; |
b89203f7 | 4021 | |
ec9ef7a1 LZ |
4022 | if (i == sub_stripes - 1) |
4023 | stripe_offset = 0; | |
fce3bb9a | 4024 | } else |
a1d3c478 | 4025 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
4026 | |
4027 | stripe_index++; | |
4028 | if (stripe_index == map->num_stripes) { | |
4029 | /* This could only happen for RAID0/10 */ | |
4030 | stripe_index = 0; | |
4031 | stripe_nr++; | |
4032 | } | |
4033 | } | |
4034 | } else { | |
4035 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 4036 | bbio->stripes[i].physical = |
212a17ab LT |
4037 | map->stripes[stripe_index].physical + |
4038 | stripe_offset + | |
4039 | stripe_nr * map->stripe_len; | |
a1d3c478 | 4040 | bbio->stripes[i].dev = |
212a17ab | 4041 | map->stripes[stripe_index].dev; |
fce3bb9a | 4042 | stripe_index++; |
f2d8d74d | 4043 | } |
593060d7 | 4044 | } |
de11cc12 LZ |
4045 | |
4046 | if (rw & REQ_WRITE) { | |
4047 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4048 | BTRFS_BLOCK_GROUP_RAID10 | | |
4049 | BTRFS_BLOCK_GROUP_DUP)) { | |
4050 | max_errors = 1; | |
4051 | } | |
f2d8d74d | 4052 | } |
de11cc12 LZ |
4053 | |
4054 | *bbio_ret = bbio; | |
4055 | bbio->num_stripes = num_stripes; | |
4056 | bbio->max_errors = max_errors; | |
4057 | bbio->mirror_num = mirror_num; | |
cea9e445 | 4058 | out: |
0b86a832 | 4059 | free_extent_map(em); |
de11cc12 | 4060 | return ret; |
0b86a832 CM |
4061 | } |
4062 | ||
f2d8d74d CM |
4063 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
4064 | u64 logical, u64 *length, | |
a1d3c478 | 4065 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 4066 | { |
a1d3c478 | 4067 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 4068 | mirror_num); |
f2d8d74d CM |
4069 | } |
4070 | ||
a512bbf8 YZ |
4071 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
4072 | u64 chunk_start, u64 physical, u64 devid, | |
4073 | u64 **logical, int *naddrs, int *stripe_len) | |
4074 | { | |
4075 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4076 | struct extent_map *em; | |
4077 | struct map_lookup *map; | |
4078 | u64 *buf; | |
4079 | u64 bytenr; | |
4080 | u64 length; | |
4081 | u64 stripe_nr; | |
4082 | int i, j, nr = 0; | |
4083 | ||
890871be | 4084 | read_lock(&em_tree->lock); |
a512bbf8 | 4085 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 4086 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
4087 | |
4088 | BUG_ON(!em || em->start != chunk_start); | |
4089 | map = (struct map_lookup *)em->bdev; | |
4090 | ||
4091 | length = em->len; | |
4092 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
4093 | do_div(length, map->num_stripes / map->sub_stripes); | |
4094 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
4095 | do_div(length, map->num_stripes); | |
4096 | ||
4097 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
79787eaa | 4098 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
4099 | |
4100 | for (i = 0; i < map->num_stripes; i++) { | |
4101 | if (devid && map->stripes[i].dev->devid != devid) | |
4102 | continue; | |
4103 | if (map->stripes[i].physical > physical || | |
4104 | map->stripes[i].physical + length <= physical) | |
4105 | continue; | |
4106 | ||
4107 | stripe_nr = physical - map->stripes[i].physical; | |
4108 | do_div(stripe_nr, map->stripe_len); | |
4109 | ||
4110 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
4111 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4112 | do_div(stripe_nr, map->sub_stripes); | |
4113 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
4114 | stripe_nr = stripe_nr * map->num_stripes + i; | |
4115 | } | |
4116 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 4117 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
4118 | for (j = 0; j < nr; j++) { |
4119 | if (buf[j] == bytenr) | |
4120 | break; | |
4121 | } | |
934d375b CM |
4122 | if (j == nr) { |
4123 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 4124 | buf[nr++] = bytenr; |
934d375b | 4125 | } |
a512bbf8 YZ |
4126 | } |
4127 | ||
a512bbf8 YZ |
4128 | *logical = buf; |
4129 | *naddrs = nr; | |
4130 | *stripe_len = map->stripe_len; | |
4131 | ||
4132 | free_extent_map(em); | |
4133 | return 0; | |
f2d8d74d CM |
4134 | } |
4135 | ||
442a4f63 SB |
4136 | static void *merge_stripe_index_into_bio_private(void *bi_private, |
4137 | unsigned int stripe_index) | |
4138 | { | |
4139 | /* | |
4140 | * with single, dup, RAID0, RAID1 and RAID10, stripe_index is | |
4141 | * at most 1. | |
4142 | * The alternative solution (instead of stealing bits from the | |
4143 | * pointer) would be to allocate an intermediate structure | |
4144 | * that contains the old private pointer plus the stripe_index. | |
4145 | */ | |
4146 | BUG_ON((((uintptr_t)bi_private) & 3) != 0); | |
4147 | BUG_ON(stripe_index > 3); | |
4148 | return (void *)(((uintptr_t)bi_private) | stripe_index); | |
4149 | } | |
4150 | ||
4151 | static struct btrfs_bio *extract_bbio_from_bio_private(void *bi_private) | |
4152 | { | |
4153 | return (struct btrfs_bio *)(((uintptr_t)bi_private) & ~((uintptr_t)3)); | |
4154 | } | |
4155 | ||
4156 | static unsigned int extract_stripe_index_from_bio_private(void *bi_private) | |
4157 | { | |
4158 | return (unsigned int)((uintptr_t)bi_private) & 3; | |
4159 | } | |
4160 | ||
a1d3c478 | 4161 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 4162 | { |
442a4f63 | 4163 | struct btrfs_bio *bbio = extract_bbio_from_bio_private(bio->bi_private); |
7d2b4daa | 4164 | int is_orig_bio = 0; |
8790d502 | 4165 | |
442a4f63 | 4166 | if (err) { |
a1d3c478 | 4167 | atomic_inc(&bbio->error); |
442a4f63 SB |
4168 | if (err == -EIO || err == -EREMOTEIO) { |
4169 | unsigned int stripe_index = | |
4170 | extract_stripe_index_from_bio_private( | |
4171 | bio->bi_private); | |
4172 | struct btrfs_device *dev; | |
4173 | ||
4174 | BUG_ON(stripe_index >= bbio->num_stripes); | |
4175 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
4176 | if (dev->bdev) { |
4177 | if (bio->bi_rw & WRITE) | |
4178 | btrfs_dev_stat_inc(dev, | |
4179 | BTRFS_DEV_STAT_WRITE_ERRS); | |
4180 | else | |
4181 | btrfs_dev_stat_inc(dev, | |
4182 | BTRFS_DEV_STAT_READ_ERRS); | |
4183 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
4184 | btrfs_dev_stat_inc(dev, | |
4185 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
4186 | btrfs_dev_stat_print_on_error(dev); | |
4187 | } | |
442a4f63 SB |
4188 | } |
4189 | } | |
8790d502 | 4190 | |
a1d3c478 | 4191 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
4192 | is_orig_bio = 1; |
4193 | ||
a1d3c478 | 4194 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
4195 | if (!is_orig_bio) { |
4196 | bio_put(bio); | |
a1d3c478 | 4197 | bio = bbio->orig_bio; |
7d2b4daa | 4198 | } |
a1d3c478 JS |
4199 | bio->bi_private = bbio->private; |
4200 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
4201 | bio->bi_bdev = (struct block_device *) |
4202 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
4203 | /* only send an error to the higher layers if it is |
4204 | * beyond the tolerance of the multi-bio | |
4205 | */ | |
a1d3c478 | 4206 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 4207 | err = -EIO; |
5dbc8fca | 4208 | } else { |
1259ab75 CM |
4209 | /* |
4210 | * this bio is actually up to date, we didn't | |
4211 | * go over the max number of errors | |
4212 | */ | |
4213 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 4214 | err = 0; |
1259ab75 | 4215 | } |
a1d3c478 | 4216 | kfree(bbio); |
8790d502 CM |
4217 | |
4218 | bio_endio(bio, err); | |
7d2b4daa | 4219 | } else if (!is_orig_bio) { |
8790d502 CM |
4220 | bio_put(bio); |
4221 | } | |
8790d502 CM |
4222 | } |
4223 | ||
8b712842 CM |
4224 | struct async_sched { |
4225 | struct bio *bio; | |
4226 | int rw; | |
4227 | struct btrfs_fs_info *info; | |
4228 | struct btrfs_work work; | |
4229 | }; | |
4230 | ||
4231 | /* | |
4232 | * see run_scheduled_bios for a description of why bios are collected for | |
4233 | * async submit. | |
4234 | * | |
4235 | * This will add one bio to the pending list for a device and make sure | |
4236 | * the work struct is scheduled. | |
4237 | */ | |
143bede5 | 4238 | static noinline void schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
4239 | struct btrfs_device *device, |
4240 | int rw, struct bio *bio) | |
8b712842 CM |
4241 | { |
4242 | int should_queue = 1; | |
ffbd517d | 4243 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
4244 | |
4245 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 4246 | if (!(rw & REQ_WRITE)) { |
492bb6de | 4247 | bio_get(bio); |
21adbd5c | 4248 | btrfsic_submit_bio(rw, bio); |
492bb6de | 4249 | bio_put(bio); |
143bede5 | 4250 | return; |
8b712842 CM |
4251 | } |
4252 | ||
4253 | /* | |
0986fe9e | 4254 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
4255 | * higher layers. Otherwise, the async bio makes it appear we have |
4256 | * made progress against dirty pages when we've really just put it | |
4257 | * on a queue for later | |
4258 | */ | |
0986fe9e | 4259 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 4260 | WARN_ON(bio->bi_next); |
8b712842 CM |
4261 | bio->bi_next = NULL; |
4262 | bio->bi_rw |= rw; | |
4263 | ||
4264 | spin_lock(&device->io_lock); | |
7b6d91da | 4265 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
4266 | pending_bios = &device->pending_sync_bios; |
4267 | else | |
4268 | pending_bios = &device->pending_bios; | |
8b712842 | 4269 | |
ffbd517d CM |
4270 | if (pending_bios->tail) |
4271 | pending_bios->tail->bi_next = bio; | |
8b712842 | 4272 | |
ffbd517d CM |
4273 | pending_bios->tail = bio; |
4274 | if (!pending_bios->head) | |
4275 | pending_bios->head = bio; | |
8b712842 CM |
4276 | if (device->running_pending) |
4277 | should_queue = 0; | |
4278 | ||
4279 | spin_unlock(&device->io_lock); | |
4280 | ||
4281 | if (should_queue) | |
1cc127b5 CM |
4282 | btrfs_queue_worker(&root->fs_info->submit_workers, |
4283 | &device->work); | |
8b712842 CM |
4284 | } |
4285 | ||
de1ee92a JB |
4286 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
4287 | sector_t sector) | |
4288 | { | |
4289 | struct bio_vec *prev; | |
4290 | struct request_queue *q = bdev_get_queue(bdev); | |
4291 | unsigned short max_sectors = queue_max_sectors(q); | |
4292 | struct bvec_merge_data bvm = { | |
4293 | .bi_bdev = bdev, | |
4294 | .bi_sector = sector, | |
4295 | .bi_rw = bio->bi_rw, | |
4296 | }; | |
4297 | ||
4298 | if (bio->bi_vcnt == 0) { | |
4299 | WARN_ON(1); | |
4300 | return 1; | |
4301 | } | |
4302 | ||
4303 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
4304 | if ((bio->bi_size >> 9) > max_sectors) | |
4305 | return 0; | |
4306 | ||
4307 | if (!q->merge_bvec_fn) | |
4308 | return 1; | |
4309 | ||
4310 | bvm.bi_size = bio->bi_size - prev->bv_len; | |
4311 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) | |
4312 | return 0; | |
4313 | return 1; | |
4314 | } | |
4315 | ||
4316 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
4317 | struct bio *bio, u64 physical, int dev_nr, | |
4318 | int rw, int async) | |
4319 | { | |
4320 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
4321 | ||
4322 | bio->bi_private = bbio; | |
4323 | bio->bi_private = merge_stripe_index_into_bio_private( | |
4324 | bio->bi_private, (unsigned int)dev_nr); | |
4325 | bio->bi_end_io = btrfs_end_bio; | |
4326 | bio->bi_sector = physical >> 9; | |
4327 | #ifdef DEBUG | |
4328 | { | |
4329 | struct rcu_string *name; | |
4330 | ||
4331 | rcu_read_lock(); | |
4332 | name = rcu_dereference(dev->name); | |
d1423248 | 4333 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a JB |
4334 | "(%s id %llu), size=%u\n", rw, |
4335 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
4336 | name->str, dev->devid, bio->bi_size); | |
4337 | rcu_read_unlock(); | |
4338 | } | |
4339 | #endif | |
4340 | bio->bi_bdev = dev->bdev; | |
4341 | if (async) | |
4342 | schedule_bio(root, dev, rw, bio); | |
4343 | else | |
4344 | btrfsic_submit_bio(rw, bio); | |
4345 | } | |
4346 | ||
4347 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
4348 | struct bio *first_bio, struct btrfs_device *dev, | |
4349 | int dev_nr, int rw, int async) | |
4350 | { | |
4351 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
4352 | struct bio *bio; | |
4353 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
4354 | u64 physical = bbio->stripes[dev_nr].physical; | |
4355 | ||
4356 | again: | |
4357 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
4358 | if (!bio) | |
4359 | return -ENOMEM; | |
4360 | ||
4361 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
4362 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
4363 | bvec->bv_offset) < bvec->bv_len) { | |
4364 | u64 len = bio->bi_size; | |
4365 | ||
4366 | atomic_inc(&bbio->stripes_pending); | |
4367 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
4368 | rw, async); | |
4369 | physical += len; | |
4370 | goto again; | |
4371 | } | |
4372 | bvec++; | |
4373 | } | |
4374 | ||
4375 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
4376 | return 0; | |
4377 | } | |
4378 | ||
4379 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
4380 | { | |
4381 | atomic_inc(&bbio->error); | |
4382 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
4383 | bio->bi_private = bbio->private; | |
4384 | bio->bi_end_io = bbio->end_io; | |
4385 | bio->bi_bdev = (struct block_device *) | |
4386 | (unsigned long)bbio->mirror_num; | |
4387 | bio->bi_sector = logical >> 9; | |
4388 | kfree(bbio); | |
4389 | bio_endio(bio, -EIO); | |
4390 | } | |
4391 | } | |
4392 | ||
f188591e | 4393 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 4394 | int mirror_num, int async_submit) |
0b86a832 CM |
4395 | { |
4396 | struct btrfs_mapping_tree *map_tree; | |
4397 | struct btrfs_device *dev; | |
8790d502 | 4398 | struct bio *first_bio = bio; |
a62b9401 | 4399 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
4400 | u64 length = 0; |
4401 | u64 map_length; | |
0b86a832 | 4402 | int ret; |
8790d502 CM |
4403 | int dev_nr = 0; |
4404 | int total_devs = 1; | |
a1d3c478 | 4405 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 4406 | |
f2d8d74d | 4407 | length = bio->bi_size; |
0b86a832 CM |
4408 | map_tree = &root->fs_info->mapping_tree; |
4409 | map_length = length; | |
cea9e445 | 4410 | |
a1d3c478 | 4411 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 4412 | mirror_num); |
79787eaa JM |
4413 | if (ret) /* -ENOMEM */ |
4414 | return ret; | |
cea9e445 | 4415 | |
a1d3c478 | 4416 | total_devs = bbio->num_stripes; |
cea9e445 | 4417 | if (map_length < length) { |
48940662 | 4418 | printk(KERN_CRIT "btrfs: mapping failed logical %llu bio len %llu " |
d397712b CM |
4419 | "len %llu\n", (unsigned long long)logical, |
4420 | (unsigned long long)length, | |
4421 | (unsigned long long)map_length); | |
cea9e445 CM |
4422 | BUG(); |
4423 | } | |
a1d3c478 JS |
4424 | |
4425 | bbio->orig_bio = first_bio; | |
4426 | bbio->private = first_bio->bi_private; | |
4427 | bbio->end_io = first_bio->bi_end_io; | |
4428 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 4429 | |
d397712b | 4430 | while (dev_nr < total_devs) { |
de1ee92a JB |
4431 | dev = bbio->stripes[dev_nr].dev; |
4432 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
4433 | bbio_error(bbio, first_bio, logical); | |
4434 | dev_nr++; | |
4435 | continue; | |
4436 | } | |
4437 | ||
4438 | /* | |
4439 | * Check and see if we're ok with this bio based on it's size | |
4440 | * and offset with the given device. | |
4441 | */ | |
4442 | if (!bio_size_ok(dev->bdev, first_bio, | |
4443 | bbio->stripes[dev_nr].physical >> 9)) { | |
4444 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
4445 | dev_nr, rw, async_submit); | |
4446 | BUG_ON(ret); | |
4447 | dev_nr++; | |
4448 | continue; | |
4449 | } | |
4450 | ||
a1d3c478 JS |
4451 | if (dev_nr < total_devs - 1) { |
4452 | bio = bio_clone(first_bio, GFP_NOFS); | |
79787eaa | 4453 | BUG_ON(!bio); /* -ENOMEM */ |
a1d3c478 JS |
4454 | } else { |
4455 | bio = first_bio; | |
8790d502 | 4456 | } |
de1ee92a JB |
4457 | |
4458 | submit_stripe_bio(root, bbio, bio, | |
4459 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
4460 | async_submit); | |
8790d502 CM |
4461 | dev_nr++; |
4462 | } | |
0b86a832 CM |
4463 | return 0; |
4464 | } | |
4465 | ||
a443755f | 4466 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 4467 | u8 *uuid, u8 *fsid) |
0b86a832 | 4468 | { |
2b82032c YZ |
4469 | struct btrfs_device *device; |
4470 | struct btrfs_fs_devices *cur_devices; | |
4471 | ||
4472 | cur_devices = root->fs_info->fs_devices; | |
4473 | while (cur_devices) { | |
4474 | if (!fsid || | |
4475 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4476 | device = __find_device(&cur_devices->devices, | |
4477 | devid, uuid); | |
4478 | if (device) | |
4479 | return device; | |
4480 | } | |
4481 | cur_devices = cur_devices->seed; | |
4482 | } | |
4483 | return NULL; | |
0b86a832 CM |
4484 | } |
4485 | ||
dfe25020 CM |
4486 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
4487 | u64 devid, u8 *dev_uuid) | |
4488 | { | |
4489 | struct btrfs_device *device; | |
4490 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4491 | ||
4492 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 4493 | if (!device) |
4494 | return NULL; | |
dfe25020 CM |
4495 | list_add(&device->dev_list, |
4496 | &fs_devices->devices); | |
dfe25020 CM |
4497 | device->dev_root = root->fs_info->dev_root; |
4498 | device->devid = devid; | |
8b712842 | 4499 | device->work.func = pending_bios_fn; |
e4404d6e | 4500 | device->fs_devices = fs_devices; |
cd02dca5 | 4501 | device->missing = 1; |
dfe25020 | 4502 | fs_devices->num_devices++; |
cd02dca5 | 4503 | fs_devices->missing_devices++; |
dfe25020 | 4504 | spin_lock_init(&device->io_lock); |
d20f7043 | 4505 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
4506 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
4507 | return device; | |
4508 | } | |
4509 | ||
0b86a832 CM |
4510 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
4511 | struct extent_buffer *leaf, | |
4512 | struct btrfs_chunk *chunk) | |
4513 | { | |
4514 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4515 | struct map_lookup *map; | |
4516 | struct extent_map *em; | |
4517 | u64 logical; | |
4518 | u64 length; | |
4519 | u64 devid; | |
a443755f | 4520 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 4521 | int num_stripes; |
0b86a832 | 4522 | int ret; |
593060d7 | 4523 | int i; |
0b86a832 | 4524 | |
e17cade2 CM |
4525 | logical = key->offset; |
4526 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 4527 | |
890871be | 4528 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 4529 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 4530 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
4531 | |
4532 | /* already mapped? */ | |
4533 | if (em && em->start <= logical && em->start + em->len > logical) { | |
4534 | free_extent_map(em); | |
0b86a832 CM |
4535 | return 0; |
4536 | } else if (em) { | |
4537 | free_extent_map(em); | |
4538 | } | |
0b86a832 | 4539 | |
172ddd60 | 4540 | em = alloc_extent_map(); |
0b86a832 CM |
4541 | if (!em) |
4542 | return -ENOMEM; | |
593060d7 CM |
4543 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
4544 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
4545 | if (!map) { |
4546 | free_extent_map(em); | |
4547 | return -ENOMEM; | |
4548 | } | |
4549 | ||
4550 | em->bdev = (struct block_device *)map; | |
4551 | em->start = logical; | |
4552 | em->len = length; | |
4553 | em->block_start = 0; | |
c8b97818 | 4554 | em->block_len = em->len; |
0b86a832 | 4555 | |
593060d7 CM |
4556 | map->num_stripes = num_stripes; |
4557 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
4558 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
4559 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
4560 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
4561 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 4562 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
4563 | for (i = 0; i < num_stripes; i++) { |
4564 | map->stripes[i].physical = | |
4565 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
4566 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
4567 | read_extent_buffer(leaf, uuid, (unsigned long) |
4568 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
4569 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4570 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
4571 | NULL); | |
dfe25020 | 4572 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
4573 | kfree(map); |
4574 | free_extent_map(em); | |
4575 | return -EIO; | |
4576 | } | |
dfe25020 CM |
4577 | if (!map->stripes[i].dev) { |
4578 | map->stripes[i].dev = | |
4579 | add_missing_dev(root, devid, uuid); | |
4580 | if (!map->stripes[i].dev) { | |
4581 | kfree(map); | |
4582 | free_extent_map(em); | |
4583 | return -EIO; | |
4584 | } | |
4585 | } | |
4586 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
4587 | } |
4588 | ||
890871be | 4589 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 4590 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 4591 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 4592 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
4593 | free_extent_map(em); |
4594 | ||
4595 | return 0; | |
4596 | } | |
4597 | ||
143bede5 | 4598 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
4599 | struct btrfs_dev_item *dev_item, |
4600 | struct btrfs_device *device) | |
4601 | { | |
4602 | unsigned long ptr; | |
0b86a832 CM |
4603 | |
4604 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
4605 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
4606 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
4607 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
4608 | device->type = btrfs_device_type(leaf, dev_item); | |
4609 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
4610 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
4611 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
4612 | |
4613 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 4614 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
4615 | } |
4616 | ||
2b82032c YZ |
4617 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
4618 | { | |
4619 | struct btrfs_fs_devices *fs_devices; | |
4620 | int ret; | |
4621 | ||
b367e47f | 4622 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
4623 | |
4624 | fs_devices = root->fs_info->fs_devices->seed; | |
4625 | while (fs_devices) { | |
4626 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4627 | ret = 0; | |
4628 | goto out; | |
4629 | } | |
4630 | fs_devices = fs_devices->seed; | |
4631 | } | |
4632 | ||
4633 | fs_devices = find_fsid(fsid); | |
4634 | if (!fs_devices) { | |
4635 | ret = -ENOENT; | |
4636 | goto out; | |
4637 | } | |
e4404d6e YZ |
4638 | |
4639 | fs_devices = clone_fs_devices(fs_devices); | |
4640 | if (IS_ERR(fs_devices)) { | |
4641 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
4642 | goto out; |
4643 | } | |
4644 | ||
97288f2c | 4645 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 4646 | root->fs_info->bdev_holder); |
48d28232 JL |
4647 | if (ret) { |
4648 | free_fs_devices(fs_devices); | |
2b82032c | 4649 | goto out; |
48d28232 | 4650 | } |
2b82032c YZ |
4651 | |
4652 | if (!fs_devices->seeding) { | |
4653 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 4654 | free_fs_devices(fs_devices); |
2b82032c YZ |
4655 | ret = -EINVAL; |
4656 | goto out; | |
4657 | } | |
4658 | ||
4659 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
4660 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 4661 | out: |
2b82032c YZ |
4662 | return ret; |
4663 | } | |
4664 | ||
0d81ba5d | 4665 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
4666 | struct extent_buffer *leaf, |
4667 | struct btrfs_dev_item *dev_item) | |
4668 | { | |
4669 | struct btrfs_device *device; | |
4670 | u64 devid; | |
4671 | int ret; | |
2b82032c | 4672 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
4673 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
4674 | ||
0b86a832 | 4675 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
4676 | read_extent_buffer(leaf, dev_uuid, |
4677 | (unsigned long)btrfs_device_uuid(dev_item), | |
4678 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4679 | read_extent_buffer(leaf, fs_uuid, |
4680 | (unsigned long)btrfs_device_fsid(dev_item), | |
4681 | BTRFS_UUID_SIZE); | |
4682 | ||
4683 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
4684 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 4685 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 4686 | return ret; |
2b82032c YZ |
4687 | } |
4688 | ||
4689 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
4690 | if (!device || !device->bdev) { | |
e4404d6e | 4691 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
4692 | return -EIO; |
4693 | ||
4694 | if (!device) { | |
d397712b CM |
4695 | printk(KERN_WARNING "warning devid %llu missing\n", |
4696 | (unsigned long long)devid); | |
2b82032c YZ |
4697 | device = add_missing_dev(root, devid, dev_uuid); |
4698 | if (!device) | |
4699 | return -ENOMEM; | |
cd02dca5 CM |
4700 | } else if (!device->missing) { |
4701 | /* | |
4702 | * this happens when a device that was properly setup | |
4703 | * in the device info lists suddenly goes bad. | |
4704 | * device->bdev is NULL, and so we have to set | |
4705 | * device->missing to one here | |
4706 | */ | |
4707 | root->fs_info->fs_devices->missing_devices++; | |
4708 | device->missing = 1; | |
2b82032c YZ |
4709 | } |
4710 | } | |
4711 | ||
4712 | if (device->fs_devices != root->fs_info->fs_devices) { | |
4713 | BUG_ON(device->writeable); | |
4714 | if (device->generation != | |
4715 | btrfs_device_generation(leaf, dev_item)) | |
4716 | return -EINVAL; | |
6324fbf3 | 4717 | } |
0b86a832 CM |
4718 | |
4719 | fill_device_from_item(leaf, dev_item, device); | |
4720 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 4721 | device->in_fs_metadata = 1; |
2bf64758 | 4722 | if (device->writeable) { |
2b82032c | 4723 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
4724 | spin_lock(&root->fs_info->free_chunk_lock); |
4725 | root->fs_info->free_chunk_space += device->total_bytes - | |
4726 | device->bytes_used; | |
4727 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4728 | } | |
0b86a832 | 4729 | ret = 0; |
0b86a832 CM |
4730 | return ret; |
4731 | } | |
4732 | ||
e4404d6e | 4733 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 4734 | { |
6c41761f | 4735 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 4736 | struct extent_buffer *sb; |
0b86a832 | 4737 | struct btrfs_disk_key *disk_key; |
0b86a832 | 4738 | struct btrfs_chunk *chunk; |
84eed90f CM |
4739 | u8 *ptr; |
4740 | unsigned long sb_ptr; | |
4741 | int ret = 0; | |
0b86a832 CM |
4742 | u32 num_stripes; |
4743 | u32 array_size; | |
4744 | u32 len = 0; | |
0b86a832 | 4745 | u32 cur; |
84eed90f | 4746 | struct btrfs_key key; |
0b86a832 | 4747 | |
e4404d6e | 4748 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
4749 | BTRFS_SUPER_INFO_SIZE); |
4750 | if (!sb) | |
4751 | return -ENOMEM; | |
4752 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 4753 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
4754 | /* |
4755 | * The sb extent buffer is artifical and just used to read the system array. | |
4756 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
4757 | * pages up-to-date when the page is larger: extent does not cover the | |
4758 | * whole page and consequently check_page_uptodate does not find all | |
4759 | * the page's extents up-to-date (the hole beyond sb), | |
4760 | * write_extent_buffer then triggers a WARN_ON. | |
4761 | * | |
4762 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
4763 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
4764 | * to silence the warning eg. on PowerPC 64. | |
4765 | */ | |
4766 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 4767 | SetPageUptodate(sb->pages[0]); |
4008c04a | 4768 | |
a061fc8d | 4769 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
4770 | array_size = btrfs_super_sys_array_size(super_copy); |
4771 | ||
0b86a832 CM |
4772 | ptr = super_copy->sys_chunk_array; |
4773 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
4774 | cur = 0; | |
4775 | ||
4776 | while (cur < array_size) { | |
4777 | disk_key = (struct btrfs_disk_key *)ptr; | |
4778 | btrfs_disk_key_to_cpu(&key, disk_key); | |
4779 | ||
a061fc8d | 4780 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
4781 | sb_ptr += len; |
4782 | cur += len; | |
4783 | ||
0d81ba5d | 4784 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 4785 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 4786 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
4787 | if (ret) |
4788 | break; | |
0b86a832 CM |
4789 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
4790 | len = btrfs_chunk_item_size(num_stripes); | |
4791 | } else { | |
84eed90f CM |
4792 | ret = -EIO; |
4793 | break; | |
0b86a832 CM |
4794 | } |
4795 | ptr += len; | |
4796 | sb_ptr += len; | |
4797 | cur += len; | |
4798 | } | |
a061fc8d | 4799 | free_extent_buffer(sb); |
84eed90f | 4800 | return ret; |
0b86a832 CM |
4801 | } |
4802 | ||
4803 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
4804 | { | |
4805 | struct btrfs_path *path; | |
4806 | struct extent_buffer *leaf; | |
4807 | struct btrfs_key key; | |
4808 | struct btrfs_key found_key; | |
4809 | int ret; | |
4810 | int slot; | |
4811 | ||
4812 | root = root->fs_info->chunk_root; | |
4813 | ||
4814 | path = btrfs_alloc_path(); | |
4815 | if (!path) | |
4816 | return -ENOMEM; | |
4817 | ||
b367e47f LZ |
4818 | mutex_lock(&uuid_mutex); |
4819 | lock_chunks(root); | |
4820 | ||
0b86a832 CM |
4821 | /* first we search for all of the device items, and then we |
4822 | * read in all of the chunk items. This way we can create chunk | |
4823 | * mappings that reference all of the devices that are afound | |
4824 | */ | |
4825 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
4826 | key.offset = 0; | |
4827 | key.type = 0; | |
4828 | again: | |
4829 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
4830 | if (ret < 0) |
4831 | goto error; | |
d397712b | 4832 | while (1) { |
0b86a832 CM |
4833 | leaf = path->nodes[0]; |
4834 | slot = path->slots[0]; | |
4835 | if (slot >= btrfs_header_nritems(leaf)) { | |
4836 | ret = btrfs_next_leaf(root, path); | |
4837 | if (ret == 0) | |
4838 | continue; | |
4839 | if (ret < 0) | |
4840 | goto error; | |
4841 | break; | |
4842 | } | |
4843 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4844 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4845 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
4846 | break; | |
4847 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
4848 | struct btrfs_dev_item *dev_item; | |
4849 | dev_item = btrfs_item_ptr(leaf, slot, | |
4850 | struct btrfs_dev_item); | |
0d81ba5d | 4851 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
4852 | if (ret) |
4853 | goto error; | |
0b86a832 CM |
4854 | } |
4855 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
4856 | struct btrfs_chunk *chunk; | |
4857 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
4858 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
4859 | if (ret) |
4860 | goto error; | |
0b86a832 CM |
4861 | } |
4862 | path->slots[0]++; | |
4863 | } | |
4864 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4865 | key.objectid = 0; | |
b3b4aa74 | 4866 | btrfs_release_path(path); |
0b86a832 CM |
4867 | goto again; |
4868 | } | |
0b86a832 CM |
4869 | ret = 0; |
4870 | error: | |
b367e47f LZ |
4871 | unlock_chunks(root); |
4872 | mutex_unlock(&uuid_mutex); | |
4873 | ||
2b82032c | 4874 | btrfs_free_path(path); |
0b86a832 CM |
4875 | return ret; |
4876 | } | |
442a4f63 | 4877 | |
733f4fbb SB |
4878 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
4879 | { | |
4880 | int i; | |
4881 | ||
4882 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4883 | btrfs_dev_stat_reset(dev, i); | |
4884 | } | |
4885 | ||
4886 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
4887 | { | |
4888 | struct btrfs_key key; | |
4889 | struct btrfs_key found_key; | |
4890 | struct btrfs_root *dev_root = fs_info->dev_root; | |
4891 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
4892 | struct extent_buffer *eb; | |
4893 | int slot; | |
4894 | int ret = 0; | |
4895 | struct btrfs_device *device; | |
4896 | struct btrfs_path *path = NULL; | |
4897 | int i; | |
4898 | ||
4899 | path = btrfs_alloc_path(); | |
4900 | if (!path) { | |
4901 | ret = -ENOMEM; | |
4902 | goto out; | |
4903 | } | |
4904 | ||
4905 | mutex_lock(&fs_devices->device_list_mutex); | |
4906 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
4907 | int item_size; | |
4908 | struct btrfs_dev_stats_item *ptr; | |
4909 | ||
4910 | key.objectid = 0; | |
4911 | key.type = BTRFS_DEV_STATS_KEY; | |
4912 | key.offset = device->devid; | |
4913 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
4914 | if (ret) { | |
733f4fbb SB |
4915 | __btrfs_reset_dev_stats(device); |
4916 | device->dev_stats_valid = 1; | |
4917 | btrfs_release_path(path); | |
4918 | continue; | |
4919 | } | |
4920 | slot = path->slots[0]; | |
4921 | eb = path->nodes[0]; | |
4922 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
4923 | item_size = btrfs_item_size_nr(eb, slot); | |
4924 | ||
4925 | ptr = btrfs_item_ptr(eb, slot, | |
4926 | struct btrfs_dev_stats_item); | |
4927 | ||
4928 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
4929 | if (item_size >= (1 + i) * sizeof(__le64)) | |
4930 | btrfs_dev_stat_set(device, i, | |
4931 | btrfs_dev_stats_value(eb, ptr, i)); | |
4932 | else | |
4933 | btrfs_dev_stat_reset(device, i); | |
4934 | } | |
4935 | ||
4936 | device->dev_stats_valid = 1; | |
4937 | btrfs_dev_stat_print_on_load(device); | |
4938 | btrfs_release_path(path); | |
4939 | } | |
4940 | mutex_unlock(&fs_devices->device_list_mutex); | |
4941 | ||
4942 | out: | |
4943 | btrfs_free_path(path); | |
4944 | return ret < 0 ? ret : 0; | |
4945 | } | |
4946 | ||
4947 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
4948 | struct btrfs_root *dev_root, | |
4949 | struct btrfs_device *device) | |
4950 | { | |
4951 | struct btrfs_path *path; | |
4952 | struct btrfs_key key; | |
4953 | struct extent_buffer *eb; | |
4954 | struct btrfs_dev_stats_item *ptr; | |
4955 | int ret; | |
4956 | int i; | |
4957 | ||
4958 | key.objectid = 0; | |
4959 | key.type = BTRFS_DEV_STATS_KEY; | |
4960 | key.offset = device->devid; | |
4961 | ||
4962 | path = btrfs_alloc_path(); | |
4963 | BUG_ON(!path); | |
4964 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
4965 | if (ret < 0) { | |
606686ee JB |
4966 | printk_in_rcu(KERN_WARNING "btrfs: error %d while searching for dev_stats item for device %s!\n", |
4967 | ret, rcu_str_deref(device->name)); | |
733f4fbb SB |
4968 | goto out; |
4969 | } | |
4970 | ||
4971 | if (ret == 0 && | |
4972 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
4973 | /* need to delete old one and insert a new one */ | |
4974 | ret = btrfs_del_item(trans, dev_root, path); | |
4975 | if (ret != 0) { | |
606686ee JB |
4976 | printk_in_rcu(KERN_WARNING "btrfs: delete too small dev_stats item for device %s failed %d!\n", |
4977 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4978 | goto out; |
4979 | } | |
4980 | ret = 1; | |
4981 | } | |
4982 | ||
4983 | if (ret == 1) { | |
4984 | /* need to insert a new item */ | |
4985 | btrfs_release_path(path); | |
4986 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
4987 | &key, sizeof(*ptr)); | |
4988 | if (ret < 0) { | |
606686ee JB |
4989 | printk_in_rcu(KERN_WARNING "btrfs: insert dev_stats item for device %s failed %d!\n", |
4990 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
4991 | goto out; |
4992 | } | |
4993 | } | |
4994 | ||
4995 | eb = path->nodes[0]; | |
4996 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
4997 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
4998 | btrfs_set_dev_stats_value(eb, ptr, i, | |
4999 | btrfs_dev_stat_read(device, i)); | |
5000 | btrfs_mark_buffer_dirty(eb); | |
5001 | ||
5002 | out: | |
5003 | btrfs_free_path(path); | |
5004 | return ret; | |
5005 | } | |
5006 | ||
5007 | /* | |
5008 | * called from commit_transaction. Writes all changed device stats to disk. | |
5009 | */ | |
5010 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
5011 | struct btrfs_fs_info *fs_info) | |
5012 | { | |
5013 | struct btrfs_root *dev_root = fs_info->dev_root; | |
5014 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
5015 | struct btrfs_device *device; | |
5016 | int ret = 0; | |
5017 | ||
5018 | mutex_lock(&fs_devices->device_list_mutex); | |
5019 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
5020 | if (!device->dev_stats_valid || !device->dev_stats_dirty) | |
5021 | continue; | |
5022 | ||
5023 | ret = update_dev_stat_item(trans, dev_root, device); | |
5024 | if (!ret) | |
5025 | device->dev_stats_dirty = 0; | |
5026 | } | |
5027 | mutex_unlock(&fs_devices->device_list_mutex); | |
5028 | ||
5029 | return ret; | |
5030 | } | |
5031 | ||
442a4f63 SB |
5032 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
5033 | { | |
5034 | btrfs_dev_stat_inc(dev, index); | |
5035 | btrfs_dev_stat_print_on_error(dev); | |
5036 | } | |
5037 | ||
5038 | void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) | |
5039 | { | |
733f4fbb SB |
5040 | if (!dev->dev_stats_valid) |
5041 | return; | |
606686ee | 5042 | printk_ratelimited_in_rcu(KERN_ERR |
442a4f63 | 5043 | "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
606686ee | 5044 | rcu_str_deref(dev->name), |
442a4f63 SB |
5045 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
5046 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
5047 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
5048 | btrfs_dev_stat_read(dev, | |
5049 | BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
5050 | btrfs_dev_stat_read(dev, | |
5051 | BTRFS_DEV_STAT_GENERATION_ERRS)); | |
5052 | } | |
c11d2c23 | 5053 | |
733f4fbb SB |
5054 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
5055 | { | |
a98cdb85 SB |
5056 | int i; |
5057 | ||
5058 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5059 | if (btrfs_dev_stat_read(dev, i) != 0) | |
5060 | break; | |
5061 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
5062 | return; /* all values == 0, suppress message */ | |
5063 | ||
606686ee JB |
5064 | printk_in_rcu(KERN_INFO "btrfs: bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", |
5065 | rcu_str_deref(dev->name), | |
733f4fbb SB |
5066 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
5067 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
5068 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
5069 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
5070 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
5071 | } | |
5072 | ||
c11d2c23 | 5073 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 5074 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
5075 | { |
5076 | struct btrfs_device *dev; | |
5077 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
5078 | int i; | |
5079 | ||
5080 | mutex_lock(&fs_devices->device_list_mutex); | |
5081 | dev = btrfs_find_device(root, stats->devid, NULL, NULL); | |
5082 | mutex_unlock(&fs_devices->device_list_mutex); | |
5083 | ||
5084 | if (!dev) { | |
5085 | printk(KERN_WARNING | |
5086 | "btrfs: get dev_stats failed, device not found\n"); | |
5087 | return -ENODEV; | |
733f4fbb SB |
5088 | } else if (!dev->dev_stats_valid) { |
5089 | printk(KERN_WARNING | |
5090 | "btrfs: get dev_stats failed, not yet valid\n"); | |
5091 | return -ENODEV; | |
b27f7c0c | 5092 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
5093 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
5094 | if (stats->nr_items > i) | |
5095 | stats->values[i] = | |
5096 | btrfs_dev_stat_read_and_reset(dev, i); | |
5097 | else | |
5098 | btrfs_dev_stat_reset(dev, i); | |
5099 | } | |
5100 | } else { | |
5101 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
5102 | if (stats->nr_items > i) | |
5103 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
5104 | } | |
5105 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
5106 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
5107 | return 0; | |
5108 | } |