]>
Commit | Line | Data |
---|---|---|
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> | |
20 | #include <linux/slab.h> | |
21 | #include <linux/buffer_head.h> | |
22 | #include <linux/blkdev.h> | |
23 | #include <linux/random.h> | |
24 | #include <linux/iocontext.h> | |
25 | #include <linux/capability.h> | |
26 | #include <linux/ratelimit.h> | |
27 | #include <linux/kthread.h> | |
28 | #include <linux/raid/pq.h> | |
29 | #include <linux/semaphore.h> | |
30 | #include <asm/div64.h> | |
31 | #include "ctree.h" | |
32 | #include "extent_map.h" | |
33 | #include "disk-io.h" | |
34 | #include "transaction.h" | |
35 | #include "print-tree.h" | |
36 | #include "volumes.h" | |
37 | #include "raid56.h" | |
38 | #include "async-thread.h" | |
39 | #include "check-integrity.h" | |
40 | #include "rcu-string.h" | |
41 | #include "math.h" | |
42 | #include "dev-replace.h" | |
43 | #include "sysfs.h" | |
44 | ||
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { | |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
51 | .tolerated_failures = 1, | |
52 | .devs_increment = 2, | |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
60 | .tolerated_failures = 1, | |
61 | .devs_increment = 2, | |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
69 | .tolerated_failures = 0, | |
70 | .devs_increment = 1, | |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
78 | .tolerated_failures = 0, | |
79 | .devs_increment = 1, | |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
87 | .tolerated_failures = 0, | |
88 | .devs_increment = 1, | |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
96 | .tolerated_failures = 1, | |
97 | .devs_increment = 1, | |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
105 | .tolerated_failures = 2, | |
106 | .devs_increment = 1, | |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
111 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { | |
112 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, | |
113 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
114 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
115 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
116 | [BTRFS_RAID_SINGLE] = 0, | |
117 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
118 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
119 | }; | |
120 | ||
121 | static int init_first_rw_device(struct btrfs_trans_handle *trans, | |
122 | struct btrfs_root *root, | |
123 | struct btrfs_device *device); | |
124 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
125 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); | |
126 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); | |
127 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); | |
128 | static void btrfs_close_one_device(struct btrfs_device *device); | |
129 | ||
130 | DEFINE_MUTEX(uuid_mutex); | |
131 | static LIST_HEAD(fs_uuids); | |
132 | struct list_head *btrfs_get_fs_uuids(void) | |
133 | { | |
134 | return &fs_uuids; | |
135 | } | |
136 | ||
137 | static struct btrfs_fs_devices *__alloc_fs_devices(void) | |
138 | { | |
139 | struct btrfs_fs_devices *fs_devs; | |
140 | ||
141 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); | |
142 | if (!fs_devs) | |
143 | return ERR_PTR(-ENOMEM); | |
144 | ||
145 | mutex_init(&fs_devs->device_list_mutex); | |
146 | ||
147 | INIT_LIST_HEAD(&fs_devs->devices); | |
148 | INIT_LIST_HEAD(&fs_devs->resized_devices); | |
149 | INIT_LIST_HEAD(&fs_devs->alloc_list); | |
150 | INIT_LIST_HEAD(&fs_devs->list); | |
151 | ||
152 | return fs_devs; | |
153 | } | |
154 | ||
155 | /** | |
156 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
157 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
158 | * generated. | |
159 | * | |
160 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
161 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
162 | * can be destroyed with kfree() right away. | |
163 | */ | |
164 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
165 | { | |
166 | struct btrfs_fs_devices *fs_devs; | |
167 | ||
168 | fs_devs = __alloc_fs_devices(); | |
169 | if (IS_ERR(fs_devs)) | |
170 | return fs_devs; | |
171 | ||
172 | if (fsid) | |
173 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
174 | else | |
175 | generate_random_uuid(fs_devs->fsid); | |
176 | ||
177 | return fs_devs; | |
178 | } | |
179 | ||
180 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) | |
181 | { | |
182 | struct btrfs_device *device; | |
183 | WARN_ON(fs_devices->opened); | |
184 | while (!list_empty(&fs_devices->devices)) { | |
185 | device = list_entry(fs_devices->devices.next, | |
186 | struct btrfs_device, dev_list); | |
187 | list_del(&device->dev_list); | |
188 | rcu_string_free(device->name); | |
189 | kfree(device); | |
190 | } | |
191 | kfree(fs_devices); | |
192 | } | |
193 | ||
194 | static void btrfs_kobject_uevent(struct block_device *bdev, | |
195 | enum kobject_action action) | |
196 | { | |
197 | int ret; | |
198 | ||
199 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
200 | if (ret) | |
201 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", | |
202 | action, | |
203 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
204 | &disk_to_dev(bdev->bd_disk)->kobj); | |
205 | } | |
206 | ||
207 | void btrfs_cleanup_fs_uuids(void) | |
208 | { | |
209 | struct btrfs_fs_devices *fs_devices; | |
210 | ||
211 | while (!list_empty(&fs_uuids)) { | |
212 | fs_devices = list_entry(fs_uuids.next, | |
213 | struct btrfs_fs_devices, list); | |
214 | list_del(&fs_devices->list); | |
215 | free_fs_devices(fs_devices); | |
216 | } | |
217 | } | |
218 | ||
219 | static struct btrfs_device *__alloc_device(void) | |
220 | { | |
221 | struct btrfs_device *dev; | |
222 | ||
223 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); | |
224 | if (!dev) | |
225 | return ERR_PTR(-ENOMEM); | |
226 | ||
227 | INIT_LIST_HEAD(&dev->dev_list); | |
228 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
229 | INIT_LIST_HEAD(&dev->resized_list); | |
230 | ||
231 | spin_lock_init(&dev->io_lock); | |
232 | ||
233 | spin_lock_init(&dev->reada_lock); | |
234 | atomic_set(&dev->reada_in_flight, 0); | |
235 | atomic_set(&dev->dev_stats_ccnt, 0); | |
236 | btrfs_device_data_ordered_init(dev); | |
237 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
238 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
239 | ||
240 | return dev; | |
241 | } | |
242 | ||
243 | static noinline struct btrfs_device *__find_device(struct list_head *head, | |
244 | u64 devid, u8 *uuid) | |
245 | { | |
246 | struct btrfs_device *dev; | |
247 | ||
248 | list_for_each_entry(dev, head, dev_list) { | |
249 | if (dev->devid == devid && | |
250 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { | |
251 | return dev; | |
252 | } | |
253 | } | |
254 | return NULL; | |
255 | } | |
256 | ||
257 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) | |
258 | { | |
259 | struct btrfs_fs_devices *fs_devices; | |
260 | ||
261 | list_for_each_entry(fs_devices, &fs_uuids, list) { | |
262 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) | |
263 | return fs_devices; | |
264 | } | |
265 | return NULL; | |
266 | } | |
267 | ||
268 | static int | |
269 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
270 | int flush, struct block_device **bdev, | |
271 | struct buffer_head **bh) | |
272 | { | |
273 | int ret; | |
274 | ||
275 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
276 | ||
277 | if (IS_ERR(*bdev)) { | |
278 | ret = PTR_ERR(*bdev); | |
279 | goto error; | |
280 | } | |
281 | ||
282 | if (flush) | |
283 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
284 | ret = set_blocksize(*bdev, 4096); | |
285 | if (ret) { | |
286 | blkdev_put(*bdev, flags); | |
287 | goto error; | |
288 | } | |
289 | invalidate_bdev(*bdev); | |
290 | *bh = btrfs_read_dev_super(*bdev); | |
291 | if (IS_ERR(*bh)) { | |
292 | ret = PTR_ERR(*bh); | |
293 | blkdev_put(*bdev, flags); | |
294 | goto error; | |
295 | } | |
296 | ||
297 | return 0; | |
298 | ||
299 | error: | |
300 | *bdev = NULL; | |
301 | *bh = NULL; | |
302 | return ret; | |
303 | } | |
304 | ||
305 | static void requeue_list(struct btrfs_pending_bios *pending_bios, | |
306 | struct bio *head, struct bio *tail) | |
307 | { | |
308 | ||
309 | struct bio *old_head; | |
310 | ||
311 | old_head = pending_bios->head; | |
312 | pending_bios->head = head; | |
313 | if (pending_bios->tail) | |
314 | tail->bi_next = old_head; | |
315 | else | |
316 | pending_bios->tail = tail; | |
317 | } | |
318 | ||
319 | /* | |
320 | * we try to collect pending bios for a device so we don't get a large | |
321 | * number of procs sending bios down to the same device. This greatly | |
322 | * improves the schedulers ability to collect and merge the bios. | |
323 | * | |
324 | * But, it also turns into a long list of bios to process and that is sure | |
325 | * to eventually make the worker thread block. The solution here is to | |
326 | * make some progress and then put this work struct back at the end of | |
327 | * the list if the block device is congested. This way, multiple devices | |
328 | * can make progress from a single worker thread. | |
329 | */ | |
330 | static noinline void run_scheduled_bios(struct btrfs_device *device) | |
331 | { | |
332 | struct bio *pending; | |
333 | struct backing_dev_info *bdi; | |
334 | struct btrfs_fs_info *fs_info; | |
335 | struct btrfs_pending_bios *pending_bios; | |
336 | struct bio *tail; | |
337 | struct bio *cur; | |
338 | int again = 0; | |
339 | unsigned long num_run; | |
340 | unsigned long batch_run = 0; | |
341 | unsigned long limit; | |
342 | unsigned long last_waited = 0; | |
343 | int force_reg = 0; | |
344 | int sync_pending = 0; | |
345 | struct blk_plug plug; | |
346 | ||
347 | /* | |
348 | * this function runs all the bios we've collected for | |
349 | * a particular device. We don't want to wander off to | |
350 | * another device without first sending all of these down. | |
351 | * So, setup a plug here and finish it off before we return | |
352 | */ | |
353 | blk_start_plug(&plug); | |
354 | ||
355 | bdi = blk_get_backing_dev_info(device->bdev); | |
356 | fs_info = device->dev_root->fs_info; | |
357 | limit = btrfs_async_submit_limit(fs_info); | |
358 | limit = limit * 2 / 3; | |
359 | ||
360 | loop: | |
361 | spin_lock(&device->io_lock); | |
362 | ||
363 | loop_lock: | |
364 | num_run = 0; | |
365 | ||
366 | /* take all the bios off the list at once and process them | |
367 | * later on (without the lock held). But, remember the | |
368 | * tail and other pointers so the bios can be properly reinserted | |
369 | * into the list if we hit congestion | |
370 | */ | |
371 | if (!force_reg && device->pending_sync_bios.head) { | |
372 | pending_bios = &device->pending_sync_bios; | |
373 | force_reg = 1; | |
374 | } else { | |
375 | pending_bios = &device->pending_bios; | |
376 | force_reg = 0; | |
377 | } | |
378 | ||
379 | pending = pending_bios->head; | |
380 | tail = pending_bios->tail; | |
381 | WARN_ON(pending && !tail); | |
382 | ||
383 | /* | |
384 | * if pending was null this time around, no bios need processing | |
385 | * at all and we can stop. Otherwise it'll loop back up again | |
386 | * and do an additional check so no bios are missed. | |
387 | * | |
388 | * device->running_pending is used to synchronize with the | |
389 | * schedule_bio code. | |
390 | */ | |
391 | if (device->pending_sync_bios.head == NULL && | |
392 | device->pending_bios.head == NULL) { | |
393 | again = 0; | |
394 | device->running_pending = 0; | |
395 | } else { | |
396 | again = 1; | |
397 | device->running_pending = 1; | |
398 | } | |
399 | ||
400 | pending_bios->head = NULL; | |
401 | pending_bios->tail = NULL; | |
402 | ||
403 | spin_unlock(&device->io_lock); | |
404 | ||
405 | while (pending) { | |
406 | ||
407 | rmb(); | |
408 | /* we want to work on both lists, but do more bios on the | |
409 | * sync list than the regular list | |
410 | */ | |
411 | if ((num_run > 32 && | |
412 | pending_bios != &device->pending_sync_bios && | |
413 | device->pending_sync_bios.head) || | |
414 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
415 | device->pending_bios.head)) { | |
416 | spin_lock(&device->io_lock); | |
417 | requeue_list(pending_bios, pending, tail); | |
418 | goto loop_lock; | |
419 | } | |
420 | ||
421 | cur = pending; | |
422 | pending = pending->bi_next; | |
423 | cur->bi_next = NULL; | |
424 | ||
425 | /* | |
426 | * atomic_dec_return implies a barrier for waitqueue_active | |
427 | */ | |
428 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && | |
429 | waitqueue_active(&fs_info->async_submit_wait)) | |
430 | wake_up(&fs_info->async_submit_wait); | |
431 | ||
432 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); | |
433 | ||
434 | /* | |
435 | * if we're doing the sync list, record that our | |
436 | * plug has some sync requests on it | |
437 | * | |
438 | * If we're doing the regular list and there are | |
439 | * sync requests sitting around, unplug before | |
440 | * we add more | |
441 | */ | |
442 | if (pending_bios == &device->pending_sync_bios) { | |
443 | sync_pending = 1; | |
444 | } else if (sync_pending) { | |
445 | blk_finish_plug(&plug); | |
446 | blk_start_plug(&plug); | |
447 | sync_pending = 0; | |
448 | } | |
449 | ||
450 | btrfsic_submit_bio(cur->bi_rw, cur); | |
451 | num_run++; | |
452 | batch_run++; | |
453 | ||
454 | cond_resched(); | |
455 | ||
456 | /* | |
457 | * we made progress, there is more work to do and the bdi | |
458 | * is now congested. Back off and let other work structs | |
459 | * run instead | |
460 | */ | |
461 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && | |
462 | fs_info->fs_devices->open_devices > 1) { | |
463 | struct io_context *ioc; | |
464 | ||
465 | ioc = current->io_context; | |
466 | ||
467 | /* | |
468 | * the main goal here is that we don't want to | |
469 | * block if we're going to be able to submit | |
470 | * more requests without blocking. | |
471 | * | |
472 | * This code does two great things, it pokes into | |
473 | * the elevator code from a filesystem _and_ | |
474 | * it makes assumptions about how batching works. | |
475 | */ | |
476 | if (ioc && ioc->nr_batch_requests > 0 && | |
477 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
478 | (last_waited == 0 || | |
479 | ioc->last_waited == last_waited)) { | |
480 | /* | |
481 | * we want to go through our batch of | |
482 | * requests and stop. So, we copy out | |
483 | * the ioc->last_waited time and test | |
484 | * against it before looping | |
485 | */ | |
486 | last_waited = ioc->last_waited; | |
487 | cond_resched(); | |
488 | continue; | |
489 | } | |
490 | spin_lock(&device->io_lock); | |
491 | requeue_list(pending_bios, pending, tail); | |
492 | device->running_pending = 1; | |
493 | ||
494 | spin_unlock(&device->io_lock); | |
495 | btrfs_queue_work(fs_info->submit_workers, | |
496 | &device->work); | |
497 | goto done; | |
498 | } | |
499 | /* unplug every 64 requests just for good measure */ | |
500 | if (batch_run % 64 == 0) { | |
501 | blk_finish_plug(&plug); | |
502 | blk_start_plug(&plug); | |
503 | sync_pending = 0; | |
504 | } | |
505 | } | |
506 | ||
507 | cond_resched(); | |
508 | if (again) | |
509 | goto loop; | |
510 | ||
511 | spin_lock(&device->io_lock); | |
512 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
513 | goto loop_lock; | |
514 | spin_unlock(&device->io_lock); | |
515 | ||
516 | done: | |
517 | blk_finish_plug(&plug); | |
518 | } | |
519 | ||
520 | static void pending_bios_fn(struct btrfs_work *work) | |
521 | { | |
522 | struct btrfs_device *device; | |
523 | ||
524 | device = container_of(work, struct btrfs_device, work); | |
525 | run_scheduled_bios(device); | |
526 | } | |
527 | ||
528 | ||
529 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
530 | { | |
531 | struct btrfs_fs_devices *fs_devs; | |
532 | struct btrfs_device *dev; | |
533 | ||
534 | if (!cur_dev->name) | |
535 | return; | |
536 | ||
537 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
538 | int del = 1; | |
539 | ||
540 | if (fs_devs->opened) | |
541 | continue; | |
542 | if (fs_devs->seeding) | |
543 | continue; | |
544 | ||
545 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
546 | ||
547 | if (dev == cur_dev) | |
548 | continue; | |
549 | if (!dev->name) | |
550 | continue; | |
551 | ||
552 | /* | |
553 | * Todo: This won't be enough. What if the same device | |
554 | * comes back (with new uuid and) with its mapper path? | |
555 | * But for now, this does help as mostly an admin will | |
556 | * either use mapper or non mapper path throughout. | |
557 | */ | |
558 | rcu_read_lock(); | |
559 | del = strcmp(rcu_str_deref(dev->name), | |
560 | rcu_str_deref(cur_dev->name)); | |
561 | rcu_read_unlock(); | |
562 | if (!del) | |
563 | break; | |
564 | } | |
565 | ||
566 | if (!del) { | |
567 | /* delete the stale device */ | |
568 | if (fs_devs->num_devices == 1) { | |
569 | btrfs_sysfs_remove_fsid(fs_devs); | |
570 | list_del(&fs_devs->list); | |
571 | free_fs_devices(fs_devs); | |
572 | } else { | |
573 | fs_devs->num_devices--; | |
574 | list_del(&dev->dev_list); | |
575 | rcu_string_free(dev->name); | |
576 | kfree(dev); | |
577 | } | |
578 | break; | |
579 | } | |
580 | } | |
581 | } | |
582 | ||
583 | /* | |
584 | * Add new device to list of registered devices | |
585 | * | |
586 | * Returns: | |
587 | * 1 - first time device is seen | |
588 | * 0 - device already known | |
589 | * < 0 - error | |
590 | */ | |
591 | static noinline int device_list_add(const char *path, | |
592 | struct btrfs_super_block *disk_super, | |
593 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
594 | { | |
595 | struct btrfs_device *device; | |
596 | struct btrfs_fs_devices *fs_devices; | |
597 | struct rcu_string *name; | |
598 | int ret = 0; | |
599 | u64 found_transid = btrfs_super_generation(disk_super); | |
600 | ||
601 | fs_devices = find_fsid(disk_super->fsid); | |
602 | if (!fs_devices) { | |
603 | fs_devices = alloc_fs_devices(disk_super->fsid); | |
604 | if (IS_ERR(fs_devices)) | |
605 | return PTR_ERR(fs_devices); | |
606 | ||
607 | list_add(&fs_devices->list, &fs_uuids); | |
608 | ||
609 | device = NULL; | |
610 | } else { | |
611 | device = __find_device(&fs_devices->devices, devid, | |
612 | disk_super->dev_item.uuid); | |
613 | } | |
614 | ||
615 | if (!device) { | |
616 | if (fs_devices->opened) | |
617 | return -EBUSY; | |
618 | ||
619 | device = btrfs_alloc_device(NULL, &devid, | |
620 | disk_super->dev_item.uuid); | |
621 | if (IS_ERR(device)) { | |
622 | /* we can safely leave the fs_devices entry around */ | |
623 | return PTR_ERR(device); | |
624 | } | |
625 | ||
626 | name = rcu_string_strdup(path, GFP_NOFS); | |
627 | if (!name) { | |
628 | kfree(device); | |
629 | return -ENOMEM; | |
630 | } | |
631 | rcu_assign_pointer(device->name, name); | |
632 | ||
633 | mutex_lock(&fs_devices->device_list_mutex); | |
634 | list_add_rcu(&device->dev_list, &fs_devices->devices); | |
635 | fs_devices->num_devices++; | |
636 | mutex_unlock(&fs_devices->device_list_mutex); | |
637 | ||
638 | ret = 1; | |
639 | device->fs_devices = fs_devices; | |
640 | } else if (!device->name || strcmp(device->name->str, path)) { | |
641 | /* | |
642 | * When FS is already mounted. | |
643 | * 1. If you are here and if the device->name is NULL that | |
644 | * means this device was missing at time of FS mount. | |
645 | * 2. If you are here and if the device->name is different | |
646 | * from 'path' that means either | |
647 | * a. The same device disappeared and reappeared with | |
648 | * different name. or | |
649 | * b. The missing-disk-which-was-replaced, has | |
650 | * reappeared now. | |
651 | * | |
652 | * We must allow 1 and 2a above. But 2b would be a spurious | |
653 | * and unintentional. | |
654 | * | |
655 | * Further in case of 1 and 2a above, the disk at 'path' | |
656 | * would have missed some transaction when it was away and | |
657 | * in case of 2a the stale bdev has to be updated as well. | |
658 | * 2b must not be allowed at all time. | |
659 | */ | |
660 | ||
661 | /* | |
662 | * For now, we do allow update to btrfs_fs_device through the | |
663 | * btrfs dev scan cli after FS has been mounted. We're still | |
664 | * tracking a problem where systems fail mount by subvolume id | |
665 | * when we reject replacement on a mounted FS. | |
666 | */ | |
667 | if (!fs_devices->opened && found_transid < device->generation) { | |
668 | /* | |
669 | * That is if the FS is _not_ mounted and if you | |
670 | * are here, that means there is more than one | |
671 | * disk with same uuid and devid.We keep the one | |
672 | * with larger generation number or the last-in if | |
673 | * generation are equal. | |
674 | */ | |
675 | return -EEXIST; | |
676 | } | |
677 | ||
678 | name = rcu_string_strdup(path, GFP_NOFS); | |
679 | if (!name) | |
680 | return -ENOMEM; | |
681 | rcu_string_free(device->name); | |
682 | rcu_assign_pointer(device->name, name); | |
683 | if (device->missing) { | |
684 | fs_devices->missing_devices--; | |
685 | device->missing = 0; | |
686 | } | |
687 | } | |
688 | ||
689 | /* | |
690 | * Unmount does not free the btrfs_device struct but would zero | |
691 | * generation along with most of the other members. So just update | |
692 | * it back. We need it to pick the disk with largest generation | |
693 | * (as above). | |
694 | */ | |
695 | if (!fs_devices->opened) | |
696 | device->generation = found_transid; | |
697 | ||
698 | /* | |
699 | * if there is new btrfs on an already registered device, | |
700 | * then remove the stale device entry. | |
701 | */ | |
702 | if (ret > 0) | |
703 | btrfs_free_stale_device(device); | |
704 | ||
705 | *fs_devices_ret = fs_devices; | |
706 | ||
707 | return ret; | |
708 | } | |
709 | ||
710 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) | |
711 | { | |
712 | struct btrfs_fs_devices *fs_devices; | |
713 | struct btrfs_device *device; | |
714 | struct btrfs_device *orig_dev; | |
715 | ||
716 | fs_devices = alloc_fs_devices(orig->fsid); | |
717 | if (IS_ERR(fs_devices)) | |
718 | return fs_devices; | |
719 | ||
720 | mutex_lock(&orig->device_list_mutex); | |
721 | fs_devices->total_devices = orig->total_devices; | |
722 | ||
723 | /* We have held the volume lock, it is safe to get the devices. */ | |
724 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { | |
725 | struct rcu_string *name; | |
726 | ||
727 | device = btrfs_alloc_device(NULL, &orig_dev->devid, | |
728 | orig_dev->uuid); | |
729 | if (IS_ERR(device)) | |
730 | goto error; | |
731 | ||
732 | /* | |
733 | * This is ok to do without rcu read locked because we hold the | |
734 | * uuid mutex so nothing we touch in here is going to disappear. | |
735 | */ | |
736 | if (orig_dev->name) { | |
737 | name = rcu_string_strdup(orig_dev->name->str, | |
738 | GFP_KERNEL); | |
739 | if (!name) { | |
740 | kfree(device); | |
741 | goto error; | |
742 | } | |
743 | rcu_assign_pointer(device->name, name); | |
744 | } | |
745 | ||
746 | list_add(&device->dev_list, &fs_devices->devices); | |
747 | device->fs_devices = fs_devices; | |
748 | fs_devices->num_devices++; | |
749 | } | |
750 | mutex_unlock(&orig->device_list_mutex); | |
751 | return fs_devices; | |
752 | error: | |
753 | mutex_unlock(&orig->device_list_mutex); | |
754 | free_fs_devices(fs_devices); | |
755 | return ERR_PTR(-ENOMEM); | |
756 | } | |
757 | ||
758 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) | |
759 | { | |
760 | struct btrfs_device *device, *next; | |
761 | struct btrfs_device *latest_dev = NULL; | |
762 | ||
763 | mutex_lock(&uuid_mutex); | |
764 | again: | |
765 | /* This is the initialized path, it is safe to release the devices. */ | |
766 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { | |
767 | if (device->in_fs_metadata) { | |
768 | if (!device->is_tgtdev_for_dev_replace && | |
769 | (!latest_dev || | |
770 | device->generation > latest_dev->generation)) { | |
771 | latest_dev = device; | |
772 | } | |
773 | continue; | |
774 | } | |
775 | ||
776 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { | |
777 | /* | |
778 | * In the first step, keep the device which has | |
779 | * the correct fsid and the devid that is used | |
780 | * for the dev_replace procedure. | |
781 | * In the second step, the dev_replace state is | |
782 | * read from the device tree and it is known | |
783 | * whether the procedure is really active or | |
784 | * not, which means whether this device is | |
785 | * used or whether it should be removed. | |
786 | */ | |
787 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
788 | continue; | |
789 | } | |
790 | } | |
791 | if (device->bdev) { | |
792 | blkdev_put(device->bdev, device->mode); | |
793 | device->bdev = NULL; | |
794 | fs_devices->open_devices--; | |
795 | } | |
796 | if (device->writeable) { | |
797 | list_del_init(&device->dev_alloc_list); | |
798 | device->writeable = 0; | |
799 | if (!device->is_tgtdev_for_dev_replace) | |
800 | fs_devices->rw_devices--; | |
801 | } | |
802 | list_del_init(&device->dev_list); | |
803 | fs_devices->num_devices--; | |
804 | rcu_string_free(device->name); | |
805 | kfree(device); | |
806 | } | |
807 | ||
808 | if (fs_devices->seed) { | |
809 | fs_devices = fs_devices->seed; | |
810 | goto again; | |
811 | } | |
812 | ||
813 | fs_devices->latest_bdev = latest_dev->bdev; | |
814 | ||
815 | mutex_unlock(&uuid_mutex); | |
816 | } | |
817 | ||
818 | static void __free_device(struct work_struct *work) | |
819 | { | |
820 | struct btrfs_device *device; | |
821 | ||
822 | device = container_of(work, struct btrfs_device, rcu_work); | |
823 | ||
824 | if (device->bdev) | |
825 | blkdev_put(device->bdev, device->mode); | |
826 | ||
827 | rcu_string_free(device->name); | |
828 | kfree(device); | |
829 | } | |
830 | ||
831 | static void free_device(struct rcu_head *head) | |
832 | { | |
833 | struct btrfs_device *device; | |
834 | ||
835 | device = container_of(head, struct btrfs_device, rcu); | |
836 | ||
837 | INIT_WORK(&device->rcu_work, __free_device); | |
838 | schedule_work(&device->rcu_work); | |
839 | } | |
840 | ||
841 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) | |
842 | { | |
843 | struct btrfs_device *device, *tmp; | |
844 | ||
845 | if (--fs_devices->opened > 0) | |
846 | return 0; | |
847 | ||
848 | mutex_lock(&fs_devices->device_list_mutex); | |
849 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { | |
850 | btrfs_close_one_device(device); | |
851 | } | |
852 | mutex_unlock(&fs_devices->device_list_mutex); | |
853 | ||
854 | WARN_ON(fs_devices->open_devices); | |
855 | WARN_ON(fs_devices->rw_devices); | |
856 | fs_devices->opened = 0; | |
857 | fs_devices->seeding = 0; | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
862 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) | |
863 | { | |
864 | struct btrfs_fs_devices *seed_devices = NULL; | |
865 | int ret; | |
866 | ||
867 | mutex_lock(&uuid_mutex); | |
868 | ret = __btrfs_close_devices(fs_devices); | |
869 | if (!fs_devices->opened) { | |
870 | seed_devices = fs_devices->seed; | |
871 | fs_devices->seed = NULL; | |
872 | } | |
873 | mutex_unlock(&uuid_mutex); | |
874 | ||
875 | while (seed_devices) { | |
876 | fs_devices = seed_devices; | |
877 | seed_devices = fs_devices->seed; | |
878 | __btrfs_close_devices(fs_devices); | |
879 | free_fs_devices(fs_devices); | |
880 | } | |
881 | /* | |
882 | * Wait for rcu kworkers under __btrfs_close_devices | |
883 | * to finish all blkdev_puts so device is really | |
884 | * free when umount is done. | |
885 | */ | |
886 | rcu_barrier(); | |
887 | return ret; | |
888 | } | |
889 | ||
890 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
891 | fmode_t flags, void *holder) | |
892 | { | |
893 | struct request_queue *q; | |
894 | struct block_device *bdev; | |
895 | struct list_head *head = &fs_devices->devices; | |
896 | struct btrfs_device *device; | |
897 | struct btrfs_device *latest_dev = NULL; | |
898 | struct buffer_head *bh; | |
899 | struct btrfs_super_block *disk_super; | |
900 | u64 devid; | |
901 | int seeding = 1; | |
902 | int ret = 0; | |
903 | ||
904 | flags |= FMODE_EXCL; | |
905 | ||
906 | list_for_each_entry(device, head, dev_list) { | |
907 | if (device->bdev) | |
908 | continue; | |
909 | if (!device->name) | |
910 | continue; | |
911 | ||
912 | /* Just open everything we can; ignore failures here */ | |
913 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
914 | &bdev, &bh)) | |
915 | continue; | |
916 | ||
917 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
918 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
919 | if (devid != device->devid) | |
920 | goto error_brelse; | |
921 | ||
922 | if (memcmp(device->uuid, disk_super->dev_item.uuid, | |
923 | BTRFS_UUID_SIZE)) | |
924 | goto error_brelse; | |
925 | ||
926 | device->generation = btrfs_super_generation(disk_super); | |
927 | if (!latest_dev || | |
928 | device->generation > latest_dev->generation) | |
929 | latest_dev = device; | |
930 | ||
931 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { | |
932 | device->writeable = 0; | |
933 | } else { | |
934 | device->writeable = !bdev_read_only(bdev); | |
935 | seeding = 0; | |
936 | } | |
937 | ||
938 | q = bdev_get_queue(bdev); | |
939 | if (blk_queue_discard(q)) | |
940 | device->can_discard = 1; | |
941 | ||
942 | device->bdev = bdev; | |
943 | device->in_fs_metadata = 0; | |
944 | device->mode = flags; | |
945 | ||
946 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) | |
947 | fs_devices->rotating = 1; | |
948 | ||
949 | fs_devices->open_devices++; | |
950 | if (device->writeable && | |
951 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
952 | fs_devices->rw_devices++; | |
953 | list_add(&device->dev_alloc_list, | |
954 | &fs_devices->alloc_list); | |
955 | } | |
956 | brelse(bh); | |
957 | continue; | |
958 | ||
959 | error_brelse: | |
960 | brelse(bh); | |
961 | blkdev_put(bdev, flags); | |
962 | continue; | |
963 | } | |
964 | if (fs_devices->open_devices == 0) { | |
965 | ret = -EINVAL; | |
966 | goto out; | |
967 | } | |
968 | fs_devices->seeding = seeding; | |
969 | fs_devices->opened = 1; | |
970 | fs_devices->latest_bdev = latest_dev->bdev; | |
971 | fs_devices->total_rw_bytes = 0; | |
972 | out: | |
973 | return ret; | |
974 | } | |
975 | ||
976 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
977 | fmode_t flags, void *holder) | |
978 | { | |
979 | int ret; | |
980 | ||
981 | mutex_lock(&uuid_mutex); | |
982 | if (fs_devices->opened) { | |
983 | fs_devices->opened++; | |
984 | ret = 0; | |
985 | } else { | |
986 | ret = __btrfs_open_devices(fs_devices, flags, holder); | |
987 | } | |
988 | mutex_unlock(&uuid_mutex); | |
989 | return ret; | |
990 | } | |
991 | ||
992 | void btrfs_release_disk_super(struct page *page) | |
993 | { | |
994 | kunmap(page); | |
995 | put_page(page); | |
996 | } | |
997 | ||
998 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
999 | struct page **page, struct btrfs_super_block **disk_super) | |
1000 | { | |
1001 | void *p; | |
1002 | pgoff_t index; | |
1003 | ||
1004 | /* make sure our super fits in the device */ | |
1005 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1006 | return 1; | |
1007 | ||
1008 | /* make sure our super fits in the page */ | |
1009 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1010 | return 1; | |
1011 | ||
1012 | /* make sure our super doesn't straddle pages on disk */ | |
1013 | index = bytenr >> PAGE_SHIFT; | |
1014 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1015 | return 1; | |
1016 | ||
1017 | /* pull in the page with our super */ | |
1018 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1019 | index, GFP_KERNEL); | |
1020 | ||
1021 | if (IS_ERR_OR_NULL(*page)) | |
1022 | return 1; | |
1023 | ||
1024 | p = kmap(*page); | |
1025 | ||
1026 | /* align our pointer to the offset of the super block */ | |
1027 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1028 | ||
1029 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1030 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1031 | btrfs_release_disk_super(*page); | |
1032 | return 1; | |
1033 | } | |
1034 | ||
1035 | if ((*disk_super)->label[0] && | |
1036 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1037 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1038 | ||
1039 | return 0; | |
1040 | } | |
1041 | ||
1042 | /* | |
1043 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1044 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1045 | * is read via pagecache | |
1046 | */ | |
1047 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, | |
1048 | struct btrfs_fs_devices **fs_devices_ret) | |
1049 | { | |
1050 | struct btrfs_super_block *disk_super; | |
1051 | struct block_device *bdev; | |
1052 | struct page *page; | |
1053 | int ret = -EINVAL; | |
1054 | u64 devid; | |
1055 | u64 transid; | |
1056 | u64 total_devices; | |
1057 | u64 bytenr; | |
1058 | ||
1059 | /* | |
1060 | * we would like to check all the supers, but that would make | |
1061 | * a btrfs mount succeed after a mkfs from a different FS. | |
1062 | * So, we need to add a special mount option to scan for | |
1063 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1064 | */ | |
1065 | bytenr = btrfs_sb_offset(0); | |
1066 | flags |= FMODE_EXCL; | |
1067 | mutex_lock(&uuid_mutex); | |
1068 | ||
1069 | bdev = blkdev_get_by_path(path, flags, holder); | |
1070 | if (IS_ERR(bdev)) { | |
1071 | ret = PTR_ERR(bdev); | |
1072 | goto error; | |
1073 | } | |
1074 | ||
1075 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) | |
1076 | goto error_bdev_put; | |
1077 | ||
1078 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1079 | transid = btrfs_super_generation(disk_super); | |
1080 | total_devices = btrfs_super_num_devices(disk_super); | |
1081 | ||
1082 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); | |
1083 | if (ret > 0) { | |
1084 | if (disk_super->label[0]) { | |
1085 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
1086 | } else { | |
1087 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
1088 | } | |
1089 | ||
1090 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
1091 | ret = 0; | |
1092 | } | |
1093 | if (!ret && fs_devices_ret) | |
1094 | (*fs_devices_ret)->total_devices = total_devices; | |
1095 | ||
1096 | btrfs_release_disk_super(page); | |
1097 | ||
1098 | error_bdev_put: | |
1099 | blkdev_put(bdev, flags); | |
1100 | error: | |
1101 | mutex_unlock(&uuid_mutex); | |
1102 | return ret; | |
1103 | } | |
1104 | ||
1105 | /* helper to account the used device space in the range */ | |
1106 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1107 | u64 end, u64 *length) | |
1108 | { | |
1109 | struct btrfs_key key; | |
1110 | struct btrfs_root *root = device->dev_root; | |
1111 | struct btrfs_dev_extent *dev_extent; | |
1112 | struct btrfs_path *path; | |
1113 | u64 extent_end; | |
1114 | int ret; | |
1115 | int slot; | |
1116 | struct extent_buffer *l; | |
1117 | ||
1118 | *length = 0; | |
1119 | ||
1120 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) | |
1121 | return 0; | |
1122 | ||
1123 | path = btrfs_alloc_path(); | |
1124 | if (!path) | |
1125 | return -ENOMEM; | |
1126 | path->reada = READA_FORWARD; | |
1127 | ||
1128 | key.objectid = device->devid; | |
1129 | key.offset = start; | |
1130 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1131 | ||
1132 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1133 | if (ret < 0) | |
1134 | goto out; | |
1135 | if (ret > 0) { | |
1136 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1137 | if (ret < 0) | |
1138 | goto out; | |
1139 | } | |
1140 | ||
1141 | while (1) { | |
1142 | l = path->nodes[0]; | |
1143 | slot = path->slots[0]; | |
1144 | if (slot >= btrfs_header_nritems(l)) { | |
1145 | ret = btrfs_next_leaf(root, path); | |
1146 | if (ret == 0) | |
1147 | continue; | |
1148 | if (ret < 0) | |
1149 | goto out; | |
1150 | ||
1151 | break; | |
1152 | } | |
1153 | btrfs_item_key_to_cpu(l, &key, slot); | |
1154 | ||
1155 | if (key.objectid < device->devid) | |
1156 | goto next; | |
1157 | ||
1158 | if (key.objectid > device->devid) | |
1159 | break; | |
1160 | ||
1161 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
1162 | goto next; | |
1163 | ||
1164 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1165 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1166 | dev_extent); | |
1167 | if (key.offset <= start && extent_end > end) { | |
1168 | *length = end - start + 1; | |
1169 | break; | |
1170 | } else if (key.offset <= start && extent_end > start) | |
1171 | *length += extent_end - start; | |
1172 | else if (key.offset > start && extent_end <= end) | |
1173 | *length += extent_end - key.offset; | |
1174 | else if (key.offset > start && key.offset <= end) { | |
1175 | *length += end - key.offset + 1; | |
1176 | break; | |
1177 | } else if (key.offset > end) | |
1178 | break; | |
1179 | ||
1180 | next: | |
1181 | path->slots[0]++; | |
1182 | } | |
1183 | ret = 0; | |
1184 | out: | |
1185 | btrfs_free_path(path); | |
1186 | return ret; | |
1187 | } | |
1188 | ||
1189 | static int contains_pending_extent(struct btrfs_transaction *transaction, | |
1190 | struct btrfs_device *device, | |
1191 | u64 *start, u64 len) | |
1192 | { | |
1193 | struct btrfs_fs_info *fs_info = device->dev_root->fs_info; | |
1194 | struct extent_map *em; | |
1195 | struct list_head *search_list = &fs_info->pinned_chunks; | |
1196 | int ret = 0; | |
1197 | u64 physical_start = *start; | |
1198 | ||
1199 | if (transaction) | |
1200 | search_list = &transaction->pending_chunks; | |
1201 | again: | |
1202 | list_for_each_entry(em, search_list, list) { | |
1203 | struct map_lookup *map; | |
1204 | int i; | |
1205 | ||
1206 | map = em->map_lookup; | |
1207 | for (i = 0; i < map->num_stripes; i++) { | |
1208 | u64 end; | |
1209 | ||
1210 | if (map->stripes[i].dev != device) | |
1211 | continue; | |
1212 | if (map->stripes[i].physical >= physical_start + len || | |
1213 | map->stripes[i].physical + em->orig_block_len <= | |
1214 | physical_start) | |
1215 | continue; | |
1216 | /* | |
1217 | * Make sure that while processing the pinned list we do | |
1218 | * not override our *start with a lower value, because | |
1219 | * we can have pinned chunks that fall within this | |
1220 | * device hole and that have lower physical addresses | |
1221 | * than the pending chunks we processed before. If we | |
1222 | * do not take this special care we can end up getting | |
1223 | * 2 pending chunks that start at the same physical | |
1224 | * device offsets because the end offset of a pinned | |
1225 | * chunk can be equal to the start offset of some | |
1226 | * pending chunk. | |
1227 | */ | |
1228 | end = map->stripes[i].physical + em->orig_block_len; | |
1229 | if (end > *start) { | |
1230 | *start = end; | |
1231 | ret = 1; | |
1232 | } | |
1233 | } | |
1234 | } | |
1235 | if (search_list != &fs_info->pinned_chunks) { | |
1236 | search_list = &fs_info->pinned_chunks; | |
1237 | goto again; | |
1238 | } | |
1239 | ||
1240 | return ret; | |
1241 | } | |
1242 | ||
1243 | ||
1244 | /* | |
1245 | * find_free_dev_extent_start - find free space in the specified device | |
1246 | * @device: the device which we search the free space in | |
1247 | * @num_bytes: the size of the free space that we need | |
1248 | * @search_start: the position from which to begin the search | |
1249 | * @start: store the start of the free space. | |
1250 | * @len: the size of the free space. that we find, or the size | |
1251 | * of the max free space if we don't find suitable free space | |
1252 | * | |
1253 | * this uses a pretty simple search, the expectation is that it is | |
1254 | * called very infrequently and that a given device has a small number | |
1255 | * of extents | |
1256 | * | |
1257 | * @start is used to store the start of the free space if we find. But if we | |
1258 | * don't find suitable free space, it will be used to store the start position | |
1259 | * of the max free space. | |
1260 | * | |
1261 | * @len is used to store the size of the free space that we find. | |
1262 | * But if we don't find suitable free space, it is used to store the size of | |
1263 | * the max free space. | |
1264 | */ | |
1265 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, | |
1266 | struct btrfs_device *device, u64 num_bytes, | |
1267 | u64 search_start, u64 *start, u64 *len) | |
1268 | { | |
1269 | struct btrfs_key key; | |
1270 | struct btrfs_root *root = device->dev_root; | |
1271 | struct btrfs_dev_extent *dev_extent; | |
1272 | struct btrfs_path *path; | |
1273 | u64 hole_size; | |
1274 | u64 max_hole_start; | |
1275 | u64 max_hole_size; | |
1276 | u64 extent_end; | |
1277 | u64 search_end = device->total_bytes; | |
1278 | int ret; | |
1279 | int slot; | |
1280 | struct extent_buffer *l; | |
1281 | u64 min_search_start; | |
1282 | ||
1283 | /* | |
1284 | * We don't want to overwrite the superblock on the drive nor any area | |
1285 | * used by the boot loader (grub for example), so we make sure to start | |
1286 | * at an offset of at least 1MB. | |
1287 | */ | |
1288 | min_search_start = max(root->fs_info->alloc_start, 1024ull * 1024); | |
1289 | search_start = max(search_start, min_search_start); | |
1290 | ||
1291 | path = btrfs_alloc_path(); | |
1292 | if (!path) | |
1293 | return -ENOMEM; | |
1294 | ||
1295 | max_hole_start = search_start; | |
1296 | max_hole_size = 0; | |
1297 | ||
1298 | again: | |
1299 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { | |
1300 | ret = -ENOSPC; | |
1301 | goto out; | |
1302 | } | |
1303 | ||
1304 | path->reada = READA_FORWARD; | |
1305 | path->search_commit_root = 1; | |
1306 | path->skip_locking = 1; | |
1307 | ||
1308 | key.objectid = device->devid; | |
1309 | key.offset = search_start; | |
1310 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1311 | ||
1312 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1313 | if (ret < 0) | |
1314 | goto out; | |
1315 | if (ret > 0) { | |
1316 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1317 | if (ret < 0) | |
1318 | goto out; | |
1319 | } | |
1320 | ||
1321 | while (1) { | |
1322 | l = path->nodes[0]; | |
1323 | slot = path->slots[0]; | |
1324 | if (slot >= btrfs_header_nritems(l)) { | |
1325 | ret = btrfs_next_leaf(root, path); | |
1326 | if (ret == 0) | |
1327 | continue; | |
1328 | if (ret < 0) | |
1329 | goto out; | |
1330 | ||
1331 | break; | |
1332 | } | |
1333 | btrfs_item_key_to_cpu(l, &key, slot); | |
1334 | ||
1335 | if (key.objectid < device->devid) | |
1336 | goto next; | |
1337 | ||
1338 | if (key.objectid > device->devid) | |
1339 | break; | |
1340 | ||
1341 | if (key.type != BTRFS_DEV_EXTENT_KEY) | |
1342 | goto next; | |
1343 | ||
1344 | if (key.offset > search_start) { | |
1345 | hole_size = key.offset - search_start; | |
1346 | ||
1347 | /* | |
1348 | * Have to check before we set max_hole_start, otherwise | |
1349 | * we could end up sending back this offset anyway. | |
1350 | */ | |
1351 | if (contains_pending_extent(transaction, device, | |
1352 | &search_start, | |
1353 | hole_size)) { | |
1354 | if (key.offset >= search_start) { | |
1355 | hole_size = key.offset - search_start; | |
1356 | } else { | |
1357 | WARN_ON_ONCE(1); | |
1358 | hole_size = 0; | |
1359 | } | |
1360 | } | |
1361 | ||
1362 | if (hole_size > max_hole_size) { | |
1363 | max_hole_start = search_start; | |
1364 | max_hole_size = hole_size; | |
1365 | } | |
1366 | ||
1367 | /* | |
1368 | * If this free space is greater than which we need, | |
1369 | * it must be the max free space that we have found | |
1370 | * until now, so max_hole_start must point to the start | |
1371 | * of this free space and the length of this free space | |
1372 | * is stored in max_hole_size. Thus, we return | |
1373 | * max_hole_start and max_hole_size and go back to the | |
1374 | * caller. | |
1375 | */ | |
1376 | if (hole_size >= num_bytes) { | |
1377 | ret = 0; | |
1378 | goto out; | |
1379 | } | |
1380 | } | |
1381 | ||
1382 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1383 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1384 | dev_extent); | |
1385 | if (extent_end > search_start) | |
1386 | search_start = extent_end; | |
1387 | next: | |
1388 | path->slots[0]++; | |
1389 | cond_resched(); | |
1390 | } | |
1391 | ||
1392 | /* | |
1393 | * At this point, search_start should be the end of | |
1394 | * allocated dev extents, and when shrinking the device, | |
1395 | * search_end may be smaller than search_start. | |
1396 | */ | |
1397 | if (search_end > search_start) { | |
1398 | hole_size = search_end - search_start; | |
1399 | ||
1400 | if (contains_pending_extent(transaction, device, &search_start, | |
1401 | hole_size)) { | |
1402 | btrfs_release_path(path); | |
1403 | goto again; | |
1404 | } | |
1405 | ||
1406 | if (hole_size > max_hole_size) { | |
1407 | max_hole_start = search_start; | |
1408 | max_hole_size = hole_size; | |
1409 | } | |
1410 | } | |
1411 | ||
1412 | /* See above. */ | |
1413 | if (max_hole_size < num_bytes) | |
1414 | ret = -ENOSPC; | |
1415 | else | |
1416 | ret = 0; | |
1417 | ||
1418 | out: | |
1419 | btrfs_free_path(path); | |
1420 | *start = max_hole_start; | |
1421 | if (len) | |
1422 | *len = max_hole_size; | |
1423 | return ret; | |
1424 | } | |
1425 | ||
1426 | int find_free_dev_extent(struct btrfs_trans_handle *trans, | |
1427 | struct btrfs_device *device, u64 num_bytes, | |
1428 | u64 *start, u64 *len) | |
1429 | { | |
1430 | /* FIXME use last free of some kind */ | |
1431 | return find_free_dev_extent_start(trans->transaction, device, | |
1432 | num_bytes, 0, start, len); | |
1433 | } | |
1434 | ||
1435 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, | |
1436 | struct btrfs_device *device, | |
1437 | u64 start, u64 *dev_extent_len) | |
1438 | { | |
1439 | int ret; | |
1440 | struct btrfs_path *path; | |
1441 | struct btrfs_root *root = device->dev_root; | |
1442 | struct btrfs_key key; | |
1443 | struct btrfs_key found_key; | |
1444 | struct extent_buffer *leaf = NULL; | |
1445 | struct btrfs_dev_extent *extent = NULL; | |
1446 | ||
1447 | path = btrfs_alloc_path(); | |
1448 | if (!path) | |
1449 | return -ENOMEM; | |
1450 | ||
1451 | key.objectid = device->devid; | |
1452 | key.offset = start; | |
1453 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1454 | again: | |
1455 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1456 | if (ret > 0) { | |
1457 | ret = btrfs_previous_item(root, path, key.objectid, | |
1458 | BTRFS_DEV_EXTENT_KEY); | |
1459 | if (ret) | |
1460 | goto out; | |
1461 | leaf = path->nodes[0]; | |
1462 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1463 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1464 | struct btrfs_dev_extent); | |
1465 | BUG_ON(found_key.offset > start || found_key.offset + | |
1466 | btrfs_dev_extent_length(leaf, extent) < start); | |
1467 | key = found_key; | |
1468 | btrfs_release_path(path); | |
1469 | goto again; | |
1470 | } else if (ret == 0) { | |
1471 | leaf = path->nodes[0]; | |
1472 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1473 | struct btrfs_dev_extent); | |
1474 | } else { | |
1475 | btrfs_std_error(root->fs_info, ret, "Slot search failed"); | |
1476 | goto out; | |
1477 | } | |
1478 | ||
1479 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); | |
1480 | ||
1481 | ret = btrfs_del_item(trans, root, path); | |
1482 | if (ret) { | |
1483 | btrfs_std_error(root->fs_info, ret, | |
1484 | "Failed to remove dev extent item"); | |
1485 | } else { | |
1486 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); | |
1487 | } | |
1488 | out: | |
1489 | btrfs_free_path(path); | |
1490 | return ret; | |
1491 | } | |
1492 | ||
1493 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, | |
1494 | struct btrfs_device *device, | |
1495 | u64 chunk_tree, u64 chunk_objectid, | |
1496 | u64 chunk_offset, u64 start, u64 num_bytes) | |
1497 | { | |
1498 | int ret; | |
1499 | struct btrfs_path *path; | |
1500 | struct btrfs_root *root = device->dev_root; | |
1501 | struct btrfs_dev_extent *extent; | |
1502 | struct extent_buffer *leaf; | |
1503 | struct btrfs_key key; | |
1504 | ||
1505 | WARN_ON(!device->in_fs_metadata); | |
1506 | WARN_ON(device->is_tgtdev_for_dev_replace); | |
1507 | path = btrfs_alloc_path(); | |
1508 | if (!path) | |
1509 | return -ENOMEM; | |
1510 | ||
1511 | key.objectid = device->devid; | |
1512 | key.offset = start; | |
1513 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1514 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1515 | sizeof(*extent)); | |
1516 | if (ret) | |
1517 | goto out; | |
1518 | ||
1519 | leaf = path->nodes[0]; | |
1520 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1521 | struct btrfs_dev_extent); | |
1522 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); | |
1523 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1524 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1525 | ||
1526 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
1527 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); | |
1528 | ||
1529 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); | |
1530 | btrfs_mark_buffer_dirty(leaf); | |
1531 | out: | |
1532 | btrfs_free_path(path); | |
1533 | return ret; | |
1534 | } | |
1535 | ||
1536 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) | |
1537 | { | |
1538 | struct extent_map_tree *em_tree; | |
1539 | struct extent_map *em; | |
1540 | struct rb_node *n; | |
1541 | u64 ret = 0; | |
1542 | ||
1543 | em_tree = &fs_info->mapping_tree.map_tree; | |
1544 | read_lock(&em_tree->lock); | |
1545 | n = rb_last(&em_tree->map); | |
1546 | if (n) { | |
1547 | em = rb_entry(n, struct extent_map, rb_node); | |
1548 | ret = em->start + em->len; | |
1549 | } | |
1550 | read_unlock(&em_tree->lock); | |
1551 | ||
1552 | return ret; | |
1553 | } | |
1554 | ||
1555 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, | |
1556 | u64 *devid_ret) | |
1557 | { | |
1558 | int ret; | |
1559 | struct btrfs_key key; | |
1560 | struct btrfs_key found_key; | |
1561 | struct btrfs_path *path; | |
1562 | ||
1563 | path = btrfs_alloc_path(); | |
1564 | if (!path) | |
1565 | return -ENOMEM; | |
1566 | ||
1567 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1568 | key.type = BTRFS_DEV_ITEM_KEY; | |
1569 | key.offset = (u64)-1; | |
1570 | ||
1571 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); | |
1572 | if (ret < 0) | |
1573 | goto error; | |
1574 | ||
1575 | BUG_ON(ret == 0); /* Corruption */ | |
1576 | ||
1577 | ret = btrfs_previous_item(fs_info->chunk_root, path, | |
1578 | BTRFS_DEV_ITEMS_OBJECTID, | |
1579 | BTRFS_DEV_ITEM_KEY); | |
1580 | if (ret) { | |
1581 | *devid_ret = 1; | |
1582 | } else { | |
1583 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1584 | path->slots[0]); | |
1585 | *devid_ret = found_key.offset + 1; | |
1586 | } | |
1587 | ret = 0; | |
1588 | error: | |
1589 | btrfs_free_path(path); | |
1590 | return ret; | |
1591 | } | |
1592 | ||
1593 | /* | |
1594 | * the device information is stored in the chunk root | |
1595 | * the btrfs_device struct should be fully filled in | |
1596 | */ | |
1597 | static int btrfs_add_device(struct btrfs_trans_handle *trans, | |
1598 | struct btrfs_root *root, | |
1599 | struct btrfs_device *device) | |
1600 | { | |
1601 | int ret; | |
1602 | struct btrfs_path *path; | |
1603 | struct btrfs_dev_item *dev_item; | |
1604 | struct extent_buffer *leaf; | |
1605 | struct btrfs_key key; | |
1606 | unsigned long ptr; | |
1607 | ||
1608 | root = root->fs_info->chunk_root; | |
1609 | ||
1610 | path = btrfs_alloc_path(); | |
1611 | if (!path) | |
1612 | return -ENOMEM; | |
1613 | ||
1614 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1615 | key.type = BTRFS_DEV_ITEM_KEY; | |
1616 | key.offset = device->devid; | |
1617 | ||
1618 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1619 | sizeof(*dev_item)); | |
1620 | if (ret) | |
1621 | goto out; | |
1622 | ||
1623 | leaf = path->nodes[0]; | |
1624 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1625 | ||
1626 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
1627 | btrfs_set_device_generation(leaf, dev_item, 0); | |
1628 | btrfs_set_device_type(leaf, dev_item, device->type); | |
1629 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1630 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1631 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
1632 | btrfs_set_device_total_bytes(leaf, dev_item, | |
1633 | btrfs_device_get_disk_total_bytes(device)); | |
1634 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1635 | btrfs_device_get_bytes_used(device)); | |
1636 | btrfs_set_device_group(leaf, dev_item, 0); | |
1637 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1638 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
1639 | btrfs_set_device_start_offset(leaf, dev_item, 0); | |
1640 | ||
1641 | ptr = btrfs_device_uuid(dev_item); | |
1642 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); | |
1643 | ptr = btrfs_device_fsid(dev_item); | |
1644 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); | |
1645 | btrfs_mark_buffer_dirty(leaf); | |
1646 | ||
1647 | ret = 0; | |
1648 | out: | |
1649 | btrfs_free_path(path); | |
1650 | return ret; | |
1651 | } | |
1652 | ||
1653 | /* | |
1654 | * Function to update ctime/mtime for a given device path. | |
1655 | * Mainly used for ctime/mtime based probe like libblkid. | |
1656 | */ | |
1657 | static void update_dev_time(char *path_name) | |
1658 | { | |
1659 | struct file *filp; | |
1660 | ||
1661 | filp = filp_open(path_name, O_RDWR, 0); | |
1662 | if (IS_ERR(filp)) | |
1663 | return; | |
1664 | file_update_time(filp); | |
1665 | filp_close(filp, NULL); | |
1666 | } | |
1667 | ||
1668 | static int btrfs_rm_dev_item(struct btrfs_root *root, | |
1669 | struct btrfs_device *device) | |
1670 | { | |
1671 | int ret; | |
1672 | struct btrfs_path *path; | |
1673 | struct btrfs_key key; | |
1674 | struct btrfs_trans_handle *trans; | |
1675 | ||
1676 | root = root->fs_info->chunk_root; | |
1677 | ||
1678 | path = btrfs_alloc_path(); | |
1679 | if (!path) | |
1680 | return -ENOMEM; | |
1681 | ||
1682 | trans = btrfs_start_transaction(root, 0); | |
1683 | if (IS_ERR(trans)) { | |
1684 | btrfs_free_path(path); | |
1685 | return PTR_ERR(trans); | |
1686 | } | |
1687 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1688 | key.type = BTRFS_DEV_ITEM_KEY; | |
1689 | key.offset = device->devid; | |
1690 | ||
1691 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1692 | if (ret < 0) | |
1693 | goto out; | |
1694 | ||
1695 | if (ret > 0) { | |
1696 | ret = -ENOENT; | |
1697 | goto out; | |
1698 | } | |
1699 | ||
1700 | ret = btrfs_del_item(trans, root, path); | |
1701 | if (ret) | |
1702 | goto out; | |
1703 | out: | |
1704 | btrfs_free_path(path); | |
1705 | btrfs_commit_transaction(trans, root); | |
1706 | return ret; | |
1707 | } | |
1708 | ||
1709 | /* | |
1710 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1711 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1712 | * replace. | |
1713 | */ | |
1714 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1715 | u64 num_devices) | |
1716 | { | |
1717 | u64 all_avail; | |
1718 | unsigned seq; | |
1719 | ||
1720 | do { | |
1721 | seq = read_seqbegin(&fs_info->profiles_lock); | |
1722 | ||
1723 | all_avail = fs_info->avail_data_alloc_bits | | |
1724 | fs_info->avail_system_alloc_bits | | |
1725 | fs_info->avail_metadata_alloc_bits; | |
1726 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
1727 | ||
1728 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices < 4) { | |
1729 | return BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; | |
1730 | } | |
1731 | ||
1732 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices < 2) { | |
1733 | return BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; | |
1734 | } | |
1735 | ||
1736 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && | |
1737 | fs_info->fs_devices->rw_devices < 2) { | |
1738 | return BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; | |
1739 | } | |
1740 | ||
1741 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1742 | fs_info->fs_devices->rw_devices < 3) { | |
1743 | return BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; | |
1744 | } | |
1745 | ||
1746 | return 0; | |
1747 | } | |
1748 | ||
1749 | int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid) | |
1750 | { | |
1751 | struct btrfs_device *device; | |
1752 | struct btrfs_device *next_device; | |
1753 | struct btrfs_fs_devices *cur_devices; | |
1754 | u64 num_devices; | |
1755 | int ret = 0; | |
1756 | bool clear_super = false; | |
1757 | char *dev_name = NULL; | |
1758 | ||
1759 | mutex_lock(&uuid_mutex); | |
1760 | ||
1761 | num_devices = root->fs_info->fs_devices->num_devices; | |
1762 | btrfs_dev_replace_lock(&root->fs_info->dev_replace, 0); | |
1763 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1764 | WARN_ON(num_devices < 1); | |
1765 | num_devices--; | |
1766 | } | |
1767 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace, 0); | |
1768 | ||
1769 | ret = btrfs_check_raid_min_devices(root->fs_info, num_devices - 1); | |
1770 | if (ret) | |
1771 | goto out; | |
1772 | ||
1773 | ret = btrfs_find_device_by_user_input(root, devid, device_path, | |
1774 | &device); | |
1775 | if (ret) | |
1776 | goto out; | |
1777 | ||
1778 | if (device->is_tgtdev_for_dev_replace) { | |
1779 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; | |
1780 | goto out; | |
1781 | } | |
1782 | ||
1783 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { | |
1784 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; | |
1785 | goto out; | |
1786 | } | |
1787 | ||
1788 | if (device->writeable) { | |
1789 | lock_chunks(root); | |
1790 | list_del_init(&device->dev_alloc_list); | |
1791 | device->fs_devices->rw_devices--; | |
1792 | unlock_chunks(root); | |
1793 | dev_name = kstrdup(device->name->str, GFP_KERNEL); | |
1794 | if (!dev_name) { | |
1795 | ret = -ENOMEM; | |
1796 | goto error_undo; | |
1797 | } | |
1798 | clear_super = true; | |
1799 | } | |
1800 | ||
1801 | mutex_unlock(&uuid_mutex); | |
1802 | ret = btrfs_shrink_device(device, 0); | |
1803 | mutex_lock(&uuid_mutex); | |
1804 | if (ret) | |
1805 | goto error_undo; | |
1806 | ||
1807 | /* | |
1808 | * TODO: the superblock still includes this device in its num_devices | |
1809 | * counter although write_all_supers() is not locked out. This | |
1810 | * could give a filesystem state which requires a degraded mount. | |
1811 | */ | |
1812 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); | |
1813 | if (ret) | |
1814 | goto error_undo; | |
1815 | ||
1816 | device->in_fs_metadata = 0; | |
1817 | btrfs_scrub_cancel_dev(root->fs_info, device); | |
1818 | ||
1819 | /* | |
1820 | * the device list mutex makes sure that we don't change | |
1821 | * the device list while someone else is writing out all | |
1822 | * the device supers. Whoever is writing all supers, should | |
1823 | * lock the device list mutex before getting the number of | |
1824 | * devices in the super block (super_copy). Conversely, | |
1825 | * whoever updates the number of devices in the super block | |
1826 | * (super_copy) should hold the device list mutex. | |
1827 | */ | |
1828 | ||
1829 | cur_devices = device->fs_devices; | |
1830 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1831 | list_del_rcu(&device->dev_list); | |
1832 | ||
1833 | device->fs_devices->num_devices--; | |
1834 | device->fs_devices->total_devices--; | |
1835 | ||
1836 | if (device->missing) | |
1837 | device->fs_devices->missing_devices--; | |
1838 | ||
1839 | next_device = list_entry(root->fs_info->fs_devices->devices.next, | |
1840 | struct btrfs_device, dev_list); | |
1841 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1842 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1843 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1844 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1845 | ||
1846 | if (device->bdev) { | |
1847 | device->fs_devices->open_devices--; | |
1848 | /* remove sysfs entry */ | |
1849 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); | |
1850 | } | |
1851 | ||
1852 | call_rcu(&device->rcu, free_device); | |
1853 | ||
1854 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; | |
1855 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
1856 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
1857 | ||
1858 | if (cur_devices->open_devices == 0) { | |
1859 | struct btrfs_fs_devices *fs_devices; | |
1860 | fs_devices = root->fs_info->fs_devices; | |
1861 | while (fs_devices) { | |
1862 | if (fs_devices->seed == cur_devices) { | |
1863 | fs_devices->seed = cur_devices->seed; | |
1864 | break; | |
1865 | } | |
1866 | fs_devices = fs_devices->seed; | |
1867 | } | |
1868 | cur_devices->seed = NULL; | |
1869 | __btrfs_close_devices(cur_devices); | |
1870 | free_fs_devices(cur_devices); | |
1871 | } | |
1872 | ||
1873 | root->fs_info->num_tolerated_disk_barrier_failures = | |
1874 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1875 | ||
1876 | /* | |
1877 | * at this point, the device is zero sized. We want to | |
1878 | * remove it from the devices list and zero out the old super | |
1879 | */ | |
1880 | if (clear_super) { | |
1881 | struct block_device *bdev; | |
1882 | ||
1883 | bdev = blkdev_get_by_path(dev_name, FMODE_READ | FMODE_EXCL, | |
1884 | root->fs_info->bdev_holder); | |
1885 | if (!IS_ERR(bdev)) { | |
1886 | btrfs_scratch_superblocks(bdev, dev_name); | |
1887 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); | |
1888 | } | |
1889 | } | |
1890 | ||
1891 | out: | |
1892 | kfree(dev_name); | |
1893 | ||
1894 | mutex_unlock(&uuid_mutex); | |
1895 | return ret; | |
1896 | ||
1897 | error_undo: | |
1898 | if (device->writeable) { | |
1899 | lock_chunks(root); | |
1900 | list_add(&device->dev_alloc_list, | |
1901 | &root->fs_info->fs_devices->alloc_list); | |
1902 | device->fs_devices->rw_devices++; | |
1903 | unlock_chunks(root); | |
1904 | } | |
1905 | goto out; | |
1906 | } | |
1907 | ||
1908 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, | |
1909 | struct btrfs_device *srcdev) | |
1910 | { | |
1911 | struct btrfs_fs_devices *fs_devices; | |
1912 | ||
1913 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); | |
1914 | ||
1915 | /* | |
1916 | * in case of fs with no seed, srcdev->fs_devices will point | |
1917 | * to fs_devices of fs_info. However when the dev being replaced is | |
1918 | * a seed dev it will point to the seed's local fs_devices. In short | |
1919 | * srcdev will have its correct fs_devices in both the cases. | |
1920 | */ | |
1921 | fs_devices = srcdev->fs_devices; | |
1922 | ||
1923 | list_del_rcu(&srcdev->dev_list); | |
1924 | list_del_rcu(&srcdev->dev_alloc_list); | |
1925 | fs_devices->num_devices--; | |
1926 | if (srcdev->missing) | |
1927 | fs_devices->missing_devices--; | |
1928 | ||
1929 | if (srcdev->writeable) { | |
1930 | fs_devices->rw_devices--; | |
1931 | /* zero out the old super if it is writable */ | |
1932 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
1933 | } | |
1934 | ||
1935 | if (srcdev->bdev) | |
1936 | fs_devices->open_devices--; | |
1937 | } | |
1938 | ||
1939 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
1940 | struct btrfs_device *srcdev) | |
1941 | { | |
1942 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
1943 | ||
1944 | call_rcu(&srcdev->rcu, free_device); | |
1945 | ||
1946 | /* | |
1947 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1948 | * zero | |
1949 | */ | |
1950 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1951 | ||
1952 | /* if this is no devs we rather delete the fs_devices */ | |
1953 | if (!fs_devices->num_devices) { | |
1954 | struct btrfs_fs_devices *tmp_fs_devices; | |
1955 | ||
1956 | tmp_fs_devices = fs_info->fs_devices; | |
1957 | while (tmp_fs_devices) { | |
1958 | if (tmp_fs_devices->seed == fs_devices) { | |
1959 | tmp_fs_devices->seed = fs_devices->seed; | |
1960 | break; | |
1961 | } | |
1962 | tmp_fs_devices = tmp_fs_devices->seed; | |
1963 | } | |
1964 | fs_devices->seed = NULL; | |
1965 | __btrfs_close_devices(fs_devices); | |
1966 | free_fs_devices(fs_devices); | |
1967 | } | |
1968 | } | |
1969 | ||
1970 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1971 | struct btrfs_device *tgtdev) | |
1972 | { | |
1973 | struct btrfs_device *next_device; | |
1974 | ||
1975 | mutex_lock(&uuid_mutex); | |
1976 | WARN_ON(!tgtdev); | |
1977 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
1978 | ||
1979 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); | |
1980 | ||
1981 | if (tgtdev->bdev) { | |
1982 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
1983 | fs_info->fs_devices->open_devices--; | |
1984 | } | |
1985 | fs_info->fs_devices->num_devices--; | |
1986 | ||
1987 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1988 | struct btrfs_device, dev_list); | |
1989 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1990 | fs_info->sb->s_bdev = next_device->bdev; | |
1991 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1992 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1993 | list_del_rcu(&tgtdev->dev_list); | |
1994 | ||
1995 | call_rcu(&tgtdev->rcu, free_device); | |
1996 | ||
1997 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
1998 | mutex_unlock(&uuid_mutex); | |
1999 | } | |
2000 | ||
2001 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, | |
2002 | struct btrfs_device **device) | |
2003 | { | |
2004 | int ret = 0; | |
2005 | struct btrfs_super_block *disk_super; | |
2006 | u64 devid; | |
2007 | u8 *dev_uuid; | |
2008 | struct block_device *bdev; | |
2009 | struct buffer_head *bh; | |
2010 | ||
2011 | *device = NULL; | |
2012 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
2013 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
2014 | if (ret) | |
2015 | return ret; | |
2016 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2017 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2018 | dev_uuid = disk_super->dev_item.uuid; | |
2019 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, | |
2020 | disk_super->fsid); | |
2021 | brelse(bh); | |
2022 | if (!*device) | |
2023 | ret = -ENOENT; | |
2024 | blkdev_put(bdev, FMODE_READ); | |
2025 | return ret; | |
2026 | } | |
2027 | ||
2028 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2029 | char *device_path, | |
2030 | struct btrfs_device **device) | |
2031 | { | |
2032 | *device = NULL; | |
2033 | if (strcmp(device_path, "missing") == 0) { | |
2034 | struct list_head *devices; | |
2035 | struct btrfs_device *tmp; | |
2036 | ||
2037 | devices = &root->fs_info->fs_devices->devices; | |
2038 | /* | |
2039 | * It is safe to read the devices since the volume_mutex | |
2040 | * is held by the caller. | |
2041 | */ | |
2042 | list_for_each_entry(tmp, devices, dev_list) { | |
2043 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2044 | *device = tmp; | |
2045 | break; | |
2046 | } | |
2047 | } | |
2048 | ||
2049 | if (!*device) | |
2050 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
2051 | ||
2052 | return 0; | |
2053 | } else { | |
2054 | return btrfs_find_device_by_path(root, device_path, device); | |
2055 | } | |
2056 | } | |
2057 | ||
2058 | int btrfs_find_device_by_user_input(struct btrfs_root *root, u64 srcdevid, | |
2059 | char *srcdev_name, | |
2060 | struct btrfs_device **device) | |
2061 | { | |
2062 | int ret; | |
2063 | ||
2064 | if (srcdevid) { | |
2065 | ret = 0; | |
2066 | *device = btrfs_find_device(root->fs_info, srcdevid, NULL, | |
2067 | NULL); | |
2068 | if (!*device) | |
2069 | ret = -ENOENT; | |
2070 | } else { | |
2071 | if (!srcdev_name || !srcdev_name[0]) | |
2072 | return -EINVAL; | |
2073 | ||
2074 | ret = btrfs_find_device_missing_or_by_path(root, srcdev_name, | |
2075 | device); | |
2076 | } | |
2077 | return ret; | |
2078 | } | |
2079 | ||
2080 | /* | |
2081 | * does all the dirty work required for changing file system's UUID. | |
2082 | */ | |
2083 | static int btrfs_prepare_sprout(struct btrfs_root *root) | |
2084 | { | |
2085 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2086 | struct btrfs_fs_devices *old_devices; | |
2087 | struct btrfs_fs_devices *seed_devices; | |
2088 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; | |
2089 | struct btrfs_device *device; | |
2090 | u64 super_flags; | |
2091 | ||
2092 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
2093 | if (!fs_devices->seeding) | |
2094 | return -EINVAL; | |
2095 | ||
2096 | seed_devices = __alloc_fs_devices(); | |
2097 | if (IS_ERR(seed_devices)) | |
2098 | return PTR_ERR(seed_devices); | |
2099 | ||
2100 | old_devices = clone_fs_devices(fs_devices); | |
2101 | if (IS_ERR(old_devices)) { | |
2102 | kfree(seed_devices); | |
2103 | return PTR_ERR(old_devices); | |
2104 | } | |
2105 | ||
2106 | list_add(&old_devices->list, &fs_uuids); | |
2107 | ||
2108 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); | |
2109 | seed_devices->opened = 1; | |
2110 | INIT_LIST_HEAD(&seed_devices->devices); | |
2111 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
2112 | mutex_init(&seed_devices->device_list_mutex); | |
2113 | ||
2114 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2115 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, | |
2116 | synchronize_rcu); | |
2117 | list_for_each_entry(device, &seed_devices->devices, dev_list) | |
2118 | device->fs_devices = seed_devices; | |
2119 | ||
2120 | lock_chunks(root); | |
2121 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); | |
2122 | unlock_chunks(root); | |
2123 | ||
2124 | fs_devices->seeding = 0; | |
2125 | fs_devices->num_devices = 0; | |
2126 | fs_devices->open_devices = 0; | |
2127 | fs_devices->missing_devices = 0; | |
2128 | fs_devices->rotating = 0; | |
2129 | fs_devices->seed = seed_devices; | |
2130 | ||
2131 | generate_random_uuid(fs_devices->fsid); | |
2132 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2133 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2134 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2135 | ||
2136 | super_flags = btrfs_super_flags(disk_super) & | |
2137 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2138 | btrfs_set_super_flags(disk_super, super_flags); | |
2139 | ||
2140 | return 0; | |
2141 | } | |
2142 | ||
2143 | /* | |
2144 | * strore the expected generation for seed devices in device items. | |
2145 | */ | |
2146 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2147 | struct btrfs_root *root) | |
2148 | { | |
2149 | struct btrfs_path *path; | |
2150 | struct extent_buffer *leaf; | |
2151 | struct btrfs_dev_item *dev_item; | |
2152 | struct btrfs_device *device; | |
2153 | struct btrfs_key key; | |
2154 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2155 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2156 | u64 devid; | |
2157 | int ret; | |
2158 | ||
2159 | path = btrfs_alloc_path(); | |
2160 | if (!path) | |
2161 | return -ENOMEM; | |
2162 | ||
2163 | root = root->fs_info->chunk_root; | |
2164 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2165 | key.offset = 0; | |
2166 | key.type = BTRFS_DEV_ITEM_KEY; | |
2167 | ||
2168 | while (1) { | |
2169 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2170 | if (ret < 0) | |
2171 | goto error; | |
2172 | ||
2173 | leaf = path->nodes[0]; | |
2174 | next_slot: | |
2175 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2176 | ret = btrfs_next_leaf(root, path); | |
2177 | if (ret > 0) | |
2178 | break; | |
2179 | if (ret < 0) | |
2180 | goto error; | |
2181 | leaf = path->nodes[0]; | |
2182 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2183 | btrfs_release_path(path); | |
2184 | continue; | |
2185 | } | |
2186 | ||
2187 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2188 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2189 | key.type != BTRFS_DEV_ITEM_KEY) | |
2190 | break; | |
2191 | ||
2192 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2193 | struct btrfs_dev_item); | |
2194 | devid = btrfs_device_id(leaf, dev_item); | |
2195 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), | |
2196 | BTRFS_UUID_SIZE); | |
2197 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), | |
2198 | BTRFS_UUID_SIZE); | |
2199 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, | |
2200 | fs_uuid); | |
2201 | BUG_ON(!device); /* Logic error */ | |
2202 | ||
2203 | if (device->fs_devices->seeding) { | |
2204 | btrfs_set_device_generation(leaf, dev_item, | |
2205 | device->generation); | |
2206 | btrfs_mark_buffer_dirty(leaf); | |
2207 | } | |
2208 | ||
2209 | path->slots[0]++; | |
2210 | goto next_slot; | |
2211 | } | |
2212 | ret = 0; | |
2213 | error: | |
2214 | btrfs_free_path(path); | |
2215 | return ret; | |
2216 | } | |
2217 | ||
2218 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) | |
2219 | { | |
2220 | struct request_queue *q; | |
2221 | struct btrfs_trans_handle *trans; | |
2222 | struct btrfs_device *device; | |
2223 | struct block_device *bdev; | |
2224 | struct list_head *devices; | |
2225 | struct super_block *sb = root->fs_info->sb; | |
2226 | struct rcu_string *name; | |
2227 | u64 tmp; | |
2228 | int seeding_dev = 0; | |
2229 | int ret = 0; | |
2230 | ||
2231 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) | |
2232 | return -EROFS; | |
2233 | ||
2234 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2235 | root->fs_info->bdev_holder); | |
2236 | if (IS_ERR(bdev)) | |
2237 | return PTR_ERR(bdev); | |
2238 | ||
2239 | if (root->fs_info->fs_devices->seeding) { | |
2240 | seeding_dev = 1; | |
2241 | down_write(&sb->s_umount); | |
2242 | mutex_lock(&uuid_mutex); | |
2243 | } | |
2244 | ||
2245 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2246 | ||
2247 | devices = &root->fs_info->fs_devices->devices; | |
2248 | ||
2249 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2250 | list_for_each_entry(device, devices, dev_list) { | |
2251 | if (device->bdev == bdev) { | |
2252 | ret = -EEXIST; | |
2253 | mutex_unlock( | |
2254 | &root->fs_info->fs_devices->device_list_mutex); | |
2255 | goto error; | |
2256 | } | |
2257 | } | |
2258 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2259 | ||
2260 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); | |
2261 | if (IS_ERR(device)) { | |
2262 | /* we can safely leave the fs_devices entry around */ | |
2263 | ret = PTR_ERR(device); | |
2264 | goto error; | |
2265 | } | |
2266 | ||
2267 | name = rcu_string_strdup(device_path, GFP_KERNEL); | |
2268 | if (!name) { | |
2269 | kfree(device); | |
2270 | ret = -ENOMEM; | |
2271 | goto error; | |
2272 | } | |
2273 | rcu_assign_pointer(device->name, name); | |
2274 | ||
2275 | trans = btrfs_start_transaction(root, 0); | |
2276 | if (IS_ERR(trans)) { | |
2277 | rcu_string_free(device->name); | |
2278 | kfree(device); | |
2279 | ret = PTR_ERR(trans); | |
2280 | goto error; | |
2281 | } | |
2282 | ||
2283 | q = bdev_get_queue(bdev); | |
2284 | if (blk_queue_discard(q)) | |
2285 | device->can_discard = 1; | |
2286 | device->writeable = 1; | |
2287 | device->generation = trans->transid; | |
2288 | device->io_width = root->sectorsize; | |
2289 | device->io_align = root->sectorsize; | |
2290 | device->sector_size = root->sectorsize; | |
2291 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2292 | device->disk_total_bytes = device->total_bytes; | |
2293 | device->commit_total_bytes = device->total_bytes; | |
2294 | device->dev_root = root->fs_info->dev_root; | |
2295 | device->bdev = bdev; | |
2296 | device->in_fs_metadata = 1; | |
2297 | device->is_tgtdev_for_dev_replace = 0; | |
2298 | device->mode = FMODE_EXCL; | |
2299 | device->dev_stats_valid = 1; | |
2300 | set_blocksize(device->bdev, 4096); | |
2301 | ||
2302 | if (seeding_dev) { | |
2303 | sb->s_flags &= ~MS_RDONLY; | |
2304 | ret = btrfs_prepare_sprout(root); | |
2305 | BUG_ON(ret); /* -ENOMEM */ | |
2306 | } | |
2307 | ||
2308 | device->fs_devices = root->fs_info->fs_devices; | |
2309 | ||
2310 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2311 | lock_chunks(root); | |
2312 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); | |
2313 | list_add(&device->dev_alloc_list, | |
2314 | &root->fs_info->fs_devices->alloc_list); | |
2315 | root->fs_info->fs_devices->num_devices++; | |
2316 | root->fs_info->fs_devices->open_devices++; | |
2317 | root->fs_info->fs_devices->rw_devices++; | |
2318 | root->fs_info->fs_devices->total_devices++; | |
2319 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
2320 | ||
2321 | spin_lock(&root->fs_info->free_chunk_lock); | |
2322 | root->fs_info->free_chunk_space += device->total_bytes; | |
2323 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2324 | ||
2325 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) | |
2326 | root->fs_info->fs_devices->rotating = 1; | |
2327 | ||
2328 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); | |
2329 | btrfs_set_super_total_bytes(root->fs_info->super_copy, | |
2330 | tmp + device->total_bytes); | |
2331 | ||
2332 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); | |
2333 | btrfs_set_super_num_devices(root->fs_info->super_copy, | |
2334 | tmp + 1); | |
2335 | ||
2336 | /* add sysfs device entry */ | |
2337 | btrfs_sysfs_add_device_link(root->fs_info->fs_devices, device); | |
2338 | ||
2339 | /* | |
2340 | * we've got more storage, clear any full flags on the space | |
2341 | * infos | |
2342 | */ | |
2343 | btrfs_clear_space_info_full(root->fs_info); | |
2344 | ||
2345 | unlock_chunks(root); | |
2346 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2347 | ||
2348 | if (seeding_dev) { | |
2349 | lock_chunks(root); | |
2350 | ret = init_first_rw_device(trans, root, device); | |
2351 | unlock_chunks(root); | |
2352 | if (ret) { | |
2353 | btrfs_abort_transaction(trans, root, ret); | |
2354 | goto error_trans; | |
2355 | } | |
2356 | } | |
2357 | ||
2358 | ret = btrfs_add_device(trans, root, device); | |
2359 | if (ret) { | |
2360 | btrfs_abort_transaction(trans, root, ret); | |
2361 | goto error_trans; | |
2362 | } | |
2363 | ||
2364 | if (seeding_dev) { | |
2365 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2366 | ||
2367 | ret = btrfs_finish_sprout(trans, root); | |
2368 | if (ret) { | |
2369 | btrfs_abort_transaction(trans, root, ret); | |
2370 | goto error_trans; | |
2371 | } | |
2372 | ||
2373 | /* Sprouting would change fsid of the mounted root, | |
2374 | * so rename the fsid on the sysfs | |
2375 | */ | |
2376 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2377 | root->fs_info->fsid); | |
2378 | if (kobject_rename(&root->fs_info->fs_devices->fsid_kobj, | |
2379 | fsid_buf)) | |
2380 | btrfs_warn(root->fs_info, | |
2381 | "sysfs: failed to create fsid for sprout"); | |
2382 | } | |
2383 | ||
2384 | root->fs_info->num_tolerated_disk_barrier_failures = | |
2385 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
2386 | ret = btrfs_commit_transaction(trans, root); | |
2387 | ||
2388 | if (seeding_dev) { | |
2389 | mutex_unlock(&uuid_mutex); | |
2390 | up_write(&sb->s_umount); | |
2391 | ||
2392 | if (ret) /* transaction commit */ | |
2393 | return ret; | |
2394 | ||
2395 | ret = btrfs_relocate_sys_chunks(root); | |
2396 | if (ret < 0) | |
2397 | btrfs_std_error(root->fs_info, ret, | |
2398 | "Failed to relocate sys chunks after " | |
2399 | "device initialization. This can be fixed " | |
2400 | "using the \"btrfs balance\" command."); | |
2401 | trans = btrfs_attach_transaction(root); | |
2402 | if (IS_ERR(trans)) { | |
2403 | if (PTR_ERR(trans) == -ENOENT) | |
2404 | return 0; | |
2405 | return PTR_ERR(trans); | |
2406 | } | |
2407 | ret = btrfs_commit_transaction(trans, root); | |
2408 | } | |
2409 | ||
2410 | /* Update ctime/mtime for libblkid */ | |
2411 | update_dev_time(device_path); | |
2412 | return ret; | |
2413 | ||
2414 | error_trans: | |
2415 | btrfs_end_transaction(trans, root); | |
2416 | rcu_string_free(device->name); | |
2417 | btrfs_sysfs_rm_device_link(root->fs_info->fs_devices, device); | |
2418 | kfree(device); | |
2419 | error: | |
2420 | blkdev_put(bdev, FMODE_EXCL); | |
2421 | if (seeding_dev) { | |
2422 | mutex_unlock(&uuid_mutex); | |
2423 | up_write(&sb->s_umount); | |
2424 | } | |
2425 | return ret; | |
2426 | } | |
2427 | ||
2428 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, | |
2429 | struct btrfs_device *srcdev, | |
2430 | struct btrfs_device **device_out) | |
2431 | { | |
2432 | struct request_queue *q; | |
2433 | struct btrfs_device *device; | |
2434 | struct block_device *bdev; | |
2435 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2436 | struct list_head *devices; | |
2437 | struct rcu_string *name; | |
2438 | u64 devid = BTRFS_DEV_REPLACE_DEVID; | |
2439 | int ret = 0; | |
2440 | ||
2441 | *device_out = NULL; | |
2442 | if (fs_info->fs_devices->seeding) { | |
2443 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
2444 | return -EINVAL; | |
2445 | } | |
2446 | ||
2447 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2448 | fs_info->bdev_holder); | |
2449 | if (IS_ERR(bdev)) { | |
2450 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
2451 | return PTR_ERR(bdev); | |
2452 | } | |
2453 | ||
2454 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2455 | ||
2456 | devices = &fs_info->fs_devices->devices; | |
2457 | list_for_each_entry(device, devices, dev_list) { | |
2458 | if (device->bdev == bdev) { | |
2459 | btrfs_err(fs_info, "target device is in the filesystem!"); | |
2460 | ret = -EEXIST; | |
2461 | goto error; | |
2462 | } | |
2463 | } | |
2464 | ||
2465 | ||
2466 | if (i_size_read(bdev->bd_inode) < | |
2467 | btrfs_device_get_total_bytes(srcdev)) { | |
2468 | btrfs_err(fs_info, "target device is smaller than source device!"); | |
2469 | ret = -EINVAL; | |
2470 | goto error; | |
2471 | } | |
2472 | ||
2473 | ||
2474 | device = btrfs_alloc_device(NULL, &devid, NULL); | |
2475 | if (IS_ERR(device)) { | |
2476 | ret = PTR_ERR(device); | |
2477 | goto error; | |
2478 | } | |
2479 | ||
2480 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2481 | if (!name) { | |
2482 | kfree(device); | |
2483 | ret = -ENOMEM; | |
2484 | goto error; | |
2485 | } | |
2486 | rcu_assign_pointer(device->name, name); | |
2487 | ||
2488 | q = bdev_get_queue(bdev); | |
2489 | if (blk_queue_discard(q)) | |
2490 | device->can_discard = 1; | |
2491 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2492 | device->writeable = 1; | |
2493 | device->generation = 0; | |
2494 | device->io_width = root->sectorsize; | |
2495 | device->io_align = root->sectorsize; | |
2496 | device->sector_size = root->sectorsize; | |
2497 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); | |
2498 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2499 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
2500 | ASSERT(list_empty(&srcdev->resized_list)); | |
2501 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
2502 | device->commit_bytes_used = device->bytes_used; | |
2503 | device->dev_root = fs_info->dev_root; | |
2504 | device->bdev = bdev; | |
2505 | device->in_fs_metadata = 1; | |
2506 | device->is_tgtdev_for_dev_replace = 1; | |
2507 | device->mode = FMODE_EXCL; | |
2508 | device->dev_stats_valid = 1; | |
2509 | set_blocksize(device->bdev, 4096); | |
2510 | device->fs_devices = fs_info->fs_devices; | |
2511 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2512 | fs_info->fs_devices->num_devices++; | |
2513 | fs_info->fs_devices->open_devices++; | |
2514 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); | |
2515 | ||
2516 | *device_out = device; | |
2517 | return ret; | |
2518 | ||
2519 | error: | |
2520 | blkdev_put(bdev, FMODE_EXCL); | |
2521 | return ret; | |
2522 | } | |
2523 | ||
2524 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2525 | struct btrfs_device *tgtdev) | |
2526 | { | |
2527 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2528 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2529 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2530 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2531 | tgtdev->dev_root = fs_info->dev_root; | |
2532 | tgtdev->in_fs_metadata = 1; | |
2533 | } | |
2534 | ||
2535 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, | |
2536 | struct btrfs_device *device) | |
2537 | { | |
2538 | int ret; | |
2539 | struct btrfs_path *path; | |
2540 | struct btrfs_root *root; | |
2541 | struct btrfs_dev_item *dev_item; | |
2542 | struct extent_buffer *leaf; | |
2543 | struct btrfs_key key; | |
2544 | ||
2545 | root = device->dev_root->fs_info->chunk_root; | |
2546 | ||
2547 | path = btrfs_alloc_path(); | |
2548 | if (!path) | |
2549 | return -ENOMEM; | |
2550 | ||
2551 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2552 | key.type = BTRFS_DEV_ITEM_KEY; | |
2553 | key.offset = device->devid; | |
2554 | ||
2555 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2556 | if (ret < 0) | |
2557 | goto out; | |
2558 | ||
2559 | if (ret > 0) { | |
2560 | ret = -ENOENT; | |
2561 | goto out; | |
2562 | } | |
2563 | ||
2564 | leaf = path->nodes[0]; | |
2565 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2566 | ||
2567 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2568 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2569 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2570 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2571 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
2572 | btrfs_set_device_total_bytes(leaf, dev_item, | |
2573 | btrfs_device_get_disk_total_bytes(device)); | |
2574 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2575 | btrfs_device_get_bytes_used(device)); | |
2576 | btrfs_mark_buffer_dirty(leaf); | |
2577 | ||
2578 | out: | |
2579 | btrfs_free_path(path); | |
2580 | return ret; | |
2581 | } | |
2582 | ||
2583 | int btrfs_grow_device(struct btrfs_trans_handle *trans, | |
2584 | struct btrfs_device *device, u64 new_size) | |
2585 | { | |
2586 | struct btrfs_super_block *super_copy = | |
2587 | device->dev_root->fs_info->super_copy; | |
2588 | struct btrfs_fs_devices *fs_devices; | |
2589 | u64 old_total; | |
2590 | u64 diff; | |
2591 | ||
2592 | if (!device->writeable) | |
2593 | return -EACCES; | |
2594 | ||
2595 | lock_chunks(device->dev_root); | |
2596 | old_total = btrfs_super_total_bytes(super_copy); | |
2597 | diff = new_size - device->total_bytes; | |
2598 | ||
2599 | if (new_size <= device->total_bytes || | |
2600 | device->is_tgtdev_for_dev_replace) { | |
2601 | unlock_chunks(device->dev_root); | |
2602 | return -EINVAL; | |
2603 | } | |
2604 | ||
2605 | fs_devices = device->dev_root->fs_info->fs_devices; | |
2606 | ||
2607 | btrfs_set_super_total_bytes(super_copy, old_total + diff); | |
2608 | device->fs_devices->total_rw_bytes += diff; | |
2609 | ||
2610 | btrfs_device_set_total_bytes(device, new_size); | |
2611 | btrfs_device_set_disk_total_bytes(device, new_size); | |
2612 | btrfs_clear_space_info_full(device->dev_root->fs_info); | |
2613 | if (list_empty(&device->resized_list)) | |
2614 | list_add_tail(&device->resized_list, | |
2615 | &fs_devices->resized_devices); | |
2616 | unlock_chunks(device->dev_root); | |
2617 | ||
2618 | return btrfs_update_device(trans, device); | |
2619 | } | |
2620 | ||
2621 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
2622 | struct btrfs_root *root, u64 chunk_objectid, | |
2623 | u64 chunk_offset) | |
2624 | { | |
2625 | int ret; | |
2626 | struct btrfs_path *path; | |
2627 | struct btrfs_key key; | |
2628 | ||
2629 | root = root->fs_info->chunk_root; | |
2630 | path = btrfs_alloc_path(); | |
2631 | if (!path) | |
2632 | return -ENOMEM; | |
2633 | ||
2634 | key.objectid = chunk_objectid; | |
2635 | key.offset = chunk_offset; | |
2636 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2637 | ||
2638 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2639 | if (ret < 0) | |
2640 | goto out; | |
2641 | else if (ret > 0) { /* Logic error or corruption */ | |
2642 | btrfs_std_error(root->fs_info, -ENOENT, | |
2643 | "Failed lookup while freeing chunk."); | |
2644 | ret = -ENOENT; | |
2645 | goto out; | |
2646 | } | |
2647 | ||
2648 | ret = btrfs_del_item(trans, root, path); | |
2649 | if (ret < 0) | |
2650 | btrfs_std_error(root->fs_info, ret, | |
2651 | "Failed to delete chunk item."); | |
2652 | out: | |
2653 | btrfs_free_path(path); | |
2654 | return ret; | |
2655 | } | |
2656 | ||
2657 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 | |
2658 | chunk_offset) | |
2659 | { | |
2660 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; | |
2661 | struct btrfs_disk_key *disk_key; | |
2662 | struct btrfs_chunk *chunk; | |
2663 | u8 *ptr; | |
2664 | int ret = 0; | |
2665 | u32 num_stripes; | |
2666 | u32 array_size; | |
2667 | u32 len = 0; | |
2668 | u32 cur; | |
2669 | struct btrfs_key key; | |
2670 | ||
2671 | lock_chunks(root); | |
2672 | array_size = btrfs_super_sys_array_size(super_copy); | |
2673 | ||
2674 | ptr = super_copy->sys_chunk_array; | |
2675 | cur = 0; | |
2676 | ||
2677 | while (cur < array_size) { | |
2678 | disk_key = (struct btrfs_disk_key *)ptr; | |
2679 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2680 | ||
2681 | len = sizeof(*disk_key); | |
2682 | ||
2683 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2684 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2685 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2686 | len += btrfs_chunk_item_size(num_stripes); | |
2687 | } else { | |
2688 | ret = -EIO; | |
2689 | break; | |
2690 | } | |
2691 | if (key.objectid == chunk_objectid && | |
2692 | key.offset == chunk_offset) { | |
2693 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2694 | array_size -= len; | |
2695 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2696 | } else { | |
2697 | ptr += len; | |
2698 | cur += len; | |
2699 | } | |
2700 | } | |
2701 | unlock_chunks(root); | |
2702 | return ret; | |
2703 | } | |
2704 | ||
2705 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, | |
2706 | struct btrfs_root *root, u64 chunk_offset) | |
2707 | { | |
2708 | struct extent_map_tree *em_tree; | |
2709 | struct extent_map *em; | |
2710 | struct btrfs_root *extent_root = root->fs_info->extent_root; | |
2711 | struct map_lookup *map; | |
2712 | u64 dev_extent_len = 0; | |
2713 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
2714 | int i, ret = 0; | |
2715 | ||
2716 | /* Just in case */ | |
2717 | root = root->fs_info->chunk_root; | |
2718 | em_tree = &root->fs_info->mapping_tree.map_tree; | |
2719 | ||
2720 | read_lock(&em_tree->lock); | |
2721 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); | |
2722 | read_unlock(&em_tree->lock); | |
2723 | ||
2724 | if (!em || em->start > chunk_offset || | |
2725 | em->start + em->len < chunk_offset) { | |
2726 | /* | |
2727 | * This is a logic error, but we don't want to just rely on the | |
2728 | * user having built with ASSERT enabled, so if ASSERT doesn't | |
2729 | * do anything we still error out. | |
2730 | */ | |
2731 | ASSERT(0); | |
2732 | if (em) | |
2733 | free_extent_map(em); | |
2734 | return -EINVAL; | |
2735 | } | |
2736 | map = em->map_lookup; | |
2737 | lock_chunks(root->fs_info->chunk_root); | |
2738 | check_system_chunk(trans, extent_root, map->type); | |
2739 | unlock_chunks(root->fs_info->chunk_root); | |
2740 | ||
2741 | for (i = 0; i < map->num_stripes; i++) { | |
2742 | struct btrfs_device *device = map->stripes[i].dev; | |
2743 | ret = btrfs_free_dev_extent(trans, device, | |
2744 | map->stripes[i].physical, | |
2745 | &dev_extent_len); | |
2746 | if (ret) { | |
2747 | btrfs_abort_transaction(trans, root, ret); | |
2748 | goto out; | |
2749 | } | |
2750 | ||
2751 | if (device->bytes_used > 0) { | |
2752 | lock_chunks(root); | |
2753 | btrfs_device_set_bytes_used(device, | |
2754 | device->bytes_used - dev_extent_len); | |
2755 | spin_lock(&root->fs_info->free_chunk_lock); | |
2756 | root->fs_info->free_chunk_space += dev_extent_len; | |
2757 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2758 | btrfs_clear_space_info_full(root->fs_info); | |
2759 | unlock_chunks(root); | |
2760 | } | |
2761 | ||
2762 | if (map->stripes[i].dev) { | |
2763 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
2764 | if (ret) { | |
2765 | btrfs_abort_transaction(trans, root, ret); | |
2766 | goto out; | |
2767 | } | |
2768 | } | |
2769 | } | |
2770 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); | |
2771 | if (ret) { | |
2772 | btrfs_abort_transaction(trans, root, ret); | |
2773 | goto out; | |
2774 | } | |
2775 | ||
2776 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); | |
2777 | ||
2778 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2779 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
2780 | if (ret) { | |
2781 | btrfs_abort_transaction(trans, root, ret); | |
2782 | goto out; | |
2783 | } | |
2784 | } | |
2785 | ||
2786 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); | |
2787 | if (ret) { | |
2788 | btrfs_abort_transaction(trans, extent_root, ret); | |
2789 | goto out; | |
2790 | } | |
2791 | ||
2792 | out: | |
2793 | /* once for us */ | |
2794 | free_extent_map(em); | |
2795 | return ret; | |
2796 | } | |
2797 | ||
2798 | static int btrfs_relocate_chunk(struct btrfs_root *root, u64 chunk_offset) | |
2799 | { | |
2800 | struct btrfs_root *extent_root; | |
2801 | struct btrfs_trans_handle *trans; | |
2802 | int ret; | |
2803 | ||
2804 | root = root->fs_info->chunk_root; | |
2805 | extent_root = root->fs_info->extent_root; | |
2806 | ||
2807 | /* | |
2808 | * Prevent races with automatic removal of unused block groups. | |
2809 | * After we relocate and before we remove the chunk with offset | |
2810 | * chunk_offset, automatic removal of the block group can kick in, | |
2811 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2812 | * | |
2813 | * Make sure to acquire this mutex before doing a tree search (dev | |
2814 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2815 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2816 | * we release the path used to search the chunk/dev tree and before | |
2817 | * the current task acquires this mutex and calls us. | |
2818 | */ | |
2819 | ASSERT(mutex_is_locked(&root->fs_info->delete_unused_bgs_mutex)); | |
2820 | ||
2821 | ret = btrfs_can_relocate(extent_root, chunk_offset); | |
2822 | if (ret) | |
2823 | return -ENOSPC; | |
2824 | ||
2825 | /* step one, relocate all the extents inside this chunk */ | |
2826 | btrfs_scrub_pause(root); | |
2827 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); | |
2828 | btrfs_scrub_continue(root); | |
2829 | if (ret) | |
2830 | return ret; | |
2831 | ||
2832 | trans = btrfs_start_trans_remove_block_group(root->fs_info, | |
2833 | chunk_offset); | |
2834 | if (IS_ERR(trans)) { | |
2835 | ret = PTR_ERR(trans); | |
2836 | btrfs_std_error(root->fs_info, ret, NULL); | |
2837 | return ret; | |
2838 | } | |
2839 | ||
2840 | /* | |
2841 | * step two, delete the device extents and the | |
2842 | * chunk tree entries | |
2843 | */ | |
2844 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2845 | btrfs_end_transaction(trans, root); | |
2846 | return ret; | |
2847 | } | |
2848 | ||
2849 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2850 | { | |
2851 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2852 | struct btrfs_path *path; | |
2853 | struct extent_buffer *leaf; | |
2854 | struct btrfs_chunk *chunk; | |
2855 | struct btrfs_key key; | |
2856 | struct btrfs_key found_key; | |
2857 | u64 chunk_type; | |
2858 | bool retried = false; | |
2859 | int failed = 0; | |
2860 | int ret; | |
2861 | ||
2862 | path = btrfs_alloc_path(); | |
2863 | if (!path) | |
2864 | return -ENOMEM; | |
2865 | ||
2866 | again: | |
2867 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
2868 | key.offset = (u64)-1; | |
2869 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2870 | ||
2871 | while (1) { | |
2872 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); | |
2873 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2874 | if (ret < 0) { | |
2875 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2876 | goto error; | |
2877 | } | |
2878 | BUG_ON(ret == 0); /* Corruption */ | |
2879 | ||
2880 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2881 | key.type); | |
2882 | if (ret) | |
2883 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2884 | if (ret < 0) | |
2885 | goto error; | |
2886 | if (ret > 0) | |
2887 | break; | |
2888 | ||
2889 | leaf = path->nodes[0]; | |
2890 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
2891 | ||
2892 | chunk = btrfs_item_ptr(leaf, path->slots[0], | |
2893 | struct btrfs_chunk); | |
2894 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
2895 | btrfs_release_path(path); | |
2896 | ||
2897 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
2898 | ret = btrfs_relocate_chunk(chunk_root, | |
2899 | found_key.offset); | |
2900 | if (ret == -ENOSPC) | |
2901 | failed++; | |
2902 | else | |
2903 | BUG_ON(ret); | |
2904 | } | |
2905 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
2906 | ||
2907 | if (found_key.offset == 0) | |
2908 | break; | |
2909 | key.offset = found_key.offset - 1; | |
2910 | } | |
2911 | ret = 0; | |
2912 | if (failed && !retried) { | |
2913 | failed = 0; | |
2914 | retried = true; | |
2915 | goto again; | |
2916 | } else if (WARN_ON(failed && retried)) { | |
2917 | ret = -ENOSPC; | |
2918 | } | |
2919 | error: | |
2920 | btrfs_free_path(path); | |
2921 | return ret; | |
2922 | } | |
2923 | ||
2924 | static int insert_balance_item(struct btrfs_root *root, | |
2925 | struct btrfs_balance_control *bctl) | |
2926 | { | |
2927 | struct btrfs_trans_handle *trans; | |
2928 | struct btrfs_balance_item *item; | |
2929 | struct btrfs_disk_balance_args disk_bargs; | |
2930 | struct btrfs_path *path; | |
2931 | struct extent_buffer *leaf; | |
2932 | struct btrfs_key key; | |
2933 | int ret, err; | |
2934 | ||
2935 | path = btrfs_alloc_path(); | |
2936 | if (!path) | |
2937 | return -ENOMEM; | |
2938 | ||
2939 | trans = btrfs_start_transaction(root, 0); | |
2940 | if (IS_ERR(trans)) { | |
2941 | btrfs_free_path(path); | |
2942 | return PTR_ERR(trans); | |
2943 | } | |
2944 | ||
2945 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2946 | key.type = BTRFS_TEMPORARY_ITEM_KEY; | |
2947 | key.offset = 0; | |
2948 | ||
2949 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2950 | sizeof(*item)); | |
2951 | if (ret) | |
2952 | goto out; | |
2953 | ||
2954 | leaf = path->nodes[0]; | |
2955 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2956 | ||
2957 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2958 | ||
2959 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2960 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2961 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2962 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2963 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2964 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2965 | ||
2966 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2967 | ||
2968 | btrfs_mark_buffer_dirty(leaf); | |
2969 | out: | |
2970 | btrfs_free_path(path); | |
2971 | err = btrfs_commit_transaction(trans, root); | |
2972 | if (err && !ret) | |
2973 | ret = err; | |
2974 | return ret; | |
2975 | } | |
2976 | ||
2977 | static int del_balance_item(struct btrfs_root *root) | |
2978 | { | |
2979 | struct btrfs_trans_handle *trans; | |
2980 | struct btrfs_path *path; | |
2981 | struct btrfs_key key; | |
2982 | int ret, err; | |
2983 | ||
2984 | path = btrfs_alloc_path(); | |
2985 | if (!path) | |
2986 | return -ENOMEM; | |
2987 | ||
2988 | trans = btrfs_start_transaction(root, 0); | |
2989 | if (IS_ERR(trans)) { | |
2990 | btrfs_free_path(path); | |
2991 | return PTR_ERR(trans); | |
2992 | } | |
2993 | ||
2994 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2995 | key.type = BTRFS_TEMPORARY_ITEM_KEY; | |
2996 | key.offset = 0; | |
2997 | ||
2998 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2999 | if (ret < 0) | |
3000 | goto out; | |
3001 | if (ret > 0) { | |
3002 | ret = -ENOENT; | |
3003 | goto out; | |
3004 | } | |
3005 | ||
3006 | ret = btrfs_del_item(trans, root, path); | |
3007 | out: | |
3008 | btrfs_free_path(path); | |
3009 | err = btrfs_commit_transaction(trans, root); | |
3010 | if (err && !ret) | |
3011 | ret = err; | |
3012 | return ret; | |
3013 | } | |
3014 | ||
3015 | /* | |
3016 | * This is a heuristic used to reduce the number of chunks balanced on | |
3017 | * resume after balance was interrupted. | |
3018 | */ | |
3019 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3020 | { | |
3021 | /* | |
3022 | * Turn on soft mode for chunk types that were being converted. | |
3023 | */ | |
3024 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3025 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3026 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3027 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3028 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3029 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3030 | ||
3031 | /* | |
3032 | * Turn on usage filter if is not already used. The idea is | |
3033 | * that chunks that we have already balanced should be | |
3034 | * reasonably full. Don't do it for chunks that are being | |
3035 | * converted - that will keep us from relocating unconverted | |
3036 | * (albeit full) chunks. | |
3037 | */ | |
3038 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3039 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && | |
3040 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3041 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3042 | bctl->data.usage = 90; | |
3043 | } | |
3044 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3045 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && | |
3046 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3047 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3048 | bctl->sys.usage = 90; | |
3049 | } | |
3050 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3051 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && | |
3052 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
3053 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3054 | bctl->meta.usage = 90; | |
3055 | } | |
3056 | } | |
3057 | ||
3058 | /* | |
3059 | * Should be called with both balance and volume mutexes held to | |
3060 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3061 | * restriper. Same goes for unset_balance_control. | |
3062 | */ | |
3063 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3064 | { | |
3065 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3066 | ||
3067 | BUG_ON(fs_info->balance_ctl); | |
3068 | ||
3069 | spin_lock(&fs_info->balance_lock); | |
3070 | fs_info->balance_ctl = bctl; | |
3071 | spin_unlock(&fs_info->balance_lock); | |
3072 | } | |
3073 | ||
3074 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3075 | { | |
3076 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3077 | ||
3078 | BUG_ON(!fs_info->balance_ctl); | |
3079 | ||
3080 | spin_lock(&fs_info->balance_lock); | |
3081 | fs_info->balance_ctl = NULL; | |
3082 | spin_unlock(&fs_info->balance_lock); | |
3083 | ||
3084 | kfree(bctl); | |
3085 | } | |
3086 | ||
3087 | /* | |
3088 | * Balance filters. Return 1 if chunk should be filtered out | |
3089 | * (should not be balanced). | |
3090 | */ | |
3091 | static int chunk_profiles_filter(u64 chunk_type, | |
3092 | struct btrfs_balance_args *bargs) | |
3093 | { | |
3094 | chunk_type = chunk_to_extended(chunk_type) & | |
3095 | BTRFS_EXTENDED_PROFILE_MASK; | |
3096 | ||
3097 | if (bargs->profiles & chunk_type) | |
3098 | return 0; | |
3099 | ||
3100 | return 1; | |
3101 | } | |
3102 | ||
3103 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, | |
3104 | struct btrfs_balance_args *bargs) | |
3105 | { | |
3106 | struct btrfs_block_group_cache *cache; | |
3107 | u64 chunk_used; | |
3108 | u64 user_thresh_min; | |
3109 | u64 user_thresh_max; | |
3110 | int ret = 1; | |
3111 | ||
3112 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3113 | chunk_used = btrfs_block_group_used(&cache->item); | |
3114 | ||
3115 | if (bargs->usage_min == 0) | |
3116 | user_thresh_min = 0; | |
3117 | else | |
3118 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3119 | bargs->usage_min); | |
3120 | ||
3121 | if (bargs->usage_max == 0) | |
3122 | user_thresh_max = 1; | |
3123 | else if (bargs->usage_max > 100) | |
3124 | user_thresh_max = cache->key.offset; | |
3125 | else | |
3126 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3127 | bargs->usage_max); | |
3128 | ||
3129 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3130 | ret = 0; | |
3131 | ||
3132 | btrfs_put_block_group(cache); | |
3133 | return ret; | |
3134 | } | |
3135 | ||
3136 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, | |
3137 | u64 chunk_offset, struct btrfs_balance_args *bargs) | |
3138 | { | |
3139 | struct btrfs_block_group_cache *cache; | |
3140 | u64 chunk_used, user_thresh; | |
3141 | int ret = 1; | |
3142 | ||
3143 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3144 | chunk_used = btrfs_block_group_used(&cache->item); | |
3145 | ||
3146 | if (bargs->usage_min == 0) | |
3147 | user_thresh = 1; | |
3148 | else if (bargs->usage > 100) | |
3149 | user_thresh = cache->key.offset; | |
3150 | else | |
3151 | user_thresh = div_factor_fine(cache->key.offset, | |
3152 | bargs->usage); | |
3153 | ||
3154 | if (chunk_used < user_thresh) | |
3155 | ret = 0; | |
3156 | ||
3157 | btrfs_put_block_group(cache); | |
3158 | return ret; | |
3159 | } | |
3160 | ||
3161 | static int chunk_devid_filter(struct extent_buffer *leaf, | |
3162 | struct btrfs_chunk *chunk, | |
3163 | struct btrfs_balance_args *bargs) | |
3164 | { | |
3165 | struct btrfs_stripe *stripe; | |
3166 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3167 | int i; | |
3168 | ||
3169 | for (i = 0; i < num_stripes; i++) { | |
3170 | stripe = btrfs_stripe_nr(chunk, i); | |
3171 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3172 | return 0; | |
3173 | } | |
3174 | ||
3175 | return 1; | |
3176 | } | |
3177 | ||
3178 | /* [pstart, pend) */ | |
3179 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3180 | struct btrfs_chunk *chunk, | |
3181 | u64 chunk_offset, | |
3182 | struct btrfs_balance_args *bargs) | |
3183 | { | |
3184 | struct btrfs_stripe *stripe; | |
3185 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3186 | u64 stripe_offset; | |
3187 | u64 stripe_length; | |
3188 | int factor; | |
3189 | int i; | |
3190 | ||
3191 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3192 | return 0; | |
3193 | ||
3194 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
3195 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { | |
3196 | factor = num_stripes / 2; | |
3197 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3198 | factor = num_stripes - 1; | |
3199 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3200 | factor = num_stripes - 2; | |
3201 | } else { | |
3202 | factor = num_stripes; | |
3203 | } | |
3204 | ||
3205 | for (i = 0; i < num_stripes; i++) { | |
3206 | stripe = btrfs_stripe_nr(chunk, i); | |
3207 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3208 | continue; | |
3209 | ||
3210 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3211 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
3212 | stripe_length = div_u64(stripe_length, factor); | |
3213 | ||
3214 | if (stripe_offset < bargs->pend && | |
3215 | stripe_offset + stripe_length > bargs->pstart) | |
3216 | return 0; | |
3217 | } | |
3218 | ||
3219 | return 1; | |
3220 | } | |
3221 | ||
3222 | /* [vstart, vend) */ | |
3223 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3224 | struct btrfs_chunk *chunk, | |
3225 | u64 chunk_offset, | |
3226 | struct btrfs_balance_args *bargs) | |
3227 | { | |
3228 | if (chunk_offset < bargs->vend && | |
3229 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3230 | /* at least part of the chunk is inside this vrange */ | |
3231 | return 0; | |
3232 | ||
3233 | return 1; | |
3234 | } | |
3235 | ||
3236 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, | |
3237 | struct btrfs_chunk *chunk, | |
3238 | struct btrfs_balance_args *bargs) | |
3239 | { | |
3240 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3241 | ||
3242 | if (bargs->stripes_min <= num_stripes | |
3243 | && num_stripes <= bargs->stripes_max) | |
3244 | return 0; | |
3245 | ||
3246 | return 1; | |
3247 | } | |
3248 | ||
3249 | static int chunk_soft_convert_filter(u64 chunk_type, | |
3250 | struct btrfs_balance_args *bargs) | |
3251 | { | |
3252 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3253 | return 0; | |
3254 | ||
3255 | chunk_type = chunk_to_extended(chunk_type) & | |
3256 | BTRFS_EXTENDED_PROFILE_MASK; | |
3257 | ||
3258 | if (bargs->target == chunk_type) | |
3259 | return 1; | |
3260 | ||
3261 | return 0; | |
3262 | } | |
3263 | ||
3264 | static int should_balance_chunk(struct btrfs_root *root, | |
3265 | struct extent_buffer *leaf, | |
3266 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3267 | { | |
3268 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3269 | struct btrfs_balance_args *bargs = NULL; | |
3270 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3271 | ||
3272 | /* type filter */ | |
3273 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3274 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3275 | return 0; | |
3276 | } | |
3277 | ||
3278 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3279 | bargs = &bctl->data; | |
3280 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3281 | bargs = &bctl->sys; | |
3282 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3283 | bargs = &bctl->meta; | |
3284 | ||
3285 | /* profiles filter */ | |
3286 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3287 | chunk_profiles_filter(chunk_type, bargs)) { | |
3288 | return 0; | |
3289 | } | |
3290 | ||
3291 | /* usage filter */ | |
3292 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3293 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3294 | return 0; | |
3295 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && | |
3296 | chunk_usage_range_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3297 | return 0; | |
3298 | } | |
3299 | ||
3300 | /* devid filter */ | |
3301 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3302 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3303 | return 0; | |
3304 | } | |
3305 | ||
3306 | /* drange filter, makes sense only with devid filter */ | |
3307 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3308 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3309 | return 0; | |
3310 | } | |
3311 | ||
3312 | /* vrange filter */ | |
3313 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3314 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3315 | return 0; | |
3316 | } | |
3317 | ||
3318 | /* stripes filter */ | |
3319 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3320 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3321 | return 0; | |
3322 | } | |
3323 | ||
3324 | /* soft profile changing mode */ | |
3325 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3326 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3327 | return 0; | |
3328 | } | |
3329 | ||
3330 | /* | |
3331 | * limited by count, must be the last filter | |
3332 | */ | |
3333 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3334 | if (bargs->limit == 0) | |
3335 | return 0; | |
3336 | else | |
3337 | bargs->limit--; | |
3338 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { | |
3339 | /* | |
3340 | * Same logic as the 'limit' filter; the minimum cannot be | |
3341 | * determined here because we do not have the global informatoin | |
3342 | * about the count of all chunks that satisfy the filters. | |
3343 | */ | |
3344 | if (bargs->limit_max == 0) | |
3345 | return 0; | |
3346 | else | |
3347 | bargs->limit_max--; | |
3348 | } | |
3349 | ||
3350 | return 1; | |
3351 | } | |
3352 | ||
3353 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) | |
3354 | { | |
3355 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3356 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
3357 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3358 | struct list_head *devices; | |
3359 | struct btrfs_device *device; | |
3360 | u64 old_size; | |
3361 | u64 size_to_free; | |
3362 | u64 chunk_type; | |
3363 | struct btrfs_chunk *chunk; | |
3364 | struct btrfs_path *path; | |
3365 | struct btrfs_key key; | |
3366 | struct btrfs_key found_key; | |
3367 | struct btrfs_trans_handle *trans; | |
3368 | struct extent_buffer *leaf; | |
3369 | int slot; | |
3370 | int ret; | |
3371 | int enospc_errors = 0; | |
3372 | bool counting = true; | |
3373 | /* The single value limit and min/max limits use the same bytes in the */ | |
3374 | u64 limit_data = bctl->data.limit; | |
3375 | u64 limit_meta = bctl->meta.limit; | |
3376 | u64 limit_sys = bctl->sys.limit; | |
3377 | u32 count_data = 0; | |
3378 | u32 count_meta = 0; | |
3379 | u32 count_sys = 0; | |
3380 | int chunk_reserved = 0; | |
3381 | ||
3382 | /* step one make some room on all the devices */ | |
3383 | devices = &fs_info->fs_devices->devices; | |
3384 | list_for_each_entry(device, devices, dev_list) { | |
3385 | old_size = btrfs_device_get_total_bytes(device); | |
3386 | size_to_free = div_factor(old_size, 1); | |
3387 | size_to_free = min_t(u64, size_to_free, SZ_1M); | |
3388 | if (!device->writeable || | |
3389 | btrfs_device_get_total_bytes(device) - | |
3390 | btrfs_device_get_bytes_used(device) > size_to_free || | |
3391 | device->is_tgtdev_for_dev_replace) | |
3392 | continue; | |
3393 | ||
3394 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
3395 | if (ret == -ENOSPC) | |
3396 | break; | |
3397 | BUG_ON(ret); | |
3398 | ||
3399 | trans = btrfs_start_transaction(dev_root, 0); | |
3400 | BUG_ON(IS_ERR(trans)); | |
3401 | ||
3402 | ret = btrfs_grow_device(trans, device, old_size); | |
3403 | BUG_ON(ret); | |
3404 | ||
3405 | btrfs_end_transaction(trans, dev_root); | |
3406 | } | |
3407 | ||
3408 | /* step two, relocate all the chunks */ | |
3409 | path = btrfs_alloc_path(); | |
3410 | if (!path) { | |
3411 | ret = -ENOMEM; | |
3412 | goto error; | |
3413 | } | |
3414 | ||
3415 | /* zero out stat counters */ | |
3416 | spin_lock(&fs_info->balance_lock); | |
3417 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3418 | spin_unlock(&fs_info->balance_lock); | |
3419 | again: | |
3420 | if (!counting) { | |
3421 | /* | |
3422 | * The single value limit and min/max limits use the same bytes | |
3423 | * in the | |
3424 | */ | |
3425 | bctl->data.limit = limit_data; | |
3426 | bctl->meta.limit = limit_meta; | |
3427 | bctl->sys.limit = limit_sys; | |
3428 | } | |
3429 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
3430 | key.offset = (u64)-1; | |
3431 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3432 | ||
3433 | while (1) { | |
3434 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || | |
3435 | atomic_read(&fs_info->balance_cancel_req)) { | |
3436 | ret = -ECANCELED; | |
3437 | goto error; | |
3438 | } | |
3439 | ||
3440 | mutex_lock(&fs_info->delete_unused_bgs_mutex); | |
3441 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
3442 | if (ret < 0) { | |
3443 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3444 | goto error; | |
3445 | } | |
3446 | ||
3447 | /* | |
3448 | * this shouldn't happen, it means the last relocate | |
3449 | * failed | |
3450 | */ | |
3451 | if (ret == 0) | |
3452 | BUG(); /* FIXME break ? */ | |
3453 | ||
3454 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3455 | BTRFS_CHUNK_ITEM_KEY); | |
3456 | if (ret) { | |
3457 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3458 | ret = 0; | |
3459 | break; | |
3460 | } | |
3461 | ||
3462 | leaf = path->nodes[0]; | |
3463 | slot = path->slots[0]; | |
3464 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
3465 | ||
3466 | if (found_key.objectid != key.objectid) { | |
3467 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3468 | break; | |
3469 | } | |
3470 | ||
3471 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
3472 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
3473 | ||
3474 | if (!counting) { | |
3475 | spin_lock(&fs_info->balance_lock); | |
3476 | bctl->stat.considered++; | |
3477 | spin_unlock(&fs_info->balance_lock); | |
3478 | } | |
3479 | ||
3480 | ret = should_balance_chunk(chunk_root, leaf, chunk, | |
3481 | found_key.offset); | |
3482 | ||
3483 | btrfs_release_path(path); | |
3484 | if (!ret) { | |
3485 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3486 | goto loop; | |
3487 | } | |
3488 | ||
3489 | if (counting) { | |
3490 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3491 | spin_lock(&fs_info->balance_lock); | |
3492 | bctl->stat.expected++; | |
3493 | spin_unlock(&fs_info->balance_lock); | |
3494 | ||
3495 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3496 | count_data++; | |
3497 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3498 | count_sys++; | |
3499 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3500 | count_meta++; | |
3501 | ||
3502 | goto loop; | |
3503 | } | |
3504 | ||
3505 | /* | |
3506 | * Apply limit_min filter, no need to check if the LIMITS | |
3507 | * filter is used, limit_min is 0 by default | |
3508 | */ | |
3509 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3510 | count_data < bctl->data.limit_min) | |
3511 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3512 | count_meta < bctl->meta.limit_min) | |
3513 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3514 | count_sys < bctl->sys.limit_min)) { | |
3515 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3516 | goto loop; | |
3517 | } | |
3518 | ||
3519 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && !chunk_reserved) { | |
3520 | trans = btrfs_start_transaction(chunk_root, 0); | |
3521 | if (IS_ERR(trans)) { | |
3522 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3523 | ret = PTR_ERR(trans); | |
3524 | goto error; | |
3525 | } | |
3526 | ||
3527 | ret = btrfs_force_chunk_alloc(trans, chunk_root, | |
3528 | BTRFS_BLOCK_GROUP_DATA); | |
3529 | btrfs_end_transaction(trans, chunk_root); | |
3530 | if (ret < 0) { | |
3531 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3532 | goto error; | |
3533 | } | |
3534 | chunk_reserved = 1; | |
3535 | } | |
3536 | ||
3537 | ret = btrfs_relocate_chunk(chunk_root, | |
3538 | found_key.offset); | |
3539 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3540 | if (ret && ret != -ENOSPC) | |
3541 | goto error; | |
3542 | if (ret == -ENOSPC) { | |
3543 | enospc_errors++; | |
3544 | } else { | |
3545 | spin_lock(&fs_info->balance_lock); | |
3546 | bctl->stat.completed++; | |
3547 | spin_unlock(&fs_info->balance_lock); | |
3548 | } | |
3549 | loop: | |
3550 | if (found_key.offset == 0) | |
3551 | break; | |
3552 | key.offset = found_key.offset - 1; | |
3553 | } | |
3554 | ||
3555 | if (counting) { | |
3556 | btrfs_release_path(path); | |
3557 | counting = false; | |
3558 | goto again; | |
3559 | } | |
3560 | error: | |
3561 | btrfs_free_path(path); | |
3562 | if (enospc_errors) { | |
3563 | btrfs_info(fs_info, "%d enospc errors during balance", | |
3564 | enospc_errors); | |
3565 | if (!ret) | |
3566 | ret = -ENOSPC; | |
3567 | } | |
3568 | ||
3569 | return ret; | |
3570 | } | |
3571 | ||
3572 | /** | |
3573 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3574 | * @flags: profile to validate | |
3575 | * @extended: if true @flags is treated as an extended profile | |
3576 | */ | |
3577 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3578 | { | |
3579 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3580 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3581 | ||
3582 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3583 | ||
3584 | /* 1) check that all other bits are zeroed */ | |
3585 | if (flags & ~mask) | |
3586 | return 0; | |
3587 | ||
3588 | /* 2) see if profile is reduced */ | |
3589 | if (flags == 0) | |
3590 | return !extended; /* "0" is valid for usual profiles */ | |
3591 | ||
3592 | /* true if exactly one bit set */ | |
3593 | return (flags & (flags - 1)) == 0; | |
3594 | } | |
3595 | ||
3596 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) | |
3597 | { | |
3598 | /* cancel requested || normal exit path */ | |
3599 | return atomic_read(&fs_info->balance_cancel_req) || | |
3600 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3601 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
3602 | } | |
3603 | ||
3604 | static void __cancel_balance(struct btrfs_fs_info *fs_info) | |
3605 | { | |
3606 | int ret; | |
3607 | ||
3608 | unset_balance_control(fs_info); | |
3609 | ret = del_balance_item(fs_info->tree_root); | |
3610 | if (ret) | |
3611 | btrfs_std_error(fs_info, ret, NULL); | |
3612 | ||
3613 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
3614 | } | |
3615 | ||
3616 | /* Non-zero return value signifies invalidity */ | |
3617 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3618 | u64 allowed) | |
3619 | { | |
3620 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3621 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3622 | (bctl_arg->target & ~allowed))); | |
3623 | } | |
3624 | ||
3625 | /* | |
3626 | * Should be called with both balance and volume mutexes held | |
3627 | */ | |
3628 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3629 | struct btrfs_ioctl_balance_args *bargs) | |
3630 | { | |
3631 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3632 | u64 allowed; | |
3633 | int mixed = 0; | |
3634 | int ret; | |
3635 | u64 num_devices; | |
3636 | unsigned seq; | |
3637 | ||
3638 | if (btrfs_fs_closing(fs_info) || | |
3639 | atomic_read(&fs_info->balance_pause_req) || | |
3640 | atomic_read(&fs_info->balance_cancel_req)) { | |
3641 | ret = -EINVAL; | |
3642 | goto out; | |
3643 | } | |
3644 | ||
3645 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); | |
3646 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3647 | mixed = 1; | |
3648 | ||
3649 | /* | |
3650 | * In case of mixed groups both data and meta should be picked, | |
3651 | * and identical options should be given for both of them. | |
3652 | */ | |
3653 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; | |
3654 | if (mixed && (bctl->flags & allowed)) { | |
3655 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || | |
3656 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3657 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
3658 | btrfs_err(fs_info, "with mixed groups data and " | |
3659 | "metadata balance options must be the same"); | |
3660 | ret = -EINVAL; | |
3661 | goto out; | |
3662 | } | |
3663 | } | |
3664 | ||
3665 | num_devices = fs_info->fs_devices->num_devices; | |
3666 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
3667 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3668 | BUG_ON(num_devices < 1); | |
3669 | num_devices--; | |
3670 | } | |
3671 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); | |
3672 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; | |
3673 | if (num_devices == 1) | |
3674 | allowed |= BTRFS_BLOCK_GROUP_DUP; | |
3675 | else if (num_devices > 1) | |
3676 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); | |
3677 | if (num_devices > 2) | |
3678 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3679 | if (num_devices > 3) | |
3680 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3681 | BTRFS_BLOCK_GROUP_RAID6); | |
3682 | if (validate_convert_profile(&bctl->data, allowed)) { | |
3683 | btrfs_err(fs_info, "unable to start balance with target " | |
3684 | "data profile %llu", | |
3685 | bctl->data.target); | |
3686 | ret = -EINVAL; | |
3687 | goto out; | |
3688 | } | |
3689 | if (validate_convert_profile(&bctl->meta, allowed)) { | |
3690 | btrfs_err(fs_info, | |
3691 | "unable to start balance with target metadata profile %llu", | |
3692 | bctl->meta.target); | |
3693 | ret = -EINVAL; | |
3694 | goto out; | |
3695 | } | |
3696 | if (validate_convert_profile(&bctl->sys, allowed)) { | |
3697 | btrfs_err(fs_info, | |
3698 | "unable to start balance with target system profile %llu", | |
3699 | bctl->sys.target); | |
3700 | ret = -EINVAL; | |
3701 | goto out; | |
3702 | } | |
3703 | ||
3704 | /* allow to reduce meta or sys integrity only if force set */ | |
3705 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
3706 | BTRFS_BLOCK_GROUP_RAID10 | | |
3707 | BTRFS_BLOCK_GROUP_RAID5 | | |
3708 | BTRFS_BLOCK_GROUP_RAID6; | |
3709 | do { | |
3710 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3711 | ||
3712 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3713 | (fs_info->avail_system_alloc_bits & allowed) && | |
3714 | !(bctl->sys.target & allowed)) || | |
3715 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3716 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3717 | !(bctl->meta.target & allowed))) { | |
3718 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
3719 | btrfs_info(fs_info, "force reducing metadata integrity"); | |
3720 | } else { | |
3721 | btrfs_err(fs_info, "balance will reduce metadata " | |
3722 | "integrity, use force if you want this"); | |
3723 | ret = -EINVAL; | |
3724 | goto out; | |
3725 | } | |
3726 | } | |
3727 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
3728 | ||
3729 | if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) < | |
3730 | btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) { | |
3731 | btrfs_warn(fs_info, | |
3732 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", | |
3733 | bctl->meta.target, bctl->data.target); | |
3734 | } | |
3735 | ||
3736 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3737 | fs_info->num_tolerated_disk_barrier_failures = min( | |
3738 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3739 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3740 | bctl->sys.target)); | |
3741 | } | |
3742 | ||
3743 | ret = insert_balance_item(fs_info->tree_root, bctl); | |
3744 | if (ret && ret != -EEXIST) | |
3745 | goto out; | |
3746 | ||
3747 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { | |
3748 | BUG_ON(ret == -EEXIST); | |
3749 | set_balance_control(bctl); | |
3750 | } else { | |
3751 | BUG_ON(ret != -EEXIST); | |
3752 | spin_lock(&fs_info->balance_lock); | |
3753 | update_balance_args(bctl); | |
3754 | spin_unlock(&fs_info->balance_lock); | |
3755 | } | |
3756 | ||
3757 | atomic_inc(&fs_info->balance_running); | |
3758 | mutex_unlock(&fs_info->balance_mutex); | |
3759 | ||
3760 | ret = __btrfs_balance(fs_info); | |
3761 | ||
3762 | mutex_lock(&fs_info->balance_mutex); | |
3763 | atomic_dec(&fs_info->balance_running); | |
3764 | ||
3765 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { | |
3766 | fs_info->num_tolerated_disk_barrier_failures = | |
3767 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3768 | } | |
3769 | ||
3770 | if (bargs) { | |
3771 | memset(bargs, 0, sizeof(*bargs)); | |
3772 | update_ioctl_balance_args(fs_info, 0, bargs); | |
3773 | } | |
3774 | ||
3775 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || | |
3776 | balance_need_close(fs_info)) { | |
3777 | __cancel_balance(fs_info); | |
3778 | } | |
3779 | ||
3780 | wake_up(&fs_info->balance_wait_q); | |
3781 | ||
3782 | return ret; | |
3783 | out: | |
3784 | if (bctl->flags & BTRFS_BALANCE_RESUME) | |
3785 | __cancel_balance(fs_info); | |
3786 | else { | |
3787 | kfree(bctl); | |
3788 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
3789 | } | |
3790 | return ret; | |
3791 | } | |
3792 | ||
3793 | static int balance_kthread(void *data) | |
3794 | { | |
3795 | struct btrfs_fs_info *fs_info = data; | |
3796 | int ret = 0; | |
3797 | ||
3798 | mutex_lock(&fs_info->volume_mutex); | |
3799 | mutex_lock(&fs_info->balance_mutex); | |
3800 | ||
3801 | if (fs_info->balance_ctl) { | |
3802 | btrfs_info(fs_info, "continuing balance"); | |
3803 | ret = btrfs_balance(fs_info->balance_ctl, NULL); | |
3804 | } | |
3805 | ||
3806 | mutex_unlock(&fs_info->balance_mutex); | |
3807 | mutex_unlock(&fs_info->volume_mutex); | |
3808 | ||
3809 | return ret; | |
3810 | } | |
3811 | ||
3812 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) | |
3813 | { | |
3814 | struct task_struct *tsk; | |
3815 | ||
3816 | spin_lock(&fs_info->balance_lock); | |
3817 | if (!fs_info->balance_ctl) { | |
3818 | spin_unlock(&fs_info->balance_lock); | |
3819 | return 0; | |
3820 | } | |
3821 | spin_unlock(&fs_info->balance_lock); | |
3822 | ||
3823 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
3824 | btrfs_info(fs_info, "force skipping balance"); | |
3825 | return 0; | |
3826 | } | |
3827 | ||
3828 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
3829 | return PTR_ERR_OR_ZERO(tsk); | |
3830 | } | |
3831 | ||
3832 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) | |
3833 | { | |
3834 | struct btrfs_balance_control *bctl; | |
3835 | struct btrfs_balance_item *item; | |
3836 | struct btrfs_disk_balance_args disk_bargs; | |
3837 | struct btrfs_path *path; | |
3838 | struct extent_buffer *leaf; | |
3839 | struct btrfs_key key; | |
3840 | int ret; | |
3841 | ||
3842 | path = btrfs_alloc_path(); | |
3843 | if (!path) | |
3844 | return -ENOMEM; | |
3845 | ||
3846 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
3847 | key.type = BTRFS_TEMPORARY_ITEM_KEY; | |
3848 | key.offset = 0; | |
3849 | ||
3850 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); | |
3851 | if (ret < 0) | |
3852 | goto out; | |
3853 | if (ret > 0) { /* ret = -ENOENT; */ | |
3854 | ret = 0; | |
3855 | goto out; | |
3856 | } | |
3857 | ||
3858 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3859 | if (!bctl) { | |
3860 | ret = -ENOMEM; | |
3861 | goto out; | |
3862 | } | |
3863 | ||
3864 | leaf = path->nodes[0]; | |
3865 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3866 | ||
3867 | bctl->fs_info = fs_info; | |
3868 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3869 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
3870 | ||
3871 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3872 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3873 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3874 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3875 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3876 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3877 | ||
3878 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); | |
3879 | ||
3880 | mutex_lock(&fs_info->volume_mutex); | |
3881 | mutex_lock(&fs_info->balance_mutex); | |
3882 | ||
3883 | set_balance_control(bctl); | |
3884 | ||
3885 | mutex_unlock(&fs_info->balance_mutex); | |
3886 | mutex_unlock(&fs_info->volume_mutex); | |
3887 | out: | |
3888 | btrfs_free_path(path); | |
3889 | return ret; | |
3890 | } | |
3891 | ||
3892 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) | |
3893 | { | |
3894 | int ret = 0; | |
3895 | ||
3896 | mutex_lock(&fs_info->balance_mutex); | |
3897 | if (!fs_info->balance_ctl) { | |
3898 | mutex_unlock(&fs_info->balance_mutex); | |
3899 | return -ENOTCONN; | |
3900 | } | |
3901 | ||
3902 | if (atomic_read(&fs_info->balance_running)) { | |
3903 | atomic_inc(&fs_info->balance_pause_req); | |
3904 | mutex_unlock(&fs_info->balance_mutex); | |
3905 | ||
3906 | wait_event(fs_info->balance_wait_q, | |
3907 | atomic_read(&fs_info->balance_running) == 0); | |
3908 | ||
3909 | mutex_lock(&fs_info->balance_mutex); | |
3910 | /* we are good with balance_ctl ripped off from under us */ | |
3911 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3912 | atomic_dec(&fs_info->balance_pause_req); | |
3913 | } else { | |
3914 | ret = -ENOTCONN; | |
3915 | } | |
3916 | ||
3917 | mutex_unlock(&fs_info->balance_mutex); | |
3918 | return ret; | |
3919 | } | |
3920 | ||
3921 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) | |
3922 | { | |
3923 | if (fs_info->sb->s_flags & MS_RDONLY) | |
3924 | return -EROFS; | |
3925 | ||
3926 | mutex_lock(&fs_info->balance_mutex); | |
3927 | if (!fs_info->balance_ctl) { | |
3928 | mutex_unlock(&fs_info->balance_mutex); | |
3929 | return -ENOTCONN; | |
3930 | } | |
3931 | ||
3932 | atomic_inc(&fs_info->balance_cancel_req); | |
3933 | /* | |
3934 | * if we are running just wait and return, balance item is | |
3935 | * deleted in btrfs_balance in this case | |
3936 | */ | |
3937 | if (atomic_read(&fs_info->balance_running)) { | |
3938 | mutex_unlock(&fs_info->balance_mutex); | |
3939 | wait_event(fs_info->balance_wait_q, | |
3940 | atomic_read(&fs_info->balance_running) == 0); | |
3941 | mutex_lock(&fs_info->balance_mutex); | |
3942 | } else { | |
3943 | /* __cancel_balance needs volume_mutex */ | |
3944 | mutex_unlock(&fs_info->balance_mutex); | |
3945 | mutex_lock(&fs_info->volume_mutex); | |
3946 | mutex_lock(&fs_info->balance_mutex); | |
3947 | ||
3948 | if (fs_info->balance_ctl) | |
3949 | __cancel_balance(fs_info); | |
3950 | ||
3951 | mutex_unlock(&fs_info->volume_mutex); | |
3952 | } | |
3953 | ||
3954 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3955 | atomic_dec(&fs_info->balance_cancel_req); | |
3956 | mutex_unlock(&fs_info->balance_mutex); | |
3957 | return 0; | |
3958 | } | |
3959 | ||
3960 | static int btrfs_uuid_scan_kthread(void *data) | |
3961 | { | |
3962 | struct btrfs_fs_info *fs_info = data; | |
3963 | struct btrfs_root *root = fs_info->tree_root; | |
3964 | struct btrfs_key key; | |
3965 | struct btrfs_key max_key; | |
3966 | struct btrfs_path *path = NULL; | |
3967 | int ret = 0; | |
3968 | struct extent_buffer *eb; | |
3969 | int slot; | |
3970 | struct btrfs_root_item root_item; | |
3971 | u32 item_size; | |
3972 | struct btrfs_trans_handle *trans = NULL; | |
3973 | ||
3974 | path = btrfs_alloc_path(); | |
3975 | if (!path) { | |
3976 | ret = -ENOMEM; | |
3977 | goto out; | |
3978 | } | |
3979 | ||
3980 | key.objectid = 0; | |
3981 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3982 | key.offset = 0; | |
3983 | ||
3984 | max_key.objectid = (u64)-1; | |
3985 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3986 | max_key.offset = (u64)-1; | |
3987 | ||
3988 | while (1) { | |
3989 | ret = btrfs_search_forward(root, &key, path, 0); | |
3990 | if (ret) { | |
3991 | if (ret > 0) | |
3992 | ret = 0; | |
3993 | break; | |
3994 | } | |
3995 | ||
3996 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3997 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3998 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3999 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4000 | goto skip; | |
4001 | ||
4002 | eb = path->nodes[0]; | |
4003 | slot = path->slots[0]; | |
4004 | item_size = btrfs_item_size_nr(eb, slot); | |
4005 | if (item_size < sizeof(root_item)) | |
4006 | goto skip; | |
4007 | ||
4008 | read_extent_buffer(eb, &root_item, | |
4009 | btrfs_item_ptr_offset(eb, slot), | |
4010 | (int)sizeof(root_item)); | |
4011 | if (btrfs_root_refs(&root_item) == 0) | |
4012 | goto skip; | |
4013 | ||
4014 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4015 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4016 | if (trans) | |
4017 | goto update_tree; | |
4018 | ||
4019 | btrfs_release_path(path); | |
4020 | /* | |
4021 | * 1 - subvol uuid item | |
4022 | * 1 - received_subvol uuid item | |
4023 | */ | |
4024 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4025 | if (IS_ERR(trans)) { | |
4026 | ret = PTR_ERR(trans); | |
4027 | break; | |
4028 | } | |
4029 | continue; | |
4030 | } else { | |
4031 | goto skip; | |
4032 | } | |
4033 | update_tree: | |
4034 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
4035 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
4036 | root_item.uuid, | |
4037 | BTRFS_UUID_KEY_SUBVOL, | |
4038 | key.objectid); | |
4039 | if (ret < 0) { | |
4040 | btrfs_warn(fs_info, "uuid_tree_add failed %d", | |
4041 | ret); | |
4042 | break; | |
4043 | } | |
4044 | } | |
4045 | ||
4046 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4047 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, | |
4048 | root_item.received_uuid, | |
4049 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4050 | key.objectid); | |
4051 | if (ret < 0) { | |
4052 | btrfs_warn(fs_info, "uuid_tree_add failed %d", | |
4053 | ret); | |
4054 | break; | |
4055 | } | |
4056 | } | |
4057 | ||
4058 | skip: | |
4059 | if (trans) { | |
4060 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
4061 | trans = NULL; | |
4062 | if (ret) | |
4063 | break; | |
4064 | } | |
4065 | ||
4066 | btrfs_release_path(path); | |
4067 | if (key.offset < (u64)-1) { | |
4068 | key.offset++; | |
4069 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4070 | key.offset = 0; | |
4071 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4072 | } else if (key.objectid < (u64)-1) { | |
4073 | key.offset = 0; | |
4074 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4075 | key.objectid++; | |
4076 | } else { | |
4077 | break; | |
4078 | } | |
4079 | cond_resched(); | |
4080 | } | |
4081 | ||
4082 | out: | |
4083 | btrfs_free_path(path); | |
4084 | if (trans && !IS_ERR(trans)) | |
4085 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
4086 | if (ret) | |
4087 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); | |
4088 | else | |
4089 | fs_info->update_uuid_tree_gen = 1; | |
4090 | up(&fs_info->uuid_tree_rescan_sem); | |
4091 | return 0; | |
4092 | } | |
4093 | ||
4094 | /* | |
4095 | * Callback for btrfs_uuid_tree_iterate(). | |
4096 | * returns: | |
4097 | * 0 check succeeded, the entry is not outdated. | |
4098 | * < 0 if an error occurred. | |
4099 | * > 0 if the check failed, which means the caller shall remove the entry. | |
4100 | */ | |
4101 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4102 | u8 *uuid, u8 type, u64 subid) | |
4103 | { | |
4104 | struct btrfs_key key; | |
4105 | int ret = 0; | |
4106 | struct btrfs_root *subvol_root; | |
4107 | ||
4108 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4109 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4110 | goto out; | |
4111 | ||
4112 | key.objectid = subid; | |
4113 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4114 | key.offset = (u64)-1; | |
4115 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4116 | if (IS_ERR(subvol_root)) { | |
4117 | ret = PTR_ERR(subvol_root); | |
4118 | if (ret == -ENOENT) | |
4119 | ret = 1; | |
4120 | goto out; | |
4121 | } | |
4122 | ||
4123 | switch (type) { | |
4124 | case BTRFS_UUID_KEY_SUBVOL: | |
4125 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4126 | ret = 1; | |
4127 | break; | |
4128 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4129 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4130 | BTRFS_UUID_SIZE)) | |
4131 | ret = 1; | |
4132 | break; | |
4133 | } | |
4134 | ||
4135 | out: | |
4136 | return ret; | |
4137 | } | |
4138 | ||
4139 | static int btrfs_uuid_rescan_kthread(void *data) | |
4140 | { | |
4141 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4142 | int ret; | |
4143 | ||
4144 | /* | |
4145 | * 1st step is to iterate through the existing UUID tree and | |
4146 | * to delete all entries that contain outdated data. | |
4147 | * 2nd step is to add all missing entries to the UUID tree. | |
4148 | */ | |
4149 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4150 | if (ret < 0) { | |
4151 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); | |
4152 | up(&fs_info->uuid_tree_rescan_sem); | |
4153 | return ret; | |
4154 | } | |
4155 | return btrfs_uuid_scan_kthread(data); | |
4156 | } | |
4157 | ||
4158 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) | |
4159 | { | |
4160 | struct btrfs_trans_handle *trans; | |
4161 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4162 | struct btrfs_root *uuid_root; | |
4163 | struct task_struct *task; | |
4164 | int ret; | |
4165 | ||
4166 | /* | |
4167 | * 1 - root node | |
4168 | * 1 - root item | |
4169 | */ | |
4170 | trans = btrfs_start_transaction(tree_root, 2); | |
4171 | if (IS_ERR(trans)) | |
4172 | return PTR_ERR(trans); | |
4173 | ||
4174 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4175 | BTRFS_UUID_TREE_OBJECTID); | |
4176 | if (IS_ERR(uuid_root)) { | |
4177 | ret = PTR_ERR(uuid_root); | |
4178 | btrfs_abort_transaction(trans, tree_root, ret); | |
4179 | return ret; | |
4180 | } | |
4181 | ||
4182 | fs_info->uuid_root = uuid_root; | |
4183 | ||
4184 | ret = btrfs_commit_transaction(trans, tree_root); | |
4185 | if (ret) | |
4186 | return ret; | |
4187 | ||
4188 | down(&fs_info->uuid_tree_rescan_sem); | |
4189 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4190 | if (IS_ERR(task)) { | |
4191 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
4192 | btrfs_warn(fs_info, "failed to start uuid_scan task"); | |
4193 | up(&fs_info->uuid_tree_rescan_sem); | |
4194 | return PTR_ERR(task); | |
4195 | } | |
4196 | ||
4197 | return 0; | |
4198 | } | |
4199 | ||
4200 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) | |
4201 | { | |
4202 | struct task_struct *task; | |
4203 | ||
4204 | down(&fs_info->uuid_tree_rescan_sem); | |
4205 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4206 | if (IS_ERR(task)) { | |
4207 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
4208 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); | |
4209 | up(&fs_info->uuid_tree_rescan_sem); | |
4210 | return PTR_ERR(task); | |
4211 | } | |
4212 | ||
4213 | return 0; | |
4214 | } | |
4215 | ||
4216 | /* | |
4217 | * shrinking a device means finding all of the device extents past | |
4218 | * the new size, and then following the back refs to the chunks. | |
4219 | * The chunk relocation code actually frees the device extent | |
4220 | */ | |
4221 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4222 | { | |
4223 | struct btrfs_trans_handle *trans; | |
4224 | struct btrfs_root *root = device->dev_root; | |
4225 | struct btrfs_dev_extent *dev_extent = NULL; | |
4226 | struct btrfs_path *path; | |
4227 | u64 length; | |
4228 | u64 chunk_offset; | |
4229 | int ret; | |
4230 | int slot; | |
4231 | int failed = 0; | |
4232 | bool retried = false; | |
4233 | bool checked_pending_chunks = false; | |
4234 | struct extent_buffer *l; | |
4235 | struct btrfs_key key; | |
4236 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; | |
4237 | u64 old_total = btrfs_super_total_bytes(super_copy); | |
4238 | u64 old_size = btrfs_device_get_total_bytes(device); | |
4239 | u64 diff = old_size - new_size; | |
4240 | ||
4241 | if (device->is_tgtdev_for_dev_replace) | |
4242 | return -EINVAL; | |
4243 | ||
4244 | path = btrfs_alloc_path(); | |
4245 | if (!path) | |
4246 | return -ENOMEM; | |
4247 | ||
4248 | path->reada = READA_FORWARD; | |
4249 | ||
4250 | lock_chunks(root); | |
4251 | ||
4252 | btrfs_device_set_total_bytes(device, new_size); | |
4253 | if (device->writeable) { | |
4254 | device->fs_devices->total_rw_bytes -= diff; | |
4255 | spin_lock(&root->fs_info->free_chunk_lock); | |
4256 | root->fs_info->free_chunk_space -= diff; | |
4257 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4258 | } | |
4259 | unlock_chunks(root); | |
4260 | ||
4261 | again: | |
4262 | key.objectid = device->devid; | |
4263 | key.offset = (u64)-1; | |
4264 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4265 | ||
4266 | do { | |
4267 | mutex_lock(&root->fs_info->delete_unused_bgs_mutex); | |
4268 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
4269 | if (ret < 0) { | |
4270 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
4271 | goto done; | |
4272 | } | |
4273 | ||
4274 | ret = btrfs_previous_item(root, path, 0, key.type); | |
4275 | if (ret) | |
4276 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
4277 | if (ret < 0) | |
4278 | goto done; | |
4279 | if (ret) { | |
4280 | ret = 0; | |
4281 | btrfs_release_path(path); | |
4282 | break; | |
4283 | } | |
4284 | ||
4285 | l = path->nodes[0]; | |
4286 | slot = path->slots[0]; | |
4287 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4288 | ||
4289 | if (key.objectid != device->devid) { | |
4290 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
4291 | btrfs_release_path(path); | |
4292 | break; | |
4293 | } | |
4294 | ||
4295 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4296 | length = btrfs_dev_extent_length(l, dev_extent); | |
4297 | ||
4298 | if (key.offset + length <= new_size) { | |
4299 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
4300 | btrfs_release_path(path); | |
4301 | break; | |
4302 | } | |
4303 | ||
4304 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
4305 | btrfs_release_path(path); | |
4306 | ||
4307 | ret = btrfs_relocate_chunk(root, chunk_offset); | |
4308 | mutex_unlock(&root->fs_info->delete_unused_bgs_mutex); | |
4309 | if (ret && ret != -ENOSPC) | |
4310 | goto done; | |
4311 | if (ret == -ENOSPC) | |
4312 | failed++; | |
4313 | } while (key.offset-- > 0); | |
4314 | ||
4315 | if (failed && !retried) { | |
4316 | failed = 0; | |
4317 | retried = true; | |
4318 | goto again; | |
4319 | } else if (failed && retried) { | |
4320 | ret = -ENOSPC; | |
4321 | goto done; | |
4322 | } | |
4323 | ||
4324 | /* Shrinking succeeded, else we would be at "done". */ | |
4325 | trans = btrfs_start_transaction(root, 0); | |
4326 | if (IS_ERR(trans)) { | |
4327 | ret = PTR_ERR(trans); | |
4328 | goto done; | |
4329 | } | |
4330 | ||
4331 | lock_chunks(root); | |
4332 | ||
4333 | /* | |
4334 | * We checked in the above loop all device extents that were already in | |
4335 | * the device tree. However before we have updated the device's | |
4336 | * total_bytes to the new size, we might have had chunk allocations that | |
4337 | * have not complete yet (new block groups attached to transaction | |
4338 | * handles), and therefore their device extents were not yet in the | |
4339 | * device tree and we missed them in the loop above. So if we have any | |
4340 | * pending chunk using a device extent that overlaps the device range | |
4341 | * that we can not use anymore, commit the current transaction and | |
4342 | * repeat the search on the device tree - this way we guarantee we will | |
4343 | * not have chunks using device extents that end beyond 'new_size'. | |
4344 | */ | |
4345 | if (!checked_pending_chunks) { | |
4346 | u64 start = new_size; | |
4347 | u64 len = old_size - new_size; | |
4348 | ||
4349 | if (contains_pending_extent(trans->transaction, device, | |
4350 | &start, len)) { | |
4351 | unlock_chunks(root); | |
4352 | checked_pending_chunks = true; | |
4353 | failed = 0; | |
4354 | retried = false; | |
4355 | ret = btrfs_commit_transaction(trans, root); | |
4356 | if (ret) | |
4357 | goto done; | |
4358 | goto again; | |
4359 | } | |
4360 | } | |
4361 | ||
4362 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4363 | if (list_empty(&device->resized_list)) | |
4364 | list_add_tail(&device->resized_list, | |
4365 | &root->fs_info->fs_devices->resized_devices); | |
4366 | ||
4367 | WARN_ON(diff > old_total); | |
4368 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4369 | unlock_chunks(root); | |
4370 | ||
4371 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4372 | ret = btrfs_update_device(trans, device); | |
4373 | btrfs_end_transaction(trans, root); | |
4374 | done: | |
4375 | btrfs_free_path(path); | |
4376 | if (ret) { | |
4377 | lock_chunks(root); | |
4378 | btrfs_device_set_total_bytes(device, old_size); | |
4379 | if (device->writeable) | |
4380 | device->fs_devices->total_rw_bytes += diff; | |
4381 | spin_lock(&root->fs_info->free_chunk_lock); | |
4382 | root->fs_info->free_chunk_space += diff; | |
4383 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4384 | unlock_chunks(root); | |
4385 | } | |
4386 | return ret; | |
4387 | } | |
4388 | ||
4389 | static int btrfs_add_system_chunk(struct btrfs_root *root, | |
4390 | struct btrfs_key *key, | |
4391 | struct btrfs_chunk *chunk, int item_size) | |
4392 | { | |
4393 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; | |
4394 | struct btrfs_disk_key disk_key; | |
4395 | u32 array_size; | |
4396 | u8 *ptr; | |
4397 | ||
4398 | lock_chunks(root); | |
4399 | array_size = btrfs_super_sys_array_size(super_copy); | |
4400 | if (array_size + item_size + sizeof(disk_key) | |
4401 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { | |
4402 | unlock_chunks(root); | |
4403 | return -EFBIG; | |
4404 | } | |
4405 | ||
4406 | ptr = super_copy->sys_chunk_array + array_size; | |
4407 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4408 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4409 | ptr += sizeof(disk_key); | |
4410 | memcpy(ptr, chunk, item_size); | |
4411 | item_size += sizeof(disk_key); | |
4412 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
4413 | unlock_chunks(root); | |
4414 | ||
4415 | return 0; | |
4416 | } | |
4417 | ||
4418 | /* | |
4419 | * sort the devices in descending order by max_avail, total_avail | |
4420 | */ | |
4421 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
4422 | { | |
4423 | const struct btrfs_device_info *di_a = a; | |
4424 | const struct btrfs_device_info *di_b = b; | |
4425 | ||
4426 | if (di_a->max_avail > di_b->max_avail) | |
4427 | return -1; | |
4428 | if (di_a->max_avail < di_b->max_avail) | |
4429 | return 1; | |
4430 | if (di_a->total_avail > di_b->total_avail) | |
4431 | return -1; | |
4432 | if (di_a->total_avail < di_b->total_avail) | |
4433 | return 1; | |
4434 | return 0; | |
4435 | } | |
4436 | ||
4437 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) | |
4438 | { | |
4439 | /* TODO allow them to set a preferred stripe size */ | |
4440 | return SZ_64K; | |
4441 | } | |
4442 | ||
4443 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4444 | { | |
4445 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) | |
4446 | return; | |
4447 | ||
4448 | btrfs_set_fs_incompat(info, RAID56); | |
4449 | } | |
4450 | ||
4451 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ | |
4452 | - sizeof(struct btrfs_item) \ | |
4453 | - sizeof(struct btrfs_chunk)) \ | |
4454 | / sizeof(struct btrfs_stripe) + 1) | |
4455 | ||
4456 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4457 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4458 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4459 | / sizeof(struct btrfs_stripe) + 1) | |
4460 | ||
4461 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4462 | struct btrfs_root *extent_root, u64 start, | |
4463 | u64 type) | |
4464 | { | |
4465 | struct btrfs_fs_info *info = extent_root->fs_info; | |
4466 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4467 | struct list_head *cur; | |
4468 | struct map_lookup *map = NULL; | |
4469 | struct extent_map_tree *em_tree; | |
4470 | struct extent_map *em; | |
4471 | struct btrfs_device_info *devices_info = NULL; | |
4472 | u64 total_avail; | |
4473 | int num_stripes; /* total number of stripes to allocate */ | |
4474 | int data_stripes; /* number of stripes that count for | |
4475 | block group size */ | |
4476 | int sub_stripes; /* sub_stripes info for map */ | |
4477 | int dev_stripes; /* stripes per dev */ | |
4478 | int devs_max; /* max devs to use */ | |
4479 | int devs_min; /* min devs needed */ | |
4480 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4481 | int ncopies; /* how many copies to data has */ | |
4482 | int ret; | |
4483 | u64 max_stripe_size; | |
4484 | u64 max_chunk_size; | |
4485 | u64 stripe_size; | |
4486 | u64 num_bytes; | |
4487 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; | |
4488 | int ndevs; | |
4489 | int i; | |
4490 | int j; | |
4491 | int index; | |
4492 | ||
4493 | BUG_ON(!alloc_profile_is_valid(type, 0)); | |
4494 | ||
4495 | if (list_empty(&fs_devices->alloc_list)) | |
4496 | return -ENOSPC; | |
4497 | ||
4498 | index = __get_raid_index(type); | |
4499 | ||
4500 | sub_stripes = btrfs_raid_array[index].sub_stripes; | |
4501 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4502 | devs_max = btrfs_raid_array[index].devs_max; | |
4503 | devs_min = btrfs_raid_array[index].devs_min; | |
4504 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4505 | ncopies = btrfs_raid_array[index].ncopies; | |
4506 | ||
4507 | if (type & BTRFS_BLOCK_GROUP_DATA) { | |
4508 | max_stripe_size = SZ_1G; | |
4509 | max_chunk_size = 10 * max_stripe_size; | |
4510 | if (!devs_max) | |
4511 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
4512 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { | |
4513 | /* for larger filesystems, use larger metadata chunks */ | |
4514 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) | |
4515 | max_stripe_size = SZ_1G; | |
4516 | else | |
4517 | max_stripe_size = SZ_256M; | |
4518 | max_chunk_size = max_stripe_size; | |
4519 | if (!devs_max) | |
4520 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
4521 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4522 | max_stripe_size = SZ_32M; | |
4523 | max_chunk_size = 2 * max_stripe_size; | |
4524 | if (!devs_max) | |
4525 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
4526 | } else { | |
4527 | btrfs_err(info, "invalid chunk type 0x%llx requested", | |
4528 | type); | |
4529 | BUG_ON(1); | |
4530 | } | |
4531 | ||
4532 | /* we don't want a chunk larger than 10% of writeable space */ | |
4533 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4534 | max_chunk_size); | |
4535 | ||
4536 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), | |
4537 | GFP_NOFS); | |
4538 | if (!devices_info) | |
4539 | return -ENOMEM; | |
4540 | ||
4541 | cur = fs_devices->alloc_list.next; | |
4542 | ||
4543 | /* | |
4544 | * in the first pass through the devices list, we gather information | |
4545 | * about the available holes on each device. | |
4546 | */ | |
4547 | ndevs = 0; | |
4548 | while (cur != &fs_devices->alloc_list) { | |
4549 | struct btrfs_device *device; | |
4550 | u64 max_avail; | |
4551 | u64 dev_offset; | |
4552 | ||
4553 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); | |
4554 | ||
4555 | cur = cur->next; | |
4556 | ||
4557 | if (!device->writeable) { | |
4558 | WARN(1, KERN_ERR | |
4559 | "BTRFS: read-only device in alloc_list\n"); | |
4560 | continue; | |
4561 | } | |
4562 | ||
4563 | if (!device->in_fs_metadata || | |
4564 | device->is_tgtdev_for_dev_replace) | |
4565 | continue; | |
4566 | ||
4567 | if (device->total_bytes > device->bytes_used) | |
4568 | total_avail = device->total_bytes - device->bytes_used; | |
4569 | else | |
4570 | total_avail = 0; | |
4571 | ||
4572 | /* If there is no space on this device, skip it. */ | |
4573 | if (total_avail == 0) | |
4574 | continue; | |
4575 | ||
4576 | ret = find_free_dev_extent(trans, device, | |
4577 | max_stripe_size * dev_stripes, | |
4578 | &dev_offset, &max_avail); | |
4579 | if (ret && ret != -ENOSPC) | |
4580 | goto error; | |
4581 | ||
4582 | if (ret == 0) | |
4583 | max_avail = max_stripe_size * dev_stripes; | |
4584 | ||
4585 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) | |
4586 | continue; | |
4587 | ||
4588 | if (ndevs == fs_devices->rw_devices) { | |
4589 | WARN(1, "%s: found more than %llu devices\n", | |
4590 | __func__, fs_devices->rw_devices); | |
4591 | break; | |
4592 | } | |
4593 | devices_info[ndevs].dev_offset = dev_offset; | |
4594 | devices_info[ndevs].max_avail = max_avail; | |
4595 | devices_info[ndevs].total_avail = total_avail; | |
4596 | devices_info[ndevs].dev = device; | |
4597 | ++ndevs; | |
4598 | } | |
4599 | ||
4600 | /* | |
4601 | * now sort the devices by hole size / available space | |
4602 | */ | |
4603 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4604 | btrfs_cmp_device_info, NULL); | |
4605 | ||
4606 | /* round down to number of usable stripes */ | |
4607 | ndevs -= ndevs % devs_increment; | |
4608 | ||
4609 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { | |
4610 | ret = -ENOSPC; | |
4611 | goto error; | |
4612 | } | |
4613 | ||
4614 | if (devs_max && ndevs > devs_max) | |
4615 | ndevs = devs_max; | |
4616 | /* | |
4617 | * the primary goal is to maximize the number of stripes, so use as many | |
4618 | * devices as possible, even if the stripes are not maximum sized. | |
4619 | */ | |
4620 | stripe_size = devices_info[ndevs-1].max_avail; | |
4621 | num_stripes = ndevs * dev_stripes; | |
4622 | ||
4623 | /* | |
4624 | * this will have to be fixed for RAID1 and RAID10 over | |
4625 | * more drives | |
4626 | */ | |
4627 | data_stripes = num_stripes / ncopies; | |
4628 | ||
4629 | if (type & BTRFS_BLOCK_GROUP_RAID5) { | |
4630 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4631 | btrfs_super_stripesize(info->super_copy)); | |
4632 | data_stripes = num_stripes - 1; | |
4633 | } | |
4634 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4635 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4636 | btrfs_super_stripesize(info->super_copy)); | |
4637 | data_stripes = num_stripes - 2; | |
4638 | } | |
4639 | ||
4640 | /* | |
4641 | * Use the number of data stripes to figure out how big this chunk | |
4642 | * is really going to be in terms of logical address space, | |
4643 | * and compare that answer with the max chunk size | |
4644 | */ | |
4645 | if (stripe_size * data_stripes > max_chunk_size) { | |
4646 | u64 mask = (1ULL << 24) - 1; | |
4647 | ||
4648 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
4649 | ||
4650 | /* bump the answer up to a 16MB boundary */ | |
4651 | stripe_size = (stripe_size + mask) & ~mask; | |
4652 | ||
4653 | /* but don't go higher than the limits we found | |
4654 | * while searching for free extents | |
4655 | */ | |
4656 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4657 | stripe_size = devices_info[ndevs-1].max_avail; | |
4658 | } | |
4659 | ||
4660 | stripe_size = div_u64(stripe_size, dev_stripes); | |
4661 | ||
4662 | /* align to BTRFS_STRIPE_LEN */ | |
4663 | stripe_size = div_u64(stripe_size, raid_stripe_len); | |
4664 | stripe_size *= raid_stripe_len; | |
4665 | ||
4666 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4667 | if (!map) { | |
4668 | ret = -ENOMEM; | |
4669 | goto error; | |
4670 | } | |
4671 | map->num_stripes = num_stripes; | |
4672 | ||
4673 | for (i = 0; i < ndevs; ++i) { | |
4674 | for (j = 0; j < dev_stripes; ++j) { | |
4675 | int s = i * dev_stripes + j; | |
4676 | map->stripes[s].dev = devices_info[i].dev; | |
4677 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4678 | j * stripe_size; | |
4679 | } | |
4680 | } | |
4681 | map->sector_size = extent_root->sectorsize; | |
4682 | map->stripe_len = raid_stripe_len; | |
4683 | map->io_align = raid_stripe_len; | |
4684 | map->io_width = raid_stripe_len; | |
4685 | map->type = type; | |
4686 | map->sub_stripes = sub_stripes; | |
4687 | ||
4688 | num_bytes = stripe_size * data_stripes; | |
4689 | ||
4690 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); | |
4691 | ||
4692 | em = alloc_extent_map(); | |
4693 | if (!em) { | |
4694 | kfree(map); | |
4695 | ret = -ENOMEM; | |
4696 | goto error; | |
4697 | } | |
4698 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); | |
4699 | em->map_lookup = map; | |
4700 | em->start = start; | |
4701 | em->len = num_bytes; | |
4702 | em->block_start = 0; | |
4703 | em->block_len = em->len; | |
4704 | em->orig_block_len = stripe_size; | |
4705 | ||
4706 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; | |
4707 | write_lock(&em_tree->lock); | |
4708 | ret = add_extent_mapping(em_tree, em, 0); | |
4709 | if (!ret) { | |
4710 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4711 | atomic_inc(&em->refs); | |
4712 | } | |
4713 | write_unlock(&em_tree->lock); | |
4714 | if (ret) { | |
4715 | free_extent_map(em); | |
4716 | goto error; | |
4717 | } | |
4718 | ||
4719 | ret = btrfs_make_block_group(trans, extent_root, 0, type, | |
4720 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4721 | start, num_bytes); | |
4722 | if (ret) | |
4723 | goto error_del_extent; | |
4724 | ||
4725 | for (i = 0; i < map->num_stripes; i++) { | |
4726 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4727 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4728 | } | |
4729 | ||
4730 | spin_lock(&extent_root->fs_info->free_chunk_lock); | |
4731 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4732 | map->num_stripes); | |
4733 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4734 | ||
4735 | free_extent_map(em); | |
4736 | check_raid56_incompat_flag(extent_root->fs_info, type); | |
4737 | ||
4738 | kfree(devices_info); | |
4739 | return 0; | |
4740 | ||
4741 | error_del_extent: | |
4742 | write_lock(&em_tree->lock); | |
4743 | remove_extent_mapping(em_tree, em); | |
4744 | write_unlock(&em_tree->lock); | |
4745 | ||
4746 | /* One for our allocation */ | |
4747 | free_extent_map(em); | |
4748 | /* One for the tree reference */ | |
4749 | free_extent_map(em); | |
4750 | /* One for the pending_chunks list reference */ | |
4751 | free_extent_map(em); | |
4752 | error: | |
4753 | kfree(devices_info); | |
4754 | return ret; | |
4755 | } | |
4756 | ||
4757 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, | |
4758 | struct btrfs_root *extent_root, | |
4759 | u64 chunk_offset, u64 chunk_size) | |
4760 | { | |
4761 | struct btrfs_key key; | |
4762 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4763 | struct btrfs_device *device; | |
4764 | struct btrfs_chunk *chunk; | |
4765 | struct btrfs_stripe *stripe; | |
4766 | struct extent_map_tree *em_tree; | |
4767 | struct extent_map *em; | |
4768 | struct map_lookup *map; | |
4769 | size_t item_size; | |
4770 | u64 dev_offset; | |
4771 | u64 stripe_size; | |
4772 | int i = 0; | |
4773 | int ret = 0; | |
4774 | ||
4775 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; | |
4776 | read_lock(&em_tree->lock); | |
4777 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4778 | read_unlock(&em_tree->lock); | |
4779 | ||
4780 | if (!em) { | |
4781 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4782 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4783 | return -EINVAL; | |
4784 | } | |
4785 | ||
4786 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4787 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
4788 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, | |
4789 | chunk_size, em->start, em->len); | |
4790 | free_extent_map(em); | |
4791 | return -EINVAL; | |
4792 | } | |
4793 | ||
4794 | map = em->map_lookup; | |
4795 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4796 | stripe_size = em->orig_block_len; | |
4797 | ||
4798 | chunk = kzalloc(item_size, GFP_NOFS); | |
4799 | if (!chunk) { | |
4800 | ret = -ENOMEM; | |
4801 | goto out; | |
4802 | } | |
4803 | ||
4804 | /* | |
4805 | * Take the device list mutex to prevent races with the final phase of | |
4806 | * a device replace operation that replaces the device object associated | |
4807 | * with the map's stripes, because the device object's id can change | |
4808 | * at any time during that final phase of the device replace operation | |
4809 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4810 | */ | |
4811 | mutex_lock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4812 | for (i = 0; i < map->num_stripes; i++) { | |
4813 | device = map->stripes[i].dev; | |
4814 | dev_offset = map->stripes[i].physical; | |
4815 | ||
4816 | ret = btrfs_update_device(trans, device); | |
4817 | if (ret) | |
4818 | break; | |
4819 | ret = btrfs_alloc_dev_extent(trans, device, | |
4820 | chunk_root->root_key.objectid, | |
4821 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4822 | chunk_offset, dev_offset, | |
4823 | stripe_size); | |
4824 | if (ret) | |
4825 | break; | |
4826 | } | |
4827 | if (ret) { | |
4828 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4829 | goto out; | |
4830 | } | |
4831 | ||
4832 | stripe = &chunk->stripe; | |
4833 | for (i = 0; i < map->num_stripes; i++) { | |
4834 | device = map->stripes[i].dev; | |
4835 | dev_offset = map->stripes[i].physical; | |
4836 | ||
4837 | btrfs_set_stack_stripe_devid(stripe, device->devid); | |
4838 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4839 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
4840 | stripe++; | |
4841 | } | |
4842 | mutex_unlock(&chunk_root->fs_info->fs_devices->device_list_mutex); | |
4843 | ||
4844 | btrfs_set_stack_chunk_length(chunk, chunk_size); | |
4845 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); | |
4846 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); | |
4847 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4848 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4849 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4850 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
4851 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); | |
4852 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); | |
4853 | ||
4854 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; | |
4855 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4856 | key.offset = chunk_offset; | |
4857 | ||
4858 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); | |
4859 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { | |
4860 | /* | |
4861 | * TODO: Cleanup of inserted chunk root in case of | |
4862 | * failure. | |
4863 | */ | |
4864 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, | |
4865 | item_size); | |
4866 | } | |
4867 | ||
4868 | out: | |
4869 | kfree(chunk); | |
4870 | free_extent_map(em); | |
4871 | return ret; | |
4872 | } | |
4873 | ||
4874 | /* | |
4875 | * Chunk allocation falls into two parts. The first part does works | |
4876 | * that make the new allocated chunk useable, but not do any operation | |
4877 | * that modifies the chunk tree. The second part does the works that | |
4878 | * require modifying the chunk tree. This division is important for the | |
4879 | * bootstrap process of adding storage to a seed btrfs. | |
4880 | */ | |
4881 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4882 | struct btrfs_root *extent_root, u64 type) | |
4883 | { | |
4884 | u64 chunk_offset; | |
4885 | ||
4886 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); | |
4887 | chunk_offset = find_next_chunk(extent_root->fs_info); | |
4888 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
4889 | } | |
4890 | ||
4891 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, | |
4892 | struct btrfs_root *root, | |
4893 | struct btrfs_device *device) | |
4894 | { | |
4895 | u64 chunk_offset; | |
4896 | u64 sys_chunk_offset; | |
4897 | u64 alloc_profile; | |
4898 | struct btrfs_fs_info *fs_info = root->fs_info; | |
4899 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4900 | int ret; | |
4901 | ||
4902 | chunk_offset = find_next_chunk(fs_info); | |
4903 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); | |
4904 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, | |
4905 | alloc_profile); | |
4906 | if (ret) | |
4907 | return ret; | |
4908 | ||
4909 | sys_chunk_offset = find_next_chunk(root->fs_info); | |
4910 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); | |
4911 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, | |
4912 | alloc_profile); | |
4913 | return ret; | |
4914 | } | |
4915 | ||
4916 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) | |
4917 | { | |
4918 | int max_errors; | |
4919 | ||
4920 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4921 | BTRFS_BLOCK_GROUP_RAID10 | | |
4922 | BTRFS_BLOCK_GROUP_RAID5 | | |
4923 | BTRFS_BLOCK_GROUP_DUP)) { | |
4924 | max_errors = 1; | |
4925 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4926 | max_errors = 2; | |
4927 | } else { | |
4928 | max_errors = 0; | |
4929 | } | |
4930 | ||
4931 | return max_errors; | |
4932 | } | |
4933 | ||
4934 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4935 | { | |
4936 | struct extent_map *em; | |
4937 | struct map_lookup *map; | |
4938 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4939 | int readonly = 0; | |
4940 | int miss_ndevs = 0; | |
4941 | int i; | |
4942 | ||
4943 | read_lock(&map_tree->map_tree.lock); | |
4944 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); | |
4945 | read_unlock(&map_tree->map_tree.lock); | |
4946 | if (!em) | |
4947 | return 1; | |
4948 | ||
4949 | map = em->map_lookup; | |
4950 | for (i = 0; i < map->num_stripes; i++) { | |
4951 | if (map->stripes[i].dev->missing) { | |
4952 | miss_ndevs++; | |
4953 | continue; | |
4954 | } | |
4955 | ||
4956 | if (!map->stripes[i].dev->writeable) { | |
4957 | readonly = 1; | |
4958 | goto end; | |
4959 | } | |
4960 | } | |
4961 | ||
4962 | /* | |
4963 | * If the number of missing devices is larger than max errors, | |
4964 | * we can not write the data into that chunk successfully, so | |
4965 | * set it readonly. | |
4966 | */ | |
4967 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4968 | readonly = 1; | |
4969 | end: | |
4970 | free_extent_map(em); | |
4971 | return readonly; | |
4972 | } | |
4973 | ||
4974 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4975 | { | |
4976 | extent_map_tree_init(&tree->map_tree); | |
4977 | } | |
4978 | ||
4979 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4980 | { | |
4981 | struct extent_map *em; | |
4982 | ||
4983 | while (1) { | |
4984 | write_lock(&tree->map_tree.lock); | |
4985 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); | |
4986 | if (em) | |
4987 | remove_extent_mapping(&tree->map_tree, em); | |
4988 | write_unlock(&tree->map_tree.lock); | |
4989 | if (!em) | |
4990 | break; | |
4991 | /* once for us */ | |
4992 | free_extent_map(em); | |
4993 | /* once for the tree */ | |
4994 | free_extent_map(em); | |
4995 | } | |
4996 | } | |
4997 | ||
4998 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) | |
4999 | { | |
5000 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
5001 | struct extent_map *em; | |
5002 | struct map_lookup *map; | |
5003 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5004 | int ret; | |
5005 | ||
5006 | read_lock(&em_tree->lock); | |
5007 | em = lookup_extent_mapping(em_tree, logical, len); | |
5008 | read_unlock(&em_tree->lock); | |
5009 | ||
5010 | /* | |
5011 | * We could return errors for these cases, but that could get ugly and | |
5012 | * we'd probably do the same thing which is just not do anything else | |
5013 | * and exit, so return 1 so the callers don't try to use other copies. | |
5014 | */ | |
5015 | if (!em) { | |
5016 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, | |
5017 | logical+len); | |
5018 | return 1; | |
5019 | } | |
5020 | ||
5021 | if (em->start > logical || em->start + em->len < logical) { | |
5022 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " | |
5023 | "%Lu-%Lu", logical, logical+len, em->start, | |
5024 | em->start + em->len); | |
5025 | free_extent_map(em); | |
5026 | return 1; | |
5027 | } | |
5028 | ||
5029 | map = em->map_lookup; | |
5030 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
5031 | ret = map->num_stripes; | |
5032 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5033 | ret = map->sub_stripes; | |
5034 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) | |
5035 | ret = 2; | |
5036 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5037 | ret = 3; | |
5038 | else | |
5039 | ret = 1; | |
5040 | free_extent_map(em); | |
5041 | ||
5042 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
5043 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
5044 | ret++; | |
5045 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); | |
5046 | ||
5047 | return ret; | |
5048 | } | |
5049 | ||
5050 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, | |
5051 | struct btrfs_mapping_tree *map_tree, | |
5052 | u64 logical) | |
5053 | { | |
5054 | struct extent_map *em; | |
5055 | struct map_lookup *map; | |
5056 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5057 | unsigned long len = root->sectorsize; | |
5058 | ||
5059 | read_lock(&em_tree->lock); | |
5060 | em = lookup_extent_mapping(em_tree, logical, len); | |
5061 | read_unlock(&em_tree->lock); | |
5062 | BUG_ON(!em); | |
5063 | ||
5064 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
5065 | map = em->map_lookup; | |
5066 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5067 | len = map->stripe_len * nr_data_stripes(map); | |
5068 | free_extent_map(em); | |
5069 | return len; | |
5070 | } | |
5071 | ||
5072 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
5073 | u64 logical, u64 len, int mirror_num) | |
5074 | { | |
5075 | struct extent_map *em; | |
5076 | struct map_lookup *map; | |
5077 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5078 | int ret = 0; | |
5079 | ||
5080 | read_lock(&em_tree->lock); | |
5081 | em = lookup_extent_mapping(em_tree, logical, len); | |
5082 | read_unlock(&em_tree->lock); | |
5083 | BUG_ON(!em); | |
5084 | ||
5085 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
5086 | map = em->map_lookup; | |
5087 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) | |
5088 | ret = 1; | |
5089 | free_extent_map(em); | |
5090 | return ret; | |
5091 | } | |
5092 | ||
5093 | static int find_live_mirror(struct btrfs_fs_info *fs_info, | |
5094 | struct map_lookup *map, int first, int num, | |
5095 | int optimal, int dev_replace_is_ongoing) | |
5096 | { | |
5097 | int i; | |
5098 | int tolerance; | |
5099 | struct btrfs_device *srcdev; | |
5100 | ||
5101 | if (dev_replace_is_ongoing && | |
5102 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5103 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5104 | srcdev = fs_info->dev_replace.srcdev; | |
5105 | else | |
5106 | srcdev = NULL; | |
5107 | ||
5108 | /* | |
5109 | * try to avoid the drive that is the source drive for a | |
5110 | * dev-replace procedure, only choose it if no other non-missing | |
5111 | * mirror is available | |
5112 | */ | |
5113 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5114 | if (map->stripes[optimal].dev->bdev && | |
5115 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5116 | return optimal; | |
5117 | for (i = first; i < first + num; i++) { | |
5118 | if (map->stripes[i].dev->bdev && | |
5119 | (tolerance || map->stripes[i].dev != srcdev)) | |
5120 | return i; | |
5121 | } | |
5122 | } | |
5123 | ||
5124 | /* we couldn't find one that doesn't fail. Just return something | |
5125 | * and the io error handling code will clean up eventually | |
5126 | */ | |
5127 | return optimal; | |
5128 | } | |
5129 | ||
5130 | static inline int parity_smaller(u64 a, u64 b) | |
5131 | { | |
5132 | return a > b; | |
5133 | } | |
5134 | ||
5135 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
5136 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) | |
5137 | { | |
5138 | struct btrfs_bio_stripe s; | |
5139 | int i; | |
5140 | u64 l; | |
5141 | int again = 1; | |
5142 | ||
5143 | while (again) { | |
5144 | again = 0; | |
5145 | for (i = 0; i < num_stripes - 1; i++) { | |
5146 | if (parity_smaller(bbio->raid_map[i], | |
5147 | bbio->raid_map[i+1])) { | |
5148 | s = bbio->stripes[i]; | |
5149 | l = bbio->raid_map[i]; | |
5150 | bbio->stripes[i] = bbio->stripes[i+1]; | |
5151 | bbio->raid_map[i] = bbio->raid_map[i+1]; | |
5152 | bbio->stripes[i+1] = s; | |
5153 | bbio->raid_map[i+1] = l; | |
5154 | ||
5155 | again = 1; | |
5156 | } | |
5157 | } | |
5158 | } | |
5159 | } | |
5160 | ||
5161 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) | |
5162 | { | |
5163 | struct btrfs_bio *bbio = kzalloc( | |
5164 | /* the size of the btrfs_bio */ | |
5165 | sizeof(struct btrfs_bio) + | |
5166 | /* plus the variable array for the stripes */ | |
5167 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + | |
5168 | /* plus the variable array for the tgt dev */ | |
5169 | sizeof(int) * (real_stripes) + | |
5170 | /* | |
5171 | * plus the raid_map, which includes both the tgt dev | |
5172 | * and the stripes | |
5173 | */ | |
5174 | sizeof(u64) * (total_stripes), | |
5175 | GFP_NOFS|__GFP_NOFAIL); | |
5176 | ||
5177 | atomic_set(&bbio->error, 0); | |
5178 | atomic_set(&bbio->refs, 1); | |
5179 | ||
5180 | return bbio; | |
5181 | } | |
5182 | ||
5183 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5184 | { | |
5185 | WARN_ON(!atomic_read(&bbio->refs)); | |
5186 | atomic_inc(&bbio->refs); | |
5187 | } | |
5188 | ||
5189 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5190 | { | |
5191 | if (!bbio) | |
5192 | return; | |
5193 | if (atomic_dec_and_test(&bbio->refs)) | |
5194 | kfree(bbio); | |
5195 | } | |
5196 | ||
5197 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, | |
5198 | u64 logical, u64 *length, | |
5199 | struct btrfs_bio **bbio_ret, | |
5200 | int mirror_num, int need_raid_map) | |
5201 | { | |
5202 | struct extent_map *em; | |
5203 | struct map_lookup *map; | |
5204 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
5205 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5206 | u64 offset; | |
5207 | u64 stripe_offset; | |
5208 | u64 stripe_end_offset; | |
5209 | u64 stripe_nr; | |
5210 | u64 stripe_nr_orig; | |
5211 | u64 stripe_nr_end; | |
5212 | u64 stripe_len; | |
5213 | u32 stripe_index; | |
5214 | int i; | |
5215 | int ret = 0; | |
5216 | int num_stripes; | |
5217 | int max_errors = 0; | |
5218 | int tgtdev_indexes = 0; | |
5219 | struct btrfs_bio *bbio = NULL; | |
5220 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; | |
5221 | int dev_replace_is_ongoing = 0; | |
5222 | int num_alloc_stripes; | |
5223 | int patch_the_first_stripe_for_dev_replace = 0; | |
5224 | u64 physical_to_patch_in_first_stripe = 0; | |
5225 | u64 raid56_full_stripe_start = (u64)-1; | |
5226 | ||
5227 | read_lock(&em_tree->lock); | |
5228 | em = lookup_extent_mapping(em_tree, logical, *length); | |
5229 | read_unlock(&em_tree->lock); | |
5230 | ||
5231 | if (!em) { | |
5232 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", | |
5233 | logical, *length); | |
5234 | return -EINVAL; | |
5235 | } | |
5236 | ||
5237 | if (em->start > logical || em->start + em->len < logical) { | |
5238 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
5239 | "found %Lu-%Lu", logical, em->start, | |
5240 | em->start + em->len); | |
5241 | free_extent_map(em); | |
5242 | return -EINVAL; | |
5243 | } | |
5244 | ||
5245 | map = em->map_lookup; | |
5246 | offset = logical - em->start; | |
5247 | ||
5248 | stripe_len = map->stripe_len; | |
5249 | stripe_nr = offset; | |
5250 | /* | |
5251 | * stripe_nr counts the total number of stripes we have to stride | |
5252 | * to get to this block | |
5253 | */ | |
5254 | stripe_nr = div64_u64(stripe_nr, stripe_len); | |
5255 | ||
5256 | stripe_offset = stripe_nr * stripe_len; | |
5257 | BUG_ON(offset < stripe_offset); | |
5258 | ||
5259 | /* stripe_offset is the offset of this block in its stripe*/ | |
5260 | stripe_offset = offset - stripe_offset; | |
5261 | ||
5262 | /* if we're here for raid56, we need to know the stripe aligned start */ | |
5263 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5264 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); | |
5265 | raid56_full_stripe_start = offset; | |
5266 | ||
5267 | /* allow a write of a full stripe, but make sure we don't | |
5268 | * allow straddling of stripes | |
5269 | */ | |
5270 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, | |
5271 | full_stripe_len); | |
5272 | raid56_full_stripe_start *= full_stripe_len; | |
5273 | } | |
5274 | ||
5275 | if (rw & REQ_DISCARD) { | |
5276 | /* we don't discard raid56 yet */ | |
5277 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5278 | ret = -EOPNOTSUPP; | |
5279 | goto out; | |
5280 | } | |
5281 | *length = min_t(u64, em->len - offset, *length); | |
5282 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { | |
5283 | u64 max_len; | |
5284 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5285 | For other RAID types and for RAID[56] reads, just allow a single | |
5286 | stripe (on a single disk). */ | |
5287 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && | |
5288 | (rw & REQ_WRITE)) { | |
5289 | max_len = stripe_len * nr_data_stripes(map) - | |
5290 | (offset - raid56_full_stripe_start); | |
5291 | } else { | |
5292 | /* we limit the length of each bio to what fits in a stripe */ | |
5293 | max_len = stripe_len - stripe_offset; | |
5294 | } | |
5295 | *length = min_t(u64, em->len - offset, max_len); | |
5296 | } else { | |
5297 | *length = em->len - offset; | |
5298 | } | |
5299 | ||
5300 | /* This is for when we're called from btrfs_merge_bio_hook() and all | |
5301 | it cares about is the length */ | |
5302 | if (!bbio_ret) | |
5303 | goto out; | |
5304 | ||
5305 | btrfs_dev_replace_lock(dev_replace, 0); | |
5306 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
5307 | if (!dev_replace_is_ongoing) | |
5308 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5309 | else | |
5310 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
5311 | ||
5312 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && | |
5313 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5314 | dev_replace->tgtdev != NULL) { | |
5315 | /* | |
5316 | * in dev-replace case, for repair case (that's the only | |
5317 | * case where the mirror is selected explicitly when | |
5318 | * calling btrfs_map_block), blocks left of the left cursor | |
5319 | * can also be read from the target drive. | |
5320 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5321 | * the last one to the array of stripes. For READ, it also | |
5322 | * needs to be supported using the same mirror number. | |
5323 | * If the requested block is not left of the left cursor, | |
5324 | * EIO is returned. This can happen because btrfs_num_copies() | |
5325 | * returns one more in the dev-replace case. | |
5326 | */ | |
5327 | u64 tmp_length = *length; | |
5328 | struct btrfs_bio *tmp_bbio = NULL; | |
5329 | int tmp_num_stripes; | |
5330 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5331 | int index_srcdev = 0; | |
5332 | int found = 0; | |
5333 | u64 physical_of_found = 0; | |
5334 | ||
5335 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
5336 | logical, &tmp_length, &tmp_bbio, 0, 0); | |
5337 | if (ret) { | |
5338 | WARN_ON(tmp_bbio != NULL); | |
5339 | goto out; | |
5340 | } | |
5341 | ||
5342 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5343 | if (mirror_num > tmp_num_stripes) { | |
5344 | /* | |
5345 | * REQ_GET_READ_MIRRORS does not contain this | |
5346 | * mirror, that means that the requested area | |
5347 | * is not left of the left cursor | |
5348 | */ | |
5349 | ret = -EIO; | |
5350 | btrfs_put_bbio(tmp_bbio); | |
5351 | goto out; | |
5352 | } | |
5353 | ||
5354 | /* | |
5355 | * process the rest of the function using the mirror_num | |
5356 | * of the source drive. Therefore look it up first. | |
5357 | * At the end, patch the device pointer to the one of the | |
5358 | * target drive. | |
5359 | */ | |
5360 | for (i = 0; i < tmp_num_stripes; i++) { | |
5361 | if (tmp_bbio->stripes[i].dev->devid != srcdev_devid) | |
5362 | continue; | |
5363 | ||
5364 | /* | |
5365 | * In case of DUP, in order to keep it simple, only add | |
5366 | * the mirror with the lowest physical address | |
5367 | */ | |
5368 | if (found && | |
5369 | physical_of_found <= tmp_bbio->stripes[i].physical) | |
5370 | continue; | |
5371 | ||
5372 | index_srcdev = i; | |
5373 | found = 1; | |
5374 | physical_of_found = tmp_bbio->stripes[i].physical; | |
5375 | } | |
5376 | ||
5377 | btrfs_put_bbio(tmp_bbio); | |
5378 | ||
5379 | if (!found) { | |
5380 | WARN_ON(1); | |
5381 | ret = -EIO; | |
5382 | goto out; | |
5383 | } | |
5384 | ||
5385 | mirror_num = index_srcdev + 1; | |
5386 | patch_the_first_stripe_for_dev_replace = 1; | |
5387 | physical_to_patch_in_first_stripe = physical_of_found; | |
5388 | } else if (mirror_num > map->num_stripes) { | |
5389 | mirror_num = 0; | |
5390 | } | |
5391 | ||
5392 | num_stripes = 1; | |
5393 | stripe_index = 0; | |
5394 | stripe_nr_orig = stripe_nr; | |
5395 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); | |
5396 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); | |
5397 | stripe_end_offset = stripe_nr_end * map->stripe_len - | |
5398 | (offset + *length); | |
5399 | ||
5400 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
5401 | if (rw & REQ_DISCARD) | |
5402 | num_stripes = min_t(u64, map->num_stripes, | |
5403 | stripe_nr_end - stripe_nr_orig); | |
5404 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5405 | &stripe_index); | |
5406 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) | |
5407 | mirror_num = 1; | |
5408 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { | |
5409 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) | |
5410 | num_stripes = map->num_stripes; | |
5411 | else if (mirror_num) | |
5412 | stripe_index = mirror_num - 1; | |
5413 | else { | |
5414 | stripe_index = find_live_mirror(fs_info, map, 0, | |
5415 | map->num_stripes, | |
5416 | current->pid % map->num_stripes, | |
5417 | dev_replace_is_ongoing); | |
5418 | mirror_num = stripe_index + 1; | |
5419 | } | |
5420 | ||
5421 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { | |
5422 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { | |
5423 | num_stripes = map->num_stripes; | |
5424 | } else if (mirror_num) { | |
5425 | stripe_index = mirror_num - 1; | |
5426 | } else { | |
5427 | mirror_num = 1; | |
5428 | } | |
5429 | ||
5430 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5431 | u32 factor = map->num_stripes / map->sub_stripes; | |
5432 | ||
5433 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); | |
5434 | stripe_index *= map->sub_stripes; | |
5435 | ||
5436 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) | |
5437 | num_stripes = map->sub_stripes; | |
5438 | else if (rw & REQ_DISCARD) | |
5439 | num_stripes = min_t(u64, map->sub_stripes * | |
5440 | (stripe_nr_end - stripe_nr_orig), | |
5441 | map->num_stripes); | |
5442 | else if (mirror_num) | |
5443 | stripe_index += mirror_num - 1; | |
5444 | else { | |
5445 | int old_stripe_index = stripe_index; | |
5446 | stripe_index = find_live_mirror(fs_info, map, | |
5447 | stripe_index, | |
5448 | map->sub_stripes, stripe_index + | |
5449 | current->pid % map->sub_stripes, | |
5450 | dev_replace_is_ongoing); | |
5451 | mirror_num = stripe_index - old_stripe_index + 1; | |
5452 | } | |
5453 | ||
5454 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5455 | if (need_raid_map && | |
5456 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || | |
5457 | mirror_num > 1)) { | |
5458 | /* push stripe_nr back to the start of the full stripe */ | |
5459 | stripe_nr = div_u64(raid56_full_stripe_start, | |
5460 | stripe_len * nr_data_stripes(map)); | |
5461 | ||
5462 | /* RAID[56] write or recovery. Return all stripes */ | |
5463 | num_stripes = map->num_stripes; | |
5464 | max_errors = nr_parity_stripes(map); | |
5465 | ||
5466 | *length = map->stripe_len; | |
5467 | stripe_index = 0; | |
5468 | stripe_offset = 0; | |
5469 | } else { | |
5470 | /* | |
5471 | * Mirror #0 or #1 means the original data block. | |
5472 | * Mirror #2 is RAID5 parity block. | |
5473 | * Mirror #3 is RAID6 Q block. | |
5474 | */ | |
5475 | stripe_nr = div_u64_rem(stripe_nr, | |
5476 | nr_data_stripes(map), &stripe_index); | |
5477 | if (mirror_num > 1) | |
5478 | stripe_index = nr_data_stripes(map) + | |
5479 | mirror_num - 2; | |
5480 | ||
5481 | /* We distribute the parity blocks across stripes */ | |
5482 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, | |
5483 | &stripe_index); | |
5484 | if (!(rw & (REQ_WRITE | REQ_DISCARD | | |
5485 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5486 | mirror_num = 1; | |
5487 | } | |
5488 | } else { | |
5489 | /* | |
5490 | * after this, stripe_nr is the number of stripes on this | |
5491 | * device we have to walk to find the data, and stripe_index is | |
5492 | * the number of our device in the stripe array | |
5493 | */ | |
5494 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, | |
5495 | &stripe_index); | |
5496 | mirror_num = stripe_index + 1; | |
5497 | } | |
5498 | BUG_ON(stripe_index >= map->num_stripes); | |
5499 | ||
5500 | num_alloc_stripes = num_stripes; | |
5501 | if (dev_replace_is_ongoing) { | |
5502 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5503 | num_alloc_stripes <<= 1; | |
5504 | if (rw & REQ_GET_READ_MIRRORS) | |
5505 | num_alloc_stripes++; | |
5506 | tgtdev_indexes = num_stripes; | |
5507 | } | |
5508 | ||
5509 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); | |
5510 | if (!bbio) { | |
5511 | ret = -ENOMEM; | |
5512 | goto out; | |
5513 | } | |
5514 | if (dev_replace_is_ongoing) | |
5515 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
5516 | ||
5517 | /* build raid_map */ | |
5518 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && | |
5519 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || | |
5520 | mirror_num > 1)) { | |
5521 | u64 tmp; | |
5522 | unsigned rot; | |
5523 | ||
5524 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5525 | sizeof(struct btrfs_bio_stripe) * | |
5526 | num_alloc_stripes + | |
5527 | sizeof(int) * tgtdev_indexes); | |
5528 | ||
5529 | /* Work out the disk rotation on this stripe-set */ | |
5530 | div_u64_rem(stripe_nr, num_stripes, &rot); | |
5531 | ||
5532 | /* Fill in the logical address of each stripe */ | |
5533 | tmp = stripe_nr * nr_data_stripes(map); | |
5534 | for (i = 0; i < nr_data_stripes(map); i++) | |
5535 | bbio->raid_map[(i+rot) % num_stripes] = | |
5536 | em->start + (tmp + i) * map->stripe_len; | |
5537 | ||
5538 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5539 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5540 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5541 | RAID6_Q_STRIPE; | |
5542 | } | |
5543 | ||
5544 | if (rw & REQ_DISCARD) { | |
5545 | u32 factor = 0; | |
5546 | u32 sub_stripes = 0; | |
5547 | u64 stripes_per_dev = 0; | |
5548 | u32 remaining_stripes = 0; | |
5549 | u32 last_stripe = 0; | |
5550 | ||
5551 | if (map->type & | |
5552 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5553 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5554 | sub_stripes = 1; | |
5555 | else | |
5556 | sub_stripes = map->sub_stripes; | |
5557 | ||
5558 | factor = map->num_stripes / sub_stripes; | |
5559 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5560 | stripe_nr_orig, | |
5561 | factor, | |
5562 | &remaining_stripes); | |
5563 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); | |
5564 | last_stripe *= sub_stripes; | |
5565 | } | |
5566 | ||
5567 | for (i = 0; i < num_stripes; i++) { | |
5568 | bbio->stripes[i].physical = | |
5569 | map->stripes[stripe_index].physical + | |
5570 | stripe_offset + stripe_nr * map->stripe_len; | |
5571 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; | |
5572 | ||
5573 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | | |
5574 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5575 | bbio->stripes[i].length = stripes_per_dev * | |
5576 | map->stripe_len; | |
5577 | ||
5578 | if (i / sub_stripes < remaining_stripes) | |
5579 | bbio->stripes[i].length += | |
5580 | map->stripe_len; | |
5581 | ||
5582 | /* | |
5583 | * Special for the first stripe and | |
5584 | * the last stripe: | |
5585 | * | |
5586 | * |-------|...|-------| | |
5587 | * |----------| | |
5588 | * off end_off | |
5589 | */ | |
5590 | if (i < sub_stripes) | |
5591 | bbio->stripes[i].length -= | |
5592 | stripe_offset; | |
5593 | ||
5594 | if (stripe_index >= last_stripe && | |
5595 | stripe_index <= (last_stripe + | |
5596 | sub_stripes - 1)) | |
5597 | bbio->stripes[i].length -= | |
5598 | stripe_end_offset; | |
5599 | ||
5600 | if (i == sub_stripes - 1) | |
5601 | stripe_offset = 0; | |
5602 | } else | |
5603 | bbio->stripes[i].length = *length; | |
5604 | ||
5605 | stripe_index++; | |
5606 | if (stripe_index == map->num_stripes) { | |
5607 | /* This could only happen for RAID0/10 */ | |
5608 | stripe_index = 0; | |
5609 | stripe_nr++; | |
5610 | } | |
5611 | } | |
5612 | } else { | |
5613 | for (i = 0; i < num_stripes; i++) { | |
5614 | bbio->stripes[i].physical = | |
5615 | map->stripes[stripe_index].physical + | |
5616 | stripe_offset + | |
5617 | stripe_nr * map->stripe_len; | |
5618 | bbio->stripes[i].dev = | |
5619 | map->stripes[stripe_index].dev; | |
5620 | stripe_index++; | |
5621 | } | |
5622 | } | |
5623 | ||
5624 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) | |
5625 | max_errors = btrfs_chunk_max_errors(map); | |
5626 | ||
5627 | if (bbio->raid_map) | |
5628 | sort_parity_stripes(bbio, num_stripes); | |
5629 | ||
5630 | tgtdev_indexes = 0; | |
5631 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && | |
5632 | dev_replace->tgtdev != NULL) { | |
5633 | int index_where_to_add; | |
5634 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5635 | ||
5636 | /* | |
5637 | * duplicate the write operations while the dev replace | |
5638 | * procedure is running. Since the copying of the old disk | |
5639 | * to the new disk takes place at run time while the | |
5640 | * filesystem is mounted writable, the regular write | |
5641 | * operations to the old disk have to be duplicated to go | |
5642 | * to the new disk as well. | |
5643 | * Note that device->missing is handled by the caller, and | |
5644 | * that the write to the old disk is already set up in the | |
5645 | * stripes array. | |
5646 | */ | |
5647 | index_where_to_add = num_stripes; | |
5648 | for (i = 0; i < num_stripes; i++) { | |
5649 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5650 | /* write to new disk, too */ | |
5651 | struct btrfs_bio_stripe *new = | |
5652 | bbio->stripes + index_where_to_add; | |
5653 | struct btrfs_bio_stripe *old = | |
5654 | bbio->stripes + i; | |
5655 | ||
5656 | new->physical = old->physical; | |
5657 | new->length = old->length; | |
5658 | new->dev = dev_replace->tgtdev; | |
5659 | bbio->tgtdev_map[i] = index_where_to_add; | |
5660 | index_where_to_add++; | |
5661 | max_errors++; | |
5662 | tgtdev_indexes++; | |
5663 | } | |
5664 | } | |
5665 | num_stripes = index_where_to_add; | |
5666 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && | |
5667 | dev_replace->tgtdev != NULL) { | |
5668 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5669 | int index_srcdev = 0; | |
5670 | int found = 0; | |
5671 | u64 physical_of_found = 0; | |
5672 | ||
5673 | /* | |
5674 | * During the dev-replace procedure, the target drive can | |
5675 | * also be used to read data in case it is needed to repair | |
5676 | * a corrupt block elsewhere. This is possible if the | |
5677 | * requested area is left of the left cursor. In this area, | |
5678 | * the target drive is a full copy of the source drive. | |
5679 | */ | |
5680 | for (i = 0; i < num_stripes; i++) { | |
5681 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5682 | /* | |
5683 | * In case of DUP, in order to keep it | |
5684 | * simple, only add the mirror with the | |
5685 | * lowest physical address | |
5686 | */ | |
5687 | if (found && | |
5688 | physical_of_found <= | |
5689 | bbio->stripes[i].physical) | |
5690 | continue; | |
5691 | index_srcdev = i; | |
5692 | found = 1; | |
5693 | physical_of_found = bbio->stripes[i].physical; | |
5694 | } | |
5695 | } | |
5696 | if (found) { | |
5697 | if (physical_of_found + map->stripe_len <= | |
5698 | dev_replace->cursor_left) { | |
5699 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5700 | bbio->stripes + num_stripes; | |
5701 | ||
5702 | tgtdev_stripe->physical = physical_of_found; | |
5703 | tgtdev_stripe->length = | |
5704 | bbio->stripes[index_srcdev].length; | |
5705 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5706 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
5707 | ||
5708 | tgtdev_indexes++; | |
5709 | num_stripes++; | |
5710 | } | |
5711 | } | |
5712 | } | |
5713 | ||
5714 | *bbio_ret = bbio; | |
5715 | bbio->map_type = map->type; | |
5716 | bbio->num_stripes = num_stripes; | |
5717 | bbio->max_errors = max_errors; | |
5718 | bbio->mirror_num = mirror_num; | |
5719 | bbio->num_tgtdevs = tgtdev_indexes; | |
5720 | ||
5721 | /* | |
5722 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5723 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5724 | * available as a mirror | |
5725 | */ | |
5726 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5727 | WARN_ON(num_stripes > 1); | |
5728 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5729 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5730 | bbio->mirror_num = map->num_stripes + 1; | |
5731 | } | |
5732 | out: | |
5733 | if (dev_replace_is_ongoing) { | |
5734 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5735 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5736 | } | |
5737 | free_extent_map(em); | |
5738 | return ret; | |
5739 | } | |
5740 | ||
5741 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, | |
5742 | u64 logical, u64 *length, | |
5743 | struct btrfs_bio **bbio_ret, int mirror_num) | |
5744 | { | |
5745 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
5746 | mirror_num, 0); | |
5747 | } | |
5748 | ||
5749 | /* For Scrub/replace */ | |
5750 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5751 | u64 logical, u64 *length, | |
5752 | struct btrfs_bio **bbio_ret, int mirror_num, | |
5753 | int need_raid_map) | |
5754 | { | |
5755 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
5756 | mirror_num, need_raid_map); | |
5757 | } | |
5758 | ||
5759 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, | |
5760 | u64 chunk_start, u64 physical, u64 devid, | |
5761 | u64 **logical, int *naddrs, int *stripe_len) | |
5762 | { | |
5763 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5764 | struct extent_map *em; | |
5765 | struct map_lookup *map; | |
5766 | u64 *buf; | |
5767 | u64 bytenr; | |
5768 | u64 length; | |
5769 | u64 stripe_nr; | |
5770 | u64 rmap_len; | |
5771 | int i, j, nr = 0; | |
5772 | ||
5773 | read_lock(&em_tree->lock); | |
5774 | em = lookup_extent_mapping(em_tree, chunk_start, 1); | |
5775 | read_unlock(&em_tree->lock); | |
5776 | ||
5777 | if (!em) { | |
5778 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", | |
5779 | chunk_start); | |
5780 | return -EIO; | |
5781 | } | |
5782 | ||
5783 | if (em->start != chunk_start) { | |
5784 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", | |
5785 | em->start, chunk_start); | |
5786 | free_extent_map(em); | |
5787 | return -EIO; | |
5788 | } | |
5789 | map = em->map_lookup; | |
5790 | ||
5791 | length = em->len; | |
5792 | rmap_len = map->stripe_len; | |
5793 | ||
5794 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) | |
5795 | length = div_u64(length, map->num_stripes / map->sub_stripes); | |
5796 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5797 | length = div_u64(length, map->num_stripes); | |
5798 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { | |
5799 | length = div_u64(length, nr_data_stripes(map)); | |
5800 | rmap_len = map->stripe_len * nr_data_stripes(map); | |
5801 | } | |
5802 | ||
5803 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); | |
5804 | BUG_ON(!buf); /* -ENOMEM */ | |
5805 | ||
5806 | for (i = 0; i < map->num_stripes; i++) { | |
5807 | if (devid && map->stripes[i].dev->devid != devid) | |
5808 | continue; | |
5809 | if (map->stripes[i].physical > physical || | |
5810 | map->stripes[i].physical + length <= physical) | |
5811 | continue; | |
5812 | ||
5813 | stripe_nr = physical - map->stripes[i].physical; | |
5814 | stripe_nr = div_u64(stripe_nr, map->stripe_len); | |
5815 | ||
5816 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5817 | stripe_nr = stripe_nr * map->num_stripes + i; | |
5818 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); | |
5819 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { | |
5820 | stripe_nr = stripe_nr * map->num_stripes + i; | |
5821 | } /* else if RAID[56], multiply by nr_data_stripes(). | |
5822 | * Alternatively, just use rmap_len below instead of | |
5823 | * map->stripe_len */ | |
5824 | ||
5825 | bytenr = chunk_start + stripe_nr * rmap_len; | |
5826 | WARN_ON(nr >= map->num_stripes); | |
5827 | for (j = 0; j < nr; j++) { | |
5828 | if (buf[j] == bytenr) | |
5829 | break; | |
5830 | } | |
5831 | if (j == nr) { | |
5832 | WARN_ON(nr >= map->num_stripes); | |
5833 | buf[nr++] = bytenr; | |
5834 | } | |
5835 | } | |
5836 | ||
5837 | *logical = buf; | |
5838 | *naddrs = nr; | |
5839 | *stripe_len = rmap_len; | |
5840 | ||
5841 | free_extent_map(em); | |
5842 | return 0; | |
5843 | } | |
5844 | ||
5845 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) | |
5846 | { | |
5847 | bio->bi_private = bbio->private; | |
5848 | bio->bi_end_io = bbio->end_io; | |
5849 | bio_endio(bio); | |
5850 | ||
5851 | btrfs_put_bbio(bbio); | |
5852 | } | |
5853 | ||
5854 | static void btrfs_end_bio(struct bio *bio) | |
5855 | { | |
5856 | struct btrfs_bio *bbio = bio->bi_private; | |
5857 | int is_orig_bio = 0; | |
5858 | ||
5859 | if (bio->bi_error) { | |
5860 | atomic_inc(&bbio->error); | |
5861 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { | |
5862 | unsigned int stripe_index = | |
5863 | btrfs_io_bio(bio)->stripe_index; | |
5864 | struct btrfs_device *dev; | |
5865 | ||
5866 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5867 | dev = bbio->stripes[stripe_index].dev; | |
5868 | if (dev->bdev) { | |
5869 | if (bio->bi_rw & WRITE) | |
5870 | btrfs_dev_stat_inc(dev, | |
5871 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5872 | else | |
5873 | btrfs_dev_stat_inc(dev, | |
5874 | BTRFS_DEV_STAT_READ_ERRS); | |
5875 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5876 | btrfs_dev_stat_inc(dev, | |
5877 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5878 | btrfs_dev_stat_print_on_error(dev); | |
5879 | } | |
5880 | } | |
5881 | } | |
5882 | ||
5883 | if (bio == bbio->orig_bio) | |
5884 | is_orig_bio = 1; | |
5885 | ||
5886 | btrfs_bio_counter_dec(bbio->fs_info); | |
5887 | ||
5888 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
5889 | if (!is_orig_bio) { | |
5890 | bio_put(bio); | |
5891 | bio = bbio->orig_bio; | |
5892 | } | |
5893 | ||
5894 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; | |
5895 | /* only send an error to the higher layers if it is | |
5896 | * beyond the tolerance of the btrfs bio | |
5897 | */ | |
5898 | if (atomic_read(&bbio->error) > bbio->max_errors) { | |
5899 | bio->bi_error = -EIO; | |
5900 | } else { | |
5901 | /* | |
5902 | * this bio is actually up to date, we didn't | |
5903 | * go over the max number of errors | |
5904 | */ | |
5905 | bio->bi_error = 0; | |
5906 | } | |
5907 | ||
5908 | btrfs_end_bbio(bbio, bio); | |
5909 | } else if (!is_orig_bio) { | |
5910 | bio_put(bio); | |
5911 | } | |
5912 | } | |
5913 | ||
5914 | /* | |
5915 | * see run_scheduled_bios for a description of why bios are collected for | |
5916 | * async submit. | |
5917 | * | |
5918 | * This will add one bio to the pending list for a device and make sure | |
5919 | * the work struct is scheduled. | |
5920 | */ | |
5921 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, | |
5922 | struct btrfs_device *device, | |
5923 | int rw, struct bio *bio) | |
5924 | { | |
5925 | int should_queue = 1; | |
5926 | struct btrfs_pending_bios *pending_bios; | |
5927 | ||
5928 | if (device->missing || !device->bdev) { | |
5929 | bio_io_error(bio); | |
5930 | return; | |
5931 | } | |
5932 | ||
5933 | /* don't bother with additional async steps for reads, right now */ | |
5934 | if (!(rw & REQ_WRITE)) { | |
5935 | bio_get(bio); | |
5936 | btrfsic_submit_bio(rw, bio); | |
5937 | bio_put(bio); | |
5938 | return; | |
5939 | } | |
5940 | ||
5941 | /* | |
5942 | * nr_async_bios allows us to reliably return congestion to the | |
5943 | * higher layers. Otherwise, the async bio makes it appear we have | |
5944 | * made progress against dirty pages when we've really just put it | |
5945 | * on a queue for later | |
5946 | */ | |
5947 | atomic_inc(&root->fs_info->nr_async_bios); | |
5948 | WARN_ON(bio->bi_next); | |
5949 | bio->bi_next = NULL; | |
5950 | bio->bi_rw |= rw; | |
5951 | ||
5952 | spin_lock(&device->io_lock); | |
5953 | if (bio->bi_rw & REQ_SYNC) | |
5954 | pending_bios = &device->pending_sync_bios; | |
5955 | else | |
5956 | pending_bios = &device->pending_bios; | |
5957 | ||
5958 | if (pending_bios->tail) | |
5959 | pending_bios->tail->bi_next = bio; | |
5960 | ||
5961 | pending_bios->tail = bio; | |
5962 | if (!pending_bios->head) | |
5963 | pending_bios->head = bio; | |
5964 | if (device->running_pending) | |
5965 | should_queue = 0; | |
5966 | ||
5967 | spin_unlock(&device->io_lock); | |
5968 | ||
5969 | if (should_queue) | |
5970 | btrfs_queue_work(root->fs_info->submit_workers, | |
5971 | &device->work); | |
5972 | } | |
5973 | ||
5974 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5975 | struct bio *bio, u64 physical, int dev_nr, | |
5976 | int rw, int async) | |
5977 | { | |
5978 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5979 | ||
5980 | bio->bi_private = bbio; | |
5981 | btrfs_io_bio(bio)->stripe_index = dev_nr; | |
5982 | bio->bi_end_io = btrfs_end_bio; | |
5983 | bio->bi_iter.bi_sector = physical >> 9; | |
5984 | #ifdef DEBUG | |
5985 | { | |
5986 | struct rcu_string *name; | |
5987 | ||
5988 | rcu_read_lock(); | |
5989 | name = rcu_dereference(dev->name); | |
5990 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " | |
5991 | "(%s id %llu), size=%u\n", rw, | |
5992 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, | |
5993 | name->str, dev->devid, bio->bi_iter.bi_size); | |
5994 | rcu_read_unlock(); | |
5995 | } | |
5996 | #endif | |
5997 | bio->bi_bdev = dev->bdev; | |
5998 | ||
5999 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
6000 | ||
6001 | if (async) | |
6002 | btrfs_schedule_bio(root, dev, rw, bio); | |
6003 | else | |
6004 | btrfsic_submit_bio(rw, bio); | |
6005 | } | |
6006 | ||
6007 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
6008 | { | |
6009 | atomic_inc(&bbio->error); | |
6010 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
6011 | /* Shoud be the original bio. */ | |
6012 | WARN_ON(bio != bbio->orig_bio); | |
6013 | ||
6014 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; | |
6015 | bio->bi_iter.bi_sector = logical >> 9; | |
6016 | bio->bi_error = -EIO; | |
6017 | btrfs_end_bbio(bbio, bio); | |
6018 | } | |
6019 | } | |
6020 | ||
6021 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, | |
6022 | int mirror_num, int async_submit) | |
6023 | { | |
6024 | struct btrfs_device *dev; | |
6025 | struct bio *first_bio = bio; | |
6026 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; | |
6027 | u64 length = 0; | |
6028 | u64 map_length; | |
6029 | int ret; | |
6030 | int dev_nr; | |
6031 | int total_devs; | |
6032 | struct btrfs_bio *bbio = NULL; | |
6033 | ||
6034 | length = bio->bi_iter.bi_size; | |
6035 | map_length = length; | |
6036 | ||
6037 | btrfs_bio_counter_inc_blocked(root->fs_info); | |
6038 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, | |
6039 | mirror_num, 1); | |
6040 | if (ret) { | |
6041 | btrfs_bio_counter_dec(root->fs_info); | |
6042 | return ret; | |
6043 | } | |
6044 | ||
6045 | total_devs = bbio->num_stripes; | |
6046 | bbio->orig_bio = first_bio; | |
6047 | bbio->private = first_bio->bi_private; | |
6048 | bbio->end_io = first_bio->bi_end_io; | |
6049 | bbio->fs_info = root->fs_info; | |
6050 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); | |
6051 | ||
6052 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && | |
6053 | ((rw & WRITE) || (mirror_num > 1))) { | |
6054 | /* In this case, map_length has been set to the length of | |
6055 | a single stripe; not the whole write */ | |
6056 | if (rw & WRITE) { | |
6057 | ret = raid56_parity_write(root, bio, bbio, map_length); | |
6058 | } else { | |
6059 | ret = raid56_parity_recover(root, bio, bbio, map_length, | |
6060 | mirror_num, 1); | |
6061 | } | |
6062 | ||
6063 | btrfs_bio_counter_dec(root->fs_info); | |
6064 | return ret; | |
6065 | } | |
6066 | ||
6067 | if (map_length < length) { | |
6068 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", | |
6069 | logical, length, map_length); | |
6070 | BUG(); | |
6071 | } | |
6072 | ||
6073 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { | |
6074 | dev = bbio->stripes[dev_nr].dev; | |
6075 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
6076 | bbio_error(bbio, first_bio, logical); | |
6077 | continue; | |
6078 | } | |
6079 | ||
6080 | if (dev_nr < total_devs - 1) { | |
6081 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); | |
6082 | BUG_ON(!bio); /* -ENOMEM */ | |
6083 | } else | |
6084 | bio = first_bio; | |
6085 | ||
6086 | submit_stripe_bio(root, bbio, bio, | |
6087 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
6088 | async_submit); | |
6089 | } | |
6090 | btrfs_bio_counter_dec(root->fs_info); | |
6091 | return 0; | |
6092 | } | |
6093 | ||
6094 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, | |
6095 | u8 *uuid, u8 *fsid) | |
6096 | { | |
6097 | struct btrfs_device *device; | |
6098 | struct btrfs_fs_devices *cur_devices; | |
6099 | ||
6100 | cur_devices = fs_info->fs_devices; | |
6101 | while (cur_devices) { | |
6102 | if (!fsid || | |
6103 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6104 | device = __find_device(&cur_devices->devices, | |
6105 | devid, uuid); | |
6106 | if (device) | |
6107 | return device; | |
6108 | } | |
6109 | cur_devices = cur_devices->seed; | |
6110 | } | |
6111 | return NULL; | |
6112 | } | |
6113 | ||
6114 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, | |
6115 | struct btrfs_fs_devices *fs_devices, | |
6116 | u64 devid, u8 *dev_uuid) | |
6117 | { | |
6118 | struct btrfs_device *device; | |
6119 | ||
6120 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); | |
6121 | if (IS_ERR(device)) | |
6122 | return NULL; | |
6123 | ||
6124 | list_add(&device->dev_list, &fs_devices->devices); | |
6125 | device->fs_devices = fs_devices; | |
6126 | fs_devices->num_devices++; | |
6127 | ||
6128 | device->missing = 1; | |
6129 | fs_devices->missing_devices++; | |
6130 | ||
6131 | return device; | |
6132 | } | |
6133 | ||
6134 | /** | |
6135 | * btrfs_alloc_device - allocate struct btrfs_device | |
6136 | * @fs_info: used only for generating a new devid, can be NULL if | |
6137 | * devid is provided (i.e. @devid != NULL). | |
6138 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6139 | * is generated. | |
6140 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6141 | * is generated. | |
6142 | * | |
6143 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6144 | * on error. Returned struct is not linked onto any lists and can be | |
6145 | * destroyed with kfree() right away. | |
6146 | */ | |
6147 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6148 | const u64 *devid, | |
6149 | const u8 *uuid) | |
6150 | { | |
6151 | struct btrfs_device *dev; | |
6152 | u64 tmp; | |
6153 | ||
6154 | if (WARN_ON(!devid && !fs_info)) | |
6155 | return ERR_PTR(-EINVAL); | |
6156 | ||
6157 | dev = __alloc_device(); | |
6158 | if (IS_ERR(dev)) | |
6159 | return dev; | |
6160 | ||
6161 | if (devid) | |
6162 | tmp = *devid; | |
6163 | else { | |
6164 | int ret; | |
6165 | ||
6166 | ret = find_next_devid(fs_info, &tmp); | |
6167 | if (ret) { | |
6168 | kfree(dev); | |
6169 | return ERR_PTR(ret); | |
6170 | } | |
6171 | } | |
6172 | dev->devid = tmp; | |
6173 | ||
6174 | if (uuid) | |
6175 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6176 | else | |
6177 | generate_random_uuid(dev->uuid); | |
6178 | ||
6179 | btrfs_init_work(&dev->work, btrfs_submit_helper, | |
6180 | pending_bios_fn, NULL, NULL); | |
6181 | ||
6182 | return dev; | |
6183 | } | |
6184 | ||
6185 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, | |
6186 | struct extent_buffer *leaf, | |
6187 | struct btrfs_chunk *chunk) | |
6188 | { | |
6189 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6190 | struct map_lookup *map; | |
6191 | struct extent_map *em; | |
6192 | u64 logical; | |
6193 | u64 length; | |
6194 | u64 stripe_len; | |
6195 | u64 devid; | |
6196 | u8 uuid[BTRFS_UUID_SIZE]; | |
6197 | int num_stripes; | |
6198 | int ret; | |
6199 | int i; | |
6200 | ||
6201 | logical = key->offset; | |
6202 | length = btrfs_chunk_length(leaf, chunk); | |
6203 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6204 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6205 | /* Validation check */ | |
6206 | if (!num_stripes) { | |
6207 | btrfs_err(root->fs_info, "invalid chunk num_stripes: %u", | |
6208 | num_stripes); | |
6209 | return -EIO; | |
6210 | } | |
6211 | if (!IS_ALIGNED(logical, root->sectorsize)) { | |
6212 | btrfs_err(root->fs_info, | |
6213 | "invalid chunk logical %llu", logical); | |
6214 | return -EIO; | |
6215 | } | |
6216 | if (!length || !IS_ALIGNED(length, root->sectorsize)) { | |
6217 | btrfs_err(root->fs_info, | |
6218 | "invalid chunk length %llu", length); | |
6219 | return -EIO; | |
6220 | } | |
6221 | if (!is_power_of_2(stripe_len)) { | |
6222 | btrfs_err(root->fs_info, "invalid chunk stripe length: %llu", | |
6223 | stripe_len); | |
6224 | return -EIO; | |
6225 | } | |
6226 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6227 | btrfs_chunk_type(leaf, chunk)) { | |
6228 | btrfs_err(root->fs_info, "unrecognized chunk type: %llu", | |
6229 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | | |
6230 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6231 | btrfs_chunk_type(leaf, chunk)); | |
6232 | return -EIO; | |
6233 | } | |
6234 | ||
6235 | read_lock(&map_tree->map_tree.lock); | |
6236 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); | |
6237 | read_unlock(&map_tree->map_tree.lock); | |
6238 | ||
6239 | /* already mapped? */ | |
6240 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6241 | free_extent_map(em); | |
6242 | return 0; | |
6243 | } else if (em) { | |
6244 | free_extent_map(em); | |
6245 | } | |
6246 | ||
6247 | em = alloc_extent_map(); | |
6248 | if (!em) | |
6249 | return -ENOMEM; | |
6250 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
6251 | if (!map) { | |
6252 | free_extent_map(em); | |
6253 | return -ENOMEM; | |
6254 | } | |
6255 | ||
6256 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); | |
6257 | em->map_lookup = map; | |
6258 | em->start = logical; | |
6259 | em->len = length; | |
6260 | em->orig_start = 0; | |
6261 | em->block_start = 0; | |
6262 | em->block_len = em->len; | |
6263 | ||
6264 | map->num_stripes = num_stripes; | |
6265 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6266 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6267 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6268 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6269 | map->type = btrfs_chunk_type(leaf, chunk); | |
6270 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); | |
6271 | for (i = 0; i < num_stripes; i++) { | |
6272 | map->stripes[i].physical = | |
6273 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6274 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
6275 | read_extent_buffer(leaf, uuid, (unsigned long) | |
6276 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6277 | BTRFS_UUID_SIZE); | |
6278 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, | |
6279 | uuid, NULL); | |
6280 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { | |
6281 | free_extent_map(em); | |
6282 | return -EIO; | |
6283 | } | |
6284 | if (!map->stripes[i].dev) { | |
6285 | map->stripes[i].dev = | |
6286 | add_missing_dev(root, root->fs_info->fs_devices, | |
6287 | devid, uuid); | |
6288 | if (!map->stripes[i].dev) { | |
6289 | free_extent_map(em); | |
6290 | return -EIO; | |
6291 | } | |
6292 | btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing", | |
6293 | devid, uuid); | |
6294 | } | |
6295 | map->stripes[i].dev->in_fs_metadata = 1; | |
6296 | } | |
6297 | ||
6298 | write_lock(&map_tree->map_tree.lock); | |
6299 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); | |
6300 | write_unlock(&map_tree->map_tree.lock); | |
6301 | BUG_ON(ret); /* Tree corruption */ | |
6302 | free_extent_map(em); | |
6303 | ||
6304 | return 0; | |
6305 | } | |
6306 | ||
6307 | static void fill_device_from_item(struct extent_buffer *leaf, | |
6308 | struct btrfs_dev_item *dev_item, | |
6309 | struct btrfs_device *device) | |
6310 | { | |
6311 | unsigned long ptr; | |
6312 | ||
6313 | device->devid = btrfs_device_id(leaf, dev_item); | |
6314 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); | |
6315 | device->total_bytes = device->disk_total_bytes; | |
6316 | device->commit_total_bytes = device->disk_total_bytes; | |
6317 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); | |
6318 | device->commit_bytes_used = device->bytes_used; | |
6319 | device->type = btrfs_device_type(leaf, dev_item); | |
6320 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6321 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6322 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
6323 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); | |
6324 | device->is_tgtdev_for_dev_replace = 0; | |
6325 | ||
6326 | ptr = btrfs_device_uuid(dev_item); | |
6327 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); | |
6328 | } | |
6329 | ||
6330 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, | |
6331 | u8 *fsid) | |
6332 | { | |
6333 | struct btrfs_fs_devices *fs_devices; | |
6334 | int ret; | |
6335 | ||
6336 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
6337 | ||
6338 | fs_devices = root->fs_info->fs_devices->seed; | |
6339 | while (fs_devices) { | |
6340 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) | |
6341 | return fs_devices; | |
6342 | ||
6343 | fs_devices = fs_devices->seed; | |
6344 | } | |
6345 | ||
6346 | fs_devices = find_fsid(fsid); | |
6347 | if (!fs_devices) { | |
6348 | if (!btrfs_test_opt(root, DEGRADED)) | |
6349 | return ERR_PTR(-ENOENT); | |
6350 | ||
6351 | fs_devices = alloc_fs_devices(fsid); | |
6352 | if (IS_ERR(fs_devices)) | |
6353 | return fs_devices; | |
6354 | ||
6355 | fs_devices->seeding = 1; | |
6356 | fs_devices->opened = 1; | |
6357 | return fs_devices; | |
6358 | } | |
6359 | ||
6360 | fs_devices = clone_fs_devices(fs_devices); | |
6361 | if (IS_ERR(fs_devices)) | |
6362 | return fs_devices; | |
6363 | ||
6364 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, | |
6365 | root->fs_info->bdev_holder); | |
6366 | if (ret) { | |
6367 | free_fs_devices(fs_devices); | |
6368 | fs_devices = ERR_PTR(ret); | |
6369 | goto out; | |
6370 | } | |
6371 | ||
6372 | if (!fs_devices->seeding) { | |
6373 | __btrfs_close_devices(fs_devices); | |
6374 | free_fs_devices(fs_devices); | |
6375 | fs_devices = ERR_PTR(-EINVAL); | |
6376 | goto out; | |
6377 | } | |
6378 | ||
6379 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6380 | root->fs_info->fs_devices->seed = fs_devices; | |
6381 | out: | |
6382 | return fs_devices; | |
6383 | } | |
6384 | ||
6385 | static int read_one_dev(struct btrfs_root *root, | |
6386 | struct extent_buffer *leaf, | |
6387 | struct btrfs_dev_item *dev_item) | |
6388 | { | |
6389 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6390 | struct btrfs_device *device; | |
6391 | u64 devid; | |
6392 | int ret; | |
6393 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
6394 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
6395 | ||
6396 | devid = btrfs_device_id(leaf, dev_item); | |
6397 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), | |
6398 | BTRFS_UUID_SIZE); | |
6399 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), | |
6400 | BTRFS_UUID_SIZE); | |
6401 | ||
6402 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
6403 | fs_devices = open_seed_devices(root, fs_uuid); | |
6404 | if (IS_ERR(fs_devices)) | |
6405 | return PTR_ERR(fs_devices); | |
6406 | } | |
6407 | ||
6408 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); | |
6409 | if (!device) { | |
6410 | if (!btrfs_test_opt(root, DEGRADED)) | |
6411 | return -EIO; | |
6412 | ||
6413 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); | |
6414 | if (!device) | |
6415 | return -ENOMEM; | |
6416 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", | |
6417 | devid, dev_uuid); | |
6418 | } else { | |
6419 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6420 | return -EIO; | |
6421 | ||
6422 | if(!device->bdev && !device->missing) { | |
6423 | /* | |
6424 | * this happens when a device that was properly setup | |
6425 | * in the device info lists suddenly goes bad. | |
6426 | * device->bdev is NULL, and so we have to set | |
6427 | * device->missing to one here | |
6428 | */ | |
6429 | device->fs_devices->missing_devices++; | |
6430 | device->missing = 1; | |
6431 | } | |
6432 | ||
6433 | /* Move the device to its own fs_devices */ | |
6434 | if (device->fs_devices != fs_devices) { | |
6435 | ASSERT(device->missing); | |
6436 | ||
6437 | list_move(&device->dev_list, &fs_devices->devices); | |
6438 | device->fs_devices->num_devices--; | |
6439 | fs_devices->num_devices++; | |
6440 | ||
6441 | device->fs_devices->missing_devices--; | |
6442 | fs_devices->missing_devices++; | |
6443 | ||
6444 | device->fs_devices = fs_devices; | |
6445 | } | |
6446 | } | |
6447 | ||
6448 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6449 | BUG_ON(device->writeable); | |
6450 | if (device->generation != | |
6451 | btrfs_device_generation(leaf, dev_item)) | |
6452 | return -EINVAL; | |
6453 | } | |
6454 | ||
6455 | fill_device_from_item(leaf, dev_item, device); | |
6456 | device->in_fs_metadata = 1; | |
6457 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { | |
6458 | device->fs_devices->total_rw_bytes += device->total_bytes; | |
6459 | spin_lock(&root->fs_info->free_chunk_lock); | |
6460 | root->fs_info->free_chunk_space += device->total_bytes - | |
6461 | device->bytes_used; | |
6462 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6463 | } | |
6464 | ret = 0; | |
6465 | return ret; | |
6466 | } | |
6467 | ||
6468 | int btrfs_read_sys_array(struct btrfs_root *root) | |
6469 | { | |
6470 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; | |
6471 | struct extent_buffer *sb; | |
6472 | struct btrfs_disk_key *disk_key; | |
6473 | struct btrfs_chunk *chunk; | |
6474 | u8 *array_ptr; | |
6475 | unsigned long sb_array_offset; | |
6476 | int ret = 0; | |
6477 | u32 num_stripes; | |
6478 | u32 array_size; | |
6479 | u32 len = 0; | |
6480 | u32 cur_offset; | |
6481 | struct btrfs_key key; | |
6482 | ||
6483 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); | |
6484 | /* | |
6485 | * This will create extent buffer of nodesize, superblock size is | |
6486 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6487 | * overallocate but we can keep it as-is, only the first page is used. | |
6488 | */ | |
6489 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
6490 | if (!sb) | |
6491 | return -ENOMEM; | |
6492 | set_extent_buffer_uptodate(sb); | |
6493 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); | |
6494 | /* | |
6495 | * The sb extent buffer is artifical and just used to read the system array. | |
6496 | * set_extent_buffer_uptodate() call does not properly mark all it's | |
6497 | * pages up-to-date when the page is larger: extent does not cover the | |
6498 | * whole page and consequently check_page_uptodate does not find all | |
6499 | * the page's extents up-to-date (the hole beyond sb), | |
6500 | * write_extent_buffer then triggers a WARN_ON. | |
6501 | * | |
6502 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6503 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6504 | * to silence the warning eg. on PowerPC 64. | |
6505 | */ | |
6506 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
6507 | SetPageUptodate(sb->pages[0]); | |
6508 | ||
6509 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); | |
6510 | array_size = btrfs_super_sys_array_size(super_copy); | |
6511 | ||
6512 | array_ptr = super_copy->sys_chunk_array; | |
6513 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6514 | cur_offset = 0; | |
6515 | ||
6516 | while (cur_offset < array_size) { | |
6517 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
6518 | len = sizeof(*disk_key); | |
6519 | if (cur_offset + len > array_size) | |
6520 | goto out_short_read; | |
6521 | ||
6522 | btrfs_disk_key_to_cpu(&key, disk_key); | |
6523 | ||
6524 | array_ptr += len; | |
6525 | sb_array_offset += len; | |
6526 | cur_offset += len; | |
6527 | ||
6528 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
6529 | chunk = (struct btrfs_chunk *)sb_array_offset; | |
6530 | /* | |
6531 | * At least one btrfs_chunk with one stripe must be | |
6532 | * present, exact stripe count check comes afterwards | |
6533 | */ | |
6534 | len = btrfs_chunk_item_size(1); | |
6535 | if (cur_offset + len > array_size) | |
6536 | goto out_short_read; | |
6537 | ||
6538 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
6539 | if (!num_stripes) { | |
6540 | printk(KERN_ERR | |
6541 | "BTRFS: invalid number of stripes %u in sys_array at offset %u\n", | |
6542 | num_stripes, cur_offset); | |
6543 | ret = -EIO; | |
6544 | break; | |
6545 | } | |
6546 | ||
6547 | len = btrfs_chunk_item_size(num_stripes); | |
6548 | if (cur_offset + len > array_size) | |
6549 | goto out_short_read; | |
6550 | ||
6551 | ret = read_one_chunk(root, &key, sb, chunk); | |
6552 | if (ret) | |
6553 | break; | |
6554 | } else { | |
6555 | printk(KERN_ERR | |
6556 | "BTRFS: unexpected item type %u in sys_array at offset %u\n", | |
6557 | (u32)key.type, cur_offset); | |
6558 | ret = -EIO; | |
6559 | break; | |
6560 | } | |
6561 | array_ptr += len; | |
6562 | sb_array_offset += len; | |
6563 | cur_offset += len; | |
6564 | } | |
6565 | free_extent_buffer(sb); | |
6566 | return ret; | |
6567 | ||
6568 | out_short_read: | |
6569 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6570 | len, cur_offset); | |
6571 | free_extent_buffer(sb); | |
6572 | return -EIO; | |
6573 | } | |
6574 | ||
6575 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6576 | { | |
6577 | struct btrfs_path *path; | |
6578 | struct extent_buffer *leaf; | |
6579 | struct btrfs_key key; | |
6580 | struct btrfs_key found_key; | |
6581 | int ret; | |
6582 | int slot; | |
6583 | ||
6584 | root = root->fs_info->chunk_root; | |
6585 | ||
6586 | path = btrfs_alloc_path(); | |
6587 | if (!path) | |
6588 | return -ENOMEM; | |
6589 | ||
6590 | mutex_lock(&uuid_mutex); | |
6591 | lock_chunks(root); | |
6592 | ||
6593 | /* | |
6594 | * Read all device items, and then all the chunk items. All | |
6595 | * device items are found before any chunk item (their object id | |
6596 | * is smaller than the lowest possible object id for a chunk | |
6597 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
6598 | */ | |
6599 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6600 | key.offset = 0; | |
6601 | key.type = 0; | |
6602 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
6603 | if (ret < 0) | |
6604 | goto error; | |
6605 | while (1) { | |
6606 | leaf = path->nodes[0]; | |
6607 | slot = path->slots[0]; | |
6608 | if (slot >= btrfs_header_nritems(leaf)) { | |
6609 | ret = btrfs_next_leaf(root, path); | |
6610 | if (ret == 0) | |
6611 | continue; | |
6612 | if (ret < 0) | |
6613 | goto error; | |
6614 | break; | |
6615 | } | |
6616 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
6617 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { | |
6618 | struct btrfs_dev_item *dev_item; | |
6619 | dev_item = btrfs_item_ptr(leaf, slot, | |
6620 | struct btrfs_dev_item); | |
6621 | ret = read_one_dev(root, leaf, dev_item); | |
6622 | if (ret) | |
6623 | goto error; | |
6624 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { | |
6625 | struct btrfs_chunk *chunk; | |
6626 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6627 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
6628 | if (ret) | |
6629 | goto error; | |
6630 | } | |
6631 | path->slots[0]++; | |
6632 | } | |
6633 | ret = 0; | |
6634 | error: | |
6635 | unlock_chunks(root); | |
6636 | mutex_unlock(&uuid_mutex); | |
6637 | ||
6638 | btrfs_free_path(path); | |
6639 | return ret; | |
6640 | } | |
6641 | ||
6642 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) | |
6643 | { | |
6644 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6645 | struct btrfs_device *device; | |
6646 | ||
6647 | while (fs_devices) { | |
6648 | mutex_lock(&fs_devices->device_list_mutex); | |
6649 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6650 | device->dev_root = fs_info->dev_root; | |
6651 | mutex_unlock(&fs_devices->device_list_mutex); | |
6652 | ||
6653 | fs_devices = fs_devices->seed; | |
6654 | } | |
6655 | } | |
6656 | ||
6657 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) | |
6658 | { | |
6659 | int i; | |
6660 | ||
6661 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6662 | btrfs_dev_stat_reset(dev, i); | |
6663 | } | |
6664 | ||
6665 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6666 | { | |
6667 | struct btrfs_key key; | |
6668 | struct btrfs_key found_key; | |
6669 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6670 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6671 | struct extent_buffer *eb; | |
6672 | int slot; | |
6673 | int ret = 0; | |
6674 | struct btrfs_device *device; | |
6675 | struct btrfs_path *path = NULL; | |
6676 | int i; | |
6677 | ||
6678 | path = btrfs_alloc_path(); | |
6679 | if (!path) { | |
6680 | ret = -ENOMEM; | |
6681 | goto out; | |
6682 | } | |
6683 | ||
6684 | mutex_lock(&fs_devices->device_list_mutex); | |
6685 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6686 | int item_size; | |
6687 | struct btrfs_dev_stats_item *ptr; | |
6688 | ||
6689 | key.objectid = BTRFS_DEV_STATS_OBJECTID; | |
6690 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
6691 | key.offset = device->devid; | |
6692 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6693 | if (ret) { | |
6694 | __btrfs_reset_dev_stats(device); | |
6695 | device->dev_stats_valid = 1; | |
6696 | btrfs_release_path(path); | |
6697 | continue; | |
6698 | } | |
6699 | slot = path->slots[0]; | |
6700 | eb = path->nodes[0]; | |
6701 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6702 | item_size = btrfs_item_size_nr(eb, slot); | |
6703 | ||
6704 | ptr = btrfs_item_ptr(eb, slot, | |
6705 | struct btrfs_dev_stats_item); | |
6706 | ||
6707 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6708 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6709 | btrfs_dev_stat_set(device, i, | |
6710 | btrfs_dev_stats_value(eb, ptr, i)); | |
6711 | else | |
6712 | btrfs_dev_stat_reset(device, i); | |
6713 | } | |
6714 | ||
6715 | device->dev_stats_valid = 1; | |
6716 | btrfs_dev_stat_print_on_load(device); | |
6717 | btrfs_release_path(path); | |
6718 | } | |
6719 | mutex_unlock(&fs_devices->device_list_mutex); | |
6720 | ||
6721 | out: | |
6722 | btrfs_free_path(path); | |
6723 | return ret < 0 ? ret : 0; | |
6724 | } | |
6725 | ||
6726 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6727 | struct btrfs_root *dev_root, | |
6728 | struct btrfs_device *device) | |
6729 | { | |
6730 | struct btrfs_path *path; | |
6731 | struct btrfs_key key; | |
6732 | struct extent_buffer *eb; | |
6733 | struct btrfs_dev_stats_item *ptr; | |
6734 | int ret; | |
6735 | int i; | |
6736 | ||
6737 | key.objectid = BTRFS_DEV_STATS_OBJECTID; | |
6738 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
6739 | key.offset = device->devid; | |
6740 | ||
6741 | path = btrfs_alloc_path(); | |
6742 | BUG_ON(!path); | |
6743 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6744 | if (ret < 0) { | |
6745 | btrfs_warn_in_rcu(dev_root->fs_info, | |
6746 | "error %d while searching for dev_stats item for device %s", | |
6747 | ret, rcu_str_deref(device->name)); | |
6748 | goto out; | |
6749 | } | |
6750 | ||
6751 | if (ret == 0 && | |
6752 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6753 | /* need to delete old one and insert a new one */ | |
6754 | ret = btrfs_del_item(trans, dev_root, path); | |
6755 | if (ret != 0) { | |
6756 | btrfs_warn_in_rcu(dev_root->fs_info, | |
6757 | "delete too small dev_stats item for device %s failed %d", | |
6758 | rcu_str_deref(device->name), ret); | |
6759 | goto out; | |
6760 | } | |
6761 | ret = 1; | |
6762 | } | |
6763 | ||
6764 | if (ret == 1) { | |
6765 | /* need to insert a new item */ | |
6766 | btrfs_release_path(path); | |
6767 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6768 | &key, sizeof(*ptr)); | |
6769 | if (ret < 0) { | |
6770 | btrfs_warn_in_rcu(dev_root->fs_info, | |
6771 | "insert dev_stats item for device %s failed %d", | |
6772 | rcu_str_deref(device->name), ret); | |
6773 | goto out; | |
6774 | } | |
6775 | } | |
6776 | ||
6777 | eb = path->nodes[0]; | |
6778 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6779 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6780 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6781 | btrfs_dev_stat_read(device, i)); | |
6782 | btrfs_mark_buffer_dirty(eb); | |
6783 | ||
6784 | out: | |
6785 | btrfs_free_path(path); | |
6786 | return ret; | |
6787 | } | |
6788 | ||
6789 | /* | |
6790 | * called from commit_transaction. Writes all changed device stats to disk. | |
6791 | */ | |
6792 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6793 | struct btrfs_fs_info *fs_info) | |
6794 | { | |
6795 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6796 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6797 | struct btrfs_device *device; | |
6798 | int stats_cnt; | |
6799 | int ret = 0; | |
6800 | ||
6801 | mutex_lock(&fs_devices->device_list_mutex); | |
6802 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6803 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) | |
6804 | continue; | |
6805 | ||
6806 | stats_cnt = atomic_read(&device->dev_stats_ccnt); | |
6807 | ret = update_dev_stat_item(trans, dev_root, device); | |
6808 | if (!ret) | |
6809 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); | |
6810 | } | |
6811 | mutex_unlock(&fs_devices->device_list_mutex); | |
6812 | ||
6813 | return ret; | |
6814 | } | |
6815 | ||
6816 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) | |
6817 | { | |
6818 | btrfs_dev_stat_inc(dev, index); | |
6819 | btrfs_dev_stat_print_on_error(dev); | |
6820 | } | |
6821 | ||
6822 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) | |
6823 | { | |
6824 | if (!dev->dev_stats_valid) | |
6825 | return; | |
6826 | btrfs_err_rl_in_rcu(dev->dev_root->fs_info, | |
6827 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
6828 | rcu_str_deref(dev->name), | |
6829 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), | |
6830 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6831 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6832 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6833 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6834 | } | |
6835 | ||
6836 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) | |
6837 | { | |
6838 | int i; | |
6839 | ||
6840 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6841 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6842 | break; | |
6843 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6844 | return; /* all values == 0, suppress message */ | |
6845 | ||
6846 | btrfs_info_in_rcu(dev->dev_root->fs_info, | |
6847 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", | |
6848 | rcu_str_deref(dev->name), | |
6849 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), | |
6850 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6851 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6852 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6853 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6854 | } | |
6855 | ||
6856 | int btrfs_get_dev_stats(struct btrfs_root *root, | |
6857 | struct btrfs_ioctl_get_dev_stats *stats) | |
6858 | { | |
6859 | struct btrfs_device *dev; | |
6860 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6861 | int i; | |
6862 | ||
6863 | mutex_lock(&fs_devices->device_list_mutex); | |
6864 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); | |
6865 | mutex_unlock(&fs_devices->device_list_mutex); | |
6866 | ||
6867 | if (!dev) { | |
6868 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); | |
6869 | return -ENODEV; | |
6870 | } else if (!dev->dev_stats_valid) { | |
6871 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); | |
6872 | return -ENODEV; | |
6873 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { | |
6874 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6875 | if (stats->nr_items > i) | |
6876 | stats->values[i] = | |
6877 | btrfs_dev_stat_read_and_reset(dev, i); | |
6878 | else | |
6879 | btrfs_dev_stat_reset(dev, i); | |
6880 | } | |
6881 | } else { | |
6882 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6883 | if (stats->nr_items > i) | |
6884 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6885 | } | |
6886 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6887 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6888 | return 0; | |
6889 | } | |
6890 | ||
6891 | void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path) | |
6892 | { | |
6893 | struct buffer_head *bh; | |
6894 | struct btrfs_super_block *disk_super; | |
6895 | int copy_num; | |
6896 | ||
6897 | if (!bdev) | |
6898 | return; | |
6899 | ||
6900 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; | |
6901 | copy_num++) { | |
6902 | ||
6903 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) | |
6904 | continue; | |
6905 | ||
6906 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6907 | ||
6908 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6909 | set_buffer_dirty(bh); | |
6910 | sync_dirty_buffer(bh); | |
6911 | brelse(bh); | |
6912 | } | |
6913 | ||
6914 | /* Notify udev that device has changed */ | |
6915 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
6916 | ||
6917 | /* Update ctime/mtime for device path for libblkid */ | |
6918 | update_dev_time(device_path); | |
6919 | } | |
6920 | ||
6921 | /* | |
6922 | * Update the size of all devices, which is used for writing out the | |
6923 | * super blocks. | |
6924 | */ | |
6925 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6926 | { | |
6927 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6928 | struct btrfs_device *curr, *next; | |
6929 | ||
6930 | if (list_empty(&fs_devices->resized_devices)) | |
6931 | return; | |
6932 | ||
6933 | mutex_lock(&fs_devices->device_list_mutex); | |
6934 | lock_chunks(fs_info->dev_root); | |
6935 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6936 | resized_list) { | |
6937 | list_del_init(&curr->resized_list); | |
6938 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6939 | } | |
6940 | unlock_chunks(fs_info->dev_root); | |
6941 | mutex_unlock(&fs_devices->device_list_mutex); | |
6942 | } | |
6943 | ||
6944 | /* Must be invoked during the transaction commit */ | |
6945 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6946 | struct btrfs_transaction *transaction) | |
6947 | { | |
6948 | struct extent_map *em; | |
6949 | struct map_lookup *map; | |
6950 | struct btrfs_device *dev; | |
6951 | int i; | |
6952 | ||
6953 | if (list_empty(&transaction->pending_chunks)) | |
6954 | return; | |
6955 | ||
6956 | /* In order to kick the device replace finish process */ | |
6957 | lock_chunks(root); | |
6958 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6959 | map = em->map_lookup; | |
6960 | ||
6961 | for (i = 0; i < map->num_stripes; i++) { | |
6962 | dev = map->stripes[i].dev; | |
6963 | dev->commit_bytes_used = dev->bytes_used; | |
6964 | } | |
6965 | } | |
6966 | unlock_chunks(root); | |
6967 | } | |
6968 | ||
6969 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6970 | { | |
6971 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6972 | while (fs_devices) { | |
6973 | fs_devices->fs_info = fs_info; | |
6974 | fs_devices = fs_devices->seed; | |
6975 | } | |
6976 | } | |
6977 | ||
6978 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6979 | { | |
6980 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6981 | while (fs_devices) { | |
6982 | fs_devices->fs_info = NULL; | |
6983 | fs_devices = fs_devices->seed; | |
6984 | } | |
6985 | } | |
6986 | ||
6987 | static void btrfs_close_one_device(struct btrfs_device *device) | |
6988 | { | |
6989 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
6990 | struct btrfs_device *new_device; | |
6991 | struct rcu_string *name; | |
6992 | ||
6993 | if (device->bdev) | |
6994 | fs_devices->open_devices--; | |
6995 | ||
6996 | if (device->writeable && | |
6997 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
6998 | list_del_init(&device->dev_alloc_list); | |
6999 | fs_devices->rw_devices--; | |
7000 | } | |
7001 | ||
7002 | if (device->missing) | |
7003 | fs_devices->missing_devices--; | |
7004 | ||
7005 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
7006 | device->uuid); | |
7007 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
7008 | ||
7009 | /* Safe because we are under uuid_mutex */ | |
7010 | if (device->name) { | |
7011 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
7012 | BUG_ON(!name); /* -ENOMEM */ | |
7013 | rcu_assign_pointer(new_device->name, name); | |
7014 | } | |
7015 | ||
7016 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
7017 | new_device->fs_devices = device->fs_devices; | |
7018 | ||
7019 | call_rcu(&device->rcu, free_device); | |
7020 | } |