]>
Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
59641015 | 26 | #include <linux/kthread.h> |
593060d7 | 27 | #include <asm/div64.h> |
4b4e25f2 | 28 | #include "compat.h" |
0b86a832 CM |
29 | #include "ctree.h" |
30 | #include "extent_map.h" | |
31 | #include "disk-io.h" | |
32 | #include "transaction.h" | |
33 | #include "print-tree.h" | |
34 | #include "volumes.h" | |
8b712842 | 35 | #include "async-thread.h" |
21adbd5c | 36 | #include "check-integrity.h" |
0b86a832 | 37 | |
2b82032c YZ |
38 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
39 | struct btrfs_root *root, | |
40 | struct btrfs_device *device); | |
41 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
42 | ||
8a4b83cc CM |
43 | static DEFINE_MUTEX(uuid_mutex); |
44 | static LIST_HEAD(fs_uuids); | |
45 | ||
7d9eb12c CM |
46 | static void lock_chunks(struct btrfs_root *root) |
47 | { | |
7d9eb12c CM |
48 | mutex_lock(&root->fs_info->chunk_mutex); |
49 | } | |
50 | ||
51 | static void unlock_chunks(struct btrfs_root *root) | |
52 | { | |
7d9eb12c CM |
53 | mutex_unlock(&root->fs_info->chunk_mutex); |
54 | } | |
55 | ||
e4404d6e YZ |
56 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
57 | { | |
58 | struct btrfs_device *device; | |
59 | WARN_ON(fs_devices->opened); | |
60 | while (!list_empty(&fs_devices->devices)) { | |
61 | device = list_entry(fs_devices->devices.next, | |
62 | struct btrfs_device, dev_list); | |
63 | list_del(&device->dev_list); | |
64 | kfree(device->name); | |
65 | kfree(device); | |
66 | } | |
67 | kfree(fs_devices); | |
68 | } | |
69 | ||
8a4b83cc CM |
70 | int btrfs_cleanup_fs_uuids(void) |
71 | { | |
72 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 73 | |
2b82032c YZ |
74 | while (!list_empty(&fs_uuids)) { |
75 | fs_devices = list_entry(fs_uuids.next, | |
76 | struct btrfs_fs_devices, list); | |
77 | list_del(&fs_devices->list); | |
e4404d6e | 78 | free_fs_devices(fs_devices); |
8a4b83cc CM |
79 | } |
80 | return 0; | |
81 | } | |
82 | ||
a1b32a59 CM |
83 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
84 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
85 | { |
86 | struct btrfs_device *dev; | |
8a4b83cc | 87 | |
c6e30871 | 88 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 89 | if (dev->devid == devid && |
8f18cf13 | 90 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 91 | return dev; |
a443755f | 92 | } |
8a4b83cc CM |
93 | } |
94 | return NULL; | |
95 | } | |
96 | ||
a1b32a59 | 97 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 98 | { |
8a4b83cc CM |
99 | struct btrfs_fs_devices *fs_devices; |
100 | ||
c6e30871 | 101 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
102 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
103 | return fs_devices; | |
104 | } | |
105 | return NULL; | |
106 | } | |
107 | ||
ffbd517d CM |
108 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
109 | struct bio *head, struct bio *tail) | |
110 | { | |
111 | ||
112 | struct bio *old_head; | |
113 | ||
114 | old_head = pending_bios->head; | |
115 | pending_bios->head = head; | |
116 | if (pending_bios->tail) | |
117 | tail->bi_next = old_head; | |
118 | else | |
119 | pending_bios->tail = tail; | |
120 | } | |
121 | ||
8b712842 CM |
122 | /* |
123 | * we try to collect pending bios for a device so we don't get a large | |
124 | * number of procs sending bios down to the same device. This greatly | |
125 | * improves the schedulers ability to collect and merge the bios. | |
126 | * | |
127 | * But, it also turns into a long list of bios to process and that is sure | |
128 | * to eventually make the worker thread block. The solution here is to | |
129 | * make some progress and then put this work struct back at the end of | |
130 | * the list if the block device is congested. This way, multiple devices | |
131 | * can make progress from a single worker thread. | |
132 | */ | |
d397712b | 133 | static noinline int run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
134 | { |
135 | struct bio *pending; | |
136 | struct backing_dev_info *bdi; | |
b64a2851 | 137 | struct btrfs_fs_info *fs_info; |
ffbd517d | 138 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
139 | struct bio *tail; |
140 | struct bio *cur; | |
141 | int again = 0; | |
ffbd517d | 142 | unsigned long num_run; |
d644d8a1 | 143 | unsigned long batch_run = 0; |
b64a2851 | 144 | unsigned long limit; |
b765ead5 | 145 | unsigned long last_waited = 0; |
d84275c9 | 146 | int force_reg = 0; |
0e588859 | 147 | int sync_pending = 0; |
211588ad CM |
148 | struct blk_plug plug; |
149 | ||
150 | /* | |
151 | * this function runs all the bios we've collected for | |
152 | * a particular device. We don't want to wander off to | |
153 | * another device without first sending all of these down. | |
154 | * So, setup a plug here and finish it off before we return | |
155 | */ | |
156 | blk_start_plug(&plug); | |
8b712842 | 157 | |
bedf762b | 158 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
159 | fs_info = device->dev_root->fs_info; |
160 | limit = btrfs_async_submit_limit(fs_info); | |
161 | limit = limit * 2 / 3; | |
162 | ||
8b712842 CM |
163 | loop: |
164 | spin_lock(&device->io_lock); | |
165 | ||
a6837051 | 166 | loop_lock: |
d84275c9 | 167 | num_run = 0; |
ffbd517d | 168 | |
8b712842 CM |
169 | /* take all the bios off the list at once and process them |
170 | * later on (without the lock held). But, remember the | |
171 | * tail and other pointers so the bios can be properly reinserted | |
172 | * into the list if we hit congestion | |
173 | */ | |
d84275c9 | 174 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 175 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
176 | force_reg = 1; |
177 | } else { | |
ffbd517d | 178 | pending_bios = &device->pending_bios; |
d84275c9 CM |
179 | force_reg = 0; |
180 | } | |
ffbd517d CM |
181 | |
182 | pending = pending_bios->head; | |
183 | tail = pending_bios->tail; | |
8b712842 | 184 | WARN_ON(pending && !tail); |
8b712842 CM |
185 | |
186 | /* | |
187 | * if pending was null this time around, no bios need processing | |
188 | * at all and we can stop. Otherwise it'll loop back up again | |
189 | * and do an additional check so no bios are missed. | |
190 | * | |
191 | * device->running_pending is used to synchronize with the | |
192 | * schedule_bio code. | |
193 | */ | |
ffbd517d CM |
194 | if (device->pending_sync_bios.head == NULL && |
195 | device->pending_bios.head == NULL) { | |
8b712842 CM |
196 | again = 0; |
197 | device->running_pending = 0; | |
ffbd517d CM |
198 | } else { |
199 | again = 1; | |
200 | device->running_pending = 1; | |
8b712842 | 201 | } |
ffbd517d CM |
202 | |
203 | pending_bios->head = NULL; | |
204 | pending_bios->tail = NULL; | |
205 | ||
8b712842 CM |
206 | spin_unlock(&device->io_lock); |
207 | ||
d397712b | 208 | while (pending) { |
ffbd517d CM |
209 | |
210 | rmb(); | |
d84275c9 CM |
211 | /* we want to work on both lists, but do more bios on the |
212 | * sync list than the regular list | |
213 | */ | |
214 | if ((num_run > 32 && | |
215 | pending_bios != &device->pending_sync_bios && | |
216 | device->pending_sync_bios.head) || | |
217 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
218 | device->pending_bios.head)) { | |
ffbd517d CM |
219 | spin_lock(&device->io_lock); |
220 | requeue_list(pending_bios, pending, tail); | |
221 | goto loop_lock; | |
222 | } | |
223 | ||
8b712842 CM |
224 | cur = pending; |
225 | pending = pending->bi_next; | |
226 | cur->bi_next = NULL; | |
b64a2851 CM |
227 | atomic_dec(&fs_info->nr_async_bios); |
228 | ||
229 | if (atomic_read(&fs_info->nr_async_bios) < limit && | |
230 | waitqueue_active(&fs_info->async_submit_wait)) | |
231 | wake_up(&fs_info->async_submit_wait); | |
492bb6de CM |
232 | |
233 | BUG_ON(atomic_read(&cur->bi_cnt) == 0); | |
d644d8a1 | 234 | |
2ab1ba68 CM |
235 | /* |
236 | * if we're doing the sync list, record that our | |
237 | * plug has some sync requests on it | |
238 | * | |
239 | * If we're doing the regular list and there are | |
240 | * sync requests sitting around, unplug before | |
241 | * we add more | |
242 | */ | |
243 | if (pending_bios == &device->pending_sync_bios) { | |
244 | sync_pending = 1; | |
245 | } else if (sync_pending) { | |
246 | blk_finish_plug(&plug); | |
247 | blk_start_plug(&plug); | |
248 | sync_pending = 0; | |
249 | } | |
250 | ||
21adbd5c | 251 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
252 | num_run++; |
253 | batch_run++; | |
7eaceacc | 254 | if (need_resched()) |
ffbd517d | 255 | cond_resched(); |
8b712842 CM |
256 | |
257 | /* | |
258 | * we made progress, there is more work to do and the bdi | |
259 | * is now congested. Back off and let other work structs | |
260 | * run instead | |
261 | */ | |
57fd5a5f | 262 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 263 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 264 | struct io_context *ioc; |
8b712842 | 265 | |
b765ead5 CM |
266 | ioc = current->io_context; |
267 | ||
268 | /* | |
269 | * the main goal here is that we don't want to | |
270 | * block if we're going to be able to submit | |
271 | * more requests without blocking. | |
272 | * | |
273 | * This code does two great things, it pokes into | |
274 | * the elevator code from a filesystem _and_ | |
275 | * it makes assumptions about how batching works. | |
276 | */ | |
277 | if (ioc && ioc->nr_batch_requests > 0 && | |
278 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
279 | (last_waited == 0 || | |
280 | ioc->last_waited == last_waited)) { | |
281 | /* | |
282 | * we want to go through our batch of | |
283 | * requests and stop. So, we copy out | |
284 | * the ioc->last_waited time and test | |
285 | * against it before looping | |
286 | */ | |
287 | last_waited = ioc->last_waited; | |
7eaceacc | 288 | if (need_resched()) |
ffbd517d | 289 | cond_resched(); |
b765ead5 CM |
290 | continue; |
291 | } | |
8b712842 | 292 | spin_lock(&device->io_lock); |
ffbd517d | 293 | requeue_list(pending_bios, pending, tail); |
a6837051 | 294 | device->running_pending = 1; |
8b712842 CM |
295 | |
296 | spin_unlock(&device->io_lock); | |
297 | btrfs_requeue_work(&device->work); | |
298 | goto done; | |
299 | } | |
d85c8a6f CM |
300 | /* unplug every 64 requests just for good measure */ |
301 | if (batch_run % 64 == 0) { | |
302 | blk_finish_plug(&plug); | |
303 | blk_start_plug(&plug); | |
304 | sync_pending = 0; | |
305 | } | |
8b712842 | 306 | } |
ffbd517d | 307 | |
51684082 CM |
308 | cond_resched(); |
309 | if (again) | |
310 | goto loop; | |
311 | ||
312 | spin_lock(&device->io_lock); | |
313 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
314 | goto loop_lock; | |
315 | spin_unlock(&device->io_lock); | |
316 | ||
8b712842 | 317 | done: |
211588ad | 318 | blk_finish_plug(&plug); |
8b712842 CM |
319 | return 0; |
320 | } | |
321 | ||
b2950863 | 322 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
323 | { |
324 | struct btrfs_device *device; | |
325 | ||
326 | device = container_of(work, struct btrfs_device, work); | |
327 | run_scheduled_bios(device); | |
328 | } | |
329 | ||
a1b32a59 | 330 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
331 | struct btrfs_super_block *disk_super, |
332 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
333 | { | |
334 | struct btrfs_device *device; | |
335 | struct btrfs_fs_devices *fs_devices; | |
336 | u64 found_transid = btrfs_super_generation(disk_super); | |
3a0524dc | 337 | char *name; |
8a4b83cc CM |
338 | |
339 | fs_devices = find_fsid(disk_super->fsid); | |
340 | if (!fs_devices) { | |
515dc322 | 341 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
8a4b83cc CM |
342 | if (!fs_devices) |
343 | return -ENOMEM; | |
344 | INIT_LIST_HEAD(&fs_devices->devices); | |
b3075717 | 345 | INIT_LIST_HEAD(&fs_devices->alloc_list); |
8a4b83cc CM |
346 | list_add(&fs_devices->list, &fs_uuids); |
347 | memcpy(fs_devices->fsid, disk_super->fsid, BTRFS_FSID_SIZE); | |
348 | fs_devices->latest_devid = devid; | |
349 | fs_devices->latest_trans = found_transid; | |
e5e9a520 | 350 | mutex_init(&fs_devices->device_list_mutex); |
8a4b83cc CM |
351 | device = NULL; |
352 | } else { | |
a443755f CM |
353 | device = __find_device(&fs_devices->devices, devid, |
354 | disk_super->dev_item.uuid); | |
8a4b83cc CM |
355 | } |
356 | if (!device) { | |
2b82032c YZ |
357 | if (fs_devices->opened) |
358 | return -EBUSY; | |
359 | ||
8a4b83cc CM |
360 | device = kzalloc(sizeof(*device), GFP_NOFS); |
361 | if (!device) { | |
362 | /* we can safely leave the fs_devices entry around */ | |
363 | return -ENOMEM; | |
364 | } | |
365 | device->devid = devid; | |
8b712842 | 366 | device->work.func = pending_bios_fn; |
a443755f CM |
367 | memcpy(device->uuid, disk_super->dev_item.uuid, |
368 | BTRFS_UUID_SIZE); | |
b248a415 | 369 | spin_lock_init(&device->io_lock); |
8a4b83cc CM |
370 | device->name = kstrdup(path, GFP_NOFS); |
371 | if (!device->name) { | |
372 | kfree(device); | |
373 | return -ENOMEM; | |
374 | } | |
2b82032c | 375 | INIT_LIST_HEAD(&device->dev_alloc_list); |
e5e9a520 | 376 | |
90519d66 AJ |
377 | /* init readahead state */ |
378 | spin_lock_init(&device->reada_lock); | |
379 | device->reada_curr_zone = NULL; | |
380 | atomic_set(&device->reada_in_flight, 0); | |
381 | device->reada_next = 0; | |
382 | INIT_RADIX_TREE(&device->reada_zones, GFP_NOFS & ~__GFP_WAIT); | |
383 | INIT_RADIX_TREE(&device->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
384 | ||
e5e9a520 | 385 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 386 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
e5e9a520 CM |
387 | mutex_unlock(&fs_devices->device_list_mutex); |
388 | ||
2b82032c | 389 | device->fs_devices = fs_devices; |
8a4b83cc | 390 | fs_devices->num_devices++; |
cd02dca5 | 391 | } else if (!device->name || strcmp(device->name, path)) { |
3a0524dc TH |
392 | name = kstrdup(path, GFP_NOFS); |
393 | if (!name) | |
394 | return -ENOMEM; | |
395 | kfree(device->name); | |
396 | device->name = name; | |
cd02dca5 CM |
397 | if (device->missing) { |
398 | fs_devices->missing_devices--; | |
399 | device->missing = 0; | |
400 | } | |
8a4b83cc CM |
401 | } |
402 | ||
403 | if (found_transid > fs_devices->latest_trans) { | |
404 | fs_devices->latest_devid = devid; | |
405 | fs_devices->latest_trans = found_transid; | |
406 | } | |
8a4b83cc CM |
407 | *fs_devices_ret = fs_devices; |
408 | return 0; | |
409 | } | |
410 | ||
e4404d6e YZ |
411 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
412 | { | |
413 | struct btrfs_fs_devices *fs_devices; | |
414 | struct btrfs_device *device; | |
415 | struct btrfs_device *orig_dev; | |
416 | ||
417 | fs_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); | |
418 | if (!fs_devices) | |
419 | return ERR_PTR(-ENOMEM); | |
420 | ||
421 | INIT_LIST_HEAD(&fs_devices->devices); | |
422 | INIT_LIST_HEAD(&fs_devices->alloc_list); | |
423 | INIT_LIST_HEAD(&fs_devices->list); | |
e5e9a520 | 424 | mutex_init(&fs_devices->device_list_mutex); |
e4404d6e YZ |
425 | fs_devices->latest_devid = orig->latest_devid; |
426 | fs_devices->latest_trans = orig->latest_trans; | |
427 | memcpy(fs_devices->fsid, orig->fsid, sizeof(fs_devices->fsid)); | |
428 | ||
46224705 | 429 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e YZ |
430 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
431 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
432 | if (!device) | |
433 | goto error; | |
434 | ||
435 | device->name = kstrdup(orig_dev->name, GFP_NOFS); | |
fd2696f3 JL |
436 | if (!device->name) { |
437 | kfree(device); | |
e4404d6e | 438 | goto error; |
fd2696f3 | 439 | } |
e4404d6e YZ |
440 | |
441 | device->devid = orig_dev->devid; | |
442 | device->work.func = pending_bios_fn; | |
443 | memcpy(device->uuid, orig_dev->uuid, sizeof(device->uuid)); | |
e4404d6e YZ |
444 | spin_lock_init(&device->io_lock); |
445 | INIT_LIST_HEAD(&device->dev_list); | |
446 | INIT_LIST_HEAD(&device->dev_alloc_list); | |
447 | ||
448 | list_add(&device->dev_list, &fs_devices->devices); | |
449 | device->fs_devices = fs_devices; | |
450 | fs_devices->num_devices++; | |
451 | } | |
452 | return fs_devices; | |
453 | error: | |
454 | free_fs_devices(fs_devices); | |
455 | return ERR_PTR(-ENOMEM); | |
456 | } | |
457 | ||
dfe25020 CM |
458 | int btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices) |
459 | { | |
c6e30871 | 460 | struct btrfs_device *device, *next; |
dfe25020 | 461 | |
a6b0d5c8 CM |
462 | struct block_device *latest_bdev = NULL; |
463 | u64 latest_devid = 0; | |
464 | u64 latest_transid = 0; | |
465 | ||
dfe25020 CM |
466 | mutex_lock(&uuid_mutex); |
467 | again: | |
46224705 | 468 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 469 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 CM |
470 | if (device->in_fs_metadata) { |
471 | if (!latest_transid || | |
472 | device->generation > latest_transid) { | |
473 | latest_devid = device->devid; | |
474 | latest_transid = device->generation; | |
475 | latest_bdev = device->bdev; | |
476 | } | |
2b82032c | 477 | continue; |
a6b0d5c8 | 478 | } |
2b82032c YZ |
479 | |
480 | if (device->bdev) { | |
d4d77629 | 481 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
482 | device->bdev = NULL; |
483 | fs_devices->open_devices--; | |
484 | } | |
485 | if (device->writeable) { | |
486 | list_del_init(&device->dev_alloc_list); | |
487 | device->writeable = 0; | |
488 | fs_devices->rw_devices--; | |
489 | } | |
e4404d6e YZ |
490 | list_del_init(&device->dev_list); |
491 | fs_devices->num_devices--; | |
492 | kfree(device->name); | |
493 | kfree(device); | |
dfe25020 | 494 | } |
2b82032c YZ |
495 | |
496 | if (fs_devices->seed) { | |
497 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
498 | goto again; |
499 | } | |
500 | ||
a6b0d5c8 CM |
501 | fs_devices->latest_bdev = latest_bdev; |
502 | fs_devices->latest_devid = latest_devid; | |
503 | fs_devices->latest_trans = latest_transid; | |
504 | ||
dfe25020 CM |
505 | mutex_unlock(&uuid_mutex); |
506 | return 0; | |
507 | } | |
a0af469b | 508 | |
1f78160c XG |
509 | static void __free_device(struct work_struct *work) |
510 | { | |
511 | struct btrfs_device *device; | |
512 | ||
513 | device = container_of(work, struct btrfs_device, rcu_work); | |
514 | ||
515 | if (device->bdev) | |
516 | blkdev_put(device->bdev, device->mode); | |
517 | ||
518 | kfree(device->name); | |
519 | kfree(device); | |
520 | } | |
521 | ||
522 | static void free_device(struct rcu_head *head) | |
523 | { | |
524 | struct btrfs_device *device; | |
525 | ||
526 | device = container_of(head, struct btrfs_device, rcu); | |
527 | ||
528 | INIT_WORK(&device->rcu_work, __free_device); | |
529 | schedule_work(&device->rcu_work); | |
530 | } | |
531 | ||
2b82032c | 532 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 533 | { |
8a4b83cc | 534 | struct btrfs_device *device; |
e4404d6e | 535 | |
2b82032c YZ |
536 | if (--fs_devices->opened > 0) |
537 | return 0; | |
8a4b83cc | 538 | |
c9513edb | 539 | mutex_lock(&fs_devices->device_list_mutex); |
c6e30871 | 540 | list_for_each_entry(device, &fs_devices->devices, dev_list) { |
1f78160c XG |
541 | struct btrfs_device *new_device; |
542 | ||
543 | if (device->bdev) | |
a0af469b | 544 | fs_devices->open_devices--; |
1f78160c | 545 | |
2b82032c YZ |
546 | if (device->writeable) { |
547 | list_del_init(&device->dev_alloc_list); | |
548 | fs_devices->rw_devices--; | |
549 | } | |
550 | ||
d5e2003c JB |
551 | if (device->can_discard) |
552 | fs_devices->num_can_discard--; | |
553 | ||
1f78160c XG |
554 | new_device = kmalloc(sizeof(*new_device), GFP_NOFS); |
555 | BUG_ON(!new_device); | |
556 | memcpy(new_device, device, sizeof(*new_device)); | |
557 | new_device->name = kstrdup(device->name, GFP_NOFS); | |
5f3f302a | 558 | BUG_ON(device->name && !new_device->name); |
1f78160c XG |
559 | new_device->bdev = NULL; |
560 | new_device->writeable = 0; | |
561 | new_device->in_fs_metadata = 0; | |
d5e2003c | 562 | new_device->can_discard = 0; |
1f78160c XG |
563 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
564 | ||
565 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 566 | } |
c9513edb XG |
567 | mutex_unlock(&fs_devices->device_list_mutex); |
568 | ||
e4404d6e YZ |
569 | WARN_ON(fs_devices->open_devices); |
570 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
571 | fs_devices->opened = 0; |
572 | fs_devices->seeding = 0; | |
2b82032c | 573 | |
8a4b83cc CM |
574 | return 0; |
575 | } | |
576 | ||
2b82032c YZ |
577 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
578 | { | |
e4404d6e | 579 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
580 | int ret; |
581 | ||
582 | mutex_lock(&uuid_mutex); | |
583 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
584 | if (!fs_devices->opened) { |
585 | seed_devices = fs_devices->seed; | |
586 | fs_devices->seed = NULL; | |
587 | } | |
2b82032c | 588 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
589 | |
590 | while (seed_devices) { | |
591 | fs_devices = seed_devices; | |
592 | seed_devices = fs_devices->seed; | |
593 | __btrfs_close_devices(fs_devices); | |
594 | free_fs_devices(fs_devices); | |
595 | } | |
2b82032c YZ |
596 | return ret; |
597 | } | |
598 | ||
e4404d6e YZ |
599 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
600 | fmode_t flags, void *holder) | |
8a4b83cc | 601 | { |
d5e2003c | 602 | struct request_queue *q; |
8a4b83cc CM |
603 | struct block_device *bdev; |
604 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 605 | struct btrfs_device *device; |
a0af469b CM |
606 | struct block_device *latest_bdev = NULL; |
607 | struct buffer_head *bh; | |
608 | struct btrfs_super_block *disk_super; | |
609 | u64 latest_devid = 0; | |
610 | u64 latest_transid = 0; | |
a0af469b | 611 | u64 devid; |
2b82032c | 612 | int seeding = 1; |
a0af469b | 613 | int ret = 0; |
8a4b83cc | 614 | |
d4d77629 TH |
615 | flags |= FMODE_EXCL; |
616 | ||
c6e30871 | 617 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
618 | if (device->bdev) |
619 | continue; | |
dfe25020 CM |
620 | if (!device->name) |
621 | continue; | |
622 | ||
d4d77629 | 623 | bdev = blkdev_get_by_path(device->name, flags, holder); |
8a4b83cc | 624 | if (IS_ERR(bdev)) { |
d397712b | 625 | printk(KERN_INFO "open %s failed\n", device->name); |
a0af469b | 626 | goto error; |
8a4b83cc | 627 | } |
a061fc8d | 628 | set_blocksize(bdev, 4096); |
a0af469b | 629 | |
a512bbf8 | 630 | bh = btrfs_read_dev_super(bdev); |
20bcd649 | 631 | if (!bh) |
a0af469b CM |
632 | goto error_close; |
633 | ||
634 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 635 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
636 | if (devid != device->devid) |
637 | goto error_brelse; | |
638 | ||
2b82032c YZ |
639 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
640 | BTRFS_UUID_SIZE)) | |
641 | goto error_brelse; | |
642 | ||
643 | device->generation = btrfs_super_generation(disk_super); | |
644 | if (!latest_transid || device->generation > latest_transid) { | |
a0af469b | 645 | latest_devid = devid; |
2b82032c | 646 | latest_transid = device->generation; |
a0af469b CM |
647 | latest_bdev = bdev; |
648 | } | |
649 | ||
2b82032c YZ |
650 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
651 | device->writeable = 0; | |
652 | } else { | |
653 | device->writeable = !bdev_read_only(bdev); | |
654 | seeding = 0; | |
655 | } | |
656 | ||
d5e2003c JB |
657 | q = bdev_get_queue(bdev); |
658 | if (blk_queue_discard(q)) { | |
659 | device->can_discard = 1; | |
660 | fs_devices->num_can_discard++; | |
661 | } | |
662 | ||
8a4b83cc | 663 | device->bdev = bdev; |
dfe25020 | 664 | device->in_fs_metadata = 0; |
15916de8 CM |
665 | device->mode = flags; |
666 | ||
c289811c CM |
667 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
668 | fs_devices->rotating = 1; | |
669 | ||
a0af469b | 670 | fs_devices->open_devices++; |
2b82032c YZ |
671 | if (device->writeable) { |
672 | fs_devices->rw_devices++; | |
673 | list_add(&device->dev_alloc_list, | |
674 | &fs_devices->alloc_list); | |
675 | } | |
4f6c9328 | 676 | brelse(bh); |
a0af469b | 677 | continue; |
a061fc8d | 678 | |
a0af469b CM |
679 | error_brelse: |
680 | brelse(bh); | |
681 | error_close: | |
d4d77629 | 682 | blkdev_put(bdev, flags); |
a0af469b CM |
683 | error: |
684 | continue; | |
8a4b83cc | 685 | } |
a0af469b | 686 | if (fs_devices->open_devices == 0) { |
20bcd649 | 687 | ret = -EINVAL; |
a0af469b CM |
688 | goto out; |
689 | } | |
2b82032c YZ |
690 | fs_devices->seeding = seeding; |
691 | fs_devices->opened = 1; | |
a0af469b CM |
692 | fs_devices->latest_bdev = latest_bdev; |
693 | fs_devices->latest_devid = latest_devid; | |
694 | fs_devices->latest_trans = latest_transid; | |
2b82032c | 695 | fs_devices->total_rw_bytes = 0; |
a0af469b | 696 | out: |
2b82032c YZ |
697 | return ret; |
698 | } | |
699 | ||
700 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 701 | fmode_t flags, void *holder) |
2b82032c YZ |
702 | { |
703 | int ret; | |
704 | ||
705 | mutex_lock(&uuid_mutex); | |
706 | if (fs_devices->opened) { | |
e4404d6e YZ |
707 | fs_devices->opened++; |
708 | ret = 0; | |
2b82032c | 709 | } else { |
15916de8 | 710 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 711 | } |
8a4b83cc | 712 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
713 | return ret; |
714 | } | |
715 | ||
97288f2c | 716 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
717 | struct btrfs_fs_devices **fs_devices_ret) |
718 | { | |
719 | struct btrfs_super_block *disk_super; | |
720 | struct block_device *bdev; | |
721 | struct buffer_head *bh; | |
722 | int ret; | |
723 | u64 devid; | |
f2984462 | 724 | u64 transid; |
8a4b83cc | 725 | |
d4d77629 TH |
726 | flags |= FMODE_EXCL; |
727 | bdev = blkdev_get_by_path(path, flags, holder); | |
8a4b83cc CM |
728 | |
729 | if (IS_ERR(bdev)) { | |
8a4b83cc CM |
730 | ret = PTR_ERR(bdev); |
731 | goto error; | |
732 | } | |
733 | ||
10f6327b | 734 | mutex_lock(&uuid_mutex); |
8a4b83cc CM |
735 | ret = set_blocksize(bdev, 4096); |
736 | if (ret) | |
737 | goto error_close; | |
a512bbf8 | 738 | bh = btrfs_read_dev_super(bdev); |
8a4b83cc | 739 | if (!bh) { |
20b45077 | 740 | ret = -EINVAL; |
8a4b83cc CM |
741 | goto error_close; |
742 | } | |
743 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 744 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 745 | transid = btrfs_super_generation(disk_super); |
7ae9c09d | 746 | if (disk_super->label[0]) |
d397712b | 747 | printk(KERN_INFO "device label %s ", disk_super->label); |
22b63a29 ID |
748 | else |
749 | printk(KERN_INFO "device fsid %pU ", disk_super->fsid); | |
119e10cf | 750 | printk(KERN_CONT "devid %llu transid %llu %s\n", |
d397712b | 751 | (unsigned long long)devid, (unsigned long long)transid, path); |
8a4b83cc CM |
752 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
753 | ||
8a4b83cc CM |
754 | brelse(bh); |
755 | error_close: | |
10f6327b | 756 | mutex_unlock(&uuid_mutex); |
d4d77629 | 757 | blkdev_put(bdev, flags); |
8a4b83cc | 758 | error: |
8a4b83cc CM |
759 | return ret; |
760 | } | |
0b86a832 | 761 | |
6d07bcec MX |
762 | /* helper to account the used device space in the range */ |
763 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
764 | u64 end, u64 *length) | |
765 | { | |
766 | struct btrfs_key key; | |
767 | struct btrfs_root *root = device->dev_root; | |
768 | struct btrfs_dev_extent *dev_extent; | |
769 | struct btrfs_path *path; | |
770 | u64 extent_end; | |
771 | int ret; | |
772 | int slot; | |
773 | struct extent_buffer *l; | |
774 | ||
775 | *length = 0; | |
776 | ||
777 | if (start >= device->total_bytes) | |
778 | return 0; | |
779 | ||
780 | path = btrfs_alloc_path(); | |
781 | if (!path) | |
782 | return -ENOMEM; | |
783 | path->reada = 2; | |
784 | ||
785 | key.objectid = device->devid; | |
786 | key.offset = start; | |
787 | key.type = BTRFS_DEV_EXTENT_KEY; | |
788 | ||
789 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
790 | if (ret < 0) | |
791 | goto out; | |
792 | if (ret > 0) { | |
793 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
794 | if (ret < 0) | |
795 | goto out; | |
796 | } | |
797 | ||
798 | while (1) { | |
799 | l = path->nodes[0]; | |
800 | slot = path->slots[0]; | |
801 | if (slot >= btrfs_header_nritems(l)) { | |
802 | ret = btrfs_next_leaf(root, path); | |
803 | if (ret == 0) | |
804 | continue; | |
805 | if (ret < 0) | |
806 | goto out; | |
807 | ||
808 | break; | |
809 | } | |
810 | btrfs_item_key_to_cpu(l, &key, slot); | |
811 | ||
812 | if (key.objectid < device->devid) | |
813 | goto next; | |
814 | ||
815 | if (key.objectid > device->devid) | |
816 | break; | |
817 | ||
818 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) | |
819 | goto next; | |
820 | ||
821 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
822 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
823 | dev_extent); | |
824 | if (key.offset <= start && extent_end > end) { | |
825 | *length = end - start + 1; | |
826 | break; | |
827 | } else if (key.offset <= start && extent_end > start) | |
828 | *length += extent_end - start; | |
829 | else if (key.offset > start && extent_end <= end) | |
830 | *length += extent_end - key.offset; | |
831 | else if (key.offset > start && key.offset <= end) { | |
832 | *length += end - key.offset + 1; | |
833 | break; | |
834 | } else if (key.offset > end) | |
835 | break; | |
836 | ||
837 | next: | |
838 | path->slots[0]++; | |
839 | } | |
840 | ret = 0; | |
841 | out: | |
842 | btrfs_free_path(path); | |
843 | return ret; | |
844 | } | |
845 | ||
0b86a832 | 846 | /* |
7bfc837d | 847 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
848 | * @device: the device which we search the free space in |
849 | * @num_bytes: the size of the free space that we need | |
850 | * @start: store the start of the free space. | |
851 | * @len: the size of the free space. that we find, or the size of the max | |
852 | * free space if we don't find suitable free space | |
853 | * | |
0b86a832 CM |
854 | * this uses a pretty simple search, the expectation is that it is |
855 | * called very infrequently and that a given device has a small number | |
856 | * of extents | |
7bfc837d MX |
857 | * |
858 | * @start is used to store the start of the free space if we find. But if we | |
859 | * don't find suitable free space, it will be used to store the start position | |
860 | * of the max free space. | |
861 | * | |
862 | * @len is used to store the size of the free space that we find. | |
863 | * But if we don't find suitable free space, it is used to store the size of | |
864 | * the max free space. | |
0b86a832 | 865 | */ |
125ccb0a | 866 | int find_free_dev_extent(struct btrfs_device *device, u64 num_bytes, |
7bfc837d | 867 | u64 *start, u64 *len) |
0b86a832 CM |
868 | { |
869 | struct btrfs_key key; | |
870 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 871 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 872 | struct btrfs_path *path; |
7bfc837d MX |
873 | u64 hole_size; |
874 | u64 max_hole_start; | |
875 | u64 max_hole_size; | |
876 | u64 extent_end; | |
877 | u64 search_start; | |
0b86a832 CM |
878 | u64 search_end = device->total_bytes; |
879 | int ret; | |
7bfc837d | 880 | int slot; |
0b86a832 CM |
881 | struct extent_buffer *l; |
882 | ||
0b86a832 CM |
883 | /* FIXME use last free of some kind */ |
884 | ||
8a4b83cc CM |
885 | /* we don't want to overwrite the superblock on the drive, |
886 | * so we make sure to start at an offset of at least 1MB | |
887 | */ | |
a9c9bf68 | 888 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 889 | |
7bfc837d MX |
890 | max_hole_start = search_start; |
891 | max_hole_size = 0; | |
38c01b96 | 892 | hole_size = 0; |
7bfc837d MX |
893 | |
894 | if (search_start >= search_end) { | |
895 | ret = -ENOSPC; | |
896 | goto error; | |
897 | } | |
898 | ||
899 | path = btrfs_alloc_path(); | |
900 | if (!path) { | |
901 | ret = -ENOMEM; | |
902 | goto error; | |
903 | } | |
904 | path->reada = 2; | |
905 | ||
0b86a832 CM |
906 | key.objectid = device->devid; |
907 | key.offset = search_start; | |
908 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 909 | |
125ccb0a | 910 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 911 | if (ret < 0) |
7bfc837d | 912 | goto out; |
1fcbac58 YZ |
913 | if (ret > 0) { |
914 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
915 | if (ret < 0) | |
7bfc837d | 916 | goto out; |
1fcbac58 | 917 | } |
7bfc837d | 918 | |
0b86a832 CM |
919 | while (1) { |
920 | l = path->nodes[0]; | |
921 | slot = path->slots[0]; | |
922 | if (slot >= btrfs_header_nritems(l)) { | |
923 | ret = btrfs_next_leaf(root, path); | |
924 | if (ret == 0) | |
925 | continue; | |
926 | if (ret < 0) | |
7bfc837d MX |
927 | goto out; |
928 | ||
929 | break; | |
0b86a832 CM |
930 | } |
931 | btrfs_item_key_to_cpu(l, &key, slot); | |
932 | ||
933 | if (key.objectid < device->devid) | |
934 | goto next; | |
935 | ||
936 | if (key.objectid > device->devid) | |
7bfc837d | 937 | break; |
0b86a832 | 938 | |
7bfc837d MX |
939 | if (btrfs_key_type(&key) != BTRFS_DEV_EXTENT_KEY) |
940 | goto next; | |
9779b72f | 941 | |
7bfc837d MX |
942 | if (key.offset > search_start) { |
943 | hole_size = key.offset - search_start; | |
9779b72f | 944 | |
7bfc837d MX |
945 | if (hole_size > max_hole_size) { |
946 | max_hole_start = search_start; | |
947 | max_hole_size = hole_size; | |
948 | } | |
9779b72f | 949 | |
7bfc837d MX |
950 | /* |
951 | * If this free space is greater than which we need, | |
952 | * it must be the max free space that we have found | |
953 | * until now, so max_hole_start must point to the start | |
954 | * of this free space and the length of this free space | |
955 | * is stored in max_hole_size. Thus, we return | |
956 | * max_hole_start and max_hole_size and go back to the | |
957 | * caller. | |
958 | */ | |
959 | if (hole_size >= num_bytes) { | |
960 | ret = 0; | |
961 | goto out; | |
0b86a832 CM |
962 | } |
963 | } | |
0b86a832 | 964 | |
0b86a832 | 965 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
966 | extent_end = key.offset + btrfs_dev_extent_length(l, |
967 | dev_extent); | |
968 | if (extent_end > search_start) | |
969 | search_start = extent_end; | |
0b86a832 CM |
970 | next: |
971 | path->slots[0]++; | |
972 | cond_resched(); | |
973 | } | |
0b86a832 | 974 | |
38c01b96 | 975 | /* |
976 | * At this point, search_start should be the end of | |
977 | * allocated dev extents, and when shrinking the device, | |
978 | * search_end may be smaller than search_start. | |
979 | */ | |
980 | if (search_end > search_start) | |
981 | hole_size = search_end - search_start; | |
982 | ||
7bfc837d MX |
983 | if (hole_size > max_hole_size) { |
984 | max_hole_start = search_start; | |
985 | max_hole_size = hole_size; | |
0b86a832 | 986 | } |
0b86a832 | 987 | |
7bfc837d MX |
988 | /* See above. */ |
989 | if (hole_size < num_bytes) | |
990 | ret = -ENOSPC; | |
991 | else | |
992 | ret = 0; | |
993 | ||
994 | out: | |
2b82032c | 995 | btrfs_free_path(path); |
7bfc837d MX |
996 | error: |
997 | *start = max_hole_start; | |
b2117a39 | 998 | if (len) |
7bfc837d | 999 | *len = max_hole_size; |
0b86a832 CM |
1000 | return ret; |
1001 | } | |
1002 | ||
b2950863 | 1003 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1004 | struct btrfs_device *device, |
1005 | u64 start) | |
1006 | { | |
1007 | int ret; | |
1008 | struct btrfs_path *path; | |
1009 | struct btrfs_root *root = device->dev_root; | |
1010 | struct btrfs_key key; | |
a061fc8d CM |
1011 | struct btrfs_key found_key; |
1012 | struct extent_buffer *leaf = NULL; | |
1013 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1014 | |
1015 | path = btrfs_alloc_path(); | |
1016 | if (!path) | |
1017 | return -ENOMEM; | |
1018 | ||
1019 | key.objectid = device->devid; | |
1020 | key.offset = start; | |
1021 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1022 | again: |
8f18cf13 | 1023 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1024 | if (ret > 0) { |
1025 | ret = btrfs_previous_item(root, path, key.objectid, | |
1026 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1027 | if (ret) |
1028 | goto out; | |
a061fc8d CM |
1029 | leaf = path->nodes[0]; |
1030 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1031 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1032 | struct btrfs_dev_extent); | |
1033 | BUG_ON(found_key.offset > start || found_key.offset + | |
1034 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1035 | key = found_key; |
1036 | btrfs_release_path(path); | |
1037 | goto again; | |
a061fc8d CM |
1038 | } else if (ret == 0) { |
1039 | leaf = path->nodes[0]; | |
1040 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1041 | struct btrfs_dev_extent); | |
1042 | } | |
8f18cf13 CM |
1043 | BUG_ON(ret); |
1044 | ||
2bf64758 JB |
1045 | if (device->bytes_used > 0) { |
1046 | u64 len = btrfs_dev_extent_length(leaf, extent); | |
1047 | device->bytes_used -= len; | |
1048 | spin_lock(&root->fs_info->free_chunk_lock); | |
1049 | root->fs_info->free_chunk_space += len; | |
1050 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1051 | } | |
8f18cf13 | 1052 | ret = btrfs_del_item(trans, root, path); |
8f18cf13 | 1053 | |
b0b802d7 | 1054 | out: |
8f18cf13 CM |
1055 | btrfs_free_path(path); |
1056 | return ret; | |
1057 | } | |
1058 | ||
2b82032c | 1059 | int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
0b86a832 | 1060 | struct btrfs_device *device, |
e17cade2 | 1061 | u64 chunk_tree, u64 chunk_objectid, |
2b82032c | 1062 | u64 chunk_offset, u64 start, u64 num_bytes) |
0b86a832 CM |
1063 | { |
1064 | int ret; | |
1065 | struct btrfs_path *path; | |
1066 | struct btrfs_root *root = device->dev_root; | |
1067 | struct btrfs_dev_extent *extent; | |
1068 | struct extent_buffer *leaf; | |
1069 | struct btrfs_key key; | |
1070 | ||
dfe25020 | 1071 | WARN_ON(!device->in_fs_metadata); |
0b86a832 CM |
1072 | path = btrfs_alloc_path(); |
1073 | if (!path) | |
1074 | return -ENOMEM; | |
1075 | ||
0b86a832 | 1076 | key.objectid = device->devid; |
2b82032c | 1077 | key.offset = start; |
0b86a832 CM |
1078 | key.type = BTRFS_DEV_EXTENT_KEY; |
1079 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1080 | sizeof(*extent)); | |
1081 | BUG_ON(ret); | |
1082 | ||
1083 | leaf = path->nodes[0]; | |
1084 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1085 | struct btrfs_dev_extent); | |
e17cade2 CM |
1086 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1087 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1088 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1089 | ||
1090 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1091 | (unsigned long)btrfs_dev_extent_chunk_tree_uuid(extent), | |
1092 | BTRFS_UUID_SIZE); | |
1093 | ||
0b86a832 CM |
1094 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1095 | btrfs_mark_buffer_dirty(leaf); | |
0b86a832 CM |
1096 | btrfs_free_path(path); |
1097 | return ret; | |
1098 | } | |
1099 | ||
a1b32a59 CM |
1100 | static noinline int find_next_chunk(struct btrfs_root *root, |
1101 | u64 objectid, u64 *offset) | |
0b86a832 CM |
1102 | { |
1103 | struct btrfs_path *path; | |
1104 | int ret; | |
1105 | struct btrfs_key key; | |
e17cade2 | 1106 | struct btrfs_chunk *chunk; |
0b86a832 CM |
1107 | struct btrfs_key found_key; |
1108 | ||
1109 | path = btrfs_alloc_path(); | |
92b8e897 MF |
1110 | if (!path) |
1111 | return -ENOMEM; | |
0b86a832 | 1112 | |
e17cade2 | 1113 | key.objectid = objectid; |
0b86a832 CM |
1114 | key.offset = (u64)-1; |
1115 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1116 | ||
1117 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1118 | if (ret < 0) | |
1119 | goto error; | |
1120 | ||
1121 | BUG_ON(ret == 0); | |
1122 | ||
1123 | ret = btrfs_previous_item(root, path, 0, BTRFS_CHUNK_ITEM_KEY); | |
1124 | if (ret) { | |
e17cade2 | 1125 | *offset = 0; |
0b86a832 CM |
1126 | } else { |
1127 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1128 | path->slots[0]); | |
e17cade2 CM |
1129 | if (found_key.objectid != objectid) |
1130 | *offset = 0; | |
1131 | else { | |
1132 | chunk = btrfs_item_ptr(path->nodes[0], path->slots[0], | |
1133 | struct btrfs_chunk); | |
1134 | *offset = found_key.offset + | |
1135 | btrfs_chunk_length(path->nodes[0], chunk); | |
1136 | } | |
0b86a832 CM |
1137 | } |
1138 | ret = 0; | |
1139 | error: | |
1140 | btrfs_free_path(path); | |
1141 | return ret; | |
1142 | } | |
1143 | ||
2b82032c | 1144 | static noinline int find_next_devid(struct btrfs_root *root, u64 *objectid) |
0b86a832 CM |
1145 | { |
1146 | int ret; | |
1147 | struct btrfs_key key; | |
1148 | struct btrfs_key found_key; | |
2b82032c YZ |
1149 | struct btrfs_path *path; |
1150 | ||
1151 | root = root->fs_info->chunk_root; | |
1152 | ||
1153 | path = btrfs_alloc_path(); | |
1154 | if (!path) | |
1155 | return -ENOMEM; | |
0b86a832 CM |
1156 | |
1157 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1158 | key.type = BTRFS_DEV_ITEM_KEY; | |
1159 | key.offset = (u64)-1; | |
1160 | ||
1161 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1162 | if (ret < 0) | |
1163 | goto error; | |
1164 | ||
1165 | BUG_ON(ret == 0); | |
1166 | ||
1167 | ret = btrfs_previous_item(root, path, BTRFS_DEV_ITEMS_OBJECTID, | |
1168 | BTRFS_DEV_ITEM_KEY); | |
1169 | if (ret) { | |
1170 | *objectid = 1; | |
1171 | } else { | |
1172 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1173 | path->slots[0]); | |
1174 | *objectid = found_key.offset + 1; | |
1175 | } | |
1176 | ret = 0; | |
1177 | error: | |
2b82032c | 1178 | btrfs_free_path(path); |
0b86a832 CM |
1179 | return ret; |
1180 | } | |
1181 | ||
1182 | /* | |
1183 | * the device information is stored in the chunk root | |
1184 | * the btrfs_device struct should be fully filled in | |
1185 | */ | |
1186 | int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1187 | struct btrfs_root *root, | |
1188 | struct btrfs_device *device) | |
1189 | { | |
1190 | int ret; | |
1191 | struct btrfs_path *path; | |
1192 | struct btrfs_dev_item *dev_item; | |
1193 | struct extent_buffer *leaf; | |
1194 | struct btrfs_key key; | |
1195 | unsigned long ptr; | |
0b86a832 CM |
1196 | |
1197 | root = root->fs_info->chunk_root; | |
1198 | ||
1199 | path = btrfs_alloc_path(); | |
1200 | if (!path) | |
1201 | return -ENOMEM; | |
1202 | ||
0b86a832 CM |
1203 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1204 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1205 | key.offset = device->devid; |
0b86a832 CM |
1206 | |
1207 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1208 | sizeof(*dev_item)); |
0b86a832 CM |
1209 | if (ret) |
1210 | goto out; | |
1211 | ||
1212 | leaf = path->nodes[0]; | |
1213 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1214 | ||
1215 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1216 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1217 | btrfs_set_device_type(leaf, dev_item, device->type); |
1218 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1219 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1220 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
0b86a832 CM |
1221 | btrfs_set_device_total_bytes(leaf, dev_item, device->total_bytes); |
1222 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); | |
e17cade2 CM |
1223 | btrfs_set_device_group(leaf, dev_item, 0); |
1224 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1225 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1226 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1227 | |
0b86a832 | 1228 | ptr = (unsigned long)btrfs_device_uuid(dev_item); |
e17cade2 | 1229 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
2b82032c YZ |
1230 | ptr = (unsigned long)btrfs_device_fsid(dev_item); |
1231 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
0b86a832 | 1232 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1233 | |
2b82032c | 1234 | ret = 0; |
0b86a832 CM |
1235 | out: |
1236 | btrfs_free_path(path); | |
1237 | return ret; | |
1238 | } | |
8f18cf13 | 1239 | |
a061fc8d CM |
1240 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1241 | struct btrfs_device *device) | |
1242 | { | |
1243 | int ret; | |
1244 | struct btrfs_path *path; | |
a061fc8d | 1245 | struct btrfs_key key; |
a061fc8d CM |
1246 | struct btrfs_trans_handle *trans; |
1247 | ||
1248 | root = root->fs_info->chunk_root; | |
1249 | ||
1250 | path = btrfs_alloc_path(); | |
1251 | if (!path) | |
1252 | return -ENOMEM; | |
1253 | ||
a22285a6 | 1254 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1255 | if (IS_ERR(trans)) { |
1256 | btrfs_free_path(path); | |
1257 | return PTR_ERR(trans); | |
1258 | } | |
a061fc8d CM |
1259 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1260 | key.type = BTRFS_DEV_ITEM_KEY; | |
1261 | key.offset = device->devid; | |
7d9eb12c | 1262 | lock_chunks(root); |
a061fc8d CM |
1263 | |
1264 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1265 | if (ret < 0) | |
1266 | goto out; | |
1267 | ||
1268 | if (ret > 0) { | |
1269 | ret = -ENOENT; | |
1270 | goto out; | |
1271 | } | |
1272 | ||
1273 | ret = btrfs_del_item(trans, root, path); | |
1274 | if (ret) | |
1275 | goto out; | |
a061fc8d CM |
1276 | out: |
1277 | btrfs_free_path(path); | |
7d9eb12c | 1278 | unlock_chunks(root); |
a061fc8d CM |
1279 | btrfs_commit_transaction(trans, root); |
1280 | return ret; | |
1281 | } | |
1282 | ||
1283 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1284 | { | |
1285 | struct btrfs_device *device; | |
2b82032c | 1286 | struct btrfs_device *next_device; |
a061fc8d | 1287 | struct block_device *bdev; |
dfe25020 | 1288 | struct buffer_head *bh = NULL; |
a061fc8d | 1289 | struct btrfs_super_block *disk_super; |
1f78160c | 1290 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1291 | u64 all_avail; |
1292 | u64 devid; | |
2b82032c YZ |
1293 | u64 num_devices; |
1294 | u8 *dev_uuid; | |
a061fc8d | 1295 | int ret = 0; |
1f78160c | 1296 | bool clear_super = false; |
a061fc8d | 1297 | |
a061fc8d CM |
1298 | mutex_lock(&uuid_mutex); |
1299 | ||
1300 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1301 | root->fs_info->avail_system_alloc_bits | | |
1302 | root->fs_info->avail_metadata_alloc_bits; | |
1303 | ||
1304 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && | |
035fe03a | 1305 | root->fs_info->fs_devices->num_devices <= 4) { |
d397712b CM |
1306 | printk(KERN_ERR "btrfs: unable to go below four devices " |
1307 | "on raid10\n"); | |
a061fc8d CM |
1308 | ret = -EINVAL; |
1309 | goto out; | |
1310 | } | |
1311 | ||
1312 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && | |
035fe03a | 1313 | root->fs_info->fs_devices->num_devices <= 2) { |
d397712b CM |
1314 | printk(KERN_ERR "btrfs: unable to go below two " |
1315 | "devices on raid1\n"); | |
a061fc8d CM |
1316 | ret = -EINVAL; |
1317 | goto out; | |
1318 | } | |
1319 | ||
dfe25020 | 1320 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1321 | struct list_head *devices; |
1322 | struct btrfs_device *tmp; | |
a061fc8d | 1323 | |
dfe25020 CM |
1324 | device = NULL; |
1325 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1326 | /* |
1327 | * It is safe to read the devices since the volume_mutex | |
1328 | * is held. | |
1329 | */ | |
c6e30871 | 1330 | list_for_each_entry(tmp, devices, dev_list) { |
dfe25020 CM |
1331 | if (tmp->in_fs_metadata && !tmp->bdev) { |
1332 | device = tmp; | |
1333 | break; | |
1334 | } | |
1335 | } | |
1336 | bdev = NULL; | |
1337 | bh = NULL; | |
1338 | disk_super = NULL; | |
1339 | if (!device) { | |
d397712b CM |
1340 | printk(KERN_ERR "btrfs: no missing devices found to " |
1341 | "remove\n"); | |
dfe25020 CM |
1342 | goto out; |
1343 | } | |
dfe25020 | 1344 | } else { |
d4d77629 TH |
1345 | bdev = blkdev_get_by_path(device_path, FMODE_READ | FMODE_EXCL, |
1346 | root->fs_info->bdev_holder); | |
dfe25020 CM |
1347 | if (IS_ERR(bdev)) { |
1348 | ret = PTR_ERR(bdev); | |
1349 | goto out; | |
1350 | } | |
a061fc8d | 1351 | |
2b82032c | 1352 | set_blocksize(bdev, 4096); |
a512bbf8 | 1353 | bh = btrfs_read_dev_super(bdev); |
dfe25020 | 1354 | if (!bh) { |
20b45077 | 1355 | ret = -EINVAL; |
dfe25020 CM |
1356 | goto error_close; |
1357 | } | |
1358 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 1359 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c YZ |
1360 | dev_uuid = disk_super->dev_item.uuid; |
1361 | device = btrfs_find_device(root, devid, dev_uuid, | |
1362 | disk_super->fsid); | |
dfe25020 CM |
1363 | if (!device) { |
1364 | ret = -ENOENT; | |
1365 | goto error_brelse; | |
1366 | } | |
2b82032c | 1367 | } |
dfe25020 | 1368 | |
2b82032c | 1369 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
d397712b CM |
1370 | printk(KERN_ERR "btrfs: unable to remove the only writeable " |
1371 | "device\n"); | |
2b82032c YZ |
1372 | ret = -EINVAL; |
1373 | goto error_brelse; | |
1374 | } | |
1375 | ||
1376 | if (device->writeable) { | |
0c1daee0 | 1377 | lock_chunks(root); |
2b82032c | 1378 | list_del_init(&device->dev_alloc_list); |
0c1daee0 | 1379 | unlock_chunks(root); |
2b82032c | 1380 | root->fs_info->fs_devices->rw_devices--; |
1f78160c | 1381 | clear_super = true; |
dfe25020 | 1382 | } |
a061fc8d CM |
1383 | |
1384 | ret = btrfs_shrink_device(device, 0); | |
1385 | if (ret) | |
9b3517e9 | 1386 | goto error_undo; |
a061fc8d | 1387 | |
a061fc8d CM |
1388 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1389 | if (ret) | |
9b3517e9 | 1390 | goto error_undo; |
a061fc8d | 1391 | |
2bf64758 JB |
1392 | spin_lock(&root->fs_info->free_chunk_lock); |
1393 | root->fs_info->free_chunk_space = device->total_bytes - | |
1394 | device->bytes_used; | |
1395 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1396 | ||
2b82032c | 1397 | device->in_fs_metadata = 0; |
a2de733c | 1398 | btrfs_scrub_cancel_dev(root, device); |
e5e9a520 CM |
1399 | |
1400 | /* | |
1401 | * the device list mutex makes sure that we don't change | |
1402 | * the device list while someone else is writing out all | |
1403 | * the device supers. | |
1404 | */ | |
1f78160c XG |
1405 | |
1406 | cur_devices = device->fs_devices; | |
e5e9a520 | 1407 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1408 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1409 | |
e4404d6e | 1410 | device->fs_devices->num_devices--; |
2b82032c | 1411 | |
cd02dca5 CM |
1412 | if (device->missing) |
1413 | root->fs_info->fs_devices->missing_devices--; | |
1414 | ||
2b82032c YZ |
1415 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1416 | struct btrfs_device, dev_list); | |
1417 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1418 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1419 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1420 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1421 | ||
1f78160c | 1422 | if (device->bdev) |
e4404d6e | 1423 | device->fs_devices->open_devices--; |
1f78160c XG |
1424 | |
1425 | call_rcu(&device->rcu, free_device); | |
1426 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
e4404d6e | 1427 | |
6c41761f DS |
1428 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1429 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
2b82032c | 1430 | |
1f78160c | 1431 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1432 | struct btrfs_fs_devices *fs_devices; |
1433 | fs_devices = root->fs_info->fs_devices; | |
1434 | while (fs_devices) { | |
1f78160c | 1435 | if (fs_devices->seed == cur_devices) |
e4404d6e YZ |
1436 | break; |
1437 | fs_devices = fs_devices->seed; | |
2b82032c | 1438 | } |
1f78160c XG |
1439 | fs_devices->seed = cur_devices->seed; |
1440 | cur_devices->seed = NULL; | |
0c1daee0 | 1441 | lock_chunks(root); |
1f78160c | 1442 | __btrfs_close_devices(cur_devices); |
0c1daee0 | 1443 | unlock_chunks(root); |
1f78160c | 1444 | free_fs_devices(cur_devices); |
2b82032c YZ |
1445 | } |
1446 | ||
1447 | /* | |
1448 | * at this point, the device is zero sized. We want to | |
1449 | * remove it from the devices list and zero out the old super | |
1450 | */ | |
1f78160c | 1451 | if (clear_super) { |
dfe25020 CM |
1452 | /* make sure this device isn't detected as part of |
1453 | * the FS anymore | |
1454 | */ | |
1455 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1456 | set_buffer_dirty(bh); | |
1457 | sync_dirty_buffer(bh); | |
dfe25020 | 1458 | } |
a061fc8d | 1459 | |
a061fc8d | 1460 | ret = 0; |
a061fc8d CM |
1461 | |
1462 | error_brelse: | |
1463 | brelse(bh); | |
1464 | error_close: | |
dfe25020 | 1465 | if (bdev) |
e525fd89 | 1466 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1467 | out: |
1468 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1469 | return ret; |
9b3517e9 ID |
1470 | error_undo: |
1471 | if (device->writeable) { | |
0c1daee0 | 1472 | lock_chunks(root); |
9b3517e9 ID |
1473 | list_add(&device->dev_alloc_list, |
1474 | &root->fs_info->fs_devices->alloc_list); | |
0c1daee0 | 1475 | unlock_chunks(root); |
9b3517e9 ID |
1476 | root->fs_info->fs_devices->rw_devices++; |
1477 | } | |
1478 | goto error_brelse; | |
a061fc8d CM |
1479 | } |
1480 | ||
2b82032c YZ |
1481 | /* |
1482 | * does all the dirty work required for changing file system's UUID. | |
1483 | */ | |
125ccb0a | 1484 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
1485 | { |
1486 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
1487 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 1488 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 1489 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
1490 | struct btrfs_device *device; |
1491 | u64 super_flags; | |
1492 | ||
1493 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 1494 | if (!fs_devices->seeding) |
2b82032c YZ |
1495 | return -EINVAL; |
1496 | ||
e4404d6e YZ |
1497 | seed_devices = kzalloc(sizeof(*fs_devices), GFP_NOFS); |
1498 | if (!seed_devices) | |
2b82032c YZ |
1499 | return -ENOMEM; |
1500 | ||
e4404d6e YZ |
1501 | old_devices = clone_fs_devices(fs_devices); |
1502 | if (IS_ERR(old_devices)) { | |
1503 | kfree(seed_devices); | |
1504 | return PTR_ERR(old_devices); | |
2b82032c | 1505 | } |
e4404d6e | 1506 | |
2b82032c YZ |
1507 | list_add(&old_devices->list, &fs_uuids); |
1508 | ||
e4404d6e YZ |
1509 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
1510 | seed_devices->opened = 1; | |
1511 | INIT_LIST_HEAD(&seed_devices->devices); | |
1512 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 1513 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
1514 | |
1515 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
1516 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
1517 | synchronize_rcu); | |
c9513edb XG |
1518 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
1519 | ||
e4404d6e YZ |
1520 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
1521 | list_for_each_entry(device, &seed_devices->devices, dev_list) { | |
1522 | device->fs_devices = seed_devices; | |
1523 | } | |
1524 | ||
2b82032c YZ |
1525 | fs_devices->seeding = 0; |
1526 | fs_devices->num_devices = 0; | |
1527 | fs_devices->open_devices = 0; | |
e4404d6e | 1528 | fs_devices->seed = seed_devices; |
2b82032c YZ |
1529 | |
1530 | generate_random_uuid(fs_devices->fsid); | |
1531 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1532 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
1533 | super_flags = btrfs_super_flags(disk_super) & | |
1534 | ~BTRFS_SUPER_FLAG_SEEDING; | |
1535 | btrfs_set_super_flags(disk_super, super_flags); | |
1536 | ||
1537 | return 0; | |
1538 | } | |
1539 | ||
1540 | /* | |
1541 | * strore the expected generation for seed devices in device items. | |
1542 | */ | |
1543 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
1544 | struct btrfs_root *root) | |
1545 | { | |
1546 | struct btrfs_path *path; | |
1547 | struct extent_buffer *leaf; | |
1548 | struct btrfs_dev_item *dev_item; | |
1549 | struct btrfs_device *device; | |
1550 | struct btrfs_key key; | |
1551 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
1552 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
1553 | u64 devid; | |
1554 | int ret; | |
1555 | ||
1556 | path = btrfs_alloc_path(); | |
1557 | if (!path) | |
1558 | return -ENOMEM; | |
1559 | ||
1560 | root = root->fs_info->chunk_root; | |
1561 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1562 | key.offset = 0; | |
1563 | key.type = BTRFS_DEV_ITEM_KEY; | |
1564 | ||
1565 | while (1) { | |
1566 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1567 | if (ret < 0) | |
1568 | goto error; | |
1569 | ||
1570 | leaf = path->nodes[0]; | |
1571 | next_slot: | |
1572 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
1573 | ret = btrfs_next_leaf(root, path); | |
1574 | if (ret > 0) | |
1575 | break; | |
1576 | if (ret < 0) | |
1577 | goto error; | |
1578 | leaf = path->nodes[0]; | |
1579 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 1580 | btrfs_release_path(path); |
2b82032c YZ |
1581 | continue; |
1582 | } | |
1583 | ||
1584 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
1585 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
1586 | key.type != BTRFS_DEV_ITEM_KEY) | |
1587 | break; | |
1588 | ||
1589 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
1590 | struct btrfs_dev_item); | |
1591 | devid = btrfs_device_id(leaf, dev_item); | |
1592 | read_extent_buffer(leaf, dev_uuid, | |
1593 | (unsigned long)btrfs_device_uuid(dev_item), | |
1594 | BTRFS_UUID_SIZE); | |
1595 | read_extent_buffer(leaf, fs_uuid, | |
1596 | (unsigned long)btrfs_device_fsid(dev_item), | |
1597 | BTRFS_UUID_SIZE); | |
1598 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
1599 | BUG_ON(!device); | |
1600 | ||
1601 | if (device->fs_devices->seeding) { | |
1602 | btrfs_set_device_generation(leaf, dev_item, | |
1603 | device->generation); | |
1604 | btrfs_mark_buffer_dirty(leaf); | |
1605 | } | |
1606 | ||
1607 | path->slots[0]++; | |
1608 | goto next_slot; | |
1609 | } | |
1610 | ret = 0; | |
1611 | error: | |
1612 | btrfs_free_path(path); | |
1613 | return ret; | |
1614 | } | |
1615 | ||
788f20eb CM |
1616 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
1617 | { | |
d5e2003c | 1618 | struct request_queue *q; |
788f20eb CM |
1619 | struct btrfs_trans_handle *trans; |
1620 | struct btrfs_device *device; | |
1621 | struct block_device *bdev; | |
788f20eb | 1622 | struct list_head *devices; |
2b82032c | 1623 | struct super_block *sb = root->fs_info->sb; |
788f20eb | 1624 | u64 total_bytes; |
2b82032c | 1625 | int seeding_dev = 0; |
788f20eb CM |
1626 | int ret = 0; |
1627 | ||
2b82032c YZ |
1628 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
1629 | return -EINVAL; | |
788f20eb | 1630 | |
a5d16333 | 1631 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 1632 | root->fs_info->bdev_holder); |
7f59203a JB |
1633 | if (IS_ERR(bdev)) |
1634 | return PTR_ERR(bdev); | |
a2135011 | 1635 | |
2b82032c YZ |
1636 | if (root->fs_info->fs_devices->seeding) { |
1637 | seeding_dev = 1; | |
1638 | down_write(&sb->s_umount); | |
1639 | mutex_lock(&uuid_mutex); | |
1640 | } | |
1641 | ||
8c8bee1d | 1642 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 1643 | |
788f20eb | 1644 | devices = &root->fs_info->fs_devices->devices; |
e5e9a520 CM |
1645 | /* |
1646 | * we have the volume lock, so we don't need the extra | |
1647 | * device list mutex while reading the list here. | |
1648 | */ | |
c6e30871 | 1649 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
1650 | if (device->bdev == bdev) { |
1651 | ret = -EEXIST; | |
2b82032c | 1652 | goto error; |
788f20eb CM |
1653 | } |
1654 | } | |
1655 | ||
1656 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
1657 | if (!device) { | |
1658 | /* we can safely leave the fs_devices entry around */ | |
1659 | ret = -ENOMEM; | |
2b82032c | 1660 | goto error; |
788f20eb CM |
1661 | } |
1662 | ||
788f20eb CM |
1663 | device->name = kstrdup(device_path, GFP_NOFS); |
1664 | if (!device->name) { | |
1665 | kfree(device); | |
2b82032c YZ |
1666 | ret = -ENOMEM; |
1667 | goto error; | |
788f20eb | 1668 | } |
2b82032c YZ |
1669 | |
1670 | ret = find_next_devid(root, &device->devid); | |
1671 | if (ret) { | |
67100f25 | 1672 | kfree(device->name); |
2b82032c YZ |
1673 | kfree(device); |
1674 | goto error; | |
1675 | } | |
1676 | ||
a22285a6 | 1677 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1678 | if (IS_ERR(trans)) { |
67100f25 | 1679 | kfree(device->name); |
98d5dc13 TI |
1680 | kfree(device); |
1681 | ret = PTR_ERR(trans); | |
1682 | goto error; | |
1683 | } | |
1684 | ||
2b82032c YZ |
1685 | lock_chunks(root); |
1686 | ||
d5e2003c JB |
1687 | q = bdev_get_queue(bdev); |
1688 | if (blk_queue_discard(q)) | |
1689 | device->can_discard = 1; | |
2b82032c YZ |
1690 | device->writeable = 1; |
1691 | device->work.func = pending_bios_fn; | |
1692 | generate_random_uuid(device->uuid); | |
1693 | spin_lock_init(&device->io_lock); | |
1694 | device->generation = trans->transid; | |
788f20eb CM |
1695 | device->io_width = root->sectorsize; |
1696 | device->io_align = root->sectorsize; | |
1697 | device->sector_size = root->sectorsize; | |
1698 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 1699 | device->disk_total_bytes = device->total_bytes; |
788f20eb CM |
1700 | device->dev_root = root->fs_info->dev_root; |
1701 | device->bdev = bdev; | |
dfe25020 | 1702 | device->in_fs_metadata = 1; |
fb01aa85 | 1703 | device->mode = FMODE_EXCL; |
2b82032c | 1704 | set_blocksize(device->bdev, 4096); |
788f20eb | 1705 | |
2b82032c YZ |
1706 | if (seeding_dev) { |
1707 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 1708 | ret = btrfs_prepare_sprout(root); |
2b82032c YZ |
1709 | BUG_ON(ret); |
1710 | } | |
788f20eb | 1711 | |
2b82032c | 1712 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 CM |
1713 | |
1714 | /* | |
1715 | * we don't want write_supers to jump in here with our device | |
1716 | * half setup | |
1717 | */ | |
1718 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c | 1719 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
1720 | list_add(&device->dev_alloc_list, |
1721 | &root->fs_info->fs_devices->alloc_list); | |
1722 | root->fs_info->fs_devices->num_devices++; | |
1723 | root->fs_info->fs_devices->open_devices++; | |
1724 | root->fs_info->fs_devices->rw_devices++; | |
d5e2003c JB |
1725 | if (device->can_discard) |
1726 | root->fs_info->fs_devices->num_can_discard++; | |
2b82032c | 1727 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 1728 | |
2bf64758 JB |
1729 | spin_lock(&root->fs_info->free_chunk_lock); |
1730 | root->fs_info->free_chunk_space += device->total_bytes; | |
1731 | spin_unlock(&root->fs_info->free_chunk_lock); | |
1732 | ||
c289811c CM |
1733 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1734 | root->fs_info->fs_devices->rotating = 1; | |
1735 | ||
6c41761f DS |
1736 | total_bytes = btrfs_super_total_bytes(root->fs_info->super_copy); |
1737 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
788f20eb CM |
1738 | total_bytes + device->total_bytes); |
1739 | ||
6c41761f DS |
1740 | total_bytes = btrfs_super_num_devices(root->fs_info->super_copy); |
1741 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
788f20eb | 1742 | total_bytes + 1); |
e5e9a520 | 1743 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 1744 | |
2b82032c YZ |
1745 | if (seeding_dev) { |
1746 | ret = init_first_rw_device(trans, root, device); | |
1747 | BUG_ON(ret); | |
1748 | ret = btrfs_finish_sprout(trans, root); | |
1749 | BUG_ON(ret); | |
1750 | } else { | |
1751 | ret = btrfs_add_device(trans, root, device); | |
1752 | } | |
1753 | ||
913d952e CM |
1754 | /* |
1755 | * we've got more storage, clear any full flags on the space | |
1756 | * infos | |
1757 | */ | |
1758 | btrfs_clear_space_info_full(root->fs_info); | |
1759 | ||
7d9eb12c | 1760 | unlock_chunks(root); |
2b82032c | 1761 | btrfs_commit_transaction(trans, root); |
a2135011 | 1762 | |
2b82032c YZ |
1763 | if (seeding_dev) { |
1764 | mutex_unlock(&uuid_mutex); | |
1765 | up_write(&sb->s_umount); | |
788f20eb | 1766 | |
2b82032c YZ |
1767 | ret = btrfs_relocate_sys_chunks(root); |
1768 | BUG_ON(ret); | |
1769 | } | |
c9e9f97b | 1770 | |
2b82032c YZ |
1771 | return ret; |
1772 | error: | |
e525fd89 | 1773 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
1774 | if (seeding_dev) { |
1775 | mutex_unlock(&uuid_mutex); | |
1776 | up_write(&sb->s_umount); | |
1777 | } | |
c9e9f97b | 1778 | return ret; |
788f20eb CM |
1779 | } |
1780 | ||
d397712b CM |
1781 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
1782 | struct btrfs_device *device) | |
0b86a832 CM |
1783 | { |
1784 | int ret; | |
1785 | struct btrfs_path *path; | |
1786 | struct btrfs_root *root; | |
1787 | struct btrfs_dev_item *dev_item; | |
1788 | struct extent_buffer *leaf; | |
1789 | struct btrfs_key key; | |
1790 | ||
1791 | root = device->dev_root->fs_info->chunk_root; | |
1792 | ||
1793 | path = btrfs_alloc_path(); | |
1794 | if (!path) | |
1795 | return -ENOMEM; | |
1796 | ||
1797 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1798 | key.type = BTRFS_DEV_ITEM_KEY; | |
1799 | key.offset = device->devid; | |
1800 | ||
1801 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
1802 | if (ret < 0) | |
1803 | goto out; | |
1804 | ||
1805 | if (ret > 0) { | |
1806 | ret = -ENOENT; | |
1807 | goto out; | |
1808 | } | |
1809 | ||
1810 | leaf = path->nodes[0]; | |
1811 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1812 | ||
1813 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1814 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1815 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1816 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1817 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
d6397bae | 1818 | btrfs_set_device_total_bytes(leaf, dev_item, device->disk_total_bytes); |
0b86a832 CM |
1819 | btrfs_set_device_bytes_used(leaf, dev_item, device->bytes_used); |
1820 | btrfs_mark_buffer_dirty(leaf); | |
1821 | ||
1822 | out: | |
1823 | btrfs_free_path(path); | |
1824 | return ret; | |
1825 | } | |
1826 | ||
7d9eb12c | 1827 | static int __btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
1828 | struct btrfs_device *device, u64 new_size) |
1829 | { | |
1830 | struct btrfs_super_block *super_copy = | |
6c41761f | 1831 | device->dev_root->fs_info->super_copy; |
8f18cf13 CM |
1832 | u64 old_total = btrfs_super_total_bytes(super_copy); |
1833 | u64 diff = new_size - device->total_bytes; | |
1834 | ||
2b82032c YZ |
1835 | if (!device->writeable) |
1836 | return -EACCES; | |
1837 | if (new_size <= device->total_bytes) | |
1838 | return -EINVAL; | |
1839 | ||
8f18cf13 | 1840 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
1841 | device->fs_devices->total_rw_bytes += diff; |
1842 | ||
1843 | device->total_bytes = new_size; | |
9779b72f | 1844 | device->disk_total_bytes = new_size; |
4184ea7f CM |
1845 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
1846 | ||
8f18cf13 CM |
1847 | return btrfs_update_device(trans, device); |
1848 | } | |
1849 | ||
7d9eb12c CM |
1850 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
1851 | struct btrfs_device *device, u64 new_size) | |
1852 | { | |
1853 | int ret; | |
1854 | lock_chunks(device->dev_root); | |
1855 | ret = __btrfs_grow_device(trans, device, new_size); | |
1856 | unlock_chunks(device->dev_root); | |
1857 | return ret; | |
1858 | } | |
1859 | ||
8f18cf13 CM |
1860 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, |
1861 | struct btrfs_root *root, | |
1862 | u64 chunk_tree, u64 chunk_objectid, | |
1863 | u64 chunk_offset) | |
1864 | { | |
1865 | int ret; | |
1866 | struct btrfs_path *path; | |
1867 | struct btrfs_key key; | |
1868 | ||
1869 | root = root->fs_info->chunk_root; | |
1870 | path = btrfs_alloc_path(); | |
1871 | if (!path) | |
1872 | return -ENOMEM; | |
1873 | ||
1874 | key.objectid = chunk_objectid; | |
1875 | key.offset = chunk_offset; | |
1876 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
1877 | ||
1878 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1879 | BUG_ON(ret); | |
1880 | ||
1881 | ret = btrfs_del_item(trans, root, path); | |
8f18cf13 CM |
1882 | |
1883 | btrfs_free_path(path); | |
65a246c5 | 1884 | return ret; |
8f18cf13 CM |
1885 | } |
1886 | ||
b2950863 | 1887 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
1888 | chunk_offset) |
1889 | { | |
6c41761f | 1890 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
1891 | struct btrfs_disk_key *disk_key; |
1892 | struct btrfs_chunk *chunk; | |
1893 | u8 *ptr; | |
1894 | int ret = 0; | |
1895 | u32 num_stripes; | |
1896 | u32 array_size; | |
1897 | u32 len = 0; | |
1898 | u32 cur; | |
1899 | struct btrfs_key key; | |
1900 | ||
1901 | array_size = btrfs_super_sys_array_size(super_copy); | |
1902 | ||
1903 | ptr = super_copy->sys_chunk_array; | |
1904 | cur = 0; | |
1905 | ||
1906 | while (cur < array_size) { | |
1907 | disk_key = (struct btrfs_disk_key *)ptr; | |
1908 | btrfs_disk_key_to_cpu(&key, disk_key); | |
1909 | ||
1910 | len = sizeof(*disk_key); | |
1911 | ||
1912 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
1913 | chunk = (struct btrfs_chunk *)(ptr + len); | |
1914 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
1915 | len += btrfs_chunk_item_size(num_stripes); | |
1916 | } else { | |
1917 | ret = -EIO; | |
1918 | break; | |
1919 | } | |
1920 | if (key.objectid == chunk_objectid && | |
1921 | key.offset == chunk_offset) { | |
1922 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
1923 | array_size -= len; | |
1924 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
1925 | } else { | |
1926 | ptr += len; | |
1927 | cur += len; | |
1928 | } | |
1929 | } | |
1930 | return ret; | |
1931 | } | |
1932 | ||
b2950863 | 1933 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
8f18cf13 CM |
1934 | u64 chunk_tree, u64 chunk_objectid, |
1935 | u64 chunk_offset) | |
1936 | { | |
1937 | struct extent_map_tree *em_tree; | |
1938 | struct btrfs_root *extent_root; | |
1939 | struct btrfs_trans_handle *trans; | |
1940 | struct extent_map *em; | |
1941 | struct map_lookup *map; | |
1942 | int ret; | |
1943 | int i; | |
1944 | ||
1945 | root = root->fs_info->chunk_root; | |
1946 | extent_root = root->fs_info->extent_root; | |
1947 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
1948 | ||
ba1bf481 JB |
1949 | ret = btrfs_can_relocate(extent_root, chunk_offset); |
1950 | if (ret) | |
1951 | return -ENOSPC; | |
1952 | ||
8f18cf13 | 1953 | /* step one, relocate all the extents inside this chunk */ |
1a40e23b | 1954 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); |
a22285a6 YZ |
1955 | if (ret) |
1956 | return ret; | |
8f18cf13 | 1957 | |
a22285a6 | 1958 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 1959 | BUG_ON(IS_ERR(trans)); |
8f18cf13 | 1960 | |
7d9eb12c CM |
1961 | lock_chunks(root); |
1962 | ||
8f18cf13 CM |
1963 | /* |
1964 | * step two, delete the device extents and the | |
1965 | * chunk tree entries | |
1966 | */ | |
890871be | 1967 | read_lock(&em_tree->lock); |
8f18cf13 | 1968 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 1969 | read_unlock(&em_tree->lock); |
8f18cf13 | 1970 | |
285190d9 | 1971 | BUG_ON(!em || em->start > chunk_offset || |
a061fc8d | 1972 | em->start + em->len < chunk_offset); |
8f18cf13 CM |
1973 | map = (struct map_lookup *)em->bdev; |
1974 | ||
1975 | for (i = 0; i < map->num_stripes; i++) { | |
1976 | ret = btrfs_free_dev_extent(trans, map->stripes[i].dev, | |
1977 | map->stripes[i].physical); | |
1978 | BUG_ON(ret); | |
a061fc8d | 1979 | |
dfe25020 CM |
1980 | if (map->stripes[i].dev) { |
1981 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
1982 | BUG_ON(ret); | |
1983 | } | |
8f18cf13 CM |
1984 | } |
1985 | ret = btrfs_free_chunk(trans, root, chunk_tree, chunk_objectid, | |
1986 | chunk_offset); | |
1987 | ||
1988 | BUG_ON(ret); | |
1989 | ||
1abe9b8a | 1990 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
1991 | ||
8f18cf13 CM |
1992 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
1993 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
1994 | BUG_ON(ret); | |
8f18cf13 CM |
1995 | } |
1996 | ||
2b82032c YZ |
1997 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset); |
1998 | BUG_ON(ret); | |
1999 | ||
890871be | 2000 | write_lock(&em_tree->lock); |
2b82032c | 2001 | remove_extent_mapping(em_tree, em); |
890871be | 2002 | write_unlock(&em_tree->lock); |
2b82032c YZ |
2003 | |
2004 | kfree(map); | |
2005 | em->bdev = NULL; | |
2006 | ||
2007 | /* once for the tree */ | |
2008 | free_extent_map(em); | |
2009 | /* once for us */ | |
2010 | free_extent_map(em); | |
2011 | ||
2012 | unlock_chunks(root); | |
2013 | btrfs_end_transaction(trans, root); | |
2014 | return 0; | |
2015 | } | |
2016 | ||
2017 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2018 | { | |
2019 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2020 | struct btrfs_path *path; | |
2021 | struct extent_buffer *leaf; | |
2022 | struct btrfs_chunk *chunk; | |
2023 | struct btrfs_key key; | |
2024 | struct btrfs_key found_key; | |
2025 | u64 chunk_tree = chunk_root->root_key.objectid; | |
2026 | u64 chunk_type; | |
ba1bf481 JB |
2027 | bool retried = false; |
2028 | int failed = 0; | |
2b82032c YZ |
2029 | int ret; |
2030 | ||
2031 | path = btrfs_alloc_path(); | |
2032 | if (!path) | |
2033 | return -ENOMEM; | |
2034 | ||
ba1bf481 | 2035 | again: |
2b82032c YZ |
2036 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2037 | key.offset = (u64)-1; | |
2038 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2039 | ||
2040 | while (1) { | |
2041 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2042 | if (ret < 0) | |
2043 | goto error; | |
2044 | BUG_ON(ret == 0); | |
2045 | ||
2046 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2047 | key.type); | |
2048 | if (ret < 0) | |
2049 | goto error; | |
2050 | if (ret > 0) | |
2051 | break; | |
1a40e23b | 2052 | |
2b82032c YZ |
2053 | leaf = path->nodes[0]; |
2054 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2055 | |
2b82032c YZ |
2056 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2057 | struct btrfs_chunk); | |
2058 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2059 | btrfs_release_path(path); |
8f18cf13 | 2060 | |
2b82032c YZ |
2061 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2062 | ret = btrfs_relocate_chunk(chunk_root, chunk_tree, | |
2063 | found_key.objectid, | |
2064 | found_key.offset); | |
ba1bf481 JB |
2065 | if (ret == -ENOSPC) |
2066 | failed++; | |
2067 | else if (ret) | |
2068 | BUG(); | |
2b82032c | 2069 | } |
8f18cf13 | 2070 | |
2b82032c YZ |
2071 | if (found_key.offset == 0) |
2072 | break; | |
2073 | key.offset = found_key.offset - 1; | |
2074 | } | |
2075 | ret = 0; | |
ba1bf481 JB |
2076 | if (failed && !retried) { |
2077 | failed = 0; | |
2078 | retried = true; | |
2079 | goto again; | |
2080 | } else if (failed && retried) { | |
2081 | WARN_ON(1); | |
2082 | ret = -ENOSPC; | |
2083 | } | |
2b82032c YZ |
2084 | error: |
2085 | btrfs_free_path(path); | |
2086 | return ret; | |
8f18cf13 CM |
2087 | } |
2088 | ||
0940ebf6 ID |
2089 | static int insert_balance_item(struct btrfs_root *root, |
2090 | struct btrfs_balance_control *bctl) | |
2091 | { | |
2092 | struct btrfs_trans_handle *trans; | |
2093 | struct btrfs_balance_item *item; | |
2094 | struct btrfs_disk_balance_args disk_bargs; | |
2095 | struct btrfs_path *path; | |
2096 | struct extent_buffer *leaf; | |
2097 | struct btrfs_key key; | |
2098 | int ret, err; | |
2099 | ||
2100 | path = btrfs_alloc_path(); | |
2101 | if (!path) | |
2102 | return -ENOMEM; | |
2103 | ||
2104 | trans = btrfs_start_transaction(root, 0); | |
2105 | if (IS_ERR(trans)) { | |
2106 | btrfs_free_path(path); | |
2107 | return PTR_ERR(trans); | |
2108 | } | |
2109 | ||
2110 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2111 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2112 | key.offset = 0; | |
2113 | ||
2114 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2115 | sizeof(*item)); | |
2116 | if (ret) | |
2117 | goto out; | |
2118 | ||
2119 | leaf = path->nodes[0]; | |
2120 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2121 | ||
2122 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2123 | ||
2124 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2125 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2126 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2127 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2128 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2129 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2130 | ||
2131 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2132 | ||
2133 | btrfs_mark_buffer_dirty(leaf); | |
2134 | out: | |
2135 | btrfs_free_path(path); | |
2136 | err = btrfs_commit_transaction(trans, root); | |
2137 | if (err && !ret) | |
2138 | ret = err; | |
2139 | return ret; | |
2140 | } | |
2141 | ||
2142 | static int del_balance_item(struct btrfs_root *root) | |
2143 | { | |
2144 | struct btrfs_trans_handle *trans; | |
2145 | struct btrfs_path *path; | |
2146 | struct btrfs_key key; | |
2147 | int ret, err; | |
2148 | ||
2149 | path = btrfs_alloc_path(); | |
2150 | if (!path) | |
2151 | return -ENOMEM; | |
2152 | ||
2153 | trans = btrfs_start_transaction(root, 0); | |
2154 | if (IS_ERR(trans)) { | |
2155 | btrfs_free_path(path); | |
2156 | return PTR_ERR(trans); | |
2157 | } | |
2158 | ||
2159 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2160 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2161 | key.offset = 0; | |
2162 | ||
2163 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2164 | if (ret < 0) | |
2165 | goto out; | |
2166 | if (ret > 0) { | |
2167 | ret = -ENOENT; | |
2168 | goto out; | |
2169 | } | |
2170 | ||
2171 | ret = btrfs_del_item(trans, root, path); | |
2172 | out: | |
2173 | btrfs_free_path(path); | |
2174 | err = btrfs_commit_transaction(trans, root); | |
2175 | if (err && !ret) | |
2176 | ret = err; | |
2177 | return ret; | |
2178 | } | |
2179 | ||
59641015 ID |
2180 | /* |
2181 | * This is a heuristic used to reduce the number of chunks balanced on | |
2182 | * resume after balance was interrupted. | |
2183 | */ | |
2184 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2185 | { | |
2186 | /* | |
2187 | * Turn on soft mode for chunk types that were being converted. | |
2188 | */ | |
2189 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2190 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2191 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2192 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2193 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2194 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2195 | ||
2196 | /* | |
2197 | * Turn on usage filter if is not already used. The idea is | |
2198 | * that chunks that we have already balanced should be | |
2199 | * reasonably full. Don't do it for chunks that are being | |
2200 | * converted - that will keep us from relocating unconverted | |
2201 | * (albeit full) chunks. | |
2202 | */ | |
2203 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2204 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2205 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2206 | bctl->data.usage = 90; | |
2207 | } | |
2208 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2209 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2210 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2211 | bctl->sys.usage = 90; | |
2212 | } | |
2213 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2214 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2215 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2216 | bctl->meta.usage = 90; | |
2217 | } | |
2218 | } | |
2219 | ||
c9e9f97b ID |
2220 | /* |
2221 | * Should be called with both balance and volume mutexes held to | |
2222 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2223 | * restriper. Same goes for unset_balance_control. | |
2224 | */ | |
2225 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
2226 | { | |
2227 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
2228 | ||
2229 | BUG_ON(fs_info->balance_ctl); | |
2230 | ||
2231 | spin_lock(&fs_info->balance_lock); | |
2232 | fs_info->balance_ctl = bctl; | |
2233 | spin_unlock(&fs_info->balance_lock); | |
2234 | } | |
2235 | ||
2236 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
2237 | { | |
2238 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
2239 | ||
2240 | BUG_ON(!fs_info->balance_ctl); | |
2241 | ||
2242 | spin_lock(&fs_info->balance_lock); | |
2243 | fs_info->balance_ctl = NULL; | |
2244 | spin_unlock(&fs_info->balance_lock); | |
2245 | ||
2246 | kfree(bctl); | |
2247 | } | |
2248 | ||
ed25e9b2 ID |
2249 | /* |
2250 | * Balance filters. Return 1 if chunk should be filtered out | |
2251 | * (should not be balanced). | |
2252 | */ | |
899c81ea | 2253 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
2254 | struct btrfs_balance_args *bargs) |
2255 | { | |
899c81ea ID |
2256 | chunk_type = chunk_to_extended(chunk_type) & |
2257 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 2258 | |
899c81ea | 2259 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
2260 | return 0; |
2261 | ||
2262 | return 1; | |
2263 | } | |
2264 | ||
5ce5b3c0 ID |
2265 | static u64 div_factor_fine(u64 num, int factor) |
2266 | { | |
2267 | if (factor <= 0) | |
2268 | return 0; | |
2269 | if (factor >= 100) | |
2270 | return num; | |
2271 | ||
2272 | num *= factor; | |
2273 | do_div(num, 100); | |
2274 | return num; | |
2275 | } | |
2276 | ||
2277 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, | |
2278 | struct btrfs_balance_args *bargs) | |
2279 | { | |
2280 | struct btrfs_block_group_cache *cache; | |
2281 | u64 chunk_used, user_thresh; | |
2282 | int ret = 1; | |
2283 | ||
2284 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
2285 | chunk_used = btrfs_block_group_used(&cache->item); | |
2286 | ||
2287 | user_thresh = div_factor_fine(cache->key.offset, bargs->usage); | |
2288 | if (chunk_used < user_thresh) | |
2289 | ret = 0; | |
2290 | ||
2291 | btrfs_put_block_group(cache); | |
2292 | return ret; | |
2293 | } | |
2294 | ||
409d404b ID |
2295 | static int chunk_devid_filter(struct extent_buffer *leaf, |
2296 | struct btrfs_chunk *chunk, | |
2297 | struct btrfs_balance_args *bargs) | |
2298 | { | |
2299 | struct btrfs_stripe *stripe; | |
2300 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2301 | int i; | |
2302 | ||
2303 | for (i = 0; i < num_stripes; i++) { | |
2304 | stripe = btrfs_stripe_nr(chunk, i); | |
2305 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
2306 | return 0; | |
2307 | } | |
2308 | ||
2309 | return 1; | |
2310 | } | |
2311 | ||
94e60d5a ID |
2312 | /* [pstart, pend) */ |
2313 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
2314 | struct btrfs_chunk *chunk, | |
2315 | u64 chunk_offset, | |
2316 | struct btrfs_balance_args *bargs) | |
2317 | { | |
2318 | struct btrfs_stripe *stripe; | |
2319 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
2320 | u64 stripe_offset; | |
2321 | u64 stripe_length; | |
2322 | int factor; | |
2323 | int i; | |
2324 | ||
2325 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
2326 | return 0; | |
2327 | ||
2328 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
2329 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) | |
2330 | factor = 2; | |
2331 | else | |
2332 | factor = 1; | |
2333 | factor = num_stripes / factor; | |
2334 | ||
2335 | for (i = 0; i < num_stripes; i++) { | |
2336 | stripe = btrfs_stripe_nr(chunk, i); | |
2337 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
2338 | continue; | |
2339 | ||
2340 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
2341 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
2342 | do_div(stripe_length, factor); | |
2343 | ||
2344 | if (stripe_offset < bargs->pend && | |
2345 | stripe_offset + stripe_length > bargs->pstart) | |
2346 | return 0; | |
2347 | } | |
2348 | ||
2349 | return 1; | |
2350 | } | |
2351 | ||
ea67176a ID |
2352 | /* [vstart, vend) */ |
2353 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
2354 | struct btrfs_chunk *chunk, | |
2355 | u64 chunk_offset, | |
2356 | struct btrfs_balance_args *bargs) | |
2357 | { | |
2358 | if (chunk_offset < bargs->vend && | |
2359 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
2360 | /* at least part of the chunk is inside this vrange */ | |
2361 | return 0; | |
2362 | ||
2363 | return 1; | |
2364 | } | |
2365 | ||
899c81ea | 2366 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
2367 | struct btrfs_balance_args *bargs) |
2368 | { | |
2369 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
2370 | return 0; | |
2371 | ||
899c81ea ID |
2372 | chunk_type = chunk_to_extended(chunk_type) & |
2373 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 2374 | |
899c81ea | 2375 | if (bargs->target == chunk_type) |
cfa4c961 ID |
2376 | return 1; |
2377 | ||
2378 | return 0; | |
2379 | } | |
2380 | ||
f43ffb60 ID |
2381 | static int should_balance_chunk(struct btrfs_root *root, |
2382 | struct extent_buffer *leaf, | |
2383 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
2384 | { | |
2385 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
2386 | struct btrfs_balance_args *bargs = NULL; | |
2387 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
2388 | ||
2389 | /* type filter */ | |
2390 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
2391 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
2392 | return 0; | |
2393 | } | |
2394 | ||
2395 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
2396 | bargs = &bctl->data; | |
2397 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
2398 | bargs = &bctl->sys; | |
2399 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
2400 | bargs = &bctl->meta; | |
2401 | ||
ed25e9b2 ID |
2402 | /* profiles filter */ |
2403 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
2404 | chunk_profiles_filter(chunk_type, bargs)) { | |
2405 | return 0; | |
5ce5b3c0 ID |
2406 | } |
2407 | ||
2408 | /* usage filter */ | |
2409 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2410 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
2411 | return 0; | |
409d404b ID |
2412 | } |
2413 | ||
2414 | /* devid filter */ | |
2415 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
2416 | chunk_devid_filter(leaf, chunk, bargs)) { | |
2417 | return 0; | |
94e60d5a ID |
2418 | } |
2419 | ||
2420 | /* drange filter, makes sense only with devid filter */ | |
2421 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
2422 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2423 | return 0; | |
ea67176a ID |
2424 | } |
2425 | ||
2426 | /* vrange filter */ | |
2427 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
2428 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
2429 | return 0; | |
ed25e9b2 ID |
2430 | } |
2431 | ||
cfa4c961 ID |
2432 | /* soft profile changing mode */ |
2433 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
2434 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
2435 | return 0; | |
2436 | } | |
2437 | ||
f43ffb60 ID |
2438 | return 1; |
2439 | } | |
2440 | ||
ec44a35c CM |
2441 | static u64 div_factor(u64 num, int factor) |
2442 | { | |
2443 | if (factor == 10) | |
2444 | return num; | |
2445 | num *= factor; | |
2446 | do_div(num, 10); | |
2447 | return num; | |
2448 | } | |
2449 | ||
c9e9f97b | 2450 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 2451 | { |
19a39dce | 2452 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
2453 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
2454 | struct btrfs_root *dev_root = fs_info->dev_root; | |
2455 | struct list_head *devices; | |
ec44a35c CM |
2456 | struct btrfs_device *device; |
2457 | u64 old_size; | |
2458 | u64 size_to_free; | |
f43ffb60 | 2459 | struct btrfs_chunk *chunk; |
ec44a35c CM |
2460 | struct btrfs_path *path; |
2461 | struct btrfs_key key; | |
ec44a35c | 2462 | struct btrfs_key found_key; |
c9e9f97b | 2463 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
2464 | struct extent_buffer *leaf; |
2465 | int slot; | |
c9e9f97b ID |
2466 | int ret; |
2467 | int enospc_errors = 0; | |
19a39dce | 2468 | bool counting = true; |
ec44a35c | 2469 | |
ec44a35c | 2470 | /* step one make some room on all the devices */ |
c9e9f97b | 2471 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 2472 | list_for_each_entry(device, devices, dev_list) { |
ec44a35c CM |
2473 | old_size = device->total_bytes; |
2474 | size_to_free = div_factor(old_size, 1); | |
2475 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c YZ |
2476 | if (!device->writeable || |
2477 | device->total_bytes - device->bytes_used > size_to_free) | |
ec44a35c CM |
2478 | continue; |
2479 | ||
2480 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
2481 | if (ret == -ENOSPC) |
2482 | break; | |
ec44a35c CM |
2483 | BUG_ON(ret); |
2484 | ||
a22285a6 | 2485 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 2486 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
2487 | |
2488 | ret = btrfs_grow_device(trans, device, old_size); | |
2489 | BUG_ON(ret); | |
2490 | ||
2491 | btrfs_end_transaction(trans, dev_root); | |
2492 | } | |
2493 | ||
2494 | /* step two, relocate all the chunks */ | |
2495 | path = btrfs_alloc_path(); | |
17e9f796 MF |
2496 | if (!path) { |
2497 | ret = -ENOMEM; | |
2498 | goto error; | |
2499 | } | |
19a39dce ID |
2500 | |
2501 | /* zero out stat counters */ | |
2502 | spin_lock(&fs_info->balance_lock); | |
2503 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
2504 | spin_unlock(&fs_info->balance_lock); | |
2505 | again: | |
ec44a35c CM |
2506 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2507 | key.offset = (u64)-1; | |
2508 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2509 | ||
d397712b | 2510 | while (1) { |
19a39dce | 2511 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 2512 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
2513 | ret = -ECANCELED; |
2514 | goto error; | |
2515 | } | |
2516 | ||
ec44a35c CM |
2517 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
2518 | if (ret < 0) | |
2519 | goto error; | |
2520 | ||
2521 | /* | |
2522 | * this shouldn't happen, it means the last relocate | |
2523 | * failed | |
2524 | */ | |
2525 | if (ret == 0) | |
c9e9f97b | 2526 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
2527 | |
2528 | ret = btrfs_previous_item(chunk_root, path, 0, | |
2529 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
2530 | if (ret) { |
2531 | ret = 0; | |
ec44a35c | 2532 | break; |
c9e9f97b | 2533 | } |
7d9eb12c | 2534 | |
f43ffb60 ID |
2535 | leaf = path->nodes[0]; |
2536 | slot = path->slots[0]; | |
2537 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 2538 | |
ec44a35c CM |
2539 | if (found_key.objectid != key.objectid) |
2540 | break; | |
7d9eb12c | 2541 | |
ec44a35c | 2542 | /* chunk zero is special */ |
ba1bf481 | 2543 | if (found_key.offset == 0) |
ec44a35c CM |
2544 | break; |
2545 | ||
f43ffb60 ID |
2546 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
2547 | ||
19a39dce ID |
2548 | if (!counting) { |
2549 | spin_lock(&fs_info->balance_lock); | |
2550 | bctl->stat.considered++; | |
2551 | spin_unlock(&fs_info->balance_lock); | |
2552 | } | |
2553 | ||
f43ffb60 ID |
2554 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
2555 | found_key.offset); | |
b3b4aa74 | 2556 | btrfs_release_path(path); |
f43ffb60 ID |
2557 | if (!ret) |
2558 | goto loop; | |
2559 | ||
19a39dce ID |
2560 | if (counting) { |
2561 | spin_lock(&fs_info->balance_lock); | |
2562 | bctl->stat.expected++; | |
2563 | spin_unlock(&fs_info->balance_lock); | |
2564 | goto loop; | |
2565 | } | |
2566 | ||
ec44a35c CM |
2567 | ret = btrfs_relocate_chunk(chunk_root, |
2568 | chunk_root->root_key.objectid, | |
2569 | found_key.objectid, | |
2570 | found_key.offset); | |
508794eb JB |
2571 | if (ret && ret != -ENOSPC) |
2572 | goto error; | |
19a39dce | 2573 | if (ret == -ENOSPC) { |
c9e9f97b | 2574 | enospc_errors++; |
19a39dce ID |
2575 | } else { |
2576 | spin_lock(&fs_info->balance_lock); | |
2577 | bctl->stat.completed++; | |
2578 | spin_unlock(&fs_info->balance_lock); | |
2579 | } | |
f43ffb60 | 2580 | loop: |
ba1bf481 | 2581 | key.offset = found_key.offset - 1; |
ec44a35c | 2582 | } |
c9e9f97b | 2583 | |
19a39dce ID |
2584 | if (counting) { |
2585 | btrfs_release_path(path); | |
2586 | counting = false; | |
2587 | goto again; | |
2588 | } | |
ec44a35c CM |
2589 | error: |
2590 | btrfs_free_path(path); | |
c9e9f97b ID |
2591 | if (enospc_errors) { |
2592 | printk(KERN_INFO "btrfs: %d enospc errors during balance\n", | |
2593 | enospc_errors); | |
2594 | if (!ret) | |
2595 | ret = -ENOSPC; | |
2596 | } | |
2597 | ||
ec44a35c CM |
2598 | return ret; |
2599 | } | |
2600 | ||
0c460c0d ID |
2601 | /** |
2602 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
2603 | * @flags: profile to validate | |
2604 | * @extended: if true @flags is treated as an extended profile | |
2605 | */ | |
2606 | static int alloc_profile_is_valid(u64 flags, int extended) | |
2607 | { | |
2608 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
2609 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
2610 | ||
2611 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
2612 | ||
2613 | /* 1) check that all other bits are zeroed */ | |
2614 | if (flags & ~mask) | |
2615 | return 0; | |
2616 | ||
2617 | /* 2) see if profile is reduced */ | |
2618 | if (flags == 0) | |
2619 | return !extended; /* "0" is valid for usual profiles */ | |
2620 | ||
2621 | /* true if exactly one bit set */ | |
2622 | return (flags & (flags - 1)) == 0; | |
2623 | } | |
2624 | ||
837d5b6e ID |
2625 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
2626 | { | |
a7e99c69 ID |
2627 | /* cancel requested || normal exit path */ |
2628 | return atomic_read(&fs_info->balance_cancel_req) || | |
2629 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
2630 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
2631 | } |
2632 | ||
c9e9f97b ID |
2633 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
2634 | { | |
0940ebf6 ID |
2635 | int ret; |
2636 | ||
c9e9f97b | 2637 | unset_balance_control(fs_info); |
0940ebf6 ID |
2638 | ret = del_balance_item(fs_info->tree_root); |
2639 | BUG_ON(ret); | |
c9e9f97b ID |
2640 | } |
2641 | ||
19a39dce | 2642 | void update_ioctl_balance_args(struct btrfs_fs_info *fs_info, int lock, |
c9e9f97b ID |
2643 | struct btrfs_ioctl_balance_args *bargs); |
2644 | ||
2645 | /* | |
2646 | * Should be called with both balance and volume mutexes held | |
2647 | */ | |
2648 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
2649 | struct btrfs_ioctl_balance_args *bargs) | |
2650 | { | |
2651 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 2652 | u64 allowed; |
e4837f8f | 2653 | int mixed = 0; |
c9e9f97b ID |
2654 | int ret; |
2655 | ||
837d5b6e | 2656 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
2657 | atomic_read(&fs_info->balance_pause_req) || |
2658 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
2659 | ret = -EINVAL; |
2660 | goto out; | |
2661 | } | |
2662 | ||
e4837f8f ID |
2663 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
2664 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
2665 | mixed = 1; | |
2666 | ||
f43ffb60 ID |
2667 | /* |
2668 | * In case of mixed groups both data and meta should be picked, | |
2669 | * and identical options should be given for both of them. | |
2670 | */ | |
e4837f8f ID |
2671 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
2672 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
2673 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
2674 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
2675 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
2676 | printk(KERN_ERR "btrfs: with mixed groups data and " | |
2677 | "metadata balance options must be the same\n"); | |
2678 | ret = -EINVAL; | |
2679 | goto out; | |
2680 | } | |
2681 | } | |
2682 | ||
e4d8ec0f ID |
2683 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
2684 | if (fs_info->fs_devices->num_devices == 1) | |
2685 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
2686 | else if (fs_info->fs_devices->num_devices < 4) | |
2687 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
2688 | else | |
2689 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1 | | |
2690 | BTRFS_BLOCK_GROUP_RAID10); | |
2691 | ||
6728b198 ID |
2692 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2693 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
2694 | (bctl->data.target & ~allowed))) { | |
e4d8ec0f ID |
2695 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2696 | "data profile %llu\n", | |
2697 | (unsigned long long)bctl->data.target); | |
2698 | ret = -EINVAL; | |
2699 | goto out; | |
2700 | } | |
6728b198 ID |
2701 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2702 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
2703 | (bctl->meta.target & ~allowed))) { | |
e4d8ec0f ID |
2704 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2705 | "metadata profile %llu\n", | |
2706 | (unsigned long long)bctl->meta.target); | |
2707 | ret = -EINVAL; | |
2708 | goto out; | |
2709 | } | |
6728b198 ID |
2710 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
2711 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
2712 | (bctl->sys.target & ~allowed))) { | |
e4d8ec0f ID |
2713 | printk(KERN_ERR "btrfs: unable to start balance with target " |
2714 | "system profile %llu\n", | |
2715 | (unsigned long long)bctl->sys.target); | |
2716 | ret = -EINVAL; | |
2717 | goto out; | |
2718 | } | |
2719 | ||
e4837f8f ID |
2720 | /* allow dup'ed data chunks only in mixed mode */ |
2721 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 2722 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
e4d8ec0f ID |
2723 | printk(KERN_ERR "btrfs: dup for data is not allowed\n"); |
2724 | ret = -EINVAL; | |
2725 | goto out; | |
2726 | } | |
2727 | ||
2728 | /* allow to reduce meta or sys integrity only if force set */ | |
2729 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
2730 | BTRFS_BLOCK_GROUP_RAID10; | |
2731 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2732 | (fs_info->avail_system_alloc_bits & allowed) && | |
2733 | !(bctl->sys.target & allowed)) || | |
2734 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
2735 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
2736 | !(bctl->meta.target & allowed))) { | |
2737 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
2738 | printk(KERN_INFO "btrfs: force reducing metadata " | |
2739 | "integrity\n"); | |
2740 | } else { | |
2741 | printk(KERN_ERR "btrfs: balance will reduce metadata " | |
2742 | "integrity, use force if you want this\n"); | |
2743 | ret = -EINVAL; | |
2744 | goto out; | |
2745 | } | |
2746 | } | |
2747 | ||
0940ebf6 | 2748 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 2749 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
2750 | goto out; |
2751 | ||
59641015 ID |
2752 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
2753 | BUG_ON(ret == -EEXIST); | |
2754 | set_balance_control(bctl); | |
2755 | } else { | |
2756 | BUG_ON(ret != -EEXIST); | |
2757 | spin_lock(&fs_info->balance_lock); | |
2758 | update_balance_args(bctl); | |
2759 | spin_unlock(&fs_info->balance_lock); | |
2760 | } | |
c9e9f97b | 2761 | |
837d5b6e | 2762 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
2763 | mutex_unlock(&fs_info->balance_mutex); |
2764 | ||
2765 | ret = __btrfs_balance(fs_info); | |
2766 | ||
2767 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 2768 | atomic_dec(&fs_info->balance_running); |
c9e9f97b ID |
2769 | |
2770 | if (bargs) { | |
2771 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 2772 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
2773 | } |
2774 | ||
837d5b6e ID |
2775 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
2776 | balance_need_close(fs_info)) { | |
2777 | __cancel_balance(fs_info); | |
2778 | } | |
2779 | ||
2780 | wake_up(&fs_info->balance_wait_q); | |
c9e9f97b ID |
2781 | |
2782 | return ret; | |
2783 | out: | |
59641015 ID |
2784 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
2785 | __cancel_balance(fs_info); | |
2786 | else | |
2787 | kfree(bctl); | |
2788 | return ret; | |
2789 | } | |
2790 | ||
2791 | static int balance_kthread(void *data) | |
2792 | { | |
2793 | struct btrfs_balance_control *bctl = | |
2794 | (struct btrfs_balance_control *)data; | |
2795 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
9555c6c1 | 2796 | int ret = 0; |
59641015 ID |
2797 | |
2798 | mutex_lock(&fs_info->volume_mutex); | |
2799 | mutex_lock(&fs_info->balance_mutex); | |
2800 | ||
2801 | set_balance_control(bctl); | |
2802 | ||
9555c6c1 ID |
2803 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { |
2804 | printk(KERN_INFO "btrfs: force skipping balance\n"); | |
2805 | } else { | |
2806 | printk(KERN_INFO "btrfs: continuing balance\n"); | |
2807 | ret = btrfs_balance(bctl, NULL); | |
2808 | } | |
59641015 ID |
2809 | |
2810 | mutex_unlock(&fs_info->balance_mutex); | |
2811 | mutex_unlock(&fs_info->volume_mutex); | |
2812 | return ret; | |
2813 | } | |
2814 | ||
2815 | int btrfs_recover_balance(struct btrfs_root *tree_root) | |
2816 | { | |
2817 | struct task_struct *tsk; | |
2818 | struct btrfs_balance_control *bctl; | |
2819 | struct btrfs_balance_item *item; | |
2820 | struct btrfs_disk_balance_args disk_bargs; | |
2821 | struct btrfs_path *path; | |
2822 | struct extent_buffer *leaf; | |
2823 | struct btrfs_key key; | |
2824 | int ret; | |
2825 | ||
2826 | path = btrfs_alloc_path(); | |
2827 | if (!path) | |
2828 | return -ENOMEM; | |
2829 | ||
2830 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
2831 | if (!bctl) { | |
2832 | ret = -ENOMEM; | |
2833 | goto out; | |
2834 | } | |
2835 | ||
2836 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2837 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2838 | key.offset = 0; | |
2839 | ||
2840 | ret = btrfs_search_slot(NULL, tree_root, &key, path, 0, 0); | |
2841 | if (ret < 0) | |
2842 | goto out_bctl; | |
2843 | if (ret > 0) { /* ret = -ENOENT; */ | |
2844 | ret = 0; | |
2845 | goto out_bctl; | |
2846 | } | |
2847 | ||
2848 | leaf = path->nodes[0]; | |
2849 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2850 | ||
2851 | bctl->fs_info = tree_root->fs_info; | |
2852 | bctl->flags = btrfs_balance_flags(leaf, item) | BTRFS_BALANCE_RESUME; | |
2853 | ||
2854 | btrfs_balance_data(leaf, item, &disk_bargs); | |
2855 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
2856 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
2857 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
2858 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
2859 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
2860 | ||
2861 | tsk = kthread_run(balance_kthread, bctl, "btrfs-balance"); | |
2862 | if (IS_ERR(tsk)) | |
2863 | ret = PTR_ERR(tsk); | |
2864 | else | |
2865 | goto out; | |
2866 | ||
2867 | out_bctl: | |
c9e9f97b | 2868 | kfree(bctl); |
59641015 ID |
2869 | out: |
2870 | btrfs_free_path(path); | |
ec44a35c CM |
2871 | return ret; |
2872 | } | |
2873 | ||
837d5b6e ID |
2874 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
2875 | { | |
2876 | int ret = 0; | |
2877 | ||
2878 | mutex_lock(&fs_info->balance_mutex); | |
2879 | if (!fs_info->balance_ctl) { | |
2880 | mutex_unlock(&fs_info->balance_mutex); | |
2881 | return -ENOTCONN; | |
2882 | } | |
2883 | ||
2884 | if (atomic_read(&fs_info->balance_running)) { | |
2885 | atomic_inc(&fs_info->balance_pause_req); | |
2886 | mutex_unlock(&fs_info->balance_mutex); | |
2887 | ||
2888 | wait_event(fs_info->balance_wait_q, | |
2889 | atomic_read(&fs_info->balance_running) == 0); | |
2890 | ||
2891 | mutex_lock(&fs_info->balance_mutex); | |
2892 | /* we are good with balance_ctl ripped off from under us */ | |
2893 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
2894 | atomic_dec(&fs_info->balance_pause_req); | |
2895 | } else { | |
2896 | ret = -ENOTCONN; | |
2897 | } | |
2898 | ||
2899 | mutex_unlock(&fs_info->balance_mutex); | |
2900 | return ret; | |
2901 | } | |
2902 | ||
a7e99c69 ID |
2903 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
2904 | { | |
2905 | mutex_lock(&fs_info->balance_mutex); | |
2906 | if (!fs_info->balance_ctl) { | |
2907 | mutex_unlock(&fs_info->balance_mutex); | |
2908 | return -ENOTCONN; | |
2909 | } | |
2910 | ||
2911 | atomic_inc(&fs_info->balance_cancel_req); | |
2912 | /* | |
2913 | * if we are running just wait and return, balance item is | |
2914 | * deleted in btrfs_balance in this case | |
2915 | */ | |
2916 | if (atomic_read(&fs_info->balance_running)) { | |
2917 | mutex_unlock(&fs_info->balance_mutex); | |
2918 | wait_event(fs_info->balance_wait_q, | |
2919 | atomic_read(&fs_info->balance_running) == 0); | |
2920 | mutex_lock(&fs_info->balance_mutex); | |
2921 | } else { | |
2922 | /* __cancel_balance needs volume_mutex */ | |
2923 | mutex_unlock(&fs_info->balance_mutex); | |
2924 | mutex_lock(&fs_info->volume_mutex); | |
2925 | mutex_lock(&fs_info->balance_mutex); | |
2926 | ||
2927 | if (fs_info->balance_ctl) | |
2928 | __cancel_balance(fs_info); | |
2929 | ||
2930 | mutex_unlock(&fs_info->volume_mutex); | |
2931 | } | |
2932 | ||
2933 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
2934 | atomic_dec(&fs_info->balance_cancel_req); | |
2935 | mutex_unlock(&fs_info->balance_mutex); | |
2936 | return 0; | |
2937 | } | |
2938 | ||
8f18cf13 CM |
2939 | /* |
2940 | * shrinking a device means finding all of the device extents past | |
2941 | * the new size, and then following the back refs to the chunks. | |
2942 | * The chunk relocation code actually frees the device extent | |
2943 | */ | |
2944 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
2945 | { | |
2946 | struct btrfs_trans_handle *trans; | |
2947 | struct btrfs_root *root = device->dev_root; | |
2948 | struct btrfs_dev_extent *dev_extent = NULL; | |
2949 | struct btrfs_path *path; | |
2950 | u64 length; | |
2951 | u64 chunk_tree; | |
2952 | u64 chunk_objectid; | |
2953 | u64 chunk_offset; | |
2954 | int ret; | |
2955 | int slot; | |
ba1bf481 JB |
2956 | int failed = 0; |
2957 | bool retried = false; | |
8f18cf13 CM |
2958 | struct extent_buffer *l; |
2959 | struct btrfs_key key; | |
6c41761f | 2960 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 2961 | u64 old_total = btrfs_super_total_bytes(super_copy); |
ba1bf481 | 2962 | u64 old_size = device->total_bytes; |
8f18cf13 CM |
2963 | u64 diff = device->total_bytes - new_size; |
2964 | ||
2b82032c YZ |
2965 | if (new_size >= device->total_bytes) |
2966 | return -EINVAL; | |
8f18cf13 CM |
2967 | |
2968 | path = btrfs_alloc_path(); | |
2969 | if (!path) | |
2970 | return -ENOMEM; | |
2971 | ||
8f18cf13 CM |
2972 | path->reada = 2; |
2973 | ||
7d9eb12c CM |
2974 | lock_chunks(root); |
2975 | ||
8f18cf13 | 2976 | device->total_bytes = new_size; |
2bf64758 | 2977 | if (device->writeable) { |
2b82032c | 2978 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
2979 | spin_lock(&root->fs_info->free_chunk_lock); |
2980 | root->fs_info->free_chunk_space -= diff; | |
2981 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2982 | } | |
7d9eb12c | 2983 | unlock_chunks(root); |
8f18cf13 | 2984 | |
ba1bf481 | 2985 | again: |
8f18cf13 CM |
2986 | key.objectid = device->devid; |
2987 | key.offset = (u64)-1; | |
2988 | key.type = BTRFS_DEV_EXTENT_KEY; | |
2989 | ||
2990 | while (1) { | |
2991 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
2992 | if (ret < 0) | |
2993 | goto done; | |
2994 | ||
2995 | ret = btrfs_previous_item(root, path, 0, key.type); | |
2996 | if (ret < 0) | |
2997 | goto done; | |
2998 | if (ret) { | |
2999 | ret = 0; | |
b3b4aa74 | 3000 | btrfs_release_path(path); |
bf1fb512 | 3001 | break; |
8f18cf13 CM |
3002 | } |
3003 | ||
3004 | l = path->nodes[0]; | |
3005 | slot = path->slots[0]; | |
3006 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
3007 | ||
ba1bf481 | 3008 | if (key.objectid != device->devid) { |
b3b4aa74 | 3009 | btrfs_release_path(path); |
bf1fb512 | 3010 | break; |
ba1bf481 | 3011 | } |
8f18cf13 CM |
3012 | |
3013 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
3014 | length = btrfs_dev_extent_length(l, dev_extent); | |
3015 | ||
ba1bf481 | 3016 | if (key.offset + length <= new_size) { |
b3b4aa74 | 3017 | btrfs_release_path(path); |
d6397bae | 3018 | break; |
ba1bf481 | 3019 | } |
8f18cf13 CM |
3020 | |
3021 | chunk_tree = btrfs_dev_extent_chunk_tree(l, dev_extent); | |
3022 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); | |
3023 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 3024 | btrfs_release_path(path); |
8f18cf13 CM |
3025 | |
3026 | ret = btrfs_relocate_chunk(root, chunk_tree, chunk_objectid, | |
3027 | chunk_offset); | |
ba1bf481 | 3028 | if (ret && ret != -ENOSPC) |
8f18cf13 | 3029 | goto done; |
ba1bf481 JB |
3030 | if (ret == -ENOSPC) |
3031 | failed++; | |
3032 | key.offset -= 1; | |
3033 | } | |
3034 | ||
3035 | if (failed && !retried) { | |
3036 | failed = 0; | |
3037 | retried = true; | |
3038 | goto again; | |
3039 | } else if (failed && retried) { | |
3040 | ret = -ENOSPC; | |
3041 | lock_chunks(root); | |
3042 | ||
3043 | device->total_bytes = old_size; | |
3044 | if (device->writeable) | |
3045 | device->fs_devices->total_rw_bytes += diff; | |
2bf64758 JB |
3046 | spin_lock(&root->fs_info->free_chunk_lock); |
3047 | root->fs_info->free_chunk_space += diff; | |
3048 | spin_unlock(&root->fs_info->free_chunk_lock); | |
ba1bf481 JB |
3049 | unlock_chunks(root); |
3050 | goto done; | |
8f18cf13 CM |
3051 | } |
3052 | ||
d6397bae | 3053 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 3054 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
3055 | if (IS_ERR(trans)) { |
3056 | ret = PTR_ERR(trans); | |
3057 | goto done; | |
3058 | } | |
3059 | ||
d6397bae CB |
3060 | lock_chunks(root); |
3061 | ||
3062 | device->disk_total_bytes = new_size; | |
3063 | /* Now btrfs_update_device() will change the on-disk size. */ | |
3064 | ret = btrfs_update_device(trans, device); | |
3065 | if (ret) { | |
3066 | unlock_chunks(root); | |
3067 | btrfs_end_transaction(trans, root); | |
3068 | goto done; | |
3069 | } | |
3070 | WARN_ON(diff > old_total); | |
3071 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
3072 | unlock_chunks(root); | |
3073 | btrfs_end_transaction(trans, root); | |
8f18cf13 CM |
3074 | done: |
3075 | btrfs_free_path(path); | |
3076 | return ret; | |
3077 | } | |
3078 | ||
125ccb0a | 3079 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
3080 | struct btrfs_key *key, |
3081 | struct btrfs_chunk *chunk, int item_size) | |
3082 | { | |
6c41761f | 3083 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
3084 | struct btrfs_disk_key disk_key; |
3085 | u32 array_size; | |
3086 | u8 *ptr; | |
3087 | ||
3088 | array_size = btrfs_super_sys_array_size(super_copy); | |
3089 | if (array_size + item_size > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) | |
3090 | return -EFBIG; | |
3091 | ||
3092 | ptr = super_copy->sys_chunk_array + array_size; | |
3093 | btrfs_cpu_key_to_disk(&disk_key, key); | |
3094 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
3095 | ptr += sizeof(disk_key); | |
3096 | memcpy(ptr, chunk, item_size); | |
3097 | item_size += sizeof(disk_key); | |
3098 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
3099 | return 0; | |
3100 | } | |
3101 | ||
73c5de00 AJ |
3102 | /* |
3103 | * sort the devices in descending order by max_avail, total_avail | |
3104 | */ | |
3105 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 3106 | { |
73c5de00 AJ |
3107 | const struct btrfs_device_info *di_a = a; |
3108 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 3109 | |
73c5de00 | 3110 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 3111 | return -1; |
73c5de00 | 3112 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 3113 | return 1; |
73c5de00 AJ |
3114 | if (di_a->total_avail > di_b->total_avail) |
3115 | return -1; | |
3116 | if (di_a->total_avail < di_b->total_avail) | |
3117 | return 1; | |
3118 | return 0; | |
b2117a39 | 3119 | } |
0b86a832 | 3120 | |
73c5de00 AJ |
3121 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
3122 | struct btrfs_root *extent_root, | |
3123 | struct map_lookup **map_ret, | |
3124 | u64 *num_bytes_out, u64 *stripe_size_out, | |
3125 | u64 start, u64 type) | |
b2117a39 | 3126 | { |
73c5de00 AJ |
3127 | struct btrfs_fs_info *info = extent_root->fs_info; |
3128 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
3129 | struct list_head *cur; | |
3130 | struct map_lookup *map = NULL; | |
3131 | struct extent_map_tree *em_tree; | |
3132 | struct extent_map *em; | |
3133 | struct btrfs_device_info *devices_info = NULL; | |
3134 | u64 total_avail; | |
3135 | int num_stripes; /* total number of stripes to allocate */ | |
3136 | int sub_stripes; /* sub_stripes info for map */ | |
3137 | int dev_stripes; /* stripes per dev */ | |
3138 | int devs_max; /* max devs to use */ | |
3139 | int devs_min; /* min devs needed */ | |
3140 | int devs_increment; /* ndevs has to be a multiple of this */ | |
3141 | int ncopies; /* how many copies to data has */ | |
3142 | int ret; | |
3143 | u64 max_stripe_size; | |
3144 | u64 max_chunk_size; | |
3145 | u64 stripe_size; | |
3146 | u64 num_bytes; | |
3147 | int ndevs; | |
3148 | int i; | |
3149 | int j; | |
593060d7 | 3150 | |
0c460c0d | 3151 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 3152 | |
73c5de00 AJ |
3153 | if (list_empty(&fs_devices->alloc_list)) |
3154 | return -ENOSPC; | |
b2117a39 | 3155 | |
73c5de00 AJ |
3156 | sub_stripes = 1; |
3157 | dev_stripes = 1; | |
3158 | devs_increment = 1; | |
3159 | ncopies = 1; | |
3160 | devs_max = 0; /* 0 == as many as possible */ | |
3161 | devs_min = 1; | |
3162 | ||
3163 | /* | |
3164 | * define the properties of each RAID type. | |
3165 | * FIXME: move this to a global table and use it in all RAID | |
3166 | * calculation code | |
3167 | */ | |
3168 | if (type & (BTRFS_BLOCK_GROUP_DUP)) { | |
3169 | dev_stripes = 2; | |
b2117a39 | 3170 | ncopies = 2; |
73c5de00 AJ |
3171 | devs_max = 1; |
3172 | } else if (type & (BTRFS_BLOCK_GROUP_RAID0)) { | |
3173 | devs_min = 2; | |
3174 | } else if (type & (BTRFS_BLOCK_GROUP_RAID1)) { | |
3175 | devs_increment = 2; | |
b2117a39 | 3176 | ncopies = 2; |
73c5de00 AJ |
3177 | devs_max = 2; |
3178 | devs_min = 2; | |
3179 | } else if (type & (BTRFS_BLOCK_GROUP_RAID10)) { | |
3180 | sub_stripes = 2; | |
3181 | devs_increment = 2; | |
3182 | ncopies = 2; | |
3183 | devs_min = 4; | |
3184 | } else { | |
3185 | devs_max = 1; | |
3186 | } | |
b2117a39 | 3187 | |
9b3f68b9 | 3188 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
3189 | max_stripe_size = 1024 * 1024 * 1024; |
3190 | max_chunk_size = 10 * max_stripe_size; | |
9b3f68b9 | 3191 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
3192 | /* for larger filesystems, use larger metadata chunks */ |
3193 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
3194 | max_stripe_size = 1024 * 1024 * 1024; | |
3195 | else | |
3196 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 3197 | max_chunk_size = max_stripe_size; |
a40a90a0 | 3198 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 3199 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 AJ |
3200 | max_chunk_size = 2 * max_stripe_size; |
3201 | } else { | |
3202 | printk(KERN_ERR "btrfs: invalid chunk type 0x%llx requested\n", | |
3203 | type); | |
3204 | BUG_ON(1); | |
9b3f68b9 CM |
3205 | } |
3206 | ||
2b82032c YZ |
3207 | /* we don't want a chunk larger than 10% of writeable space */ |
3208 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
3209 | max_chunk_size); | |
9b3f68b9 | 3210 | |
73c5de00 AJ |
3211 | devices_info = kzalloc(sizeof(*devices_info) * fs_devices->rw_devices, |
3212 | GFP_NOFS); | |
3213 | if (!devices_info) | |
3214 | return -ENOMEM; | |
0cad8a11 | 3215 | |
73c5de00 | 3216 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 3217 | |
9f680ce0 | 3218 | /* |
73c5de00 AJ |
3219 | * in the first pass through the devices list, we gather information |
3220 | * about the available holes on each device. | |
9f680ce0 | 3221 | */ |
73c5de00 AJ |
3222 | ndevs = 0; |
3223 | while (cur != &fs_devices->alloc_list) { | |
3224 | struct btrfs_device *device; | |
3225 | u64 max_avail; | |
3226 | u64 dev_offset; | |
b2117a39 | 3227 | |
73c5de00 | 3228 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 3229 | |
73c5de00 | 3230 | cur = cur->next; |
b2117a39 | 3231 | |
73c5de00 AJ |
3232 | if (!device->writeable) { |
3233 | printk(KERN_ERR | |
3234 | "btrfs: read-only device in alloc_list\n"); | |
3235 | WARN_ON(1); | |
3236 | continue; | |
3237 | } | |
b2117a39 | 3238 | |
73c5de00 AJ |
3239 | if (!device->in_fs_metadata) |
3240 | continue; | |
b2117a39 | 3241 | |
73c5de00 AJ |
3242 | if (device->total_bytes > device->bytes_used) |
3243 | total_avail = device->total_bytes - device->bytes_used; | |
3244 | else | |
3245 | total_avail = 0; | |
38c01b96 | 3246 | |
3247 | /* If there is no space on this device, skip it. */ | |
3248 | if (total_avail == 0) | |
3249 | continue; | |
b2117a39 | 3250 | |
125ccb0a | 3251 | ret = find_free_dev_extent(device, |
73c5de00 AJ |
3252 | max_stripe_size * dev_stripes, |
3253 | &dev_offset, &max_avail); | |
3254 | if (ret && ret != -ENOSPC) | |
3255 | goto error; | |
b2117a39 | 3256 | |
73c5de00 AJ |
3257 | if (ret == 0) |
3258 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 3259 | |
73c5de00 AJ |
3260 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
3261 | continue; | |
b2117a39 | 3262 | |
73c5de00 AJ |
3263 | devices_info[ndevs].dev_offset = dev_offset; |
3264 | devices_info[ndevs].max_avail = max_avail; | |
3265 | devices_info[ndevs].total_avail = total_avail; | |
3266 | devices_info[ndevs].dev = device; | |
3267 | ++ndevs; | |
3268 | } | |
b2117a39 | 3269 | |
73c5de00 AJ |
3270 | /* |
3271 | * now sort the devices by hole size / available space | |
3272 | */ | |
3273 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
3274 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 3275 | |
73c5de00 AJ |
3276 | /* round down to number of usable stripes */ |
3277 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 3278 | |
73c5de00 AJ |
3279 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
3280 | ret = -ENOSPC; | |
3281 | goto error; | |
b2117a39 | 3282 | } |
9f680ce0 | 3283 | |
73c5de00 AJ |
3284 | if (devs_max && ndevs > devs_max) |
3285 | ndevs = devs_max; | |
3286 | /* | |
3287 | * the primary goal is to maximize the number of stripes, so use as many | |
3288 | * devices as possible, even if the stripes are not maximum sized. | |
3289 | */ | |
3290 | stripe_size = devices_info[ndevs-1].max_avail; | |
3291 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 3292 | |
73c5de00 AJ |
3293 | if (stripe_size * num_stripes > max_chunk_size * ncopies) { |
3294 | stripe_size = max_chunk_size * ncopies; | |
3295 | do_div(stripe_size, num_stripes); | |
b2117a39 | 3296 | } |
b2117a39 | 3297 | |
73c5de00 AJ |
3298 | do_div(stripe_size, dev_stripes); |
3299 | do_div(stripe_size, BTRFS_STRIPE_LEN); | |
3300 | stripe_size *= BTRFS_STRIPE_LEN; | |
b2117a39 MX |
3301 | |
3302 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
3303 | if (!map) { | |
3304 | ret = -ENOMEM; | |
3305 | goto error; | |
3306 | } | |
3307 | map->num_stripes = num_stripes; | |
9b3f68b9 | 3308 | |
73c5de00 AJ |
3309 | for (i = 0; i < ndevs; ++i) { |
3310 | for (j = 0; j < dev_stripes; ++j) { | |
3311 | int s = i * dev_stripes + j; | |
3312 | map->stripes[s].dev = devices_info[i].dev; | |
3313 | map->stripes[s].physical = devices_info[i].dev_offset + | |
3314 | j * stripe_size; | |
6324fbf3 | 3315 | } |
6324fbf3 | 3316 | } |
2b82032c | 3317 | map->sector_size = extent_root->sectorsize; |
b2117a39 MX |
3318 | map->stripe_len = BTRFS_STRIPE_LEN; |
3319 | map->io_align = BTRFS_STRIPE_LEN; | |
3320 | map->io_width = BTRFS_STRIPE_LEN; | |
2b82032c | 3321 | map->type = type; |
2b82032c | 3322 | map->sub_stripes = sub_stripes; |
0b86a832 | 3323 | |
2b82032c | 3324 | *map_ret = map; |
73c5de00 | 3325 | num_bytes = stripe_size * (num_stripes / ncopies); |
0b86a832 | 3326 | |
73c5de00 AJ |
3327 | *stripe_size_out = stripe_size; |
3328 | *num_bytes_out = num_bytes; | |
0b86a832 | 3329 | |
73c5de00 | 3330 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 3331 | |
172ddd60 | 3332 | em = alloc_extent_map(); |
2b82032c | 3333 | if (!em) { |
b2117a39 MX |
3334 | ret = -ENOMEM; |
3335 | goto error; | |
593060d7 | 3336 | } |
2b82032c YZ |
3337 | em->bdev = (struct block_device *)map; |
3338 | em->start = start; | |
73c5de00 | 3339 | em->len = num_bytes; |
2b82032c YZ |
3340 | em->block_start = 0; |
3341 | em->block_len = em->len; | |
593060d7 | 3342 | |
2b82032c | 3343 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 3344 | write_lock(&em_tree->lock); |
2b82032c | 3345 | ret = add_extent_mapping(em_tree, em); |
890871be | 3346 | write_unlock(&em_tree->lock); |
2b82032c YZ |
3347 | BUG_ON(ret); |
3348 | free_extent_map(em); | |
0b86a832 | 3349 | |
2b82032c YZ |
3350 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
3351 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3352 | start, num_bytes); |
2b82032c | 3353 | BUG_ON(ret); |
611f0e00 | 3354 | |
73c5de00 AJ |
3355 | for (i = 0; i < map->num_stripes; ++i) { |
3356 | struct btrfs_device *device; | |
3357 | u64 dev_offset; | |
3358 | ||
3359 | device = map->stripes[i].dev; | |
3360 | dev_offset = map->stripes[i].physical; | |
0b86a832 CM |
3361 | |
3362 | ret = btrfs_alloc_dev_extent(trans, device, | |
2b82032c YZ |
3363 | info->chunk_root->root_key.objectid, |
3364 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
73c5de00 | 3365 | start, dev_offset, stripe_size); |
0b86a832 | 3366 | BUG_ON(ret); |
2b82032c YZ |
3367 | } |
3368 | ||
b2117a39 | 3369 | kfree(devices_info); |
2b82032c | 3370 | return 0; |
b2117a39 MX |
3371 | |
3372 | error: | |
3373 | kfree(map); | |
3374 | kfree(devices_info); | |
3375 | return ret; | |
2b82032c YZ |
3376 | } |
3377 | ||
3378 | static int __finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
3379 | struct btrfs_root *extent_root, | |
3380 | struct map_lookup *map, u64 chunk_offset, | |
3381 | u64 chunk_size, u64 stripe_size) | |
3382 | { | |
3383 | u64 dev_offset; | |
3384 | struct btrfs_key key; | |
3385 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3386 | struct btrfs_device *device; | |
3387 | struct btrfs_chunk *chunk; | |
3388 | struct btrfs_stripe *stripe; | |
3389 | size_t item_size = btrfs_chunk_item_size(map->num_stripes); | |
3390 | int index = 0; | |
3391 | int ret; | |
3392 | ||
3393 | chunk = kzalloc(item_size, GFP_NOFS); | |
3394 | if (!chunk) | |
3395 | return -ENOMEM; | |
3396 | ||
3397 | index = 0; | |
3398 | while (index < map->num_stripes) { | |
3399 | device = map->stripes[index].dev; | |
3400 | device->bytes_used += stripe_size; | |
0b86a832 CM |
3401 | ret = btrfs_update_device(trans, device); |
3402 | BUG_ON(ret); | |
2b82032c YZ |
3403 | index++; |
3404 | } | |
3405 | ||
2bf64758 JB |
3406 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
3407 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
3408 | map->num_stripes); | |
3409 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
3410 | ||
2b82032c YZ |
3411 | index = 0; |
3412 | stripe = &chunk->stripe; | |
3413 | while (index < map->num_stripes) { | |
3414 | device = map->stripes[index].dev; | |
3415 | dev_offset = map->stripes[index].physical; | |
0b86a832 | 3416 | |
e17cade2 CM |
3417 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
3418 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
3419 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 3420 | stripe++; |
0b86a832 CM |
3421 | index++; |
3422 | } | |
3423 | ||
2b82032c | 3424 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 3425 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
3426 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
3427 | btrfs_set_stack_chunk_type(chunk, map->type); | |
3428 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
3429 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
3430 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 3431 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 3432 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 3433 | |
2b82032c YZ |
3434 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3435 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3436 | key.offset = chunk_offset; | |
0b86a832 | 3437 | |
2b82032c YZ |
3438 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
3439 | BUG_ON(ret); | |
0b86a832 | 3440 | |
2b82032c | 3441 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
125ccb0a | 3442 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 3443 | item_size); |
8f18cf13 CM |
3444 | BUG_ON(ret); |
3445 | } | |
1abe9b8a | 3446 | |
0b86a832 | 3447 | kfree(chunk); |
2b82032c YZ |
3448 | return 0; |
3449 | } | |
0b86a832 | 3450 | |
2b82032c YZ |
3451 | /* |
3452 | * Chunk allocation falls into two parts. The first part does works | |
3453 | * that make the new allocated chunk useable, but not do any operation | |
3454 | * that modifies the chunk tree. The second part does the works that | |
3455 | * require modifying the chunk tree. This division is important for the | |
3456 | * bootstrap process of adding storage to a seed btrfs. | |
3457 | */ | |
3458 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
3459 | struct btrfs_root *extent_root, u64 type) | |
3460 | { | |
3461 | u64 chunk_offset; | |
3462 | u64 chunk_size; | |
3463 | u64 stripe_size; | |
3464 | struct map_lookup *map; | |
3465 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
3466 | int ret; | |
3467 | ||
3468 | ret = find_next_chunk(chunk_root, BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
3469 | &chunk_offset); | |
3470 | if (ret) | |
3471 | return ret; | |
3472 | ||
3473 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3474 | &stripe_size, chunk_offset, type); | |
3475 | if (ret) | |
3476 | return ret; | |
3477 | ||
3478 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3479 | chunk_size, stripe_size); | |
3480 | BUG_ON(ret); | |
3481 | return 0; | |
3482 | } | |
3483 | ||
d397712b | 3484 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
3485 | struct btrfs_root *root, |
3486 | struct btrfs_device *device) | |
3487 | { | |
3488 | u64 chunk_offset; | |
3489 | u64 sys_chunk_offset; | |
3490 | u64 chunk_size; | |
3491 | u64 sys_chunk_size; | |
3492 | u64 stripe_size; | |
3493 | u64 sys_stripe_size; | |
3494 | u64 alloc_profile; | |
3495 | struct map_lookup *map; | |
3496 | struct map_lookup *sys_map; | |
3497 | struct btrfs_fs_info *fs_info = root->fs_info; | |
3498 | struct btrfs_root *extent_root = fs_info->extent_root; | |
3499 | int ret; | |
3500 | ||
3501 | ret = find_next_chunk(fs_info->chunk_root, | |
3502 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, &chunk_offset); | |
92b8e897 MF |
3503 | if (ret) |
3504 | return ret; | |
2b82032c YZ |
3505 | |
3506 | alloc_profile = BTRFS_BLOCK_GROUP_METADATA | | |
6fef8df1 | 3507 | fs_info->avail_metadata_alloc_bits; |
2b82032c YZ |
3508 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3509 | ||
3510 | ret = __btrfs_alloc_chunk(trans, extent_root, &map, &chunk_size, | |
3511 | &stripe_size, chunk_offset, alloc_profile); | |
3512 | BUG_ON(ret); | |
3513 | ||
3514 | sys_chunk_offset = chunk_offset + chunk_size; | |
3515 | ||
3516 | alloc_profile = BTRFS_BLOCK_GROUP_SYSTEM | | |
6fef8df1 | 3517 | fs_info->avail_system_alloc_bits; |
2b82032c YZ |
3518 | alloc_profile = btrfs_reduce_alloc_profile(root, alloc_profile); |
3519 | ||
3520 | ret = __btrfs_alloc_chunk(trans, extent_root, &sys_map, | |
3521 | &sys_chunk_size, &sys_stripe_size, | |
3522 | sys_chunk_offset, alloc_profile); | |
3523 | BUG_ON(ret); | |
3524 | ||
3525 | ret = btrfs_add_device(trans, fs_info->chunk_root, device); | |
3526 | BUG_ON(ret); | |
3527 | ||
3528 | /* | |
3529 | * Modifying chunk tree needs allocating new blocks from both | |
3530 | * system block group and metadata block group. So we only can | |
3531 | * do operations require modifying the chunk tree after both | |
3532 | * block groups were created. | |
3533 | */ | |
3534 | ret = __finish_chunk_alloc(trans, extent_root, map, chunk_offset, | |
3535 | chunk_size, stripe_size); | |
3536 | BUG_ON(ret); | |
3537 | ||
3538 | ret = __finish_chunk_alloc(trans, extent_root, sys_map, | |
3539 | sys_chunk_offset, sys_chunk_size, | |
3540 | sys_stripe_size); | |
b248a415 | 3541 | BUG_ON(ret); |
2b82032c YZ |
3542 | return 0; |
3543 | } | |
3544 | ||
3545 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
3546 | { | |
3547 | struct extent_map *em; | |
3548 | struct map_lookup *map; | |
3549 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
3550 | int readonly = 0; | |
3551 | int i; | |
3552 | ||
890871be | 3553 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 3554 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 3555 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
3556 | if (!em) |
3557 | return 1; | |
3558 | ||
f48b9075 JB |
3559 | if (btrfs_test_opt(root, DEGRADED)) { |
3560 | free_extent_map(em); | |
3561 | return 0; | |
3562 | } | |
3563 | ||
2b82032c YZ |
3564 | map = (struct map_lookup *)em->bdev; |
3565 | for (i = 0; i < map->num_stripes; i++) { | |
3566 | if (!map->stripes[i].dev->writeable) { | |
3567 | readonly = 1; | |
3568 | break; | |
3569 | } | |
3570 | } | |
0b86a832 | 3571 | free_extent_map(em); |
2b82032c | 3572 | return readonly; |
0b86a832 CM |
3573 | } |
3574 | ||
3575 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
3576 | { | |
a8067e02 | 3577 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
3578 | } |
3579 | ||
3580 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
3581 | { | |
3582 | struct extent_map *em; | |
3583 | ||
d397712b | 3584 | while (1) { |
890871be | 3585 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
3586 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
3587 | if (em) | |
3588 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 3589 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
3590 | if (!em) |
3591 | break; | |
3592 | kfree(em->bdev); | |
3593 | /* once for us */ | |
3594 | free_extent_map(em); | |
3595 | /* once for the tree */ | |
3596 | free_extent_map(em); | |
3597 | } | |
3598 | } | |
3599 | ||
f188591e CM |
3600 | int btrfs_num_copies(struct btrfs_mapping_tree *map_tree, u64 logical, u64 len) |
3601 | { | |
3602 | struct extent_map *em; | |
3603 | struct map_lookup *map; | |
3604 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3605 | int ret; | |
3606 | ||
890871be | 3607 | read_lock(&em_tree->lock); |
f188591e | 3608 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 3609 | read_unlock(&em_tree->lock); |
f188591e CM |
3610 | BUG_ON(!em); |
3611 | ||
3612 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3613 | map = (struct map_lookup *)em->bdev; | |
3614 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
3615 | ret = map->num_stripes; | |
321aecc6 CM |
3616 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
3617 | ret = map->sub_stripes; | |
f188591e CM |
3618 | else |
3619 | ret = 1; | |
3620 | free_extent_map(em); | |
f188591e CM |
3621 | return ret; |
3622 | } | |
3623 | ||
dfe25020 CM |
3624 | static int find_live_mirror(struct map_lookup *map, int first, int num, |
3625 | int optimal) | |
3626 | { | |
3627 | int i; | |
3628 | if (map->stripes[optimal].dev->bdev) | |
3629 | return optimal; | |
3630 | for (i = first; i < first + num; i++) { | |
3631 | if (map->stripes[i].dev->bdev) | |
3632 | return i; | |
3633 | } | |
3634 | /* we couldn't find one that doesn't fail. Just return something | |
3635 | * and the io error handling code will clean up eventually | |
3636 | */ | |
3637 | return optimal; | |
3638 | } | |
3639 | ||
f2d8d74d CM |
3640 | static int __btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3641 | u64 logical, u64 *length, | |
a1d3c478 | 3642 | struct btrfs_bio **bbio_ret, |
7eaceacc | 3643 | int mirror_num) |
0b86a832 CM |
3644 | { |
3645 | struct extent_map *em; | |
3646 | struct map_lookup *map; | |
3647 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3648 | u64 offset; | |
593060d7 | 3649 | u64 stripe_offset; |
fce3bb9a | 3650 | u64 stripe_end_offset; |
593060d7 | 3651 | u64 stripe_nr; |
fce3bb9a LD |
3652 | u64 stripe_nr_orig; |
3653 | u64 stripe_nr_end; | |
593060d7 | 3654 | int stripe_index; |
cea9e445 | 3655 | int i; |
de11cc12 | 3656 | int ret = 0; |
f2d8d74d | 3657 | int num_stripes; |
a236aed1 | 3658 | int max_errors = 0; |
a1d3c478 | 3659 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 3660 | |
890871be | 3661 | read_lock(&em_tree->lock); |
0b86a832 | 3662 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 3663 | read_unlock(&em_tree->lock); |
f2d8d74d | 3664 | |
3b951516 | 3665 | if (!em) { |
d397712b CM |
3666 | printk(KERN_CRIT "unable to find logical %llu len %llu\n", |
3667 | (unsigned long long)logical, | |
3668 | (unsigned long long)*length); | |
f2d8d74d | 3669 | BUG(); |
3b951516 | 3670 | } |
0b86a832 CM |
3671 | |
3672 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
3673 | map = (struct map_lookup *)em->bdev; | |
3674 | offset = logical - em->start; | |
593060d7 | 3675 | |
f188591e CM |
3676 | if (mirror_num > map->num_stripes) |
3677 | mirror_num = 0; | |
3678 | ||
593060d7 CM |
3679 | stripe_nr = offset; |
3680 | /* | |
3681 | * stripe_nr counts the total number of stripes we have to stride | |
3682 | * to get to this block | |
3683 | */ | |
3684 | do_div(stripe_nr, map->stripe_len); | |
3685 | ||
3686 | stripe_offset = stripe_nr * map->stripe_len; | |
3687 | BUG_ON(offset < stripe_offset); | |
3688 | ||
3689 | /* stripe_offset is the offset of this block in its stripe*/ | |
3690 | stripe_offset = offset - stripe_offset; | |
3691 | ||
fce3bb9a LD |
3692 | if (rw & REQ_DISCARD) |
3693 | *length = min_t(u64, em->len - offset, *length); | |
52ba6929 | 3694 | else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
cea9e445 CM |
3695 | /* we limit the length of each bio to what fits in a stripe */ |
3696 | *length = min_t(u64, em->len - offset, | |
fce3bb9a | 3697 | map->stripe_len - stripe_offset); |
cea9e445 CM |
3698 | } else { |
3699 | *length = em->len - offset; | |
3700 | } | |
f2d8d74d | 3701 | |
a1d3c478 | 3702 | if (!bbio_ret) |
cea9e445 CM |
3703 | goto out; |
3704 | ||
f2d8d74d | 3705 | num_stripes = 1; |
cea9e445 | 3706 | stripe_index = 0; |
fce3bb9a LD |
3707 | stripe_nr_orig = stripe_nr; |
3708 | stripe_nr_end = (offset + *length + map->stripe_len - 1) & | |
3709 | (~(map->stripe_len - 1)); | |
3710 | do_div(stripe_nr_end, map->stripe_len); | |
3711 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
3712 | (offset + *length); | |
3713 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3714 | if (rw & REQ_DISCARD) | |
3715 | num_stripes = min_t(u64, map->num_stripes, | |
3716 | stripe_nr_end - stripe_nr_orig); | |
3717 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
3718 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
212a17ab | 3719 | if (rw & (REQ_WRITE | REQ_DISCARD)) |
f2d8d74d | 3720 | num_stripes = map->num_stripes; |
2fff734f | 3721 | else if (mirror_num) |
f188591e | 3722 | stripe_index = mirror_num - 1; |
dfe25020 CM |
3723 | else { |
3724 | stripe_index = find_live_mirror(map, 0, | |
3725 | map->num_stripes, | |
3726 | current->pid % map->num_stripes); | |
a1d3c478 | 3727 | mirror_num = stripe_index + 1; |
dfe25020 | 3728 | } |
2fff734f | 3729 | |
611f0e00 | 3730 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
a1d3c478 | 3731 | if (rw & (REQ_WRITE | REQ_DISCARD)) { |
f2d8d74d | 3732 | num_stripes = map->num_stripes; |
a1d3c478 | 3733 | } else if (mirror_num) { |
f188591e | 3734 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
3735 | } else { |
3736 | mirror_num = 1; | |
3737 | } | |
2fff734f | 3738 | |
321aecc6 CM |
3739 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
3740 | int factor = map->num_stripes / map->sub_stripes; | |
321aecc6 CM |
3741 | |
3742 | stripe_index = do_div(stripe_nr, factor); | |
3743 | stripe_index *= map->sub_stripes; | |
3744 | ||
7eaceacc | 3745 | if (rw & REQ_WRITE) |
f2d8d74d | 3746 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
3747 | else if (rw & REQ_DISCARD) |
3748 | num_stripes = min_t(u64, map->sub_stripes * | |
3749 | (stripe_nr_end - stripe_nr_orig), | |
3750 | map->num_stripes); | |
321aecc6 CM |
3751 | else if (mirror_num) |
3752 | stripe_index += mirror_num - 1; | |
dfe25020 CM |
3753 | else { |
3754 | stripe_index = find_live_mirror(map, stripe_index, | |
3755 | map->sub_stripes, stripe_index + | |
3756 | current->pid % map->sub_stripes); | |
a1d3c478 | 3757 | mirror_num = stripe_index + 1; |
dfe25020 | 3758 | } |
8790d502 CM |
3759 | } else { |
3760 | /* | |
3761 | * after this do_div call, stripe_nr is the number of stripes | |
3762 | * on this device we have to walk to find the data, and | |
3763 | * stripe_index is the number of our device in the stripe array | |
3764 | */ | |
3765 | stripe_index = do_div(stripe_nr, map->num_stripes); | |
a1d3c478 | 3766 | mirror_num = stripe_index + 1; |
8790d502 | 3767 | } |
593060d7 | 3768 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 3769 | |
de11cc12 LZ |
3770 | bbio = kzalloc(btrfs_bio_size(num_stripes), GFP_NOFS); |
3771 | if (!bbio) { | |
3772 | ret = -ENOMEM; | |
3773 | goto out; | |
3774 | } | |
3775 | atomic_set(&bbio->error, 0); | |
3776 | ||
fce3bb9a | 3777 | if (rw & REQ_DISCARD) { |
ec9ef7a1 LZ |
3778 | int factor = 0; |
3779 | int sub_stripes = 0; | |
3780 | u64 stripes_per_dev = 0; | |
3781 | u32 remaining_stripes = 0; | |
3782 | ||
3783 | if (map->type & | |
3784 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3785 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3786 | sub_stripes = 1; | |
3787 | else | |
3788 | sub_stripes = map->sub_stripes; | |
3789 | ||
3790 | factor = map->num_stripes / sub_stripes; | |
3791 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
3792 | stripe_nr_orig, | |
3793 | factor, | |
3794 | &remaining_stripes); | |
3795 | } | |
3796 | ||
fce3bb9a | 3797 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 3798 | bbio->stripes[i].physical = |
f2d8d74d CM |
3799 | map->stripes[stripe_index].physical + |
3800 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 3801 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 3802 | |
ec9ef7a1 LZ |
3803 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
3804 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3805 | bbio->stripes[i].length = stripes_per_dev * | |
3806 | map->stripe_len; | |
3807 | if (i / sub_stripes < remaining_stripes) | |
3808 | bbio->stripes[i].length += | |
3809 | map->stripe_len; | |
3810 | if (i < sub_stripes) | |
a1d3c478 | 3811 | bbio->stripes[i].length -= |
fce3bb9a | 3812 | stripe_offset; |
ec9ef7a1 LZ |
3813 | if ((i / sub_stripes + 1) % |
3814 | sub_stripes == remaining_stripes) | |
a1d3c478 | 3815 | bbio->stripes[i].length -= |
fce3bb9a | 3816 | stripe_end_offset; |
ec9ef7a1 LZ |
3817 | if (i == sub_stripes - 1) |
3818 | stripe_offset = 0; | |
fce3bb9a | 3819 | } else |
a1d3c478 | 3820 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
3821 | |
3822 | stripe_index++; | |
3823 | if (stripe_index == map->num_stripes) { | |
3824 | /* This could only happen for RAID0/10 */ | |
3825 | stripe_index = 0; | |
3826 | stripe_nr++; | |
3827 | } | |
3828 | } | |
3829 | } else { | |
3830 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 3831 | bbio->stripes[i].physical = |
212a17ab LT |
3832 | map->stripes[stripe_index].physical + |
3833 | stripe_offset + | |
3834 | stripe_nr * map->stripe_len; | |
a1d3c478 | 3835 | bbio->stripes[i].dev = |
212a17ab | 3836 | map->stripes[stripe_index].dev; |
fce3bb9a | 3837 | stripe_index++; |
f2d8d74d | 3838 | } |
593060d7 | 3839 | } |
de11cc12 LZ |
3840 | |
3841 | if (rw & REQ_WRITE) { | |
3842 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
3843 | BTRFS_BLOCK_GROUP_RAID10 | | |
3844 | BTRFS_BLOCK_GROUP_DUP)) { | |
3845 | max_errors = 1; | |
3846 | } | |
f2d8d74d | 3847 | } |
de11cc12 LZ |
3848 | |
3849 | *bbio_ret = bbio; | |
3850 | bbio->num_stripes = num_stripes; | |
3851 | bbio->max_errors = max_errors; | |
3852 | bbio->mirror_num = mirror_num; | |
cea9e445 | 3853 | out: |
0b86a832 | 3854 | free_extent_map(em); |
de11cc12 | 3855 | return ret; |
0b86a832 CM |
3856 | } |
3857 | ||
f2d8d74d CM |
3858 | int btrfs_map_block(struct btrfs_mapping_tree *map_tree, int rw, |
3859 | u64 logical, u64 *length, | |
a1d3c478 | 3860 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 3861 | { |
a1d3c478 | 3862 | return __btrfs_map_block(map_tree, rw, logical, length, bbio_ret, |
7eaceacc | 3863 | mirror_num); |
f2d8d74d CM |
3864 | } |
3865 | ||
a512bbf8 YZ |
3866 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
3867 | u64 chunk_start, u64 physical, u64 devid, | |
3868 | u64 **logical, int *naddrs, int *stripe_len) | |
3869 | { | |
3870 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
3871 | struct extent_map *em; | |
3872 | struct map_lookup *map; | |
3873 | u64 *buf; | |
3874 | u64 bytenr; | |
3875 | u64 length; | |
3876 | u64 stripe_nr; | |
3877 | int i, j, nr = 0; | |
3878 | ||
890871be | 3879 | read_lock(&em_tree->lock); |
a512bbf8 | 3880 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 3881 | read_unlock(&em_tree->lock); |
a512bbf8 YZ |
3882 | |
3883 | BUG_ON(!em || em->start != chunk_start); | |
3884 | map = (struct map_lookup *)em->bdev; | |
3885 | ||
3886 | length = em->len; | |
3887 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
3888 | do_div(length, map->num_stripes / map->sub_stripes); | |
3889 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
3890 | do_div(length, map->num_stripes); | |
3891 | ||
3892 | buf = kzalloc(sizeof(u64) * map->num_stripes, GFP_NOFS); | |
3893 | BUG_ON(!buf); | |
3894 | ||
3895 | for (i = 0; i < map->num_stripes; i++) { | |
3896 | if (devid && map->stripes[i].dev->devid != devid) | |
3897 | continue; | |
3898 | if (map->stripes[i].physical > physical || | |
3899 | map->stripes[i].physical + length <= physical) | |
3900 | continue; | |
3901 | ||
3902 | stripe_nr = physical - map->stripes[i].physical; | |
3903 | do_div(stripe_nr, map->stripe_len); | |
3904 | ||
3905 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
3906 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3907 | do_div(stripe_nr, map->sub_stripes); | |
3908 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
3909 | stripe_nr = stripe_nr * map->num_stripes + i; | |
3910 | } | |
3911 | bytenr = chunk_start + stripe_nr * map->stripe_len; | |
934d375b | 3912 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
3913 | for (j = 0; j < nr; j++) { |
3914 | if (buf[j] == bytenr) | |
3915 | break; | |
3916 | } | |
934d375b CM |
3917 | if (j == nr) { |
3918 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 3919 | buf[nr++] = bytenr; |
934d375b | 3920 | } |
a512bbf8 YZ |
3921 | } |
3922 | ||
a512bbf8 YZ |
3923 | *logical = buf; |
3924 | *naddrs = nr; | |
3925 | *stripe_len = map->stripe_len; | |
3926 | ||
3927 | free_extent_map(em); | |
3928 | return 0; | |
f2d8d74d CM |
3929 | } |
3930 | ||
a1d3c478 | 3931 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 3932 | { |
a1d3c478 | 3933 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 3934 | int is_orig_bio = 0; |
8790d502 | 3935 | |
8790d502 | 3936 | if (err) |
a1d3c478 | 3937 | atomic_inc(&bbio->error); |
8790d502 | 3938 | |
a1d3c478 | 3939 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
3940 | is_orig_bio = 1; |
3941 | ||
a1d3c478 | 3942 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
3943 | if (!is_orig_bio) { |
3944 | bio_put(bio); | |
a1d3c478 | 3945 | bio = bbio->orig_bio; |
7d2b4daa | 3946 | } |
a1d3c478 JS |
3947 | bio->bi_private = bbio->private; |
3948 | bio->bi_end_io = bbio->end_io; | |
2774b2ca JS |
3949 | bio->bi_bdev = (struct block_device *) |
3950 | (unsigned long)bbio->mirror_num; | |
a236aed1 CM |
3951 | /* only send an error to the higher layers if it is |
3952 | * beyond the tolerance of the multi-bio | |
3953 | */ | |
a1d3c478 | 3954 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 3955 | err = -EIO; |
5dbc8fca | 3956 | } else { |
1259ab75 CM |
3957 | /* |
3958 | * this bio is actually up to date, we didn't | |
3959 | * go over the max number of errors | |
3960 | */ | |
3961 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 3962 | err = 0; |
1259ab75 | 3963 | } |
a1d3c478 | 3964 | kfree(bbio); |
8790d502 CM |
3965 | |
3966 | bio_endio(bio, err); | |
7d2b4daa | 3967 | } else if (!is_orig_bio) { |
8790d502 CM |
3968 | bio_put(bio); |
3969 | } | |
8790d502 CM |
3970 | } |
3971 | ||
8b712842 CM |
3972 | struct async_sched { |
3973 | struct bio *bio; | |
3974 | int rw; | |
3975 | struct btrfs_fs_info *info; | |
3976 | struct btrfs_work work; | |
3977 | }; | |
3978 | ||
3979 | /* | |
3980 | * see run_scheduled_bios for a description of why bios are collected for | |
3981 | * async submit. | |
3982 | * | |
3983 | * This will add one bio to the pending list for a device and make sure | |
3984 | * the work struct is scheduled. | |
3985 | */ | |
d397712b | 3986 | static noinline int schedule_bio(struct btrfs_root *root, |
a1b32a59 CM |
3987 | struct btrfs_device *device, |
3988 | int rw, struct bio *bio) | |
8b712842 CM |
3989 | { |
3990 | int should_queue = 1; | |
ffbd517d | 3991 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
3992 | |
3993 | /* don't bother with additional async steps for reads, right now */ | |
7b6d91da | 3994 | if (!(rw & REQ_WRITE)) { |
492bb6de | 3995 | bio_get(bio); |
21adbd5c | 3996 | btrfsic_submit_bio(rw, bio); |
492bb6de | 3997 | bio_put(bio); |
8b712842 CM |
3998 | return 0; |
3999 | } | |
4000 | ||
4001 | /* | |
0986fe9e | 4002 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
4003 | * higher layers. Otherwise, the async bio makes it appear we have |
4004 | * made progress against dirty pages when we've really just put it | |
4005 | * on a queue for later | |
4006 | */ | |
0986fe9e | 4007 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 4008 | WARN_ON(bio->bi_next); |
8b712842 CM |
4009 | bio->bi_next = NULL; |
4010 | bio->bi_rw |= rw; | |
4011 | ||
4012 | spin_lock(&device->io_lock); | |
7b6d91da | 4013 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
4014 | pending_bios = &device->pending_sync_bios; |
4015 | else | |
4016 | pending_bios = &device->pending_bios; | |
8b712842 | 4017 | |
ffbd517d CM |
4018 | if (pending_bios->tail) |
4019 | pending_bios->tail->bi_next = bio; | |
8b712842 | 4020 | |
ffbd517d CM |
4021 | pending_bios->tail = bio; |
4022 | if (!pending_bios->head) | |
4023 | pending_bios->head = bio; | |
8b712842 CM |
4024 | if (device->running_pending) |
4025 | should_queue = 0; | |
4026 | ||
4027 | spin_unlock(&device->io_lock); | |
4028 | ||
4029 | if (should_queue) | |
1cc127b5 CM |
4030 | btrfs_queue_worker(&root->fs_info->submit_workers, |
4031 | &device->work); | |
8b712842 CM |
4032 | return 0; |
4033 | } | |
4034 | ||
f188591e | 4035 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 4036 | int mirror_num, int async_submit) |
0b86a832 CM |
4037 | { |
4038 | struct btrfs_mapping_tree *map_tree; | |
4039 | struct btrfs_device *dev; | |
8790d502 | 4040 | struct bio *first_bio = bio; |
a62b9401 | 4041 | u64 logical = (u64)bio->bi_sector << 9; |
0b86a832 CM |
4042 | u64 length = 0; |
4043 | u64 map_length; | |
0b86a832 | 4044 | int ret; |
8790d502 CM |
4045 | int dev_nr = 0; |
4046 | int total_devs = 1; | |
a1d3c478 | 4047 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 4048 | |
f2d8d74d | 4049 | length = bio->bi_size; |
0b86a832 CM |
4050 | map_tree = &root->fs_info->mapping_tree; |
4051 | map_length = length; | |
cea9e445 | 4052 | |
a1d3c478 | 4053 | ret = btrfs_map_block(map_tree, rw, logical, &map_length, &bbio, |
f188591e | 4054 | mirror_num); |
cea9e445 CM |
4055 | BUG_ON(ret); |
4056 | ||
a1d3c478 | 4057 | total_devs = bbio->num_stripes; |
cea9e445 | 4058 | if (map_length < length) { |
d397712b CM |
4059 | printk(KERN_CRIT "mapping failed logical %llu bio len %llu " |
4060 | "len %llu\n", (unsigned long long)logical, | |
4061 | (unsigned long long)length, | |
4062 | (unsigned long long)map_length); | |
cea9e445 CM |
4063 | BUG(); |
4064 | } | |
a1d3c478 JS |
4065 | |
4066 | bbio->orig_bio = first_bio; | |
4067 | bbio->private = first_bio->bi_private; | |
4068 | bbio->end_io = first_bio->bi_end_io; | |
4069 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
cea9e445 | 4070 | |
d397712b | 4071 | while (dev_nr < total_devs) { |
a1d3c478 JS |
4072 | if (dev_nr < total_devs - 1) { |
4073 | bio = bio_clone(first_bio, GFP_NOFS); | |
4074 | BUG_ON(!bio); | |
4075 | } else { | |
4076 | bio = first_bio; | |
8790d502 | 4077 | } |
a1d3c478 JS |
4078 | bio->bi_private = bbio; |
4079 | bio->bi_end_io = btrfs_end_bio; | |
4080 | bio->bi_sector = bbio->stripes[dev_nr].physical >> 9; | |
4081 | dev = bbio->stripes[dev_nr].dev; | |
18e503d6 | 4082 | if (dev && dev->bdev && (rw != WRITE || dev->writeable)) { |
a1d3c478 JS |
4083 | pr_debug("btrfs_map_bio: rw %d, secor=%llu, dev=%lu " |
4084 | "(%s id %llu), size=%u\n", rw, | |
4085 | (u64)bio->bi_sector, (u_long)dev->bdev->bd_dev, | |
4086 | dev->name, dev->devid, bio->bi_size); | |
dfe25020 | 4087 | bio->bi_bdev = dev->bdev; |
8b712842 CM |
4088 | if (async_submit) |
4089 | schedule_bio(root, dev, rw, bio); | |
4090 | else | |
21adbd5c | 4091 | btrfsic_submit_bio(rw, bio); |
dfe25020 CM |
4092 | } else { |
4093 | bio->bi_bdev = root->fs_info->fs_devices->latest_bdev; | |
4094 | bio->bi_sector = logical >> 9; | |
dfe25020 | 4095 | bio_endio(bio, -EIO); |
dfe25020 | 4096 | } |
8790d502 CM |
4097 | dev_nr++; |
4098 | } | |
0b86a832 CM |
4099 | return 0; |
4100 | } | |
4101 | ||
a443755f | 4102 | struct btrfs_device *btrfs_find_device(struct btrfs_root *root, u64 devid, |
2b82032c | 4103 | u8 *uuid, u8 *fsid) |
0b86a832 | 4104 | { |
2b82032c YZ |
4105 | struct btrfs_device *device; |
4106 | struct btrfs_fs_devices *cur_devices; | |
4107 | ||
4108 | cur_devices = root->fs_info->fs_devices; | |
4109 | while (cur_devices) { | |
4110 | if (!fsid || | |
4111 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4112 | device = __find_device(&cur_devices->devices, | |
4113 | devid, uuid); | |
4114 | if (device) | |
4115 | return device; | |
4116 | } | |
4117 | cur_devices = cur_devices->seed; | |
4118 | } | |
4119 | return NULL; | |
0b86a832 CM |
4120 | } |
4121 | ||
dfe25020 CM |
4122 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
4123 | u64 devid, u8 *dev_uuid) | |
4124 | { | |
4125 | struct btrfs_device *device; | |
4126 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
4127 | ||
4128 | device = kzalloc(sizeof(*device), GFP_NOFS); | |
7cbd8a83 | 4129 | if (!device) |
4130 | return NULL; | |
dfe25020 CM |
4131 | list_add(&device->dev_list, |
4132 | &fs_devices->devices); | |
dfe25020 CM |
4133 | device->dev_root = root->fs_info->dev_root; |
4134 | device->devid = devid; | |
8b712842 | 4135 | device->work.func = pending_bios_fn; |
e4404d6e | 4136 | device->fs_devices = fs_devices; |
cd02dca5 | 4137 | device->missing = 1; |
dfe25020 | 4138 | fs_devices->num_devices++; |
cd02dca5 | 4139 | fs_devices->missing_devices++; |
dfe25020 | 4140 | spin_lock_init(&device->io_lock); |
d20f7043 | 4141 | INIT_LIST_HEAD(&device->dev_alloc_list); |
dfe25020 CM |
4142 | memcpy(device->uuid, dev_uuid, BTRFS_UUID_SIZE); |
4143 | return device; | |
4144 | } | |
4145 | ||
0b86a832 CM |
4146 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
4147 | struct extent_buffer *leaf, | |
4148 | struct btrfs_chunk *chunk) | |
4149 | { | |
4150 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4151 | struct map_lookup *map; | |
4152 | struct extent_map *em; | |
4153 | u64 logical; | |
4154 | u64 length; | |
4155 | u64 devid; | |
a443755f | 4156 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 4157 | int num_stripes; |
0b86a832 | 4158 | int ret; |
593060d7 | 4159 | int i; |
0b86a832 | 4160 | |
e17cade2 CM |
4161 | logical = key->offset; |
4162 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 4163 | |
890871be | 4164 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 4165 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 4166 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
4167 | |
4168 | /* already mapped? */ | |
4169 | if (em && em->start <= logical && em->start + em->len > logical) { | |
4170 | free_extent_map(em); | |
0b86a832 CM |
4171 | return 0; |
4172 | } else if (em) { | |
4173 | free_extent_map(em); | |
4174 | } | |
0b86a832 | 4175 | |
172ddd60 | 4176 | em = alloc_extent_map(); |
0b86a832 CM |
4177 | if (!em) |
4178 | return -ENOMEM; | |
593060d7 CM |
4179 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
4180 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
4181 | if (!map) { |
4182 | free_extent_map(em); | |
4183 | return -ENOMEM; | |
4184 | } | |
4185 | ||
4186 | em->bdev = (struct block_device *)map; | |
4187 | em->start = logical; | |
4188 | em->len = length; | |
4189 | em->block_start = 0; | |
c8b97818 | 4190 | em->block_len = em->len; |
0b86a832 | 4191 | |
593060d7 CM |
4192 | map->num_stripes = num_stripes; |
4193 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
4194 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
4195 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
4196 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
4197 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 4198 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
4199 | for (i = 0; i < num_stripes; i++) { |
4200 | map->stripes[i].physical = | |
4201 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
4202 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
4203 | read_extent_buffer(leaf, uuid, (unsigned long) |
4204 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
4205 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4206 | map->stripes[i].dev = btrfs_find_device(root, devid, uuid, |
4207 | NULL); | |
dfe25020 | 4208 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
4209 | kfree(map); |
4210 | free_extent_map(em); | |
4211 | return -EIO; | |
4212 | } | |
dfe25020 CM |
4213 | if (!map->stripes[i].dev) { |
4214 | map->stripes[i].dev = | |
4215 | add_missing_dev(root, devid, uuid); | |
4216 | if (!map->stripes[i].dev) { | |
4217 | kfree(map); | |
4218 | free_extent_map(em); | |
4219 | return -EIO; | |
4220 | } | |
4221 | } | |
4222 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
4223 | } |
4224 | ||
890871be | 4225 | write_lock(&map_tree->map_tree.lock); |
0b86a832 | 4226 | ret = add_extent_mapping(&map_tree->map_tree, em); |
890871be | 4227 | write_unlock(&map_tree->map_tree.lock); |
b248a415 | 4228 | BUG_ON(ret); |
0b86a832 CM |
4229 | free_extent_map(em); |
4230 | ||
4231 | return 0; | |
4232 | } | |
4233 | ||
4234 | static int fill_device_from_item(struct extent_buffer *leaf, | |
4235 | struct btrfs_dev_item *dev_item, | |
4236 | struct btrfs_device *device) | |
4237 | { | |
4238 | unsigned long ptr; | |
0b86a832 CM |
4239 | |
4240 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
4241 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
4242 | device->total_bytes = device->disk_total_bytes; | |
0b86a832 CM |
4243 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
4244 | device->type = btrfs_device_type(leaf, dev_item); | |
4245 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
4246 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
4247 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
0b86a832 CM |
4248 | |
4249 | ptr = (unsigned long)btrfs_device_uuid(dev_item); | |
e17cade2 | 4250 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 4251 | |
0b86a832 CM |
4252 | return 0; |
4253 | } | |
4254 | ||
2b82032c YZ |
4255 | static int open_seed_devices(struct btrfs_root *root, u8 *fsid) |
4256 | { | |
4257 | struct btrfs_fs_devices *fs_devices; | |
4258 | int ret; | |
4259 | ||
b367e47f | 4260 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
4261 | |
4262 | fs_devices = root->fs_info->fs_devices->seed; | |
4263 | while (fs_devices) { | |
4264 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
4265 | ret = 0; | |
4266 | goto out; | |
4267 | } | |
4268 | fs_devices = fs_devices->seed; | |
4269 | } | |
4270 | ||
4271 | fs_devices = find_fsid(fsid); | |
4272 | if (!fs_devices) { | |
4273 | ret = -ENOENT; | |
4274 | goto out; | |
4275 | } | |
e4404d6e YZ |
4276 | |
4277 | fs_devices = clone_fs_devices(fs_devices); | |
4278 | if (IS_ERR(fs_devices)) { | |
4279 | ret = PTR_ERR(fs_devices); | |
2b82032c YZ |
4280 | goto out; |
4281 | } | |
4282 | ||
97288f2c | 4283 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 4284 | root->fs_info->bdev_holder); |
2b82032c YZ |
4285 | if (ret) |
4286 | goto out; | |
4287 | ||
4288 | if (!fs_devices->seeding) { | |
4289 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 4290 | free_fs_devices(fs_devices); |
2b82032c YZ |
4291 | ret = -EINVAL; |
4292 | goto out; | |
4293 | } | |
4294 | ||
4295 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
4296 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 4297 | out: |
2b82032c YZ |
4298 | return ret; |
4299 | } | |
4300 | ||
0d81ba5d | 4301 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
4302 | struct extent_buffer *leaf, |
4303 | struct btrfs_dev_item *dev_item) | |
4304 | { | |
4305 | struct btrfs_device *device; | |
4306 | u64 devid; | |
4307 | int ret; | |
2b82032c | 4308 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
4309 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
4310 | ||
0b86a832 | 4311 | devid = btrfs_device_id(leaf, dev_item); |
a443755f CM |
4312 | read_extent_buffer(leaf, dev_uuid, |
4313 | (unsigned long)btrfs_device_uuid(dev_item), | |
4314 | BTRFS_UUID_SIZE); | |
2b82032c YZ |
4315 | read_extent_buffer(leaf, fs_uuid, |
4316 | (unsigned long)btrfs_device_fsid(dev_item), | |
4317 | BTRFS_UUID_SIZE); | |
4318 | ||
4319 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
4320 | ret = open_seed_devices(root, fs_uuid); | |
e4404d6e | 4321 | if (ret && !btrfs_test_opt(root, DEGRADED)) |
2b82032c | 4322 | return ret; |
2b82032c YZ |
4323 | } |
4324 | ||
4325 | device = btrfs_find_device(root, devid, dev_uuid, fs_uuid); | |
4326 | if (!device || !device->bdev) { | |
e4404d6e | 4327 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
4328 | return -EIO; |
4329 | ||
4330 | if (!device) { | |
d397712b CM |
4331 | printk(KERN_WARNING "warning devid %llu missing\n", |
4332 | (unsigned long long)devid); | |
2b82032c YZ |
4333 | device = add_missing_dev(root, devid, dev_uuid); |
4334 | if (!device) | |
4335 | return -ENOMEM; | |
cd02dca5 CM |
4336 | } else if (!device->missing) { |
4337 | /* | |
4338 | * this happens when a device that was properly setup | |
4339 | * in the device info lists suddenly goes bad. | |
4340 | * device->bdev is NULL, and so we have to set | |
4341 | * device->missing to one here | |
4342 | */ | |
4343 | root->fs_info->fs_devices->missing_devices++; | |
4344 | device->missing = 1; | |
2b82032c YZ |
4345 | } |
4346 | } | |
4347 | ||
4348 | if (device->fs_devices != root->fs_info->fs_devices) { | |
4349 | BUG_ON(device->writeable); | |
4350 | if (device->generation != | |
4351 | btrfs_device_generation(leaf, dev_item)) | |
4352 | return -EINVAL; | |
6324fbf3 | 4353 | } |
0b86a832 CM |
4354 | |
4355 | fill_device_from_item(leaf, dev_item, device); | |
4356 | device->dev_root = root->fs_info->dev_root; | |
dfe25020 | 4357 | device->in_fs_metadata = 1; |
2bf64758 | 4358 | if (device->writeable) { |
2b82032c | 4359 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
4360 | spin_lock(&root->fs_info->free_chunk_lock); |
4361 | root->fs_info->free_chunk_space += device->total_bytes - | |
4362 | device->bytes_used; | |
4363 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4364 | } | |
0b86a832 | 4365 | ret = 0; |
0b86a832 CM |
4366 | return ret; |
4367 | } | |
4368 | ||
e4404d6e | 4369 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 4370 | { |
6c41761f | 4371 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 4372 | struct extent_buffer *sb; |
0b86a832 | 4373 | struct btrfs_disk_key *disk_key; |
0b86a832 | 4374 | struct btrfs_chunk *chunk; |
84eed90f CM |
4375 | u8 *ptr; |
4376 | unsigned long sb_ptr; | |
4377 | int ret = 0; | |
0b86a832 CM |
4378 | u32 num_stripes; |
4379 | u32 array_size; | |
4380 | u32 len = 0; | |
0b86a832 | 4381 | u32 cur; |
84eed90f | 4382 | struct btrfs_key key; |
0b86a832 | 4383 | |
e4404d6e | 4384 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET, |
a061fc8d CM |
4385 | BTRFS_SUPER_INFO_SIZE); |
4386 | if (!sb) | |
4387 | return -ENOMEM; | |
4388 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 4389 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
4390 | /* |
4391 | * The sb extent buffer is artifical and just used to read the system array. | |
4392 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
4393 | * pages up-to-date when the page is larger: extent does not cover the | |
4394 | * whole page and consequently check_page_uptodate does not find all | |
4395 | * the page's extents up-to-date (the hole beyond sb), | |
4396 | * write_extent_buffer then triggers a WARN_ON. | |
4397 | * | |
4398 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
4399 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
4400 | * to silence the warning eg. on PowerPC 64. | |
4401 | */ | |
4402 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 4403 | SetPageUptodate(sb->pages[0]); |
4008c04a | 4404 | |
a061fc8d | 4405 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
4406 | array_size = btrfs_super_sys_array_size(super_copy); |
4407 | ||
0b86a832 CM |
4408 | ptr = super_copy->sys_chunk_array; |
4409 | sb_ptr = offsetof(struct btrfs_super_block, sys_chunk_array); | |
4410 | cur = 0; | |
4411 | ||
4412 | while (cur < array_size) { | |
4413 | disk_key = (struct btrfs_disk_key *)ptr; | |
4414 | btrfs_disk_key_to_cpu(&key, disk_key); | |
4415 | ||
a061fc8d | 4416 | len = sizeof(*disk_key); ptr += len; |
0b86a832 CM |
4417 | sb_ptr += len; |
4418 | cur += len; | |
4419 | ||
0d81ba5d | 4420 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
0b86a832 | 4421 | chunk = (struct btrfs_chunk *)sb_ptr; |
0d81ba5d | 4422 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
4423 | if (ret) |
4424 | break; | |
0b86a832 CM |
4425 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); |
4426 | len = btrfs_chunk_item_size(num_stripes); | |
4427 | } else { | |
84eed90f CM |
4428 | ret = -EIO; |
4429 | break; | |
0b86a832 CM |
4430 | } |
4431 | ptr += len; | |
4432 | sb_ptr += len; | |
4433 | cur += len; | |
4434 | } | |
a061fc8d | 4435 | free_extent_buffer(sb); |
84eed90f | 4436 | return ret; |
0b86a832 CM |
4437 | } |
4438 | ||
4439 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
4440 | { | |
4441 | struct btrfs_path *path; | |
4442 | struct extent_buffer *leaf; | |
4443 | struct btrfs_key key; | |
4444 | struct btrfs_key found_key; | |
4445 | int ret; | |
4446 | int slot; | |
4447 | ||
4448 | root = root->fs_info->chunk_root; | |
4449 | ||
4450 | path = btrfs_alloc_path(); | |
4451 | if (!path) | |
4452 | return -ENOMEM; | |
4453 | ||
b367e47f LZ |
4454 | mutex_lock(&uuid_mutex); |
4455 | lock_chunks(root); | |
4456 | ||
0b86a832 CM |
4457 | /* first we search for all of the device items, and then we |
4458 | * read in all of the chunk items. This way we can create chunk | |
4459 | * mappings that reference all of the devices that are afound | |
4460 | */ | |
4461 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
4462 | key.offset = 0; | |
4463 | key.type = 0; | |
4464 | again: | |
4465 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
ab59381e ZL |
4466 | if (ret < 0) |
4467 | goto error; | |
d397712b | 4468 | while (1) { |
0b86a832 CM |
4469 | leaf = path->nodes[0]; |
4470 | slot = path->slots[0]; | |
4471 | if (slot >= btrfs_header_nritems(leaf)) { | |
4472 | ret = btrfs_next_leaf(root, path); | |
4473 | if (ret == 0) | |
4474 | continue; | |
4475 | if (ret < 0) | |
4476 | goto error; | |
4477 | break; | |
4478 | } | |
4479 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
4480 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4481 | if (found_key.objectid != BTRFS_DEV_ITEMS_OBJECTID) | |
4482 | break; | |
4483 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
4484 | struct btrfs_dev_item *dev_item; | |
4485 | dev_item = btrfs_item_ptr(leaf, slot, | |
4486 | struct btrfs_dev_item); | |
0d81ba5d | 4487 | ret = read_one_dev(root, leaf, dev_item); |
2b82032c YZ |
4488 | if (ret) |
4489 | goto error; | |
0b86a832 CM |
4490 | } |
4491 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
4492 | struct btrfs_chunk *chunk; | |
4493 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
4494 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
4495 | if (ret) |
4496 | goto error; | |
0b86a832 CM |
4497 | } |
4498 | path->slots[0]++; | |
4499 | } | |
4500 | if (key.objectid == BTRFS_DEV_ITEMS_OBJECTID) { | |
4501 | key.objectid = 0; | |
b3b4aa74 | 4502 | btrfs_release_path(path); |
0b86a832 CM |
4503 | goto again; |
4504 | } | |
0b86a832 CM |
4505 | ret = 0; |
4506 | error: | |
b367e47f LZ |
4507 | unlock_chunks(root); |
4508 | mutex_unlock(&uuid_mutex); | |
4509 | ||
2b82032c | 4510 | btrfs_free_path(path); |
0b86a832 CM |
4511 | return ret; |
4512 | } |