]>
Commit | Line | Data |
---|---|---|
0b86a832 CM |
1 | /* |
2 | * Copyright (C) 2007 Oracle. All rights reserved. | |
3 | * | |
4 | * This program is free software; you can redistribute it and/or | |
5 | * modify it under the terms of the GNU General Public | |
6 | * License v2 as published by the Free Software Foundation. | |
7 | * | |
8 | * This program is distributed in the hope that it will be useful, | |
9 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
10 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU | |
11 | * General Public License for more details. | |
12 | * | |
13 | * You should have received a copy of the GNU General Public | |
14 | * License along with this program; if not, write to the | |
15 | * Free Software Foundation, Inc., 59 Temple Place - Suite 330, | |
16 | * Boston, MA 021110-1307, USA. | |
17 | */ | |
18 | #include <linux/sched.h> | |
19 | #include <linux/bio.h> | |
5a0e3ad6 | 20 | #include <linux/slab.h> |
8a4b83cc | 21 | #include <linux/buffer_head.h> |
f2d8d74d | 22 | #include <linux/blkdev.h> |
788f20eb | 23 | #include <linux/random.h> |
b765ead5 | 24 | #include <linux/iocontext.h> |
6f88a440 | 25 | #include <linux/capability.h> |
442a4f63 | 26 | #include <linux/ratelimit.h> |
59641015 | 27 | #include <linux/kthread.h> |
53b381b3 | 28 | #include <linux/raid/pq.h> |
803b2f54 | 29 | #include <linux/semaphore.h> |
53b381b3 | 30 | #include <asm/div64.h> |
0b86a832 CM |
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" | |
53b381b3 | 37 | #include "raid56.h" |
8b712842 | 38 | #include "async-thread.h" |
21adbd5c | 39 | #include "check-integrity.h" |
606686ee | 40 | #include "rcu-string.h" |
3fed40cc | 41 | #include "math.h" |
8dabb742 | 42 | #include "dev-replace.h" |
99994cde | 43 | #include "sysfs.h" |
0b86a832 | 44 | |
2b82032c YZ |
45 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
46 | struct btrfs_root *root, | |
47 | struct btrfs_device *device); | |
48 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 49 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 50 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 51 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 52 | |
67a2c45e | 53 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 54 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
55 | struct list_head *btrfs_get_fs_uuids(void) |
56 | { | |
57 | return &fs_uuids; | |
58 | } | |
8a4b83cc | 59 | |
2208a378 ID |
60 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
61 | { | |
62 | struct btrfs_fs_devices *fs_devs; | |
63 | ||
64 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_NOFS); | |
65 | if (!fs_devs) | |
66 | return ERR_PTR(-ENOMEM); | |
67 | ||
68 | mutex_init(&fs_devs->device_list_mutex); | |
69 | ||
70 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 71 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
72 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
73 | INIT_LIST_HEAD(&fs_devs->list); | |
74 | ||
75 | return fs_devs; | |
76 | } | |
77 | ||
78 | /** | |
79 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
80 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
81 | * generated. | |
82 | * | |
83 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
84 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
85 | * can be destroyed with kfree() right away. | |
86 | */ | |
87 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
88 | { | |
89 | struct btrfs_fs_devices *fs_devs; | |
90 | ||
91 | fs_devs = __alloc_fs_devices(); | |
92 | if (IS_ERR(fs_devs)) | |
93 | return fs_devs; | |
94 | ||
95 | if (fsid) | |
96 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
97 | else | |
98 | generate_random_uuid(fs_devs->fsid); | |
99 | ||
100 | return fs_devs; | |
101 | } | |
102 | ||
e4404d6e YZ |
103 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
104 | { | |
105 | struct btrfs_device *device; | |
106 | WARN_ON(fs_devices->opened); | |
107 | while (!list_empty(&fs_devices->devices)) { | |
108 | device = list_entry(fs_devices->devices.next, | |
109 | struct btrfs_device, dev_list); | |
110 | list_del(&device->dev_list); | |
606686ee | 111 | rcu_string_free(device->name); |
e4404d6e YZ |
112 | kfree(device); |
113 | } | |
114 | kfree(fs_devices); | |
115 | } | |
116 | ||
b8b8ff59 LC |
117 | static void btrfs_kobject_uevent(struct block_device *bdev, |
118 | enum kobject_action action) | |
119 | { | |
120 | int ret; | |
121 | ||
122 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
123 | if (ret) | |
efe120a0 | 124 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
125 | action, |
126 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
127 | &disk_to_dev(bdev->bd_disk)->kobj); | |
128 | } | |
129 | ||
143bede5 | 130 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
131 | { |
132 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 133 | |
2b82032c YZ |
134 | while (!list_empty(&fs_uuids)) { |
135 | fs_devices = list_entry(fs_uuids.next, | |
136 | struct btrfs_fs_devices, list); | |
137 | list_del(&fs_devices->list); | |
e4404d6e | 138 | free_fs_devices(fs_devices); |
8a4b83cc | 139 | } |
8a4b83cc CM |
140 | } |
141 | ||
12bd2fc0 ID |
142 | static struct btrfs_device *__alloc_device(void) |
143 | { | |
144 | struct btrfs_device *dev; | |
145 | ||
146 | dev = kzalloc(sizeof(*dev), GFP_NOFS); | |
147 | if (!dev) | |
148 | return ERR_PTR(-ENOMEM); | |
149 | ||
150 | INIT_LIST_HEAD(&dev->dev_list); | |
151 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 152 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
153 | |
154 | spin_lock_init(&dev->io_lock); | |
155 | ||
156 | spin_lock_init(&dev->reada_lock); | |
157 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 158 | atomic_set(&dev->dev_stats_ccnt, 0); |
12bd2fc0 ID |
159 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_WAIT); |
160 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_WAIT); | |
161 | ||
162 | return dev; | |
163 | } | |
164 | ||
a1b32a59 CM |
165 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
166 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
167 | { |
168 | struct btrfs_device *dev; | |
8a4b83cc | 169 | |
c6e30871 | 170 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 171 | if (dev->devid == devid && |
8f18cf13 | 172 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 173 | return dev; |
a443755f | 174 | } |
8a4b83cc CM |
175 | } |
176 | return NULL; | |
177 | } | |
178 | ||
a1b32a59 | 179 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 180 | { |
8a4b83cc CM |
181 | struct btrfs_fs_devices *fs_devices; |
182 | ||
c6e30871 | 183 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
184 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
185 | return fs_devices; | |
186 | } | |
187 | return NULL; | |
188 | } | |
189 | ||
beaf8ab3 SB |
190 | static int |
191 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
192 | int flush, struct block_device **bdev, | |
193 | struct buffer_head **bh) | |
194 | { | |
195 | int ret; | |
196 | ||
197 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
198 | ||
199 | if (IS_ERR(*bdev)) { | |
200 | ret = PTR_ERR(*bdev); | |
efe120a0 | 201 | printk(KERN_INFO "BTRFS: open %s failed\n", device_path); |
beaf8ab3 SB |
202 | goto error; |
203 | } | |
204 | ||
205 | if (flush) | |
206 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
207 | ret = set_blocksize(*bdev, 4096); | |
208 | if (ret) { | |
209 | blkdev_put(*bdev, flags); | |
210 | goto error; | |
211 | } | |
212 | invalidate_bdev(*bdev); | |
213 | *bh = btrfs_read_dev_super(*bdev); | |
214 | if (!*bh) { | |
215 | ret = -EINVAL; | |
216 | blkdev_put(*bdev, flags); | |
217 | goto error; | |
218 | } | |
219 | ||
220 | return 0; | |
221 | ||
222 | error: | |
223 | *bdev = NULL; | |
224 | *bh = NULL; | |
225 | return ret; | |
226 | } | |
227 | ||
ffbd517d CM |
228 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
229 | struct bio *head, struct bio *tail) | |
230 | { | |
231 | ||
232 | struct bio *old_head; | |
233 | ||
234 | old_head = pending_bios->head; | |
235 | pending_bios->head = head; | |
236 | if (pending_bios->tail) | |
237 | tail->bi_next = old_head; | |
238 | else | |
239 | pending_bios->tail = tail; | |
240 | } | |
241 | ||
8b712842 CM |
242 | /* |
243 | * we try to collect pending bios for a device so we don't get a large | |
244 | * number of procs sending bios down to the same device. This greatly | |
245 | * improves the schedulers ability to collect and merge the bios. | |
246 | * | |
247 | * But, it also turns into a long list of bios to process and that is sure | |
248 | * to eventually make the worker thread block. The solution here is to | |
249 | * make some progress and then put this work struct back at the end of | |
250 | * the list if the block device is congested. This way, multiple devices | |
251 | * can make progress from a single worker thread. | |
252 | */ | |
143bede5 | 253 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 CM |
254 | { |
255 | struct bio *pending; | |
256 | struct backing_dev_info *bdi; | |
b64a2851 | 257 | struct btrfs_fs_info *fs_info; |
ffbd517d | 258 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
259 | struct bio *tail; |
260 | struct bio *cur; | |
261 | int again = 0; | |
ffbd517d | 262 | unsigned long num_run; |
d644d8a1 | 263 | unsigned long batch_run = 0; |
b64a2851 | 264 | unsigned long limit; |
b765ead5 | 265 | unsigned long last_waited = 0; |
d84275c9 | 266 | int force_reg = 0; |
0e588859 | 267 | int sync_pending = 0; |
211588ad CM |
268 | struct blk_plug plug; |
269 | ||
270 | /* | |
271 | * this function runs all the bios we've collected for | |
272 | * a particular device. We don't want to wander off to | |
273 | * another device without first sending all of these down. | |
274 | * So, setup a plug here and finish it off before we return | |
275 | */ | |
276 | blk_start_plug(&plug); | |
8b712842 | 277 | |
bedf762b | 278 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
279 | fs_info = device->dev_root->fs_info; |
280 | limit = btrfs_async_submit_limit(fs_info); | |
281 | limit = limit * 2 / 3; | |
282 | ||
8b712842 CM |
283 | loop: |
284 | spin_lock(&device->io_lock); | |
285 | ||
a6837051 | 286 | loop_lock: |
d84275c9 | 287 | num_run = 0; |
ffbd517d | 288 | |
8b712842 CM |
289 | /* take all the bios off the list at once and process them |
290 | * later on (without the lock held). But, remember the | |
291 | * tail and other pointers so the bios can be properly reinserted | |
292 | * into the list if we hit congestion | |
293 | */ | |
d84275c9 | 294 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 295 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
296 | force_reg = 1; |
297 | } else { | |
ffbd517d | 298 | pending_bios = &device->pending_bios; |
d84275c9 CM |
299 | force_reg = 0; |
300 | } | |
ffbd517d CM |
301 | |
302 | pending = pending_bios->head; | |
303 | tail = pending_bios->tail; | |
8b712842 | 304 | WARN_ON(pending && !tail); |
8b712842 CM |
305 | |
306 | /* | |
307 | * if pending was null this time around, no bios need processing | |
308 | * at all and we can stop. Otherwise it'll loop back up again | |
309 | * and do an additional check so no bios are missed. | |
310 | * | |
311 | * device->running_pending is used to synchronize with the | |
312 | * schedule_bio code. | |
313 | */ | |
ffbd517d CM |
314 | if (device->pending_sync_bios.head == NULL && |
315 | device->pending_bios.head == NULL) { | |
8b712842 CM |
316 | again = 0; |
317 | device->running_pending = 0; | |
ffbd517d CM |
318 | } else { |
319 | again = 1; | |
320 | device->running_pending = 1; | |
8b712842 | 321 | } |
ffbd517d CM |
322 | |
323 | pending_bios->head = NULL; | |
324 | pending_bios->tail = NULL; | |
325 | ||
8b712842 CM |
326 | spin_unlock(&device->io_lock); |
327 | ||
d397712b | 328 | while (pending) { |
ffbd517d CM |
329 | |
330 | rmb(); | |
d84275c9 CM |
331 | /* we want to work on both lists, but do more bios on the |
332 | * sync list than the regular list | |
333 | */ | |
334 | if ((num_run > 32 && | |
335 | pending_bios != &device->pending_sync_bios && | |
336 | device->pending_sync_bios.head) || | |
337 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
338 | device->pending_bios.head)) { | |
ffbd517d CM |
339 | spin_lock(&device->io_lock); |
340 | requeue_list(pending_bios, pending, tail); | |
341 | goto loop_lock; | |
342 | } | |
343 | ||
8b712842 CM |
344 | cur = pending; |
345 | pending = pending->bi_next; | |
346 | cur->bi_next = NULL; | |
b64a2851 | 347 | |
66657b31 | 348 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
349 | waitqueue_active(&fs_info->async_submit_wait)) |
350 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 351 | |
dac56212 | 352 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 353 | |
2ab1ba68 CM |
354 | /* |
355 | * if we're doing the sync list, record that our | |
356 | * plug has some sync requests on it | |
357 | * | |
358 | * If we're doing the regular list and there are | |
359 | * sync requests sitting around, unplug before | |
360 | * we add more | |
361 | */ | |
362 | if (pending_bios == &device->pending_sync_bios) { | |
363 | sync_pending = 1; | |
364 | } else if (sync_pending) { | |
365 | blk_finish_plug(&plug); | |
366 | blk_start_plug(&plug); | |
367 | sync_pending = 0; | |
368 | } | |
369 | ||
21adbd5c | 370 | btrfsic_submit_bio(cur->bi_rw, cur); |
5ff7ba3a CM |
371 | num_run++; |
372 | batch_run++; | |
853d8ec4 DS |
373 | |
374 | cond_resched(); | |
8b712842 CM |
375 | |
376 | /* | |
377 | * we made progress, there is more work to do and the bdi | |
378 | * is now congested. Back off and let other work structs | |
379 | * run instead | |
380 | */ | |
57fd5a5f | 381 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 382 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 383 | struct io_context *ioc; |
8b712842 | 384 | |
b765ead5 CM |
385 | ioc = current->io_context; |
386 | ||
387 | /* | |
388 | * the main goal here is that we don't want to | |
389 | * block if we're going to be able to submit | |
390 | * more requests without blocking. | |
391 | * | |
392 | * This code does two great things, it pokes into | |
393 | * the elevator code from a filesystem _and_ | |
394 | * it makes assumptions about how batching works. | |
395 | */ | |
396 | if (ioc && ioc->nr_batch_requests > 0 && | |
397 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
398 | (last_waited == 0 || | |
399 | ioc->last_waited == last_waited)) { | |
400 | /* | |
401 | * we want to go through our batch of | |
402 | * requests and stop. So, we copy out | |
403 | * the ioc->last_waited time and test | |
404 | * against it before looping | |
405 | */ | |
406 | last_waited = ioc->last_waited; | |
853d8ec4 | 407 | cond_resched(); |
b765ead5 CM |
408 | continue; |
409 | } | |
8b712842 | 410 | spin_lock(&device->io_lock); |
ffbd517d | 411 | requeue_list(pending_bios, pending, tail); |
a6837051 | 412 | device->running_pending = 1; |
8b712842 CM |
413 | |
414 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
415 | btrfs_queue_work(fs_info->submit_workers, |
416 | &device->work); | |
8b712842 CM |
417 | goto done; |
418 | } | |
d85c8a6f CM |
419 | /* unplug every 64 requests just for good measure */ |
420 | if (batch_run % 64 == 0) { | |
421 | blk_finish_plug(&plug); | |
422 | blk_start_plug(&plug); | |
423 | sync_pending = 0; | |
424 | } | |
8b712842 | 425 | } |
ffbd517d | 426 | |
51684082 CM |
427 | cond_resched(); |
428 | if (again) | |
429 | goto loop; | |
430 | ||
431 | spin_lock(&device->io_lock); | |
432 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
433 | goto loop_lock; | |
434 | spin_unlock(&device->io_lock); | |
435 | ||
8b712842 | 436 | done: |
211588ad | 437 | blk_finish_plug(&plug); |
8b712842 CM |
438 | } |
439 | ||
b2950863 | 440 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
441 | { |
442 | struct btrfs_device *device; | |
443 | ||
444 | device = container_of(work, struct btrfs_device, work); | |
445 | run_scheduled_bios(device); | |
446 | } | |
447 | ||
4fde46f0 AJ |
448 | |
449 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
450 | { | |
451 | struct btrfs_fs_devices *fs_devs; | |
452 | struct btrfs_device *dev; | |
453 | ||
454 | if (!cur_dev->name) | |
455 | return; | |
456 | ||
457 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
458 | int del = 1; | |
459 | ||
460 | if (fs_devs->opened) | |
461 | continue; | |
462 | if (fs_devs->seeding) | |
463 | continue; | |
464 | ||
465 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
466 | ||
467 | if (dev == cur_dev) | |
468 | continue; | |
469 | if (!dev->name) | |
470 | continue; | |
471 | ||
472 | /* | |
473 | * Todo: This won't be enough. What if the same device | |
474 | * comes back (with new uuid and) with its mapper path? | |
475 | * But for now, this does help as mostly an admin will | |
476 | * either use mapper or non mapper path throughout. | |
477 | */ | |
478 | rcu_read_lock(); | |
479 | del = strcmp(rcu_str_deref(dev->name), | |
480 | rcu_str_deref(cur_dev->name)); | |
481 | rcu_read_unlock(); | |
482 | if (!del) | |
483 | break; | |
484 | } | |
485 | ||
486 | if (!del) { | |
487 | /* delete the stale device */ | |
488 | if (fs_devs->num_devices == 1) { | |
489 | btrfs_sysfs_remove_fsid(fs_devs); | |
490 | list_del(&fs_devs->list); | |
491 | free_fs_devices(fs_devs); | |
492 | } else { | |
493 | fs_devs->num_devices--; | |
494 | list_del(&dev->dev_list); | |
495 | rcu_string_free(dev->name); | |
496 | kfree(dev); | |
497 | } | |
498 | break; | |
499 | } | |
500 | } | |
501 | } | |
502 | ||
60999ca4 DS |
503 | /* |
504 | * Add new device to list of registered devices | |
505 | * | |
506 | * Returns: | |
507 | * 1 - first time device is seen | |
508 | * 0 - device already known | |
509 | * < 0 - error | |
510 | */ | |
a1b32a59 | 511 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
512 | struct btrfs_super_block *disk_super, |
513 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
514 | { | |
515 | struct btrfs_device *device; | |
516 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 517 | struct rcu_string *name; |
60999ca4 | 518 | int ret = 0; |
8a4b83cc CM |
519 | u64 found_transid = btrfs_super_generation(disk_super); |
520 | ||
521 | fs_devices = find_fsid(disk_super->fsid); | |
522 | if (!fs_devices) { | |
2208a378 ID |
523 | fs_devices = alloc_fs_devices(disk_super->fsid); |
524 | if (IS_ERR(fs_devices)) | |
525 | return PTR_ERR(fs_devices); | |
526 | ||
8a4b83cc | 527 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 528 | |
8a4b83cc CM |
529 | device = NULL; |
530 | } else { | |
a443755f CM |
531 | device = __find_device(&fs_devices->devices, devid, |
532 | disk_super->dev_item.uuid); | |
8a4b83cc | 533 | } |
443f24fe | 534 | |
8a4b83cc | 535 | if (!device) { |
2b82032c YZ |
536 | if (fs_devices->opened) |
537 | return -EBUSY; | |
538 | ||
12bd2fc0 ID |
539 | device = btrfs_alloc_device(NULL, &devid, |
540 | disk_super->dev_item.uuid); | |
541 | if (IS_ERR(device)) { | |
8a4b83cc | 542 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 543 | return PTR_ERR(device); |
8a4b83cc | 544 | } |
606686ee JB |
545 | |
546 | name = rcu_string_strdup(path, GFP_NOFS); | |
547 | if (!name) { | |
8a4b83cc CM |
548 | kfree(device); |
549 | return -ENOMEM; | |
550 | } | |
606686ee | 551 | rcu_assign_pointer(device->name, name); |
90519d66 | 552 | |
e5e9a520 | 553 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 554 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 555 | fs_devices->num_devices++; |
e5e9a520 CM |
556 | mutex_unlock(&fs_devices->device_list_mutex); |
557 | ||
60999ca4 | 558 | ret = 1; |
2b82032c | 559 | device->fs_devices = fs_devices; |
606686ee | 560 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
561 | /* |
562 | * When FS is already mounted. | |
563 | * 1. If you are here and if the device->name is NULL that | |
564 | * means this device was missing at time of FS mount. | |
565 | * 2. If you are here and if the device->name is different | |
566 | * from 'path' that means either | |
567 | * a. The same device disappeared and reappeared with | |
568 | * different name. or | |
569 | * b. The missing-disk-which-was-replaced, has | |
570 | * reappeared now. | |
571 | * | |
572 | * We must allow 1 and 2a above. But 2b would be a spurious | |
573 | * and unintentional. | |
574 | * | |
575 | * Further in case of 1 and 2a above, the disk at 'path' | |
576 | * would have missed some transaction when it was away and | |
577 | * in case of 2a the stale bdev has to be updated as well. | |
578 | * 2b must not be allowed at all time. | |
579 | */ | |
580 | ||
581 | /* | |
0f23ae74 CM |
582 | * For now, we do allow update to btrfs_fs_device through the |
583 | * btrfs dev scan cli after FS has been mounted. We're still | |
584 | * tracking a problem where systems fail mount by subvolume id | |
585 | * when we reject replacement on a mounted FS. | |
b96de000 | 586 | */ |
0f23ae74 | 587 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
588 | /* |
589 | * That is if the FS is _not_ mounted and if you | |
590 | * are here, that means there is more than one | |
591 | * disk with same uuid and devid.We keep the one | |
592 | * with larger generation number or the last-in if | |
593 | * generation are equal. | |
594 | */ | |
0f23ae74 | 595 | return -EEXIST; |
77bdae4d | 596 | } |
b96de000 | 597 | |
606686ee | 598 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
599 | if (!name) |
600 | return -ENOMEM; | |
606686ee JB |
601 | rcu_string_free(device->name); |
602 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
603 | if (device->missing) { |
604 | fs_devices->missing_devices--; | |
605 | device->missing = 0; | |
606 | } | |
8a4b83cc CM |
607 | } |
608 | ||
77bdae4d AJ |
609 | /* |
610 | * Unmount does not free the btrfs_device struct but would zero | |
611 | * generation along with most of the other members. So just update | |
612 | * it back. We need it to pick the disk with largest generation | |
613 | * (as above). | |
614 | */ | |
615 | if (!fs_devices->opened) | |
616 | device->generation = found_transid; | |
617 | ||
4fde46f0 AJ |
618 | /* |
619 | * if there is new btrfs on an already registered device, | |
620 | * then remove the stale device entry. | |
621 | */ | |
622 | btrfs_free_stale_device(device); | |
623 | ||
8a4b83cc | 624 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
625 | |
626 | return ret; | |
8a4b83cc CM |
627 | } |
628 | ||
e4404d6e YZ |
629 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
630 | { | |
631 | struct btrfs_fs_devices *fs_devices; | |
632 | struct btrfs_device *device; | |
633 | struct btrfs_device *orig_dev; | |
634 | ||
2208a378 ID |
635 | fs_devices = alloc_fs_devices(orig->fsid); |
636 | if (IS_ERR(fs_devices)) | |
637 | return fs_devices; | |
e4404d6e | 638 | |
adbbb863 | 639 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 640 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 641 | |
46224705 | 642 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 643 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
644 | struct rcu_string *name; |
645 | ||
12bd2fc0 ID |
646 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
647 | orig_dev->uuid); | |
648 | if (IS_ERR(device)) | |
e4404d6e YZ |
649 | goto error; |
650 | ||
606686ee JB |
651 | /* |
652 | * This is ok to do without rcu read locked because we hold the | |
653 | * uuid mutex so nothing we touch in here is going to disappear. | |
654 | */ | |
e755f780 AJ |
655 | if (orig_dev->name) { |
656 | name = rcu_string_strdup(orig_dev->name->str, GFP_NOFS); | |
657 | if (!name) { | |
658 | kfree(device); | |
659 | goto error; | |
660 | } | |
661 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 662 | } |
e4404d6e | 663 | |
e4404d6e YZ |
664 | list_add(&device->dev_list, &fs_devices->devices); |
665 | device->fs_devices = fs_devices; | |
666 | fs_devices->num_devices++; | |
667 | } | |
adbbb863 | 668 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
669 | return fs_devices; |
670 | error: | |
adbbb863 | 671 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
672 | free_fs_devices(fs_devices); |
673 | return ERR_PTR(-ENOMEM); | |
674 | } | |
675 | ||
9eaed21e | 676 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 677 | { |
c6e30871 | 678 | struct btrfs_device *device, *next; |
443f24fe | 679 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 680 | |
dfe25020 CM |
681 | mutex_lock(&uuid_mutex); |
682 | again: | |
46224705 | 683 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 684 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 685 | if (device->in_fs_metadata) { |
63a212ab | 686 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
687 | (!latest_dev || |
688 | device->generation > latest_dev->generation)) { | |
689 | latest_dev = device; | |
a6b0d5c8 | 690 | } |
2b82032c | 691 | continue; |
a6b0d5c8 | 692 | } |
2b82032c | 693 | |
8dabb742 SB |
694 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
695 | /* | |
696 | * In the first step, keep the device which has | |
697 | * the correct fsid and the devid that is used | |
698 | * for the dev_replace procedure. | |
699 | * In the second step, the dev_replace state is | |
700 | * read from the device tree and it is known | |
701 | * whether the procedure is really active or | |
702 | * not, which means whether this device is | |
703 | * used or whether it should be removed. | |
704 | */ | |
705 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
706 | continue; | |
707 | } | |
708 | } | |
2b82032c | 709 | if (device->bdev) { |
d4d77629 | 710 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
711 | device->bdev = NULL; |
712 | fs_devices->open_devices--; | |
713 | } | |
714 | if (device->writeable) { | |
715 | list_del_init(&device->dev_alloc_list); | |
716 | device->writeable = 0; | |
8dabb742 SB |
717 | if (!device->is_tgtdev_for_dev_replace) |
718 | fs_devices->rw_devices--; | |
2b82032c | 719 | } |
e4404d6e YZ |
720 | list_del_init(&device->dev_list); |
721 | fs_devices->num_devices--; | |
606686ee | 722 | rcu_string_free(device->name); |
e4404d6e | 723 | kfree(device); |
dfe25020 | 724 | } |
2b82032c YZ |
725 | |
726 | if (fs_devices->seed) { | |
727 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
728 | goto again; |
729 | } | |
730 | ||
443f24fe | 731 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 732 | |
dfe25020 | 733 | mutex_unlock(&uuid_mutex); |
dfe25020 | 734 | } |
a0af469b | 735 | |
1f78160c XG |
736 | static void __free_device(struct work_struct *work) |
737 | { | |
738 | struct btrfs_device *device; | |
739 | ||
740 | device = container_of(work, struct btrfs_device, rcu_work); | |
741 | ||
742 | if (device->bdev) | |
743 | blkdev_put(device->bdev, device->mode); | |
744 | ||
606686ee | 745 | rcu_string_free(device->name); |
1f78160c XG |
746 | kfree(device); |
747 | } | |
748 | ||
749 | static void free_device(struct rcu_head *head) | |
750 | { | |
751 | struct btrfs_device *device; | |
752 | ||
753 | device = container_of(head, struct btrfs_device, rcu); | |
754 | ||
755 | INIT_WORK(&device->rcu_work, __free_device); | |
756 | schedule_work(&device->rcu_work); | |
757 | } | |
758 | ||
2b82032c | 759 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 760 | { |
2037a093 | 761 | struct btrfs_device *device, *tmp; |
e4404d6e | 762 | |
2b82032c YZ |
763 | if (--fs_devices->opened > 0) |
764 | return 0; | |
8a4b83cc | 765 | |
c9513edb | 766 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 767 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
1f78160c | 768 | struct btrfs_device *new_device; |
606686ee | 769 | struct rcu_string *name; |
1f78160c XG |
770 | |
771 | if (device->bdev) | |
a0af469b | 772 | fs_devices->open_devices--; |
1f78160c | 773 | |
f747cab7 ID |
774 | if (device->writeable && |
775 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
776 | list_del_init(&device->dev_alloc_list); |
777 | fs_devices->rw_devices--; | |
778 | } | |
779 | ||
726551eb JB |
780 | if (device->missing) |
781 | fs_devices->missing_devices--; | |
d5e2003c | 782 | |
a1e8780a ID |
783 | new_device = btrfs_alloc_device(NULL, &device->devid, |
784 | device->uuid); | |
785 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
606686ee JB |
786 | |
787 | /* Safe because we are under uuid_mutex */ | |
99f5944b JB |
788 | if (device->name) { |
789 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
a1e8780a | 790 | BUG_ON(!name); /* -ENOMEM */ |
99f5944b JB |
791 | rcu_assign_pointer(new_device->name, name); |
792 | } | |
a1e8780a | 793 | |
1f78160c | 794 | list_replace_rcu(&device->dev_list, &new_device->dev_list); |
a1e8780a | 795 | new_device->fs_devices = device->fs_devices; |
1f78160c XG |
796 | |
797 | call_rcu(&device->rcu, free_device); | |
8a4b83cc | 798 | } |
c9513edb XG |
799 | mutex_unlock(&fs_devices->device_list_mutex); |
800 | ||
e4404d6e YZ |
801 | WARN_ON(fs_devices->open_devices); |
802 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
803 | fs_devices->opened = 0; |
804 | fs_devices->seeding = 0; | |
2b82032c | 805 | |
8a4b83cc CM |
806 | return 0; |
807 | } | |
808 | ||
2b82032c YZ |
809 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
810 | { | |
e4404d6e | 811 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
812 | int ret; |
813 | ||
814 | mutex_lock(&uuid_mutex); | |
815 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
816 | if (!fs_devices->opened) { |
817 | seed_devices = fs_devices->seed; | |
818 | fs_devices->seed = NULL; | |
819 | } | |
2b82032c | 820 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
821 | |
822 | while (seed_devices) { | |
823 | fs_devices = seed_devices; | |
824 | seed_devices = fs_devices->seed; | |
825 | __btrfs_close_devices(fs_devices); | |
826 | free_fs_devices(fs_devices); | |
827 | } | |
bc178622 ES |
828 | /* |
829 | * Wait for rcu kworkers under __btrfs_close_devices | |
830 | * to finish all blkdev_puts so device is really | |
831 | * free when umount is done. | |
832 | */ | |
833 | rcu_barrier(); | |
2b82032c YZ |
834 | return ret; |
835 | } | |
836 | ||
e4404d6e YZ |
837 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
838 | fmode_t flags, void *holder) | |
8a4b83cc | 839 | { |
d5e2003c | 840 | struct request_queue *q; |
8a4b83cc CM |
841 | struct block_device *bdev; |
842 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 843 | struct btrfs_device *device; |
443f24fe | 844 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
845 | struct buffer_head *bh; |
846 | struct btrfs_super_block *disk_super; | |
a0af469b | 847 | u64 devid; |
2b82032c | 848 | int seeding = 1; |
a0af469b | 849 | int ret = 0; |
8a4b83cc | 850 | |
d4d77629 TH |
851 | flags |= FMODE_EXCL; |
852 | ||
c6e30871 | 853 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
854 | if (device->bdev) |
855 | continue; | |
dfe25020 CM |
856 | if (!device->name) |
857 | continue; | |
858 | ||
f63e0cca ES |
859 | /* Just open everything we can; ignore failures here */ |
860 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
861 | &bdev, &bh)) | |
beaf8ab3 | 862 | continue; |
a0af469b CM |
863 | |
864 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 865 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
866 | if (devid != device->devid) |
867 | goto error_brelse; | |
868 | ||
2b82032c YZ |
869 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
870 | BTRFS_UUID_SIZE)) | |
871 | goto error_brelse; | |
872 | ||
873 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
874 | if (!latest_dev || |
875 | device->generation > latest_dev->generation) | |
876 | latest_dev = device; | |
a0af469b | 877 | |
2b82032c YZ |
878 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
879 | device->writeable = 0; | |
880 | } else { | |
881 | device->writeable = !bdev_read_only(bdev); | |
882 | seeding = 0; | |
883 | } | |
884 | ||
d5e2003c | 885 | q = bdev_get_queue(bdev); |
90180da4 | 886 | if (blk_queue_discard(q)) |
d5e2003c | 887 | device->can_discard = 1; |
d5e2003c | 888 | |
8a4b83cc | 889 | device->bdev = bdev; |
dfe25020 | 890 | device->in_fs_metadata = 0; |
15916de8 CM |
891 | device->mode = flags; |
892 | ||
c289811c CM |
893 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
894 | fs_devices->rotating = 1; | |
895 | ||
a0af469b | 896 | fs_devices->open_devices++; |
55e50e45 ID |
897 | if (device->writeable && |
898 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
899 | fs_devices->rw_devices++; |
900 | list_add(&device->dev_alloc_list, | |
901 | &fs_devices->alloc_list); | |
902 | } | |
4f6c9328 | 903 | brelse(bh); |
a0af469b | 904 | continue; |
a061fc8d | 905 | |
a0af469b CM |
906 | error_brelse: |
907 | brelse(bh); | |
d4d77629 | 908 | blkdev_put(bdev, flags); |
a0af469b | 909 | continue; |
8a4b83cc | 910 | } |
a0af469b | 911 | if (fs_devices->open_devices == 0) { |
20bcd649 | 912 | ret = -EINVAL; |
a0af469b CM |
913 | goto out; |
914 | } | |
2b82032c YZ |
915 | fs_devices->seeding = seeding; |
916 | fs_devices->opened = 1; | |
443f24fe | 917 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 918 | fs_devices->total_rw_bytes = 0; |
a0af469b | 919 | out: |
2b82032c YZ |
920 | return ret; |
921 | } | |
922 | ||
923 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 924 | fmode_t flags, void *holder) |
2b82032c YZ |
925 | { |
926 | int ret; | |
927 | ||
928 | mutex_lock(&uuid_mutex); | |
929 | if (fs_devices->opened) { | |
e4404d6e YZ |
930 | fs_devices->opened++; |
931 | ret = 0; | |
2b82032c | 932 | } else { |
15916de8 | 933 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 934 | } |
8a4b83cc | 935 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
936 | return ret; |
937 | } | |
938 | ||
6f60cbd3 DS |
939 | /* |
940 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
941 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
942 | * is read via pagecache | |
943 | */ | |
97288f2c | 944 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
945 | struct btrfs_fs_devices **fs_devices_ret) |
946 | { | |
947 | struct btrfs_super_block *disk_super; | |
948 | struct block_device *bdev; | |
6f60cbd3 DS |
949 | struct page *page; |
950 | void *p; | |
951 | int ret = -EINVAL; | |
8a4b83cc | 952 | u64 devid; |
f2984462 | 953 | u64 transid; |
02db0844 | 954 | u64 total_devices; |
6f60cbd3 DS |
955 | u64 bytenr; |
956 | pgoff_t index; | |
8a4b83cc | 957 | |
6f60cbd3 DS |
958 | /* |
959 | * we would like to check all the supers, but that would make | |
960 | * a btrfs mount succeed after a mkfs from a different FS. | |
961 | * So, we need to add a special mount option to scan for | |
962 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
963 | */ | |
964 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 965 | flags |= FMODE_EXCL; |
10f6327b | 966 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
967 | |
968 | bdev = blkdev_get_by_path(path, flags, holder); | |
969 | ||
970 | if (IS_ERR(bdev)) { | |
971 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 972 | goto error; |
6f60cbd3 DS |
973 | } |
974 | ||
975 | /* make sure our super fits in the device */ | |
976 | if (bytenr + PAGE_CACHE_SIZE >= i_size_read(bdev->bd_inode)) | |
977 | goto error_bdev_put; | |
978 | ||
979 | /* make sure our super fits in the page */ | |
980 | if (sizeof(*disk_super) > PAGE_CACHE_SIZE) | |
981 | goto error_bdev_put; | |
982 | ||
983 | /* make sure our super doesn't straddle pages on disk */ | |
984 | index = bytenr >> PAGE_CACHE_SHIFT; | |
985 | if ((bytenr + sizeof(*disk_super) - 1) >> PAGE_CACHE_SHIFT != index) | |
986 | goto error_bdev_put; | |
987 | ||
988 | /* pull in the page with our super */ | |
989 | page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
990 | index, GFP_NOFS); | |
991 | ||
992 | if (IS_ERR_OR_NULL(page)) | |
993 | goto error_bdev_put; | |
994 | ||
995 | p = kmap(page); | |
996 | ||
997 | /* align our pointer to the offset of the super block */ | |
998 | disk_super = p + (bytenr & ~PAGE_CACHE_MASK); | |
999 | ||
1000 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
3cae210f | 1001 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) |
6f60cbd3 DS |
1002 | goto error_unmap; |
1003 | ||
a343832f | 1004 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1005 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1006 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1007 | |
8a4b83cc | 1008 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1009 | if (ret > 0) { |
1010 | if (disk_super->label[0]) { | |
1011 | if (disk_super->label[BTRFS_LABEL_SIZE - 1]) | |
1012 | disk_super->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1013 | printk(KERN_INFO "BTRFS: device label %s ", disk_super->label); | |
1014 | } else { | |
1015 | printk(KERN_INFO "BTRFS: device fsid %pU ", disk_super->fsid); | |
1016 | } | |
1017 | ||
1018 | printk(KERN_CONT "devid %llu transid %llu %s\n", devid, transid, path); | |
1019 | ret = 0; | |
1020 | } | |
02db0844 JB |
1021 | if (!ret && fs_devices_ret) |
1022 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 DS |
1023 | |
1024 | error_unmap: | |
1025 | kunmap(page); | |
1026 | page_cache_release(page); | |
1027 | ||
1028 | error_bdev_put: | |
d4d77629 | 1029 | blkdev_put(bdev, flags); |
8a4b83cc | 1030 | error: |
beaf8ab3 | 1031 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1032 | return ret; |
1033 | } | |
0b86a832 | 1034 | |
6d07bcec MX |
1035 | /* helper to account the used device space in the range */ |
1036 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1037 | u64 end, u64 *length) | |
1038 | { | |
1039 | struct btrfs_key key; | |
1040 | struct btrfs_root *root = device->dev_root; | |
1041 | struct btrfs_dev_extent *dev_extent; | |
1042 | struct btrfs_path *path; | |
1043 | u64 extent_end; | |
1044 | int ret; | |
1045 | int slot; | |
1046 | struct extent_buffer *l; | |
1047 | ||
1048 | *length = 0; | |
1049 | ||
63a212ab | 1050 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1051 | return 0; |
1052 | ||
1053 | path = btrfs_alloc_path(); | |
1054 | if (!path) | |
1055 | return -ENOMEM; | |
1056 | path->reada = 2; | |
1057 | ||
1058 | key.objectid = device->devid; | |
1059 | key.offset = start; | |
1060 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1061 | ||
1062 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1063 | if (ret < 0) | |
1064 | goto out; | |
1065 | if (ret > 0) { | |
1066 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1067 | if (ret < 0) | |
1068 | goto out; | |
1069 | } | |
1070 | ||
1071 | while (1) { | |
1072 | l = path->nodes[0]; | |
1073 | slot = path->slots[0]; | |
1074 | if (slot >= btrfs_header_nritems(l)) { | |
1075 | ret = btrfs_next_leaf(root, path); | |
1076 | if (ret == 0) | |
1077 | continue; | |
1078 | if (ret < 0) | |
1079 | goto out; | |
1080 | ||
1081 | break; | |
1082 | } | |
1083 | btrfs_item_key_to_cpu(l, &key, slot); | |
1084 | ||
1085 | if (key.objectid < device->devid) | |
1086 | goto next; | |
1087 | ||
1088 | if (key.objectid > device->devid) | |
1089 | break; | |
1090 | ||
962a298f | 1091 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1092 | goto next; |
1093 | ||
1094 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1095 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1096 | dev_extent); | |
1097 | if (key.offset <= start && extent_end > end) { | |
1098 | *length = end - start + 1; | |
1099 | break; | |
1100 | } else if (key.offset <= start && extent_end > start) | |
1101 | *length += extent_end - start; | |
1102 | else if (key.offset > start && extent_end <= end) | |
1103 | *length += extent_end - key.offset; | |
1104 | else if (key.offset > start && key.offset <= end) { | |
1105 | *length += end - key.offset + 1; | |
1106 | break; | |
1107 | } else if (key.offset > end) | |
1108 | break; | |
1109 | ||
1110 | next: | |
1111 | path->slots[0]++; | |
1112 | } | |
1113 | ret = 0; | |
1114 | out: | |
1115 | btrfs_free_path(path); | |
1116 | return ret; | |
1117 | } | |
1118 | ||
6df9a95e JB |
1119 | static int contains_pending_extent(struct btrfs_trans_handle *trans, |
1120 | struct btrfs_device *device, | |
1121 | u64 *start, u64 len) | |
1122 | { | |
1123 | struct extent_map *em; | |
04216820 | 1124 | struct list_head *search_list = &trans->transaction->pending_chunks; |
6df9a95e | 1125 | int ret = 0; |
1b984508 | 1126 | u64 physical_start = *start; |
6df9a95e | 1127 | |
04216820 FM |
1128 | again: |
1129 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1130 | struct map_lookup *map; |
1131 | int i; | |
1132 | ||
1133 | map = (struct map_lookup *)em->bdev; | |
1134 | for (i = 0; i < map->num_stripes; i++) { | |
c152b63e FM |
1135 | u64 end; |
1136 | ||
6df9a95e JB |
1137 | if (map->stripes[i].dev != device) |
1138 | continue; | |
1b984508 | 1139 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1140 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1141 | physical_start) |
6df9a95e | 1142 | continue; |
c152b63e FM |
1143 | /* |
1144 | * Make sure that while processing the pinned list we do | |
1145 | * not override our *start with a lower value, because | |
1146 | * we can have pinned chunks that fall within this | |
1147 | * device hole and that have lower physical addresses | |
1148 | * than the pending chunks we processed before. If we | |
1149 | * do not take this special care we can end up getting | |
1150 | * 2 pending chunks that start at the same physical | |
1151 | * device offsets because the end offset of a pinned | |
1152 | * chunk can be equal to the start offset of some | |
1153 | * pending chunk. | |
1154 | */ | |
1155 | end = map->stripes[i].physical + em->orig_block_len; | |
1156 | if (end > *start) { | |
1157 | *start = end; | |
1158 | ret = 1; | |
1159 | } | |
6df9a95e JB |
1160 | } |
1161 | } | |
04216820 FM |
1162 | if (search_list == &trans->transaction->pending_chunks) { |
1163 | search_list = &trans->root->fs_info->pinned_chunks; | |
1164 | goto again; | |
1165 | } | |
6df9a95e JB |
1166 | |
1167 | return ret; | |
1168 | } | |
1169 | ||
1170 | ||
0b86a832 | 1171 | /* |
7bfc837d | 1172 | * find_free_dev_extent - find free space in the specified device |
7bfc837d MX |
1173 | * @device: the device which we search the free space in |
1174 | * @num_bytes: the size of the free space that we need | |
1175 | * @start: store the start of the free space. | |
1176 | * @len: the size of the free space. that we find, or the size of the max | |
1177 | * free space if we don't find suitable free space | |
1178 | * | |
0b86a832 CM |
1179 | * this uses a pretty simple search, the expectation is that it is |
1180 | * called very infrequently and that a given device has a small number | |
1181 | * of extents | |
7bfc837d MX |
1182 | * |
1183 | * @start is used to store the start of the free space if we find. But if we | |
1184 | * don't find suitable free space, it will be used to store the start position | |
1185 | * of the max free space. | |
1186 | * | |
1187 | * @len is used to store the size of the free space that we find. | |
1188 | * But if we don't find suitable free space, it is used to store the size of | |
1189 | * the max free space. | |
0b86a832 | 1190 | */ |
6df9a95e JB |
1191 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1192 | struct btrfs_device *device, u64 num_bytes, | |
7bfc837d | 1193 | u64 *start, u64 *len) |
0b86a832 CM |
1194 | { |
1195 | struct btrfs_key key; | |
1196 | struct btrfs_root *root = device->dev_root; | |
7bfc837d | 1197 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1198 | struct btrfs_path *path; |
7bfc837d MX |
1199 | u64 hole_size; |
1200 | u64 max_hole_start; | |
1201 | u64 max_hole_size; | |
1202 | u64 extent_end; | |
1203 | u64 search_start; | |
0b86a832 CM |
1204 | u64 search_end = device->total_bytes; |
1205 | int ret; | |
7bfc837d | 1206 | int slot; |
0b86a832 CM |
1207 | struct extent_buffer *l; |
1208 | ||
0b86a832 CM |
1209 | /* FIXME use last free of some kind */ |
1210 | ||
8a4b83cc CM |
1211 | /* we don't want to overwrite the superblock on the drive, |
1212 | * so we make sure to start at an offset of at least 1MB | |
1213 | */ | |
a9c9bf68 | 1214 | search_start = max(root->fs_info->alloc_start, 1024ull * 1024); |
8f18cf13 | 1215 | |
6df9a95e JB |
1216 | path = btrfs_alloc_path(); |
1217 | if (!path) | |
1218 | return -ENOMEM; | |
f2ab7618 | 1219 | |
7bfc837d MX |
1220 | max_hole_start = search_start; |
1221 | max_hole_size = 0; | |
1222 | ||
f2ab7618 | 1223 | again: |
63a212ab | 1224 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1225 | ret = -ENOSPC; |
6df9a95e | 1226 | goto out; |
7bfc837d MX |
1227 | } |
1228 | ||
7bfc837d | 1229 | path->reada = 2; |
6df9a95e JB |
1230 | path->search_commit_root = 1; |
1231 | path->skip_locking = 1; | |
7bfc837d | 1232 | |
0b86a832 CM |
1233 | key.objectid = device->devid; |
1234 | key.offset = search_start; | |
1235 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1236 | |
125ccb0a | 1237 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1238 | if (ret < 0) |
7bfc837d | 1239 | goto out; |
1fcbac58 YZ |
1240 | if (ret > 0) { |
1241 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1242 | if (ret < 0) | |
7bfc837d | 1243 | goto out; |
1fcbac58 | 1244 | } |
7bfc837d | 1245 | |
0b86a832 CM |
1246 | while (1) { |
1247 | l = path->nodes[0]; | |
1248 | slot = path->slots[0]; | |
1249 | if (slot >= btrfs_header_nritems(l)) { | |
1250 | ret = btrfs_next_leaf(root, path); | |
1251 | if (ret == 0) | |
1252 | continue; | |
1253 | if (ret < 0) | |
7bfc837d MX |
1254 | goto out; |
1255 | ||
1256 | break; | |
0b86a832 CM |
1257 | } |
1258 | btrfs_item_key_to_cpu(l, &key, slot); | |
1259 | ||
1260 | if (key.objectid < device->devid) | |
1261 | goto next; | |
1262 | ||
1263 | if (key.objectid > device->devid) | |
7bfc837d | 1264 | break; |
0b86a832 | 1265 | |
962a298f | 1266 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1267 | goto next; |
9779b72f | 1268 | |
7bfc837d MX |
1269 | if (key.offset > search_start) { |
1270 | hole_size = key.offset - search_start; | |
9779b72f | 1271 | |
6df9a95e JB |
1272 | /* |
1273 | * Have to check before we set max_hole_start, otherwise | |
1274 | * we could end up sending back this offset anyway. | |
1275 | */ | |
1276 | if (contains_pending_extent(trans, device, | |
1277 | &search_start, | |
1b984508 FL |
1278 | hole_size)) { |
1279 | if (key.offset >= search_start) { | |
1280 | hole_size = key.offset - search_start; | |
1281 | } else { | |
1282 | WARN_ON_ONCE(1); | |
1283 | hole_size = 0; | |
1284 | } | |
1285 | } | |
6df9a95e | 1286 | |
7bfc837d MX |
1287 | if (hole_size > max_hole_size) { |
1288 | max_hole_start = search_start; | |
1289 | max_hole_size = hole_size; | |
1290 | } | |
9779b72f | 1291 | |
7bfc837d MX |
1292 | /* |
1293 | * If this free space is greater than which we need, | |
1294 | * it must be the max free space that we have found | |
1295 | * until now, so max_hole_start must point to the start | |
1296 | * of this free space and the length of this free space | |
1297 | * is stored in max_hole_size. Thus, we return | |
1298 | * max_hole_start and max_hole_size and go back to the | |
1299 | * caller. | |
1300 | */ | |
1301 | if (hole_size >= num_bytes) { | |
1302 | ret = 0; | |
1303 | goto out; | |
0b86a832 CM |
1304 | } |
1305 | } | |
0b86a832 | 1306 | |
0b86a832 | 1307 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1308 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1309 | dev_extent); | |
1310 | if (extent_end > search_start) | |
1311 | search_start = extent_end; | |
0b86a832 CM |
1312 | next: |
1313 | path->slots[0]++; | |
1314 | cond_resched(); | |
1315 | } | |
0b86a832 | 1316 | |
38c01b96 | 1317 | /* |
1318 | * At this point, search_start should be the end of | |
1319 | * allocated dev extents, and when shrinking the device, | |
1320 | * search_end may be smaller than search_start. | |
1321 | */ | |
f2ab7618 | 1322 | if (search_end > search_start) { |
38c01b96 | 1323 | hole_size = search_end - search_start; |
1324 | ||
f2ab7618 ZL |
1325 | if (contains_pending_extent(trans, device, &search_start, |
1326 | hole_size)) { | |
1327 | btrfs_release_path(path); | |
1328 | goto again; | |
1329 | } | |
0b86a832 | 1330 | |
f2ab7618 ZL |
1331 | if (hole_size > max_hole_size) { |
1332 | max_hole_start = search_start; | |
1333 | max_hole_size = hole_size; | |
1334 | } | |
6df9a95e JB |
1335 | } |
1336 | ||
7bfc837d | 1337 | /* See above. */ |
f2ab7618 | 1338 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1339 | ret = -ENOSPC; |
1340 | else | |
1341 | ret = 0; | |
1342 | ||
1343 | out: | |
2b82032c | 1344 | btrfs_free_path(path); |
7bfc837d | 1345 | *start = max_hole_start; |
b2117a39 | 1346 | if (len) |
7bfc837d | 1347 | *len = max_hole_size; |
0b86a832 CM |
1348 | return ret; |
1349 | } | |
1350 | ||
b2950863 | 1351 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1352 | struct btrfs_device *device, |
2196d6e8 | 1353 | u64 start, u64 *dev_extent_len) |
8f18cf13 CM |
1354 | { |
1355 | int ret; | |
1356 | struct btrfs_path *path; | |
1357 | struct btrfs_root *root = device->dev_root; | |
1358 | struct btrfs_key key; | |
a061fc8d CM |
1359 | struct btrfs_key found_key; |
1360 | struct extent_buffer *leaf = NULL; | |
1361 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1362 | |
1363 | path = btrfs_alloc_path(); | |
1364 | if (!path) | |
1365 | return -ENOMEM; | |
1366 | ||
1367 | key.objectid = device->devid; | |
1368 | key.offset = start; | |
1369 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1370 | again: |
8f18cf13 | 1371 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1372 | if (ret > 0) { |
1373 | ret = btrfs_previous_item(root, path, key.objectid, | |
1374 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1375 | if (ret) |
1376 | goto out; | |
a061fc8d CM |
1377 | leaf = path->nodes[0]; |
1378 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1379 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1380 | struct btrfs_dev_extent); | |
1381 | BUG_ON(found_key.offset > start || found_key.offset + | |
1382 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1383 | key = found_key; |
1384 | btrfs_release_path(path); | |
1385 | goto again; | |
a061fc8d CM |
1386 | } else if (ret == 0) { |
1387 | leaf = path->nodes[0]; | |
1388 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1389 | struct btrfs_dev_extent); | |
79787eaa JM |
1390 | } else { |
1391 | btrfs_error(root->fs_info, ret, "Slot search failed"); | |
1392 | goto out; | |
a061fc8d | 1393 | } |
8f18cf13 | 1394 | |
2196d6e8 MX |
1395 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1396 | ||
8f18cf13 | 1397 | ret = btrfs_del_item(trans, root, path); |
79787eaa JM |
1398 | if (ret) { |
1399 | btrfs_error(root->fs_info, ret, | |
1400 | "Failed to remove dev extent item"); | |
13212b54 ZL |
1401 | } else { |
1402 | trans->transaction->have_free_bgs = 1; | |
79787eaa | 1403 | } |
b0b802d7 | 1404 | out: |
8f18cf13 CM |
1405 | btrfs_free_path(path); |
1406 | return ret; | |
1407 | } | |
1408 | ||
48a3b636 ES |
1409 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1410 | struct btrfs_device *device, | |
1411 | u64 chunk_tree, u64 chunk_objectid, | |
1412 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1413 | { |
1414 | int ret; | |
1415 | struct btrfs_path *path; | |
1416 | struct btrfs_root *root = device->dev_root; | |
1417 | struct btrfs_dev_extent *extent; | |
1418 | struct extent_buffer *leaf; | |
1419 | struct btrfs_key key; | |
1420 | ||
dfe25020 | 1421 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1422 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1423 | path = btrfs_alloc_path(); |
1424 | if (!path) | |
1425 | return -ENOMEM; | |
1426 | ||
0b86a832 | 1427 | key.objectid = device->devid; |
2b82032c | 1428 | key.offset = start; |
0b86a832 CM |
1429 | key.type = BTRFS_DEV_EXTENT_KEY; |
1430 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1431 | sizeof(*extent)); | |
2cdcecbc MF |
1432 | if (ret) |
1433 | goto out; | |
0b86a832 CM |
1434 | |
1435 | leaf = path->nodes[0]; | |
1436 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1437 | struct btrfs_dev_extent); | |
e17cade2 CM |
1438 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1439 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1440 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1441 | ||
1442 | write_extent_buffer(leaf, root->fs_info->chunk_tree_uuid, | |
231e88f4 | 1443 | btrfs_dev_extent_chunk_tree_uuid(extent), BTRFS_UUID_SIZE); |
e17cade2 | 1444 | |
0b86a832 CM |
1445 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1446 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1447 | out: |
0b86a832 CM |
1448 | btrfs_free_path(path); |
1449 | return ret; | |
1450 | } | |
1451 | ||
6df9a95e | 1452 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1453 | { |
6df9a95e JB |
1454 | struct extent_map_tree *em_tree; |
1455 | struct extent_map *em; | |
1456 | struct rb_node *n; | |
1457 | u64 ret = 0; | |
0b86a832 | 1458 | |
6df9a95e JB |
1459 | em_tree = &fs_info->mapping_tree.map_tree; |
1460 | read_lock(&em_tree->lock); | |
1461 | n = rb_last(&em_tree->map); | |
1462 | if (n) { | |
1463 | em = rb_entry(n, struct extent_map, rb_node); | |
1464 | ret = em->start + em->len; | |
0b86a832 | 1465 | } |
6df9a95e JB |
1466 | read_unlock(&em_tree->lock); |
1467 | ||
0b86a832 CM |
1468 | return ret; |
1469 | } | |
1470 | ||
53f10659 ID |
1471 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1472 | u64 *devid_ret) | |
0b86a832 CM |
1473 | { |
1474 | int ret; | |
1475 | struct btrfs_key key; | |
1476 | struct btrfs_key found_key; | |
2b82032c YZ |
1477 | struct btrfs_path *path; |
1478 | ||
2b82032c YZ |
1479 | path = btrfs_alloc_path(); |
1480 | if (!path) | |
1481 | return -ENOMEM; | |
0b86a832 CM |
1482 | |
1483 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1484 | key.type = BTRFS_DEV_ITEM_KEY; | |
1485 | key.offset = (u64)-1; | |
1486 | ||
53f10659 | 1487 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1488 | if (ret < 0) |
1489 | goto error; | |
1490 | ||
79787eaa | 1491 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1492 | |
53f10659 ID |
1493 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1494 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1495 | BTRFS_DEV_ITEM_KEY); |
1496 | if (ret) { | |
53f10659 | 1497 | *devid_ret = 1; |
0b86a832 CM |
1498 | } else { |
1499 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1500 | path->slots[0]); | |
53f10659 | 1501 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1502 | } |
1503 | ret = 0; | |
1504 | error: | |
2b82032c | 1505 | btrfs_free_path(path); |
0b86a832 CM |
1506 | return ret; |
1507 | } | |
1508 | ||
1509 | /* | |
1510 | * the device information is stored in the chunk root | |
1511 | * the btrfs_device struct should be fully filled in | |
1512 | */ | |
48a3b636 ES |
1513 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
1514 | struct btrfs_root *root, | |
1515 | struct btrfs_device *device) | |
0b86a832 CM |
1516 | { |
1517 | int ret; | |
1518 | struct btrfs_path *path; | |
1519 | struct btrfs_dev_item *dev_item; | |
1520 | struct extent_buffer *leaf; | |
1521 | struct btrfs_key key; | |
1522 | unsigned long ptr; | |
0b86a832 CM |
1523 | |
1524 | root = root->fs_info->chunk_root; | |
1525 | ||
1526 | path = btrfs_alloc_path(); | |
1527 | if (!path) | |
1528 | return -ENOMEM; | |
1529 | ||
0b86a832 CM |
1530 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1531 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1532 | key.offset = device->devid; |
0b86a832 CM |
1533 | |
1534 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1535 | sizeof(*dev_item)); |
0b86a832 CM |
1536 | if (ret) |
1537 | goto out; | |
1538 | ||
1539 | leaf = path->nodes[0]; | |
1540 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1541 | ||
1542 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1543 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1544 | btrfs_set_device_type(leaf, dev_item, device->type); |
1545 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1546 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1547 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1548 | btrfs_set_device_total_bytes(leaf, dev_item, |
1549 | btrfs_device_get_disk_total_bytes(device)); | |
1550 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1551 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1552 | btrfs_set_device_group(leaf, dev_item, 0); |
1553 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1554 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1555 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1556 | |
410ba3a2 | 1557 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1558 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1559 | ptr = btrfs_device_fsid(dev_item); |
2b82032c | 1560 | write_extent_buffer(leaf, root->fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1561 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1562 | |
2b82032c | 1563 | ret = 0; |
0b86a832 CM |
1564 | out: |
1565 | btrfs_free_path(path); | |
1566 | return ret; | |
1567 | } | |
8f18cf13 | 1568 | |
5a1972bd QW |
1569 | /* |
1570 | * Function to update ctime/mtime for a given device path. | |
1571 | * Mainly used for ctime/mtime based probe like libblkid. | |
1572 | */ | |
1573 | static void update_dev_time(char *path_name) | |
1574 | { | |
1575 | struct file *filp; | |
1576 | ||
1577 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1578 | if (IS_ERR(filp)) |
5a1972bd QW |
1579 | return; |
1580 | file_update_time(filp); | |
1581 | filp_close(filp, NULL); | |
1582 | return; | |
1583 | } | |
1584 | ||
a061fc8d CM |
1585 | static int btrfs_rm_dev_item(struct btrfs_root *root, |
1586 | struct btrfs_device *device) | |
1587 | { | |
1588 | int ret; | |
1589 | struct btrfs_path *path; | |
a061fc8d | 1590 | struct btrfs_key key; |
a061fc8d CM |
1591 | struct btrfs_trans_handle *trans; |
1592 | ||
1593 | root = root->fs_info->chunk_root; | |
1594 | ||
1595 | path = btrfs_alloc_path(); | |
1596 | if (!path) | |
1597 | return -ENOMEM; | |
1598 | ||
a22285a6 | 1599 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1600 | if (IS_ERR(trans)) { |
1601 | btrfs_free_path(path); | |
1602 | return PTR_ERR(trans); | |
1603 | } | |
a061fc8d CM |
1604 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1605 | key.type = BTRFS_DEV_ITEM_KEY; | |
1606 | key.offset = device->devid; | |
1607 | ||
1608 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1609 | if (ret < 0) | |
1610 | goto out; | |
1611 | ||
1612 | if (ret > 0) { | |
1613 | ret = -ENOENT; | |
1614 | goto out; | |
1615 | } | |
1616 | ||
1617 | ret = btrfs_del_item(trans, root, path); | |
1618 | if (ret) | |
1619 | goto out; | |
a061fc8d CM |
1620 | out: |
1621 | btrfs_free_path(path); | |
1622 | btrfs_commit_transaction(trans, root); | |
1623 | return ret; | |
1624 | } | |
1625 | ||
1626 | int btrfs_rm_device(struct btrfs_root *root, char *device_path) | |
1627 | { | |
1628 | struct btrfs_device *device; | |
2b82032c | 1629 | struct btrfs_device *next_device; |
a061fc8d | 1630 | struct block_device *bdev; |
dfe25020 | 1631 | struct buffer_head *bh = NULL; |
a061fc8d | 1632 | struct btrfs_super_block *disk_super; |
1f78160c | 1633 | struct btrfs_fs_devices *cur_devices; |
a061fc8d CM |
1634 | u64 all_avail; |
1635 | u64 devid; | |
2b82032c YZ |
1636 | u64 num_devices; |
1637 | u8 *dev_uuid; | |
de98ced9 | 1638 | unsigned seq; |
a061fc8d | 1639 | int ret = 0; |
1f78160c | 1640 | bool clear_super = false; |
a061fc8d | 1641 | |
a061fc8d CM |
1642 | mutex_lock(&uuid_mutex); |
1643 | ||
de98ced9 MX |
1644 | do { |
1645 | seq = read_seqbegin(&root->fs_info->profiles_lock); | |
1646 | ||
1647 | all_avail = root->fs_info->avail_data_alloc_bits | | |
1648 | root->fs_info->avail_system_alloc_bits | | |
1649 | root->fs_info->avail_metadata_alloc_bits; | |
1650 | } while (read_seqretry(&root->fs_info->profiles_lock, seq)); | |
a061fc8d | 1651 | |
8dabb742 SB |
1652 | num_devices = root->fs_info->fs_devices->num_devices; |
1653 | btrfs_dev_replace_lock(&root->fs_info->dev_replace); | |
1654 | if (btrfs_dev_replace_is_ongoing(&root->fs_info->dev_replace)) { | |
1655 | WARN_ON(num_devices < 1); | |
1656 | num_devices--; | |
1657 | } | |
1658 | btrfs_dev_replace_unlock(&root->fs_info->dev_replace); | |
1659 | ||
1660 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID10) && num_devices <= 4) { | |
183860f6 | 1661 | ret = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET; |
a061fc8d CM |
1662 | goto out; |
1663 | } | |
1664 | ||
8dabb742 | 1665 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID1) && num_devices <= 2) { |
183860f6 | 1666 | ret = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET; |
a061fc8d CM |
1667 | goto out; |
1668 | } | |
1669 | ||
53b381b3 DW |
1670 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID5) && |
1671 | root->fs_info->fs_devices->rw_devices <= 2) { | |
183860f6 | 1672 | ret = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET; |
53b381b3 DW |
1673 | goto out; |
1674 | } | |
1675 | if ((all_avail & BTRFS_BLOCK_GROUP_RAID6) && | |
1676 | root->fs_info->fs_devices->rw_devices <= 3) { | |
183860f6 | 1677 | ret = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET; |
53b381b3 DW |
1678 | goto out; |
1679 | } | |
1680 | ||
dfe25020 | 1681 | if (strcmp(device_path, "missing") == 0) { |
dfe25020 CM |
1682 | struct list_head *devices; |
1683 | struct btrfs_device *tmp; | |
a061fc8d | 1684 | |
dfe25020 CM |
1685 | device = NULL; |
1686 | devices = &root->fs_info->fs_devices->devices; | |
46224705 XG |
1687 | /* |
1688 | * It is safe to read the devices since the volume_mutex | |
1689 | * is held. | |
1690 | */ | |
c6e30871 | 1691 | list_for_each_entry(tmp, devices, dev_list) { |
63a212ab SB |
1692 | if (tmp->in_fs_metadata && |
1693 | !tmp->is_tgtdev_for_dev_replace && | |
1694 | !tmp->bdev) { | |
dfe25020 CM |
1695 | device = tmp; |
1696 | break; | |
1697 | } | |
1698 | } | |
1699 | bdev = NULL; | |
1700 | bh = NULL; | |
1701 | disk_super = NULL; | |
1702 | if (!device) { | |
183860f6 | 1703 | ret = BTRFS_ERROR_DEV_MISSING_NOT_FOUND; |
dfe25020 CM |
1704 | goto out; |
1705 | } | |
dfe25020 | 1706 | } else { |
beaf8ab3 | 1707 | ret = btrfs_get_bdev_and_sb(device_path, |
cc975eb4 | 1708 | FMODE_WRITE | FMODE_EXCL, |
beaf8ab3 SB |
1709 | root->fs_info->bdev_holder, 0, |
1710 | &bdev, &bh); | |
1711 | if (ret) | |
dfe25020 | 1712 | goto out; |
dfe25020 | 1713 | disk_super = (struct btrfs_super_block *)bh->b_data; |
a343832f | 1714 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
2b82032c | 1715 | dev_uuid = disk_super->dev_item.uuid; |
aa1b8cd4 | 1716 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2b82032c | 1717 | disk_super->fsid); |
dfe25020 CM |
1718 | if (!device) { |
1719 | ret = -ENOENT; | |
1720 | goto error_brelse; | |
1721 | } | |
2b82032c | 1722 | } |
dfe25020 | 1723 | |
63a212ab | 1724 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1725 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
63a212ab SB |
1726 | goto error_brelse; |
1727 | } | |
1728 | ||
2b82032c | 1729 | if (device->writeable && root->fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1730 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
2b82032c YZ |
1731 | goto error_brelse; |
1732 | } | |
1733 | ||
1734 | if (device->writeable) { | |
0c1daee0 | 1735 | lock_chunks(root); |
2b82032c | 1736 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1737 | device->fs_devices->rw_devices--; |
0c1daee0 | 1738 | unlock_chunks(root); |
1f78160c | 1739 | clear_super = true; |
dfe25020 | 1740 | } |
a061fc8d | 1741 | |
d7901554 | 1742 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1743 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1744 | mutex_lock(&uuid_mutex); |
a061fc8d | 1745 | if (ret) |
9b3517e9 | 1746 | goto error_undo; |
a061fc8d | 1747 | |
63a212ab SB |
1748 | /* |
1749 | * TODO: the superblock still includes this device in its num_devices | |
1750 | * counter although write_all_supers() is not locked out. This | |
1751 | * could give a filesystem state which requires a degraded mount. | |
1752 | */ | |
a061fc8d CM |
1753 | ret = btrfs_rm_dev_item(root->fs_info->chunk_root, device); |
1754 | if (ret) | |
9b3517e9 | 1755 | goto error_undo; |
a061fc8d | 1756 | |
2b82032c | 1757 | device->in_fs_metadata = 0; |
aa1b8cd4 | 1758 | btrfs_scrub_cancel_dev(root->fs_info, device); |
e5e9a520 CM |
1759 | |
1760 | /* | |
1761 | * the device list mutex makes sure that we don't change | |
1762 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1763 | * the device supers. Whoever is writing all supers, should |
1764 | * lock the device list mutex before getting the number of | |
1765 | * devices in the super block (super_copy). Conversely, | |
1766 | * whoever updates the number of devices in the super block | |
1767 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1768 | */ |
1f78160c XG |
1769 | |
1770 | cur_devices = device->fs_devices; | |
e5e9a520 | 1771 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
1f78160c | 1772 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1773 | |
e4404d6e | 1774 | device->fs_devices->num_devices--; |
02db0844 | 1775 | device->fs_devices->total_devices--; |
2b82032c | 1776 | |
cd02dca5 | 1777 | if (device->missing) |
3a7d55c8 | 1778 | device->fs_devices->missing_devices--; |
cd02dca5 | 1779 | |
2b82032c YZ |
1780 | next_device = list_entry(root->fs_info->fs_devices->devices.next, |
1781 | struct btrfs_device, dev_list); | |
1782 | if (device->bdev == root->fs_info->sb->s_bdev) | |
1783 | root->fs_info->sb->s_bdev = next_device->bdev; | |
1784 | if (device->bdev == root->fs_info->fs_devices->latest_bdev) | |
1785 | root->fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1786 | ||
0bfaa9c5 | 1787 | if (device->bdev) { |
e4404d6e | 1788 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1789 | /* remove sysfs entry */ |
6c14a164 | 1790 | btrfs_kobj_rm_device(root->fs_info->fs_devices, device); |
0bfaa9c5 | 1791 | } |
99994cde | 1792 | |
1f78160c | 1793 | call_rcu(&device->rcu, free_device); |
e4404d6e | 1794 | |
6c41761f DS |
1795 | num_devices = btrfs_super_num_devices(root->fs_info->super_copy) - 1; |
1796 | btrfs_set_super_num_devices(root->fs_info->super_copy, num_devices); | |
d7306801 | 1797 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2b82032c | 1798 | |
1f78160c | 1799 | if (cur_devices->open_devices == 0) { |
e4404d6e YZ |
1800 | struct btrfs_fs_devices *fs_devices; |
1801 | fs_devices = root->fs_info->fs_devices; | |
1802 | while (fs_devices) { | |
8321cf25 RS |
1803 | if (fs_devices->seed == cur_devices) { |
1804 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1805 | break; |
8321cf25 | 1806 | } |
e4404d6e | 1807 | fs_devices = fs_devices->seed; |
2b82032c | 1808 | } |
1f78160c | 1809 | cur_devices->seed = NULL; |
1f78160c | 1810 | __btrfs_close_devices(cur_devices); |
1f78160c | 1811 | free_fs_devices(cur_devices); |
2b82032c YZ |
1812 | } |
1813 | ||
5af3e8cc SB |
1814 | root->fs_info->num_tolerated_disk_barrier_failures = |
1815 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
1816 | ||
2b82032c YZ |
1817 | /* |
1818 | * at this point, the device is zero sized. We want to | |
1819 | * remove it from the devices list and zero out the old super | |
1820 | */ | |
aa1b8cd4 | 1821 | if (clear_super && disk_super) { |
4d90d28b AJ |
1822 | u64 bytenr; |
1823 | int i; | |
1824 | ||
dfe25020 CM |
1825 | /* make sure this device isn't detected as part of |
1826 | * the FS anymore | |
1827 | */ | |
1828 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
1829 | set_buffer_dirty(bh); | |
1830 | sync_dirty_buffer(bh); | |
4d90d28b AJ |
1831 | |
1832 | /* clear the mirror copies of super block on the disk | |
1833 | * being removed, 0th copy is been taken care above and | |
1834 | * the below would take of the rest | |
1835 | */ | |
1836 | for (i = 1; i < BTRFS_SUPER_MIRROR_MAX; i++) { | |
1837 | bytenr = btrfs_sb_offset(i); | |
1838 | if (bytenr + BTRFS_SUPER_INFO_SIZE >= | |
1839 | i_size_read(bdev->bd_inode)) | |
1840 | break; | |
1841 | ||
1842 | brelse(bh); | |
1843 | bh = __bread(bdev, bytenr / 4096, | |
1844 | BTRFS_SUPER_INFO_SIZE); | |
1845 | if (!bh) | |
1846 | continue; | |
1847 | ||
1848 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1849 | ||
1850 | if (btrfs_super_bytenr(disk_super) != bytenr || | |
1851 | btrfs_super_magic(disk_super) != BTRFS_MAGIC) { | |
1852 | continue; | |
1853 | } | |
1854 | memset(&disk_super->magic, 0, | |
1855 | sizeof(disk_super->magic)); | |
1856 | set_buffer_dirty(bh); | |
1857 | sync_dirty_buffer(bh); | |
1858 | } | |
dfe25020 | 1859 | } |
a061fc8d | 1860 | |
a061fc8d | 1861 | ret = 0; |
a061fc8d | 1862 | |
5a1972bd QW |
1863 | if (bdev) { |
1864 | /* Notify udev that device has changed */ | |
3c911608 | 1865 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); |
b8b8ff59 | 1866 | |
5a1972bd QW |
1867 | /* Update ctime/mtime for device path for libblkid */ |
1868 | update_dev_time(device_path); | |
1869 | } | |
1870 | ||
a061fc8d CM |
1871 | error_brelse: |
1872 | brelse(bh); | |
dfe25020 | 1873 | if (bdev) |
e525fd89 | 1874 | blkdev_put(bdev, FMODE_READ | FMODE_EXCL); |
a061fc8d CM |
1875 | out: |
1876 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1877 | return ret; |
9b3517e9 ID |
1878 | error_undo: |
1879 | if (device->writeable) { | |
0c1daee0 | 1880 | lock_chunks(root); |
9b3517e9 ID |
1881 | list_add(&device->dev_alloc_list, |
1882 | &root->fs_info->fs_devices->alloc_list); | |
c3929c36 | 1883 | device->fs_devices->rw_devices++; |
0c1daee0 | 1884 | unlock_chunks(root); |
9b3517e9 ID |
1885 | } |
1886 | goto error_brelse; | |
a061fc8d CM |
1887 | } |
1888 | ||
084b6e7c QW |
1889 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1890 | struct btrfs_device *srcdev) | |
e93c89c1 | 1891 | { |
d51908ce AJ |
1892 | struct btrfs_fs_devices *fs_devices; |
1893 | ||
e93c89c1 | 1894 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 1895 | |
25e8e911 AJ |
1896 | /* |
1897 | * in case of fs with no seed, srcdev->fs_devices will point | |
1898 | * to fs_devices of fs_info. However when the dev being replaced is | |
1899 | * a seed dev it will point to the seed's local fs_devices. In short | |
1900 | * srcdev will have its correct fs_devices in both the cases. | |
1901 | */ | |
1902 | fs_devices = srcdev->fs_devices; | |
d51908ce | 1903 | |
e93c89c1 SB |
1904 | list_del_rcu(&srcdev->dev_list); |
1905 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 1906 | fs_devices->num_devices--; |
82372bc8 | 1907 | if (srcdev->missing) |
d51908ce | 1908 | fs_devices->missing_devices--; |
e93c89c1 | 1909 | |
82372bc8 MX |
1910 | if (srcdev->writeable) { |
1911 | fs_devices->rw_devices--; | |
1912 | /* zero out the old super if it is writable */ | |
1913 | btrfs_scratch_superblock(srcdev); | |
1357272f ID |
1914 | } |
1915 | ||
82372bc8 | 1916 | if (srcdev->bdev) |
d51908ce | 1917 | fs_devices->open_devices--; |
084b6e7c QW |
1918 | } |
1919 | ||
1920 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
1921 | struct btrfs_device *srcdev) | |
1922 | { | |
1923 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 SB |
1924 | |
1925 | call_rcu(&srcdev->rcu, free_device); | |
94d5f0c2 AJ |
1926 | |
1927 | /* | |
1928 | * unless fs_devices is seed fs, num_devices shouldn't go | |
1929 | * zero | |
1930 | */ | |
1931 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
1932 | ||
1933 | /* if this is no devs we rather delete the fs_devices */ | |
1934 | if (!fs_devices->num_devices) { | |
1935 | struct btrfs_fs_devices *tmp_fs_devices; | |
1936 | ||
1937 | tmp_fs_devices = fs_info->fs_devices; | |
1938 | while (tmp_fs_devices) { | |
1939 | if (tmp_fs_devices->seed == fs_devices) { | |
1940 | tmp_fs_devices->seed = fs_devices->seed; | |
1941 | break; | |
1942 | } | |
1943 | tmp_fs_devices = tmp_fs_devices->seed; | |
1944 | } | |
1945 | fs_devices->seed = NULL; | |
8bef8401 AJ |
1946 | __btrfs_close_devices(fs_devices); |
1947 | free_fs_devices(fs_devices); | |
94d5f0c2 | 1948 | } |
e93c89c1 SB |
1949 | } |
1950 | ||
1951 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
1952 | struct btrfs_device *tgtdev) | |
1953 | { | |
1954 | struct btrfs_device *next_device; | |
1955 | ||
67a2c45e | 1956 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
1957 | WARN_ON(!tgtdev); |
1958 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 AJ |
1959 | |
1960 | btrfs_kobj_rm_device(fs_info->fs_devices, tgtdev); | |
1961 | ||
e93c89c1 SB |
1962 | if (tgtdev->bdev) { |
1963 | btrfs_scratch_superblock(tgtdev); | |
1964 | fs_info->fs_devices->open_devices--; | |
1965 | } | |
1966 | fs_info->fs_devices->num_devices--; | |
e93c89c1 SB |
1967 | |
1968 | next_device = list_entry(fs_info->fs_devices->devices.next, | |
1969 | struct btrfs_device, dev_list); | |
1970 | if (tgtdev->bdev == fs_info->sb->s_bdev) | |
1971 | fs_info->sb->s_bdev = next_device->bdev; | |
1972 | if (tgtdev->bdev == fs_info->fs_devices->latest_bdev) | |
1973 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1974 | list_del_rcu(&tgtdev->dev_list); | |
1975 | ||
1976 | call_rcu(&tgtdev->rcu, free_device); | |
1977 | ||
1978 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 1979 | mutex_unlock(&uuid_mutex); |
e93c89c1 SB |
1980 | } |
1981 | ||
48a3b636 ES |
1982 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
1983 | struct btrfs_device **device) | |
7ba15b7d SB |
1984 | { |
1985 | int ret = 0; | |
1986 | struct btrfs_super_block *disk_super; | |
1987 | u64 devid; | |
1988 | u8 *dev_uuid; | |
1989 | struct block_device *bdev; | |
1990 | struct buffer_head *bh; | |
1991 | ||
1992 | *device = NULL; | |
1993 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
1994 | root->fs_info->bdev_holder, 0, &bdev, &bh); | |
1995 | if (ret) | |
1996 | return ret; | |
1997 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
1998 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
1999 | dev_uuid = disk_super->dev_item.uuid; | |
aa1b8cd4 | 2000 | *device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
7ba15b7d SB |
2001 | disk_super->fsid); |
2002 | brelse(bh); | |
2003 | if (!*device) | |
2004 | ret = -ENOENT; | |
2005 | blkdev_put(bdev, FMODE_READ); | |
2006 | return ret; | |
2007 | } | |
2008 | ||
2009 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2010 | char *device_path, | |
2011 | struct btrfs_device **device) | |
2012 | { | |
2013 | *device = NULL; | |
2014 | if (strcmp(device_path, "missing") == 0) { | |
2015 | struct list_head *devices; | |
2016 | struct btrfs_device *tmp; | |
2017 | ||
2018 | devices = &root->fs_info->fs_devices->devices; | |
2019 | /* | |
2020 | * It is safe to read the devices since the volume_mutex | |
2021 | * is held by the caller. | |
2022 | */ | |
2023 | list_for_each_entry(tmp, devices, dev_list) { | |
2024 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2025 | *device = tmp; | |
2026 | break; | |
2027 | } | |
2028 | } | |
2029 | ||
2030 | if (!*device) { | |
efe120a0 | 2031 | btrfs_err(root->fs_info, "no missing device found"); |
7ba15b7d SB |
2032 | return -ENOENT; |
2033 | } | |
2034 | ||
2035 | return 0; | |
2036 | } else { | |
2037 | return btrfs_find_device_by_path(root, device_path, device); | |
2038 | } | |
2039 | } | |
2040 | ||
2b82032c YZ |
2041 | /* |
2042 | * does all the dirty work required for changing file system's UUID. | |
2043 | */ | |
125ccb0a | 2044 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c YZ |
2045 | { |
2046 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
2047 | struct btrfs_fs_devices *old_devices; | |
e4404d6e | 2048 | struct btrfs_fs_devices *seed_devices; |
6c41761f | 2049 | struct btrfs_super_block *disk_super = root->fs_info->super_copy; |
2b82032c YZ |
2050 | struct btrfs_device *device; |
2051 | u64 super_flags; | |
2052 | ||
2053 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2054 | if (!fs_devices->seeding) |
2b82032c YZ |
2055 | return -EINVAL; |
2056 | ||
2208a378 ID |
2057 | seed_devices = __alloc_fs_devices(); |
2058 | if (IS_ERR(seed_devices)) | |
2059 | return PTR_ERR(seed_devices); | |
2b82032c | 2060 | |
e4404d6e YZ |
2061 | old_devices = clone_fs_devices(fs_devices); |
2062 | if (IS_ERR(old_devices)) { | |
2063 | kfree(seed_devices); | |
2064 | return PTR_ERR(old_devices); | |
2b82032c | 2065 | } |
e4404d6e | 2066 | |
2b82032c YZ |
2067 | list_add(&old_devices->list, &fs_uuids); |
2068 | ||
e4404d6e YZ |
2069 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2070 | seed_devices->opened = 1; | |
2071 | INIT_LIST_HEAD(&seed_devices->devices); | |
2072 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2073 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb XG |
2074 | |
2075 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
1f78160c XG |
2076 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2077 | synchronize_rcu); | |
2196d6e8 MX |
2078 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2079 | device->fs_devices = seed_devices; | |
c9513edb | 2080 | |
2196d6e8 | 2081 | lock_chunks(root); |
e4404d6e | 2082 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
2196d6e8 | 2083 | unlock_chunks(root); |
e4404d6e | 2084 | |
2b82032c YZ |
2085 | fs_devices->seeding = 0; |
2086 | fs_devices->num_devices = 0; | |
2087 | fs_devices->open_devices = 0; | |
69611ac8 | 2088 | fs_devices->missing_devices = 0; |
69611ac8 | 2089 | fs_devices->rotating = 0; |
e4404d6e | 2090 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2091 | |
2092 | generate_random_uuid(fs_devices->fsid); | |
2093 | memcpy(root->fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
2094 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); | |
f7171750 FDBM |
2095 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2096 | ||
2b82032c YZ |
2097 | super_flags = btrfs_super_flags(disk_super) & |
2098 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2099 | btrfs_set_super_flags(disk_super, super_flags); | |
2100 | ||
2101 | return 0; | |
2102 | } | |
2103 | ||
2104 | /* | |
2105 | * strore the expected generation for seed devices in device items. | |
2106 | */ | |
2107 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
2108 | struct btrfs_root *root) | |
2109 | { | |
2110 | struct btrfs_path *path; | |
2111 | struct extent_buffer *leaf; | |
2112 | struct btrfs_dev_item *dev_item; | |
2113 | struct btrfs_device *device; | |
2114 | struct btrfs_key key; | |
2115 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2116 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2117 | u64 devid; | |
2118 | int ret; | |
2119 | ||
2120 | path = btrfs_alloc_path(); | |
2121 | if (!path) | |
2122 | return -ENOMEM; | |
2123 | ||
2124 | root = root->fs_info->chunk_root; | |
2125 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2126 | key.offset = 0; | |
2127 | key.type = BTRFS_DEV_ITEM_KEY; | |
2128 | ||
2129 | while (1) { | |
2130 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2131 | if (ret < 0) | |
2132 | goto error; | |
2133 | ||
2134 | leaf = path->nodes[0]; | |
2135 | next_slot: | |
2136 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2137 | ret = btrfs_next_leaf(root, path); | |
2138 | if (ret > 0) | |
2139 | break; | |
2140 | if (ret < 0) | |
2141 | goto error; | |
2142 | leaf = path->nodes[0]; | |
2143 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2144 | btrfs_release_path(path); |
2b82032c YZ |
2145 | continue; |
2146 | } | |
2147 | ||
2148 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2149 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2150 | key.type != BTRFS_DEV_ITEM_KEY) | |
2151 | break; | |
2152 | ||
2153 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2154 | struct btrfs_dev_item); | |
2155 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2156 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2157 | BTRFS_UUID_SIZE); |
1473b24e | 2158 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2159 | BTRFS_UUID_SIZE); |
aa1b8cd4 SB |
2160 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, |
2161 | fs_uuid); | |
79787eaa | 2162 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2163 | |
2164 | if (device->fs_devices->seeding) { | |
2165 | btrfs_set_device_generation(leaf, dev_item, | |
2166 | device->generation); | |
2167 | btrfs_mark_buffer_dirty(leaf); | |
2168 | } | |
2169 | ||
2170 | path->slots[0]++; | |
2171 | goto next_slot; | |
2172 | } | |
2173 | ret = 0; | |
2174 | error: | |
2175 | btrfs_free_path(path); | |
2176 | return ret; | |
2177 | } | |
2178 | ||
788f20eb CM |
2179 | int btrfs_init_new_device(struct btrfs_root *root, char *device_path) |
2180 | { | |
d5e2003c | 2181 | struct request_queue *q; |
788f20eb CM |
2182 | struct btrfs_trans_handle *trans; |
2183 | struct btrfs_device *device; | |
2184 | struct block_device *bdev; | |
788f20eb | 2185 | struct list_head *devices; |
2b82032c | 2186 | struct super_block *sb = root->fs_info->sb; |
606686ee | 2187 | struct rcu_string *name; |
3c1dbdf5 | 2188 | u64 tmp; |
2b82032c | 2189 | int seeding_dev = 0; |
788f20eb CM |
2190 | int ret = 0; |
2191 | ||
2b82032c | 2192 | if ((sb->s_flags & MS_RDONLY) && !root->fs_info->fs_devices->seeding) |
f8c5d0b4 | 2193 | return -EROFS; |
788f20eb | 2194 | |
a5d16333 | 2195 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
d4d77629 | 2196 | root->fs_info->bdev_holder); |
7f59203a JB |
2197 | if (IS_ERR(bdev)) |
2198 | return PTR_ERR(bdev); | |
a2135011 | 2199 | |
2b82032c YZ |
2200 | if (root->fs_info->fs_devices->seeding) { |
2201 | seeding_dev = 1; | |
2202 | down_write(&sb->s_umount); | |
2203 | mutex_lock(&uuid_mutex); | |
2204 | } | |
2205 | ||
8c8bee1d | 2206 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2207 | |
788f20eb | 2208 | devices = &root->fs_info->fs_devices->devices; |
d25628bd LB |
2209 | |
2210 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
c6e30871 | 2211 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2212 | if (device->bdev == bdev) { |
2213 | ret = -EEXIST; | |
d25628bd LB |
2214 | mutex_unlock( |
2215 | &root->fs_info->fs_devices->device_list_mutex); | |
2b82032c | 2216 | goto error; |
788f20eb CM |
2217 | } |
2218 | } | |
d25628bd | 2219 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2220 | |
12bd2fc0 ID |
2221 | device = btrfs_alloc_device(root->fs_info, NULL, NULL); |
2222 | if (IS_ERR(device)) { | |
788f20eb | 2223 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2224 | ret = PTR_ERR(device); |
2b82032c | 2225 | goto error; |
788f20eb CM |
2226 | } |
2227 | ||
606686ee JB |
2228 | name = rcu_string_strdup(device_path, GFP_NOFS); |
2229 | if (!name) { | |
788f20eb | 2230 | kfree(device); |
2b82032c YZ |
2231 | ret = -ENOMEM; |
2232 | goto error; | |
788f20eb | 2233 | } |
606686ee | 2234 | rcu_assign_pointer(device->name, name); |
2b82032c | 2235 | |
a22285a6 | 2236 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2237 | if (IS_ERR(trans)) { |
606686ee | 2238 | rcu_string_free(device->name); |
98d5dc13 TI |
2239 | kfree(device); |
2240 | ret = PTR_ERR(trans); | |
2241 | goto error; | |
2242 | } | |
2243 | ||
d5e2003c JB |
2244 | q = bdev_get_queue(bdev); |
2245 | if (blk_queue_discard(q)) | |
2246 | device->can_discard = 1; | |
2b82032c | 2247 | device->writeable = 1; |
2b82032c | 2248 | device->generation = trans->transid; |
788f20eb CM |
2249 | device->io_width = root->sectorsize; |
2250 | device->io_align = root->sectorsize; | |
2251 | device->sector_size = root->sectorsize; | |
2252 | device->total_bytes = i_size_read(bdev->bd_inode); | |
2cc3c559 | 2253 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2254 | device->commit_total_bytes = device->total_bytes; |
788f20eb CM |
2255 | device->dev_root = root->fs_info->dev_root; |
2256 | device->bdev = bdev; | |
dfe25020 | 2257 | device->in_fs_metadata = 1; |
63a212ab | 2258 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2259 | device->mode = FMODE_EXCL; |
27087f37 | 2260 | device->dev_stats_valid = 1; |
2b82032c | 2261 | set_blocksize(device->bdev, 4096); |
788f20eb | 2262 | |
2b82032c YZ |
2263 | if (seeding_dev) { |
2264 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2265 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2266 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2267 | } |
788f20eb | 2268 | |
2b82032c | 2269 | device->fs_devices = root->fs_info->fs_devices; |
e5e9a520 | 2270 | |
e5e9a520 | 2271 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); |
2196d6e8 | 2272 | lock_chunks(root); |
1f78160c | 2273 | list_add_rcu(&device->dev_list, &root->fs_info->fs_devices->devices); |
2b82032c YZ |
2274 | list_add(&device->dev_alloc_list, |
2275 | &root->fs_info->fs_devices->alloc_list); | |
2276 | root->fs_info->fs_devices->num_devices++; | |
2277 | root->fs_info->fs_devices->open_devices++; | |
2278 | root->fs_info->fs_devices->rw_devices++; | |
02db0844 | 2279 | root->fs_info->fs_devices->total_devices++; |
2b82032c | 2280 | root->fs_info->fs_devices->total_rw_bytes += device->total_bytes; |
325cd4ba | 2281 | |
2bf64758 JB |
2282 | spin_lock(&root->fs_info->free_chunk_lock); |
2283 | root->fs_info->free_chunk_space += device->total_bytes; | |
2284 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2285 | ||
c289811c CM |
2286 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
2287 | root->fs_info->fs_devices->rotating = 1; | |
2288 | ||
3c1dbdf5 | 2289 | tmp = btrfs_super_total_bytes(root->fs_info->super_copy); |
6c41761f | 2290 | btrfs_set_super_total_bytes(root->fs_info->super_copy, |
3c1dbdf5 | 2291 | tmp + device->total_bytes); |
788f20eb | 2292 | |
3c1dbdf5 | 2293 | tmp = btrfs_super_num_devices(root->fs_info->super_copy); |
6c41761f | 2294 | btrfs_set_super_num_devices(root->fs_info->super_copy, |
3c1dbdf5 | 2295 | tmp + 1); |
0d39376a AJ |
2296 | |
2297 | /* add sysfs device entry */ | |
1ba43816 | 2298 | btrfs_kobj_add_device(root->fs_info->fs_devices, device); |
0d39376a | 2299 | |
2196d6e8 MX |
2300 | /* |
2301 | * we've got more storage, clear any full flags on the space | |
2302 | * infos | |
2303 | */ | |
2304 | btrfs_clear_space_info_full(root->fs_info); | |
2305 | ||
2306 | unlock_chunks(root); | |
e5e9a520 | 2307 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
788f20eb | 2308 | |
2b82032c | 2309 | if (seeding_dev) { |
2196d6e8 | 2310 | lock_chunks(root); |
2b82032c | 2311 | ret = init_first_rw_device(trans, root, device); |
2196d6e8 | 2312 | unlock_chunks(root); |
005d6427 DS |
2313 | if (ret) { |
2314 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2315 | goto error_trans; |
005d6427 | 2316 | } |
2196d6e8 MX |
2317 | } |
2318 | ||
2319 | ret = btrfs_add_device(trans, root, device); | |
2320 | if (ret) { | |
2321 | btrfs_abort_transaction(trans, root, ret); | |
2322 | goto error_trans; | |
2323 | } | |
2324 | ||
2325 | if (seeding_dev) { | |
2326 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2327 | ||
2b82032c | 2328 | ret = btrfs_finish_sprout(trans, root); |
005d6427 DS |
2329 | if (ret) { |
2330 | btrfs_abort_transaction(trans, root, ret); | |
79787eaa | 2331 | goto error_trans; |
005d6427 | 2332 | } |
b2373f25 AJ |
2333 | |
2334 | /* Sprouting would change fsid of the mounted root, | |
2335 | * so rename the fsid on the sysfs | |
2336 | */ | |
2337 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
2338 | root->fs_info->fsid); | |
2e7910d6 AJ |
2339 | if (kobject_rename(&root->fs_info->fs_devices->super_kobj, |
2340 | fsid_buf)) | |
2421a8cd | 2341 | pr_warn("BTRFS: sysfs: failed to create fsid for sprout\n"); |
2b82032c YZ |
2342 | } |
2343 | ||
5af3e8cc SB |
2344 | root->fs_info->num_tolerated_disk_barrier_failures = |
2345 | btrfs_calc_num_tolerated_disk_barrier_failures(root->fs_info); | |
79787eaa | 2346 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2347 | |
2b82032c YZ |
2348 | if (seeding_dev) { |
2349 | mutex_unlock(&uuid_mutex); | |
2350 | up_write(&sb->s_umount); | |
788f20eb | 2351 | |
79787eaa JM |
2352 | if (ret) /* transaction commit */ |
2353 | return ret; | |
2354 | ||
2b82032c | 2355 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa JM |
2356 | if (ret < 0) |
2357 | btrfs_error(root->fs_info, ret, | |
2358 | "Failed to relocate sys chunks after " | |
2359 | "device initialization. This can be fixed " | |
2360 | "using the \"btrfs balance\" command."); | |
671415b7 MX |
2361 | trans = btrfs_attach_transaction(root); |
2362 | if (IS_ERR(trans)) { | |
2363 | if (PTR_ERR(trans) == -ENOENT) | |
2364 | return 0; | |
2365 | return PTR_ERR(trans); | |
2366 | } | |
2367 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2368 | } |
c9e9f97b | 2369 | |
5a1972bd QW |
2370 | /* Update ctime/mtime for libblkid */ |
2371 | update_dev_time(device_path); | |
2b82032c | 2372 | return ret; |
79787eaa JM |
2373 | |
2374 | error_trans: | |
79787eaa | 2375 | btrfs_end_transaction(trans, root); |
606686ee | 2376 | rcu_string_free(device->name); |
6c14a164 | 2377 | btrfs_kobj_rm_device(root->fs_info->fs_devices, device); |
79787eaa | 2378 | kfree(device); |
2b82032c | 2379 | error: |
e525fd89 | 2380 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2381 | if (seeding_dev) { |
2382 | mutex_unlock(&uuid_mutex); | |
2383 | up_write(&sb->s_umount); | |
2384 | } | |
c9e9f97b | 2385 | return ret; |
788f20eb CM |
2386 | } |
2387 | ||
e93c89c1 | 2388 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2389 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2390 | struct btrfs_device **device_out) |
2391 | { | |
2392 | struct request_queue *q; | |
2393 | struct btrfs_device *device; | |
2394 | struct block_device *bdev; | |
2395 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2396 | struct list_head *devices; | |
2397 | struct rcu_string *name; | |
12bd2fc0 | 2398 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2399 | int ret = 0; |
2400 | ||
2401 | *device_out = NULL; | |
1c43366d MX |
2402 | if (fs_info->fs_devices->seeding) { |
2403 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2404 | return -EINVAL; |
1c43366d | 2405 | } |
e93c89c1 SB |
2406 | |
2407 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2408 | fs_info->bdev_holder); | |
1c43366d MX |
2409 | if (IS_ERR(bdev)) { |
2410 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2411 | return PTR_ERR(bdev); |
1c43366d | 2412 | } |
e93c89c1 SB |
2413 | |
2414 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2415 | ||
2416 | devices = &fs_info->fs_devices->devices; | |
2417 | list_for_each_entry(device, devices, dev_list) { | |
2418 | if (device->bdev == bdev) { | |
1c43366d | 2419 | btrfs_err(fs_info, "target device is in the filesystem!"); |
e93c89c1 SB |
2420 | ret = -EEXIST; |
2421 | goto error; | |
2422 | } | |
2423 | } | |
2424 | ||
1c43366d | 2425 | |
7cc8e58d MX |
2426 | if (i_size_read(bdev->bd_inode) < |
2427 | btrfs_device_get_total_bytes(srcdev)) { | |
1c43366d MX |
2428 | btrfs_err(fs_info, "target device is smaller than source device!"); |
2429 | ret = -EINVAL; | |
2430 | goto error; | |
2431 | } | |
2432 | ||
2433 | ||
12bd2fc0 ID |
2434 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2435 | if (IS_ERR(device)) { | |
2436 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2437 | goto error; |
2438 | } | |
2439 | ||
2440 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2441 | if (!name) { | |
2442 | kfree(device); | |
2443 | ret = -ENOMEM; | |
2444 | goto error; | |
2445 | } | |
2446 | rcu_assign_pointer(device->name, name); | |
2447 | ||
2448 | q = bdev_get_queue(bdev); | |
2449 | if (blk_queue_discard(q)) | |
2450 | device->can_discard = 1; | |
2451 | mutex_lock(&root->fs_info->fs_devices->device_list_mutex); | |
2452 | device->writeable = 1; | |
e93c89c1 SB |
2453 | device->generation = 0; |
2454 | device->io_width = root->sectorsize; | |
2455 | device->io_align = root->sectorsize; | |
2456 | device->sector_size = root->sectorsize; | |
7cc8e58d MX |
2457 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2458 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2459 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2460 | ASSERT(list_empty(&srcdev->resized_list)); |
2461 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2462 | device->commit_bytes_used = device->bytes_used; |
e93c89c1 SB |
2463 | device->dev_root = fs_info->dev_root; |
2464 | device->bdev = bdev; | |
2465 | device->in_fs_metadata = 1; | |
2466 | device->is_tgtdev_for_dev_replace = 1; | |
2467 | device->mode = FMODE_EXCL; | |
27087f37 | 2468 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2469 | set_blocksize(device->bdev, 4096); |
2470 | device->fs_devices = fs_info->fs_devices; | |
2471 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2472 | fs_info->fs_devices->num_devices++; | |
2473 | fs_info->fs_devices->open_devices++; | |
e93c89c1 SB |
2474 | mutex_unlock(&root->fs_info->fs_devices->device_list_mutex); |
2475 | ||
2476 | *device_out = device; | |
2477 | return ret; | |
2478 | ||
2479 | error: | |
2480 | blkdev_put(bdev, FMODE_EXCL); | |
2481 | return ret; | |
2482 | } | |
2483 | ||
2484 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2485 | struct btrfs_device *tgtdev) | |
2486 | { | |
2487 | WARN_ON(fs_info->fs_devices->rw_devices == 0); | |
2488 | tgtdev->io_width = fs_info->dev_root->sectorsize; | |
2489 | tgtdev->io_align = fs_info->dev_root->sectorsize; | |
2490 | tgtdev->sector_size = fs_info->dev_root->sectorsize; | |
2491 | tgtdev->dev_root = fs_info->dev_root; | |
2492 | tgtdev->in_fs_metadata = 1; | |
2493 | } | |
2494 | ||
d397712b CM |
2495 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2496 | struct btrfs_device *device) | |
0b86a832 CM |
2497 | { |
2498 | int ret; | |
2499 | struct btrfs_path *path; | |
2500 | struct btrfs_root *root; | |
2501 | struct btrfs_dev_item *dev_item; | |
2502 | struct extent_buffer *leaf; | |
2503 | struct btrfs_key key; | |
2504 | ||
2505 | root = device->dev_root->fs_info->chunk_root; | |
2506 | ||
2507 | path = btrfs_alloc_path(); | |
2508 | if (!path) | |
2509 | return -ENOMEM; | |
2510 | ||
2511 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2512 | key.type = BTRFS_DEV_ITEM_KEY; | |
2513 | key.offset = device->devid; | |
2514 | ||
2515 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2516 | if (ret < 0) | |
2517 | goto out; | |
2518 | ||
2519 | if (ret > 0) { | |
2520 | ret = -ENOENT; | |
2521 | goto out; | |
2522 | } | |
2523 | ||
2524 | leaf = path->nodes[0]; | |
2525 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2526 | ||
2527 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2528 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2529 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2530 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2531 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2532 | btrfs_set_device_total_bytes(leaf, dev_item, |
2533 | btrfs_device_get_disk_total_bytes(device)); | |
2534 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2535 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2536 | btrfs_mark_buffer_dirty(leaf); |
2537 | ||
2538 | out: | |
2539 | btrfs_free_path(path); | |
2540 | return ret; | |
2541 | } | |
2542 | ||
2196d6e8 | 2543 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2544 | struct btrfs_device *device, u64 new_size) |
2545 | { | |
2546 | struct btrfs_super_block *super_copy = | |
6c41761f | 2547 | device->dev_root->fs_info->super_copy; |
935e5cc9 | 2548 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2549 | u64 old_total; |
2550 | u64 diff; | |
8f18cf13 | 2551 | |
2b82032c YZ |
2552 | if (!device->writeable) |
2553 | return -EACCES; | |
2196d6e8 MX |
2554 | |
2555 | lock_chunks(device->dev_root); | |
2556 | old_total = btrfs_super_total_bytes(super_copy); | |
2557 | diff = new_size - device->total_bytes; | |
2558 | ||
63a212ab | 2559 | if (new_size <= device->total_bytes || |
2196d6e8 MX |
2560 | device->is_tgtdev_for_dev_replace) { |
2561 | unlock_chunks(device->dev_root); | |
2b82032c | 2562 | return -EINVAL; |
2196d6e8 | 2563 | } |
2b82032c | 2564 | |
935e5cc9 | 2565 | fs_devices = device->dev_root->fs_info->fs_devices; |
2b82032c | 2566 | |
8f18cf13 | 2567 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2568 | device->fs_devices->total_rw_bytes += diff; |
2569 | ||
7cc8e58d MX |
2570 | btrfs_device_set_total_bytes(device, new_size); |
2571 | btrfs_device_set_disk_total_bytes(device, new_size); | |
4184ea7f | 2572 | btrfs_clear_space_info_full(device->dev_root->fs_info); |
935e5cc9 MX |
2573 | if (list_empty(&device->resized_list)) |
2574 | list_add_tail(&device->resized_list, | |
2575 | &fs_devices->resized_devices); | |
2196d6e8 | 2576 | unlock_chunks(device->dev_root); |
4184ea7f | 2577 | |
8f18cf13 CM |
2578 | return btrfs_update_device(trans, device); |
2579 | } | |
2580 | ||
2581 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
a688a04a | 2582 | struct btrfs_root *root, u64 chunk_objectid, |
8f18cf13 CM |
2583 | u64 chunk_offset) |
2584 | { | |
2585 | int ret; | |
2586 | struct btrfs_path *path; | |
2587 | struct btrfs_key key; | |
2588 | ||
2589 | root = root->fs_info->chunk_root; | |
2590 | path = btrfs_alloc_path(); | |
2591 | if (!path) | |
2592 | return -ENOMEM; | |
2593 | ||
2594 | key.objectid = chunk_objectid; | |
2595 | key.offset = chunk_offset; | |
2596 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2597 | ||
2598 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2599 | if (ret < 0) |
2600 | goto out; | |
2601 | else if (ret > 0) { /* Logic error or corruption */ | |
2602 | btrfs_error(root->fs_info, -ENOENT, | |
2603 | "Failed lookup while freeing chunk."); | |
2604 | ret = -ENOENT; | |
2605 | goto out; | |
2606 | } | |
8f18cf13 CM |
2607 | |
2608 | ret = btrfs_del_item(trans, root, path); | |
79787eaa JM |
2609 | if (ret < 0) |
2610 | btrfs_error(root->fs_info, ret, | |
2611 | "Failed to delete chunk item."); | |
2612 | out: | |
8f18cf13 | 2613 | btrfs_free_path(path); |
65a246c5 | 2614 | return ret; |
8f18cf13 CM |
2615 | } |
2616 | ||
b2950863 | 2617 | static int btrfs_del_sys_chunk(struct btrfs_root *root, u64 chunk_objectid, u64 |
8f18cf13 CM |
2618 | chunk_offset) |
2619 | { | |
6c41761f | 2620 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 CM |
2621 | struct btrfs_disk_key *disk_key; |
2622 | struct btrfs_chunk *chunk; | |
2623 | u8 *ptr; | |
2624 | int ret = 0; | |
2625 | u32 num_stripes; | |
2626 | u32 array_size; | |
2627 | u32 len = 0; | |
2628 | u32 cur; | |
2629 | struct btrfs_key key; | |
2630 | ||
2196d6e8 | 2631 | lock_chunks(root); |
8f18cf13 CM |
2632 | array_size = btrfs_super_sys_array_size(super_copy); |
2633 | ||
2634 | ptr = super_copy->sys_chunk_array; | |
2635 | cur = 0; | |
2636 | ||
2637 | while (cur < array_size) { | |
2638 | disk_key = (struct btrfs_disk_key *)ptr; | |
2639 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2640 | ||
2641 | len = sizeof(*disk_key); | |
2642 | ||
2643 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2644 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2645 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2646 | len += btrfs_chunk_item_size(num_stripes); | |
2647 | } else { | |
2648 | ret = -EIO; | |
2649 | break; | |
2650 | } | |
2651 | if (key.objectid == chunk_objectid && | |
2652 | key.offset == chunk_offset) { | |
2653 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2654 | array_size -= len; | |
2655 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2656 | } else { | |
2657 | ptr += len; | |
2658 | cur += len; | |
2659 | } | |
2660 | } | |
2196d6e8 | 2661 | unlock_chunks(root); |
8f18cf13 CM |
2662 | return ret; |
2663 | } | |
2664 | ||
47ab2a6c JB |
2665 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
2666 | struct btrfs_root *root, u64 chunk_offset) | |
8f18cf13 CM |
2667 | { |
2668 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2669 | struct extent_map *em; |
47ab2a6c | 2670 | struct btrfs_root *extent_root = root->fs_info->extent_root; |
8f18cf13 | 2671 | struct map_lookup *map; |
2196d6e8 | 2672 | u64 dev_extent_len = 0; |
47ab2a6c | 2673 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2674 | int i, ret = 0; |
8f18cf13 | 2675 | |
47ab2a6c | 2676 | /* Just in case */ |
8f18cf13 | 2677 | root = root->fs_info->chunk_root; |
8f18cf13 CM |
2678 | em_tree = &root->fs_info->mapping_tree.map_tree; |
2679 | ||
890871be | 2680 | read_lock(&em_tree->lock); |
8f18cf13 | 2681 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2682 | read_unlock(&em_tree->lock); |
8f18cf13 | 2683 | |
47ab2a6c JB |
2684 | if (!em || em->start > chunk_offset || |
2685 | em->start + em->len < chunk_offset) { | |
2686 | /* | |
2687 | * This is a logic error, but we don't want to just rely on the | |
2688 | * user having built with ASSERT enabled, so if ASSERT doens't | |
2689 | * do anything we still error out. | |
2690 | */ | |
2691 | ASSERT(0); | |
2692 | if (em) | |
2693 | free_extent_map(em); | |
2694 | return -EINVAL; | |
2695 | } | |
8f18cf13 | 2696 | map = (struct map_lookup *)em->bdev; |
39c2d7fa | 2697 | lock_chunks(root->fs_info->chunk_root); |
4617ea3a | 2698 | check_system_chunk(trans, extent_root, map->type); |
39c2d7fa | 2699 | unlock_chunks(root->fs_info->chunk_root); |
8f18cf13 CM |
2700 | |
2701 | for (i = 0; i < map->num_stripes; i++) { | |
47ab2a6c | 2702 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2703 | ret = btrfs_free_dev_extent(trans, device, |
2704 | map->stripes[i].physical, | |
2705 | &dev_extent_len); | |
47ab2a6c JB |
2706 | if (ret) { |
2707 | btrfs_abort_transaction(trans, root, ret); | |
2708 | goto out; | |
2709 | } | |
a061fc8d | 2710 | |
2196d6e8 MX |
2711 | if (device->bytes_used > 0) { |
2712 | lock_chunks(root); | |
2713 | btrfs_device_set_bytes_used(device, | |
2714 | device->bytes_used - dev_extent_len); | |
2715 | spin_lock(&root->fs_info->free_chunk_lock); | |
2716 | root->fs_info->free_chunk_space += dev_extent_len; | |
2717 | spin_unlock(&root->fs_info->free_chunk_lock); | |
2718 | btrfs_clear_space_info_full(root->fs_info); | |
2719 | unlock_chunks(root); | |
2720 | } | |
a061fc8d | 2721 | |
dfe25020 CM |
2722 | if (map->stripes[i].dev) { |
2723 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c JB |
2724 | if (ret) { |
2725 | btrfs_abort_transaction(trans, root, ret); | |
2726 | goto out; | |
2727 | } | |
dfe25020 | 2728 | } |
8f18cf13 | 2729 | } |
a688a04a | 2730 | ret = btrfs_free_chunk(trans, root, chunk_objectid, chunk_offset); |
47ab2a6c JB |
2731 | if (ret) { |
2732 | btrfs_abort_transaction(trans, root, ret); | |
2733 | goto out; | |
2734 | } | |
8f18cf13 | 2735 | |
1abe9b8a | 2736 | trace_btrfs_chunk_free(root, map, chunk_offset, em->len); |
2737 | ||
8f18cf13 CM |
2738 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
2739 | ret = btrfs_del_sys_chunk(root, chunk_objectid, chunk_offset); | |
47ab2a6c JB |
2740 | if (ret) { |
2741 | btrfs_abort_transaction(trans, root, ret); | |
2742 | goto out; | |
2743 | } | |
8f18cf13 CM |
2744 | } |
2745 | ||
04216820 | 2746 | ret = btrfs_remove_block_group(trans, extent_root, chunk_offset, em); |
47ab2a6c JB |
2747 | if (ret) { |
2748 | btrfs_abort_transaction(trans, extent_root, ret); | |
2749 | goto out; | |
2750 | } | |
2b82032c | 2751 | |
47ab2a6c | 2752 | out: |
2b82032c YZ |
2753 | /* once for us */ |
2754 | free_extent_map(em); | |
47ab2a6c JB |
2755 | return ret; |
2756 | } | |
2b82032c | 2757 | |
47ab2a6c | 2758 | static int btrfs_relocate_chunk(struct btrfs_root *root, |
a688a04a ZL |
2759 | u64 chunk_objectid, |
2760 | u64 chunk_offset) | |
47ab2a6c JB |
2761 | { |
2762 | struct btrfs_root *extent_root; | |
2763 | struct btrfs_trans_handle *trans; | |
2764 | int ret; | |
2b82032c | 2765 | |
47ab2a6c JB |
2766 | root = root->fs_info->chunk_root; |
2767 | extent_root = root->fs_info->extent_root; | |
2768 | ||
2769 | ret = btrfs_can_relocate(extent_root, chunk_offset); | |
2770 | if (ret) | |
2771 | return -ENOSPC; | |
2772 | ||
2773 | /* step one, relocate all the extents inside this chunk */ | |
2774 | ret = btrfs_relocate_block_group(extent_root, chunk_offset); | |
2775 | if (ret) | |
2776 | return ret; | |
2777 | ||
2778 | trans = btrfs_start_transaction(root, 0); | |
2779 | if (IS_ERR(trans)) { | |
2780 | ret = PTR_ERR(trans); | |
2781 | btrfs_std_error(root->fs_info, ret); | |
2782 | return ret; | |
2783 | } | |
2784 | ||
2785 | /* | |
2786 | * step two, delete the device extents and the | |
2787 | * chunk tree entries | |
2788 | */ | |
2789 | ret = btrfs_remove_chunk(trans, root, chunk_offset); | |
2b82032c | 2790 | btrfs_end_transaction(trans, root); |
47ab2a6c | 2791 | return ret; |
2b82032c YZ |
2792 | } |
2793 | ||
2794 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2795 | { | |
2796 | struct btrfs_root *chunk_root = root->fs_info->chunk_root; | |
2797 | struct btrfs_path *path; | |
2798 | struct extent_buffer *leaf; | |
2799 | struct btrfs_chunk *chunk; | |
2800 | struct btrfs_key key; | |
2801 | struct btrfs_key found_key; | |
2b82032c | 2802 | u64 chunk_type; |
ba1bf481 JB |
2803 | bool retried = false; |
2804 | int failed = 0; | |
2b82032c YZ |
2805 | int ret; |
2806 | ||
2807 | path = btrfs_alloc_path(); | |
2808 | if (!path) | |
2809 | return -ENOMEM; | |
2810 | ||
ba1bf481 | 2811 | again: |
2b82032c YZ |
2812 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2813 | key.offset = (u64)-1; | |
2814 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2815 | ||
2816 | while (1) { | |
2817 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); | |
2818 | if (ret < 0) | |
2819 | goto error; | |
79787eaa | 2820 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2821 | |
2822 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2823 | key.type); | |
2824 | if (ret < 0) | |
2825 | goto error; | |
2826 | if (ret > 0) | |
2827 | break; | |
1a40e23b | 2828 | |
2b82032c YZ |
2829 | leaf = path->nodes[0]; |
2830 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2831 | |
2b82032c YZ |
2832 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
2833 | struct btrfs_chunk); | |
2834 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 2835 | btrfs_release_path(path); |
8f18cf13 | 2836 | |
2b82032c | 2837 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
a688a04a | 2838 | ret = btrfs_relocate_chunk(chunk_root, |
2b82032c YZ |
2839 | found_key.objectid, |
2840 | found_key.offset); | |
ba1bf481 JB |
2841 | if (ret == -ENOSPC) |
2842 | failed++; | |
14586651 HS |
2843 | else |
2844 | BUG_ON(ret); | |
2b82032c | 2845 | } |
8f18cf13 | 2846 | |
2b82032c YZ |
2847 | if (found_key.offset == 0) |
2848 | break; | |
2849 | key.offset = found_key.offset - 1; | |
2850 | } | |
2851 | ret = 0; | |
ba1bf481 JB |
2852 | if (failed && !retried) { |
2853 | failed = 0; | |
2854 | retried = true; | |
2855 | goto again; | |
fae7f21c | 2856 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
2857 | ret = -ENOSPC; |
2858 | } | |
2b82032c YZ |
2859 | error: |
2860 | btrfs_free_path(path); | |
2861 | return ret; | |
8f18cf13 CM |
2862 | } |
2863 | ||
0940ebf6 ID |
2864 | static int insert_balance_item(struct btrfs_root *root, |
2865 | struct btrfs_balance_control *bctl) | |
2866 | { | |
2867 | struct btrfs_trans_handle *trans; | |
2868 | struct btrfs_balance_item *item; | |
2869 | struct btrfs_disk_balance_args disk_bargs; | |
2870 | struct btrfs_path *path; | |
2871 | struct extent_buffer *leaf; | |
2872 | struct btrfs_key key; | |
2873 | int ret, err; | |
2874 | ||
2875 | path = btrfs_alloc_path(); | |
2876 | if (!path) | |
2877 | return -ENOMEM; | |
2878 | ||
2879 | trans = btrfs_start_transaction(root, 0); | |
2880 | if (IS_ERR(trans)) { | |
2881 | btrfs_free_path(path); | |
2882 | return PTR_ERR(trans); | |
2883 | } | |
2884 | ||
2885 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2886 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2887 | key.offset = 0; | |
2888 | ||
2889 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
2890 | sizeof(*item)); | |
2891 | if (ret) | |
2892 | goto out; | |
2893 | ||
2894 | leaf = path->nodes[0]; | |
2895 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
2896 | ||
2897 | memset_extent_buffer(leaf, 0, (unsigned long)item, sizeof(*item)); | |
2898 | ||
2899 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
2900 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
2901 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
2902 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
2903 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
2904 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
2905 | ||
2906 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
2907 | ||
2908 | btrfs_mark_buffer_dirty(leaf); | |
2909 | out: | |
2910 | btrfs_free_path(path); | |
2911 | err = btrfs_commit_transaction(trans, root); | |
2912 | if (err && !ret) | |
2913 | ret = err; | |
2914 | return ret; | |
2915 | } | |
2916 | ||
2917 | static int del_balance_item(struct btrfs_root *root) | |
2918 | { | |
2919 | struct btrfs_trans_handle *trans; | |
2920 | struct btrfs_path *path; | |
2921 | struct btrfs_key key; | |
2922 | int ret, err; | |
2923 | ||
2924 | path = btrfs_alloc_path(); | |
2925 | if (!path) | |
2926 | return -ENOMEM; | |
2927 | ||
2928 | trans = btrfs_start_transaction(root, 0); | |
2929 | if (IS_ERR(trans)) { | |
2930 | btrfs_free_path(path); | |
2931 | return PTR_ERR(trans); | |
2932 | } | |
2933 | ||
2934 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
2935 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
2936 | key.offset = 0; | |
2937 | ||
2938 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
2939 | if (ret < 0) | |
2940 | goto out; | |
2941 | if (ret > 0) { | |
2942 | ret = -ENOENT; | |
2943 | goto out; | |
2944 | } | |
2945 | ||
2946 | ret = btrfs_del_item(trans, root, path); | |
2947 | out: | |
2948 | btrfs_free_path(path); | |
2949 | err = btrfs_commit_transaction(trans, root); | |
2950 | if (err && !ret) | |
2951 | ret = err; | |
2952 | return ret; | |
2953 | } | |
2954 | ||
59641015 ID |
2955 | /* |
2956 | * This is a heuristic used to reduce the number of chunks balanced on | |
2957 | * resume after balance was interrupted. | |
2958 | */ | |
2959 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
2960 | { | |
2961 | /* | |
2962 | * Turn on soft mode for chunk types that were being converted. | |
2963 | */ | |
2964 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2965 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2966 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2967 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2968 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
2969 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
2970 | ||
2971 | /* | |
2972 | * Turn on usage filter if is not already used. The idea is | |
2973 | * that chunks that we have already balanced should be | |
2974 | * reasonably full. Don't do it for chunks that are being | |
2975 | * converted - that will keep us from relocating unconverted | |
2976 | * (albeit full) chunks. | |
2977 | */ | |
2978 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2979 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2980 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2981 | bctl->data.usage = 90; | |
2982 | } | |
2983 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2984 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2985 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2986 | bctl->sys.usage = 90; | |
2987 | } | |
2988 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
2989 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { | |
2990 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
2991 | bctl->meta.usage = 90; | |
2992 | } | |
2993 | } | |
2994 | ||
c9e9f97b ID |
2995 | /* |
2996 | * Should be called with both balance and volume mutexes held to | |
2997 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
2998 | * restriper. Same goes for unset_balance_control. | |
2999 | */ | |
3000 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3001 | { | |
3002 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3003 | ||
3004 | BUG_ON(fs_info->balance_ctl); | |
3005 | ||
3006 | spin_lock(&fs_info->balance_lock); | |
3007 | fs_info->balance_ctl = bctl; | |
3008 | spin_unlock(&fs_info->balance_lock); | |
3009 | } | |
3010 | ||
3011 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3012 | { | |
3013 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3014 | ||
3015 | BUG_ON(!fs_info->balance_ctl); | |
3016 | ||
3017 | spin_lock(&fs_info->balance_lock); | |
3018 | fs_info->balance_ctl = NULL; | |
3019 | spin_unlock(&fs_info->balance_lock); | |
3020 | ||
3021 | kfree(bctl); | |
3022 | } | |
3023 | ||
ed25e9b2 ID |
3024 | /* |
3025 | * Balance filters. Return 1 if chunk should be filtered out | |
3026 | * (should not be balanced). | |
3027 | */ | |
899c81ea | 3028 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3029 | struct btrfs_balance_args *bargs) |
3030 | { | |
899c81ea ID |
3031 | chunk_type = chunk_to_extended(chunk_type) & |
3032 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3033 | |
899c81ea | 3034 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3035 | return 0; |
3036 | ||
3037 | return 1; | |
3038 | } | |
3039 | ||
5ce5b3c0 ID |
3040 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
3041 | struct btrfs_balance_args *bargs) | |
3042 | { | |
3043 | struct btrfs_block_group_cache *cache; | |
3044 | u64 chunk_used, user_thresh; | |
3045 | int ret = 1; | |
3046 | ||
3047 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3048 | chunk_used = btrfs_block_group_used(&cache->item); | |
3049 | ||
a105bb88 | 3050 | if (bargs->usage == 0) |
3e39cea6 | 3051 | user_thresh = 1; |
a105bb88 ID |
3052 | else if (bargs->usage > 100) |
3053 | user_thresh = cache->key.offset; | |
3054 | else | |
3055 | user_thresh = div_factor_fine(cache->key.offset, | |
3056 | bargs->usage); | |
3057 | ||
5ce5b3c0 ID |
3058 | if (chunk_used < user_thresh) |
3059 | ret = 0; | |
3060 | ||
3061 | btrfs_put_block_group(cache); | |
3062 | return ret; | |
3063 | } | |
3064 | ||
409d404b ID |
3065 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3066 | struct btrfs_chunk *chunk, | |
3067 | struct btrfs_balance_args *bargs) | |
3068 | { | |
3069 | struct btrfs_stripe *stripe; | |
3070 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3071 | int i; | |
3072 | ||
3073 | for (i = 0; i < num_stripes; i++) { | |
3074 | stripe = btrfs_stripe_nr(chunk, i); | |
3075 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3076 | return 0; | |
3077 | } | |
3078 | ||
3079 | return 1; | |
3080 | } | |
3081 | ||
94e60d5a ID |
3082 | /* [pstart, pend) */ |
3083 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3084 | struct btrfs_chunk *chunk, | |
3085 | u64 chunk_offset, | |
3086 | struct btrfs_balance_args *bargs) | |
3087 | { | |
3088 | struct btrfs_stripe *stripe; | |
3089 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3090 | u64 stripe_offset; | |
3091 | u64 stripe_length; | |
3092 | int factor; | |
3093 | int i; | |
3094 | ||
3095 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3096 | return 0; | |
3097 | ||
3098 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3099 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3100 | factor = num_stripes / 2; | |
3101 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3102 | factor = num_stripes - 1; | |
3103 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3104 | factor = num_stripes - 2; | |
3105 | } else { | |
3106 | factor = num_stripes; | |
3107 | } | |
94e60d5a ID |
3108 | |
3109 | for (i = 0; i < num_stripes; i++) { | |
3110 | stripe = btrfs_stripe_nr(chunk, i); | |
3111 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3112 | continue; | |
3113 | ||
3114 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3115 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3116 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3117 | |
3118 | if (stripe_offset < bargs->pend && | |
3119 | stripe_offset + stripe_length > bargs->pstart) | |
3120 | return 0; | |
3121 | } | |
3122 | ||
3123 | return 1; | |
3124 | } | |
3125 | ||
ea67176a ID |
3126 | /* [vstart, vend) */ |
3127 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3128 | struct btrfs_chunk *chunk, | |
3129 | u64 chunk_offset, | |
3130 | struct btrfs_balance_args *bargs) | |
3131 | { | |
3132 | if (chunk_offset < bargs->vend && | |
3133 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3134 | /* at least part of the chunk is inside this vrange */ | |
3135 | return 0; | |
3136 | ||
3137 | return 1; | |
3138 | } | |
3139 | ||
899c81ea | 3140 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3141 | struct btrfs_balance_args *bargs) |
3142 | { | |
3143 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3144 | return 0; | |
3145 | ||
899c81ea ID |
3146 | chunk_type = chunk_to_extended(chunk_type) & |
3147 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3148 | |
899c81ea | 3149 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3150 | return 1; |
3151 | ||
3152 | return 0; | |
3153 | } | |
3154 | ||
f43ffb60 ID |
3155 | static int should_balance_chunk(struct btrfs_root *root, |
3156 | struct extent_buffer *leaf, | |
3157 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3158 | { | |
3159 | struct btrfs_balance_control *bctl = root->fs_info->balance_ctl; | |
3160 | struct btrfs_balance_args *bargs = NULL; | |
3161 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3162 | ||
3163 | /* type filter */ | |
3164 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3165 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3166 | return 0; | |
3167 | } | |
3168 | ||
3169 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3170 | bargs = &bctl->data; | |
3171 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3172 | bargs = &bctl->sys; | |
3173 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3174 | bargs = &bctl->meta; | |
3175 | ||
ed25e9b2 ID |
3176 | /* profiles filter */ |
3177 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3178 | chunk_profiles_filter(chunk_type, bargs)) { | |
3179 | return 0; | |
5ce5b3c0 ID |
3180 | } |
3181 | ||
3182 | /* usage filter */ | |
3183 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
3184 | chunk_usage_filter(bctl->fs_info, chunk_offset, bargs)) { | |
3185 | return 0; | |
409d404b ID |
3186 | } |
3187 | ||
3188 | /* devid filter */ | |
3189 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3190 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3191 | return 0; | |
94e60d5a ID |
3192 | } |
3193 | ||
3194 | /* drange filter, makes sense only with devid filter */ | |
3195 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3196 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3197 | return 0; | |
ea67176a ID |
3198 | } |
3199 | ||
3200 | /* vrange filter */ | |
3201 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3202 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3203 | return 0; | |
ed25e9b2 ID |
3204 | } |
3205 | ||
cfa4c961 ID |
3206 | /* soft profile changing mode */ |
3207 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3208 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3209 | return 0; | |
3210 | } | |
3211 | ||
7d824b6f DS |
3212 | /* |
3213 | * limited by count, must be the last filter | |
3214 | */ | |
3215 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3216 | if (bargs->limit == 0) | |
3217 | return 0; | |
3218 | else | |
3219 | bargs->limit--; | |
3220 | } | |
3221 | ||
f43ffb60 ID |
3222 | return 1; |
3223 | } | |
3224 | ||
c9e9f97b | 3225 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3226 | { |
19a39dce | 3227 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3228 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3229 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3230 | struct list_head *devices; | |
ec44a35c CM |
3231 | struct btrfs_device *device; |
3232 | u64 old_size; | |
3233 | u64 size_to_free; | |
f43ffb60 | 3234 | struct btrfs_chunk *chunk; |
ec44a35c CM |
3235 | struct btrfs_path *path; |
3236 | struct btrfs_key key; | |
ec44a35c | 3237 | struct btrfs_key found_key; |
c9e9f97b | 3238 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3239 | struct extent_buffer *leaf; |
3240 | int slot; | |
c9e9f97b ID |
3241 | int ret; |
3242 | int enospc_errors = 0; | |
19a39dce | 3243 | bool counting = true; |
7d824b6f DS |
3244 | u64 limit_data = bctl->data.limit; |
3245 | u64 limit_meta = bctl->meta.limit; | |
3246 | u64 limit_sys = bctl->sys.limit; | |
ec44a35c | 3247 | |
ec44a35c | 3248 | /* step one make some room on all the devices */ |
c9e9f97b | 3249 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3250 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3251 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c CM |
3252 | size_to_free = div_factor(old_size, 1); |
3253 | size_to_free = min(size_to_free, (u64)1 * 1024 * 1024); | |
2b82032c | 3254 | if (!device->writeable || |
7cc8e58d MX |
3255 | btrfs_device_get_total_bytes(device) - |
3256 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3257 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3258 | continue; |
3259 | ||
3260 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3261 | if (ret == -ENOSPC) |
3262 | break; | |
ec44a35c CM |
3263 | BUG_ON(ret); |
3264 | ||
a22285a6 | 3265 | trans = btrfs_start_transaction(dev_root, 0); |
98d5dc13 | 3266 | BUG_ON(IS_ERR(trans)); |
ec44a35c CM |
3267 | |
3268 | ret = btrfs_grow_device(trans, device, old_size); | |
3269 | BUG_ON(ret); | |
3270 | ||
3271 | btrfs_end_transaction(trans, dev_root); | |
3272 | } | |
3273 | ||
3274 | /* step two, relocate all the chunks */ | |
3275 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3276 | if (!path) { |
3277 | ret = -ENOMEM; | |
3278 | goto error; | |
3279 | } | |
19a39dce ID |
3280 | |
3281 | /* zero out stat counters */ | |
3282 | spin_lock(&fs_info->balance_lock); | |
3283 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3284 | spin_unlock(&fs_info->balance_lock); | |
3285 | again: | |
7d824b6f DS |
3286 | if (!counting) { |
3287 | bctl->data.limit = limit_data; | |
3288 | bctl->meta.limit = limit_meta; | |
3289 | bctl->sys.limit = limit_sys; | |
3290 | } | |
ec44a35c CM |
3291 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3292 | key.offset = (u64)-1; | |
3293 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3294 | ||
d397712b | 3295 | while (1) { |
19a39dce | 3296 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3297 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3298 | ret = -ECANCELED; |
3299 | goto error; | |
3300 | } | |
3301 | ||
ec44a35c CM |
3302 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
3303 | if (ret < 0) | |
3304 | goto error; | |
3305 | ||
3306 | /* | |
3307 | * this shouldn't happen, it means the last relocate | |
3308 | * failed | |
3309 | */ | |
3310 | if (ret == 0) | |
c9e9f97b | 3311 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3312 | |
3313 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3314 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b ID |
3315 | if (ret) { |
3316 | ret = 0; | |
ec44a35c | 3317 | break; |
c9e9f97b | 3318 | } |
7d9eb12c | 3319 | |
f43ffb60 ID |
3320 | leaf = path->nodes[0]; |
3321 | slot = path->slots[0]; | |
3322 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3323 | |
ec44a35c CM |
3324 | if (found_key.objectid != key.objectid) |
3325 | break; | |
7d9eb12c | 3326 | |
f43ffb60 ID |
3327 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
3328 | ||
19a39dce ID |
3329 | if (!counting) { |
3330 | spin_lock(&fs_info->balance_lock); | |
3331 | bctl->stat.considered++; | |
3332 | spin_unlock(&fs_info->balance_lock); | |
3333 | } | |
3334 | ||
f43ffb60 ID |
3335 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3336 | found_key.offset); | |
b3b4aa74 | 3337 | btrfs_release_path(path); |
f43ffb60 ID |
3338 | if (!ret) |
3339 | goto loop; | |
3340 | ||
19a39dce ID |
3341 | if (counting) { |
3342 | spin_lock(&fs_info->balance_lock); | |
3343 | bctl->stat.expected++; | |
3344 | spin_unlock(&fs_info->balance_lock); | |
3345 | goto loop; | |
3346 | } | |
3347 | ||
ec44a35c | 3348 | ret = btrfs_relocate_chunk(chunk_root, |
ec44a35c CM |
3349 | found_key.objectid, |
3350 | found_key.offset); | |
508794eb JB |
3351 | if (ret && ret != -ENOSPC) |
3352 | goto error; | |
19a39dce | 3353 | if (ret == -ENOSPC) { |
c9e9f97b | 3354 | enospc_errors++; |
19a39dce ID |
3355 | } else { |
3356 | spin_lock(&fs_info->balance_lock); | |
3357 | bctl->stat.completed++; | |
3358 | spin_unlock(&fs_info->balance_lock); | |
3359 | } | |
f43ffb60 | 3360 | loop: |
795a3321 ID |
3361 | if (found_key.offset == 0) |
3362 | break; | |
ba1bf481 | 3363 | key.offset = found_key.offset - 1; |
ec44a35c | 3364 | } |
c9e9f97b | 3365 | |
19a39dce ID |
3366 | if (counting) { |
3367 | btrfs_release_path(path); | |
3368 | counting = false; | |
3369 | goto again; | |
3370 | } | |
ec44a35c CM |
3371 | error: |
3372 | btrfs_free_path(path); | |
c9e9f97b | 3373 | if (enospc_errors) { |
efe120a0 | 3374 | btrfs_info(fs_info, "%d enospc errors during balance", |
c9e9f97b ID |
3375 | enospc_errors); |
3376 | if (!ret) | |
3377 | ret = -ENOSPC; | |
3378 | } | |
3379 | ||
ec44a35c CM |
3380 | return ret; |
3381 | } | |
3382 | ||
0c460c0d ID |
3383 | /** |
3384 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3385 | * @flags: profile to validate | |
3386 | * @extended: if true @flags is treated as an extended profile | |
3387 | */ | |
3388 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3389 | { | |
3390 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3391 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3392 | ||
3393 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3394 | ||
3395 | /* 1) check that all other bits are zeroed */ | |
3396 | if (flags & ~mask) | |
3397 | return 0; | |
3398 | ||
3399 | /* 2) see if profile is reduced */ | |
3400 | if (flags == 0) | |
3401 | return !extended; /* "0" is valid for usual profiles */ | |
3402 | ||
3403 | /* true if exactly one bit set */ | |
3404 | return (flags & (flags - 1)) == 0; | |
3405 | } | |
3406 | ||
837d5b6e ID |
3407 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3408 | { | |
a7e99c69 ID |
3409 | /* cancel requested || normal exit path */ |
3410 | return atomic_read(&fs_info->balance_cancel_req) || | |
3411 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3412 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3413 | } |
3414 | ||
c9e9f97b ID |
3415 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3416 | { | |
0940ebf6 ID |
3417 | int ret; |
3418 | ||
c9e9f97b | 3419 | unset_balance_control(fs_info); |
0940ebf6 | 3420 | ret = del_balance_item(fs_info->tree_root); |
0f788c58 LB |
3421 | if (ret) |
3422 | btrfs_std_error(fs_info, ret); | |
ed0fb78f ID |
3423 | |
3424 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3425 | } |
3426 | ||
c9e9f97b ID |
3427 | /* |
3428 | * Should be called with both balance and volume mutexes held | |
3429 | */ | |
3430 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3431 | struct btrfs_ioctl_balance_args *bargs) | |
3432 | { | |
3433 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3434 | u64 allowed; |
e4837f8f | 3435 | int mixed = 0; |
c9e9f97b | 3436 | int ret; |
8dabb742 | 3437 | u64 num_devices; |
de98ced9 | 3438 | unsigned seq; |
c9e9f97b | 3439 | |
837d5b6e | 3440 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3441 | atomic_read(&fs_info->balance_pause_req) || |
3442 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3443 | ret = -EINVAL; |
3444 | goto out; | |
3445 | } | |
3446 | ||
e4837f8f ID |
3447 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3448 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3449 | mixed = 1; | |
3450 | ||
f43ffb60 ID |
3451 | /* |
3452 | * In case of mixed groups both data and meta should be picked, | |
3453 | * and identical options should be given for both of them. | |
3454 | */ | |
e4837f8f ID |
3455 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3456 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3457 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3458 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3459 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
efe120a0 FH |
3460 | btrfs_err(fs_info, "with mixed groups data and " |
3461 | "metadata balance options must be the same"); | |
f43ffb60 ID |
3462 | ret = -EINVAL; |
3463 | goto out; | |
3464 | } | |
3465 | } | |
3466 | ||
8dabb742 SB |
3467 | num_devices = fs_info->fs_devices->num_devices; |
3468 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
3469 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
3470 | BUG_ON(num_devices < 1); | |
3471 | num_devices--; | |
3472 | } | |
3473 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
e4d8ec0f | 3474 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE; |
8dabb742 | 3475 | if (num_devices == 1) |
e4d8ec0f | 3476 | allowed |= BTRFS_BLOCK_GROUP_DUP; |
8250dabe | 3477 | else if (num_devices > 1) |
e4d8ec0f | 3478 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3479 | if (num_devices > 2) |
3480 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3481 | if (num_devices > 3) | |
3482 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3483 | BTRFS_BLOCK_GROUP_RAID6); | |
6728b198 ID |
3484 | if ((bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3485 | (!alloc_profile_is_valid(bctl->data.target, 1) || | |
3486 | (bctl->data.target & ~allowed))) { | |
efe120a0 FH |
3487 | btrfs_err(fs_info, "unable to start balance with target " |
3488 | "data profile %llu", | |
c1c9ff7c | 3489 | bctl->data.target); |
e4d8ec0f ID |
3490 | ret = -EINVAL; |
3491 | goto out; | |
3492 | } | |
6728b198 ID |
3493 | if ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3494 | (!alloc_profile_is_valid(bctl->meta.target, 1) || | |
3495 | (bctl->meta.target & ~allowed))) { | |
efe120a0 FH |
3496 | btrfs_err(fs_info, |
3497 | "unable to start balance with target metadata profile %llu", | |
c1c9ff7c | 3498 | bctl->meta.target); |
e4d8ec0f ID |
3499 | ret = -EINVAL; |
3500 | goto out; | |
3501 | } | |
6728b198 ID |
3502 | if ((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && |
3503 | (!alloc_profile_is_valid(bctl->sys.target, 1) || | |
3504 | (bctl->sys.target & ~allowed))) { | |
efe120a0 FH |
3505 | btrfs_err(fs_info, |
3506 | "unable to start balance with target system profile %llu", | |
c1c9ff7c | 3507 | bctl->sys.target); |
e4d8ec0f ID |
3508 | ret = -EINVAL; |
3509 | goto out; | |
3510 | } | |
3511 | ||
e4837f8f ID |
3512 | /* allow dup'ed data chunks only in mixed mode */ |
3513 | if (!mixed && (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
6728b198 | 3514 | (bctl->data.target & BTRFS_BLOCK_GROUP_DUP)) { |
efe120a0 | 3515 | btrfs_err(fs_info, "dup for data is not allowed"); |
e4d8ec0f ID |
3516 | ret = -EINVAL; |
3517 | goto out; | |
3518 | } | |
3519 | ||
3520 | /* allow to reduce meta or sys integrity only if force set */ | |
3521 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3522 | BTRFS_BLOCK_GROUP_RAID10 | |
3523 | BTRFS_BLOCK_GROUP_RAID5 | | |
3524 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3525 | do { |
3526 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3527 | ||
3528 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3529 | (fs_info->avail_system_alloc_bits & allowed) && | |
3530 | !(bctl->sys.target & allowed)) || | |
3531 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3532 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3533 | !(bctl->meta.target & allowed))) { | |
3534 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
efe120a0 | 3535 | btrfs_info(fs_info, "force reducing metadata integrity"); |
de98ced9 | 3536 | } else { |
efe120a0 FH |
3537 | btrfs_err(fs_info, "balance will reduce metadata " |
3538 | "integrity, use force if you want this"); | |
de98ced9 MX |
3539 | ret = -EINVAL; |
3540 | goto out; | |
3541 | } | |
e4d8ec0f | 3542 | } |
de98ced9 | 3543 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3544 | |
5af3e8cc SB |
3545 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3546 | int num_tolerated_disk_barrier_failures; | |
3547 | u64 target = bctl->sys.target; | |
3548 | ||
3549 | num_tolerated_disk_barrier_failures = | |
3550 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3551 | if (num_tolerated_disk_barrier_failures > 0 && | |
3552 | (target & | |
3553 | (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID0 | | |
3554 | BTRFS_AVAIL_ALLOC_BIT_SINGLE))) | |
3555 | num_tolerated_disk_barrier_failures = 0; | |
3556 | else if (num_tolerated_disk_barrier_failures > 1 && | |
3557 | (target & | |
3558 | (BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10))) | |
3559 | num_tolerated_disk_barrier_failures = 1; | |
3560 | ||
3561 | fs_info->num_tolerated_disk_barrier_failures = | |
3562 | num_tolerated_disk_barrier_failures; | |
3563 | } | |
3564 | ||
0940ebf6 | 3565 | ret = insert_balance_item(fs_info->tree_root, bctl); |
59641015 | 3566 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3567 | goto out; |
3568 | ||
59641015 ID |
3569 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3570 | BUG_ON(ret == -EEXIST); | |
3571 | set_balance_control(bctl); | |
3572 | } else { | |
3573 | BUG_ON(ret != -EEXIST); | |
3574 | spin_lock(&fs_info->balance_lock); | |
3575 | update_balance_args(bctl); | |
3576 | spin_unlock(&fs_info->balance_lock); | |
3577 | } | |
c9e9f97b | 3578 | |
837d5b6e | 3579 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3580 | mutex_unlock(&fs_info->balance_mutex); |
3581 | ||
3582 | ret = __btrfs_balance(fs_info); | |
3583 | ||
3584 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3585 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3586 | |
bf023ecf ID |
3587 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3588 | fs_info->num_tolerated_disk_barrier_failures = | |
3589 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3590 | } | |
3591 | ||
c9e9f97b ID |
3592 | if (bargs) { |
3593 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3594 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3595 | } |
3596 | ||
3a01aa7a ID |
3597 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3598 | balance_need_close(fs_info)) { | |
3599 | __cancel_balance(fs_info); | |
3600 | } | |
3601 | ||
837d5b6e | 3602 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3603 | |
3604 | return ret; | |
3605 | out: | |
59641015 ID |
3606 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3607 | __cancel_balance(fs_info); | |
ed0fb78f | 3608 | else { |
59641015 | 3609 | kfree(bctl); |
ed0fb78f ID |
3610 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3611 | } | |
59641015 ID |
3612 | return ret; |
3613 | } | |
3614 | ||
3615 | static int balance_kthread(void *data) | |
3616 | { | |
2b6ba629 | 3617 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3618 | int ret = 0; |
59641015 ID |
3619 | |
3620 | mutex_lock(&fs_info->volume_mutex); | |
3621 | mutex_lock(&fs_info->balance_mutex); | |
3622 | ||
2b6ba629 | 3623 | if (fs_info->balance_ctl) { |
efe120a0 | 3624 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3625 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3626 | } |
59641015 ID |
3627 | |
3628 | mutex_unlock(&fs_info->balance_mutex); | |
3629 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3630 | |
59641015 ID |
3631 | return ret; |
3632 | } | |
3633 | ||
2b6ba629 ID |
3634 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3635 | { | |
3636 | struct task_struct *tsk; | |
3637 | ||
3638 | spin_lock(&fs_info->balance_lock); | |
3639 | if (!fs_info->balance_ctl) { | |
3640 | spin_unlock(&fs_info->balance_lock); | |
3641 | return 0; | |
3642 | } | |
3643 | spin_unlock(&fs_info->balance_lock); | |
3644 | ||
3645 | if (btrfs_test_opt(fs_info->tree_root, SKIP_BALANCE)) { | |
efe120a0 | 3646 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3647 | return 0; |
3648 | } | |
3649 | ||
3650 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3651 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3652 | } |
3653 | ||
68310a5e | 3654 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3655 | { |
59641015 ID |
3656 | struct btrfs_balance_control *bctl; |
3657 | struct btrfs_balance_item *item; | |
3658 | struct btrfs_disk_balance_args disk_bargs; | |
3659 | struct btrfs_path *path; | |
3660 | struct extent_buffer *leaf; | |
3661 | struct btrfs_key key; | |
3662 | int ret; | |
3663 | ||
3664 | path = btrfs_alloc_path(); | |
3665 | if (!path) | |
3666 | return -ENOMEM; | |
3667 | ||
59641015 ID |
3668 | key.objectid = BTRFS_BALANCE_OBJECTID; |
3669 | key.type = BTRFS_BALANCE_ITEM_KEY; | |
3670 | key.offset = 0; | |
3671 | ||
68310a5e | 3672 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3673 | if (ret < 0) |
68310a5e | 3674 | goto out; |
59641015 ID |
3675 | if (ret > 0) { /* ret = -ENOENT; */ |
3676 | ret = 0; | |
68310a5e ID |
3677 | goto out; |
3678 | } | |
3679 | ||
3680 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3681 | if (!bctl) { | |
3682 | ret = -ENOMEM; | |
3683 | goto out; | |
59641015 ID |
3684 | } |
3685 | ||
3686 | leaf = path->nodes[0]; | |
3687 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3688 | ||
68310a5e ID |
3689 | bctl->fs_info = fs_info; |
3690 | bctl->flags = btrfs_balance_flags(leaf, item); | |
3691 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
3692 | |
3693 | btrfs_balance_data(leaf, item, &disk_bargs); | |
3694 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
3695 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
3696 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
3697 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
3698 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
3699 | ||
ed0fb78f ID |
3700 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
3701 | ||
68310a5e ID |
3702 | mutex_lock(&fs_info->volume_mutex); |
3703 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 3704 | |
68310a5e ID |
3705 | set_balance_control(bctl); |
3706 | ||
3707 | mutex_unlock(&fs_info->balance_mutex); | |
3708 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
3709 | out: |
3710 | btrfs_free_path(path); | |
ec44a35c CM |
3711 | return ret; |
3712 | } | |
3713 | ||
837d5b6e ID |
3714 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
3715 | { | |
3716 | int ret = 0; | |
3717 | ||
3718 | mutex_lock(&fs_info->balance_mutex); | |
3719 | if (!fs_info->balance_ctl) { | |
3720 | mutex_unlock(&fs_info->balance_mutex); | |
3721 | return -ENOTCONN; | |
3722 | } | |
3723 | ||
3724 | if (atomic_read(&fs_info->balance_running)) { | |
3725 | atomic_inc(&fs_info->balance_pause_req); | |
3726 | mutex_unlock(&fs_info->balance_mutex); | |
3727 | ||
3728 | wait_event(fs_info->balance_wait_q, | |
3729 | atomic_read(&fs_info->balance_running) == 0); | |
3730 | ||
3731 | mutex_lock(&fs_info->balance_mutex); | |
3732 | /* we are good with balance_ctl ripped off from under us */ | |
3733 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
3734 | atomic_dec(&fs_info->balance_pause_req); | |
3735 | } else { | |
3736 | ret = -ENOTCONN; | |
3737 | } | |
3738 | ||
3739 | mutex_unlock(&fs_info->balance_mutex); | |
3740 | return ret; | |
3741 | } | |
3742 | ||
a7e99c69 ID |
3743 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
3744 | { | |
e649e587 ID |
3745 | if (fs_info->sb->s_flags & MS_RDONLY) |
3746 | return -EROFS; | |
3747 | ||
a7e99c69 ID |
3748 | mutex_lock(&fs_info->balance_mutex); |
3749 | if (!fs_info->balance_ctl) { | |
3750 | mutex_unlock(&fs_info->balance_mutex); | |
3751 | return -ENOTCONN; | |
3752 | } | |
3753 | ||
3754 | atomic_inc(&fs_info->balance_cancel_req); | |
3755 | /* | |
3756 | * if we are running just wait and return, balance item is | |
3757 | * deleted in btrfs_balance in this case | |
3758 | */ | |
3759 | if (atomic_read(&fs_info->balance_running)) { | |
3760 | mutex_unlock(&fs_info->balance_mutex); | |
3761 | wait_event(fs_info->balance_wait_q, | |
3762 | atomic_read(&fs_info->balance_running) == 0); | |
3763 | mutex_lock(&fs_info->balance_mutex); | |
3764 | } else { | |
3765 | /* __cancel_balance needs volume_mutex */ | |
3766 | mutex_unlock(&fs_info->balance_mutex); | |
3767 | mutex_lock(&fs_info->volume_mutex); | |
3768 | mutex_lock(&fs_info->balance_mutex); | |
3769 | ||
3770 | if (fs_info->balance_ctl) | |
3771 | __cancel_balance(fs_info); | |
3772 | ||
3773 | mutex_unlock(&fs_info->volume_mutex); | |
3774 | } | |
3775 | ||
3776 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
3777 | atomic_dec(&fs_info->balance_cancel_req); | |
3778 | mutex_unlock(&fs_info->balance_mutex); | |
3779 | return 0; | |
3780 | } | |
3781 | ||
803b2f54 SB |
3782 | static int btrfs_uuid_scan_kthread(void *data) |
3783 | { | |
3784 | struct btrfs_fs_info *fs_info = data; | |
3785 | struct btrfs_root *root = fs_info->tree_root; | |
3786 | struct btrfs_key key; | |
3787 | struct btrfs_key max_key; | |
3788 | struct btrfs_path *path = NULL; | |
3789 | int ret = 0; | |
3790 | struct extent_buffer *eb; | |
3791 | int slot; | |
3792 | struct btrfs_root_item root_item; | |
3793 | u32 item_size; | |
f45388f3 | 3794 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
3795 | |
3796 | path = btrfs_alloc_path(); | |
3797 | if (!path) { | |
3798 | ret = -ENOMEM; | |
3799 | goto out; | |
3800 | } | |
3801 | ||
3802 | key.objectid = 0; | |
3803 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3804 | key.offset = 0; | |
3805 | ||
3806 | max_key.objectid = (u64)-1; | |
3807 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
3808 | max_key.offset = (u64)-1; | |
3809 | ||
803b2f54 | 3810 | while (1) { |
6174d3cb | 3811 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
3812 | if (ret) { |
3813 | if (ret > 0) | |
3814 | ret = 0; | |
3815 | break; | |
3816 | } | |
3817 | ||
3818 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
3819 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
3820 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
3821 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
3822 | goto skip; | |
3823 | ||
3824 | eb = path->nodes[0]; | |
3825 | slot = path->slots[0]; | |
3826 | item_size = btrfs_item_size_nr(eb, slot); | |
3827 | if (item_size < sizeof(root_item)) | |
3828 | goto skip; | |
3829 | ||
803b2f54 SB |
3830 | read_extent_buffer(eb, &root_item, |
3831 | btrfs_item_ptr_offset(eb, slot), | |
3832 | (int)sizeof(root_item)); | |
3833 | if (btrfs_root_refs(&root_item) == 0) | |
3834 | goto skip; | |
f45388f3 FDBM |
3835 | |
3836 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
3837 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
3838 | if (trans) | |
3839 | goto update_tree; | |
3840 | ||
3841 | btrfs_release_path(path); | |
803b2f54 SB |
3842 | /* |
3843 | * 1 - subvol uuid item | |
3844 | * 1 - received_subvol uuid item | |
3845 | */ | |
3846 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
3847 | if (IS_ERR(trans)) { | |
3848 | ret = PTR_ERR(trans); | |
3849 | break; | |
3850 | } | |
f45388f3 FDBM |
3851 | continue; |
3852 | } else { | |
3853 | goto skip; | |
3854 | } | |
3855 | update_tree: | |
3856 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
803b2f54 SB |
3857 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3858 | root_item.uuid, | |
3859 | BTRFS_UUID_KEY_SUBVOL, | |
3860 | key.objectid); | |
3861 | if (ret < 0) { | |
efe120a0 | 3862 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3863 | ret); |
803b2f54 SB |
3864 | break; |
3865 | } | |
3866 | } | |
3867 | ||
3868 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
803b2f54 SB |
3869 | ret = btrfs_uuid_tree_add(trans, fs_info->uuid_root, |
3870 | root_item.received_uuid, | |
3871 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
3872 | key.objectid); | |
3873 | if (ret < 0) { | |
efe120a0 | 3874 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 3875 | ret); |
803b2f54 SB |
3876 | break; |
3877 | } | |
3878 | } | |
3879 | ||
f45388f3 | 3880 | skip: |
803b2f54 SB |
3881 | if (trans) { |
3882 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 3883 | trans = NULL; |
803b2f54 SB |
3884 | if (ret) |
3885 | break; | |
3886 | } | |
3887 | ||
803b2f54 SB |
3888 | btrfs_release_path(path); |
3889 | if (key.offset < (u64)-1) { | |
3890 | key.offset++; | |
3891 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
3892 | key.offset = 0; | |
3893 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3894 | } else if (key.objectid < (u64)-1) { | |
3895 | key.offset = 0; | |
3896 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3897 | key.objectid++; | |
3898 | } else { | |
3899 | break; | |
3900 | } | |
3901 | cond_resched(); | |
3902 | } | |
3903 | ||
3904 | out: | |
3905 | btrfs_free_path(path); | |
f45388f3 FDBM |
3906 | if (trans && !IS_ERR(trans)) |
3907 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 3908 | if (ret) |
efe120a0 | 3909 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 SB |
3910 | else |
3911 | fs_info->update_uuid_tree_gen = 1; | |
803b2f54 SB |
3912 | up(&fs_info->uuid_tree_rescan_sem); |
3913 | return 0; | |
3914 | } | |
3915 | ||
70f80175 SB |
3916 | /* |
3917 | * Callback for btrfs_uuid_tree_iterate(). | |
3918 | * returns: | |
3919 | * 0 check succeeded, the entry is not outdated. | |
3920 | * < 0 if an error occured. | |
3921 | * > 0 if the check failed, which means the caller shall remove the entry. | |
3922 | */ | |
3923 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
3924 | u8 *uuid, u8 type, u64 subid) | |
3925 | { | |
3926 | struct btrfs_key key; | |
3927 | int ret = 0; | |
3928 | struct btrfs_root *subvol_root; | |
3929 | ||
3930 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
3931 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
3932 | goto out; | |
3933 | ||
3934 | key.objectid = subid; | |
3935 | key.type = BTRFS_ROOT_ITEM_KEY; | |
3936 | key.offset = (u64)-1; | |
3937 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
3938 | if (IS_ERR(subvol_root)) { | |
3939 | ret = PTR_ERR(subvol_root); | |
3940 | if (ret == -ENOENT) | |
3941 | ret = 1; | |
3942 | goto out; | |
3943 | } | |
3944 | ||
3945 | switch (type) { | |
3946 | case BTRFS_UUID_KEY_SUBVOL: | |
3947 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
3948 | ret = 1; | |
3949 | break; | |
3950 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
3951 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
3952 | BTRFS_UUID_SIZE)) | |
3953 | ret = 1; | |
3954 | break; | |
3955 | } | |
3956 | ||
3957 | out: | |
3958 | return ret; | |
3959 | } | |
3960 | ||
3961 | static int btrfs_uuid_rescan_kthread(void *data) | |
3962 | { | |
3963 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
3964 | int ret; | |
3965 | ||
3966 | /* | |
3967 | * 1st step is to iterate through the existing UUID tree and | |
3968 | * to delete all entries that contain outdated data. | |
3969 | * 2nd step is to add all missing entries to the UUID tree. | |
3970 | */ | |
3971 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
3972 | if (ret < 0) { | |
efe120a0 | 3973 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
3974 | up(&fs_info->uuid_tree_rescan_sem); |
3975 | return ret; | |
3976 | } | |
3977 | return btrfs_uuid_scan_kthread(data); | |
3978 | } | |
3979 | ||
f7a81ea4 SB |
3980 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
3981 | { | |
3982 | struct btrfs_trans_handle *trans; | |
3983 | struct btrfs_root *tree_root = fs_info->tree_root; | |
3984 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
3985 | struct task_struct *task; |
3986 | int ret; | |
f7a81ea4 SB |
3987 | |
3988 | /* | |
3989 | * 1 - root node | |
3990 | * 1 - root item | |
3991 | */ | |
3992 | trans = btrfs_start_transaction(tree_root, 2); | |
3993 | if (IS_ERR(trans)) | |
3994 | return PTR_ERR(trans); | |
3995 | ||
3996 | uuid_root = btrfs_create_tree(trans, fs_info, | |
3997 | BTRFS_UUID_TREE_OBJECTID); | |
3998 | if (IS_ERR(uuid_root)) { | |
6d13f549 DS |
3999 | ret = PTR_ERR(uuid_root); |
4000 | btrfs_abort_transaction(trans, tree_root, ret); | |
4001 | return ret; | |
f7a81ea4 SB |
4002 | } |
4003 | ||
4004 | fs_info->uuid_root = uuid_root; | |
4005 | ||
803b2f54 SB |
4006 | ret = btrfs_commit_transaction(trans, tree_root); |
4007 | if (ret) | |
4008 | return ret; | |
4009 | ||
4010 | down(&fs_info->uuid_tree_rescan_sem); | |
4011 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4012 | if (IS_ERR(task)) { | |
70f80175 | 4013 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4014 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4015 | up(&fs_info->uuid_tree_rescan_sem); |
4016 | return PTR_ERR(task); | |
4017 | } | |
4018 | ||
4019 | return 0; | |
f7a81ea4 | 4020 | } |
803b2f54 | 4021 | |
70f80175 SB |
4022 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4023 | { | |
4024 | struct task_struct *task; | |
4025 | ||
4026 | down(&fs_info->uuid_tree_rescan_sem); | |
4027 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4028 | if (IS_ERR(task)) { | |
4029 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4030 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4031 | up(&fs_info->uuid_tree_rescan_sem); |
4032 | return PTR_ERR(task); | |
4033 | } | |
4034 | ||
4035 | return 0; | |
4036 | } | |
4037 | ||
8f18cf13 CM |
4038 | /* |
4039 | * shrinking a device means finding all of the device extents past | |
4040 | * the new size, and then following the back refs to the chunks. | |
4041 | * The chunk relocation code actually frees the device extent | |
4042 | */ | |
4043 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4044 | { | |
4045 | struct btrfs_trans_handle *trans; | |
4046 | struct btrfs_root *root = device->dev_root; | |
4047 | struct btrfs_dev_extent *dev_extent = NULL; | |
4048 | struct btrfs_path *path; | |
4049 | u64 length; | |
8f18cf13 CM |
4050 | u64 chunk_objectid; |
4051 | u64 chunk_offset; | |
4052 | int ret; | |
4053 | int slot; | |
ba1bf481 JB |
4054 | int failed = 0; |
4055 | bool retried = false; | |
53e489bc | 4056 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4057 | struct extent_buffer *l; |
4058 | struct btrfs_key key; | |
6c41761f | 4059 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
8f18cf13 | 4060 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4061 | u64 old_size = btrfs_device_get_total_bytes(device); |
4062 | u64 diff = old_size - new_size; | |
8f18cf13 | 4063 | |
63a212ab SB |
4064 | if (device->is_tgtdev_for_dev_replace) |
4065 | return -EINVAL; | |
4066 | ||
8f18cf13 CM |
4067 | path = btrfs_alloc_path(); |
4068 | if (!path) | |
4069 | return -ENOMEM; | |
4070 | ||
8f18cf13 CM |
4071 | path->reada = 2; |
4072 | ||
7d9eb12c CM |
4073 | lock_chunks(root); |
4074 | ||
7cc8e58d | 4075 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4076 | if (device->writeable) { |
2b82032c | 4077 | device->fs_devices->total_rw_bytes -= diff; |
2bf64758 JB |
4078 | spin_lock(&root->fs_info->free_chunk_lock); |
4079 | root->fs_info->free_chunk_space -= diff; | |
4080 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4081 | } | |
7d9eb12c | 4082 | unlock_chunks(root); |
8f18cf13 | 4083 | |
ba1bf481 | 4084 | again: |
8f18cf13 CM |
4085 | key.objectid = device->devid; |
4086 | key.offset = (u64)-1; | |
4087 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4088 | ||
213e64da | 4089 | do { |
8f18cf13 CM |
4090 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
4091 | if (ret < 0) | |
4092 | goto done; | |
4093 | ||
4094 | ret = btrfs_previous_item(root, path, 0, key.type); | |
4095 | if (ret < 0) | |
4096 | goto done; | |
4097 | if (ret) { | |
4098 | ret = 0; | |
b3b4aa74 | 4099 | btrfs_release_path(path); |
bf1fb512 | 4100 | break; |
8f18cf13 CM |
4101 | } |
4102 | ||
4103 | l = path->nodes[0]; | |
4104 | slot = path->slots[0]; | |
4105 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4106 | ||
ba1bf481 | 4107 | if (key.objectid != device->devid) { |
b3b4aa74 | 4108 | btrfs_release_path(path); |
bf1fb512 | 4109 | break; |
ba1bf481 | 4110 | } |
8f18cf13 CM |
4111 | |
4112 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4113 | length = btrfs_dev_extent_length(l, dev_extent); | |
4114 | ||
ba1bf481 | 4115 | if (key.offset + length <= new_size) { |
b3b4aa74 | 4116 | btrfs_release_path(path); |
d6397bae | 4117 | break; |
ba1bf481 | 4118 | } |
8f18cf13 | 4119 | |
8f18cf13 CM |
4120 | chunk_objectid = btrfs_dev_extent_chunk_objectid(l, dev_extent); |
4121 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); | |
b3b4aa74 | 4122 | btrfs_release_path(path); |
8f18cf13 | 4123 | |
a688a04a | 4124 | ret = btrfs_relocate_chunk(root, chunk_objectid, chunk_offset); |
ba1bf481 | 4125 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4126 | goto done; |
ba1bf481 JB |
4127 | if (ret == -ENOSPC) |
4128 | failed++; | |
213e64da | 4129 | } while (key.offset-- > 0); |
ba1bf481 JB |
4130 | |
4131 | if (failed && !retried) { | |
4132 | failed = 0; | |
4133 | retried = true; | |
4134 | goto again; | |
4135 | } else if (failed && retried) { | |
4136 | ret = -ENOSPC; | |
ba1bf481 | 4137 | goto done; |
8f18cf13 CM |
4138 | } |
4139 | ||
d6397bae | 4140 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4141 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4142 | if (IS_ERR(trans)) { |
4143 | ret = PTR_ERR(trans); | |
4144 | goto done; | |
4145 | } | |
4146 | ||
d6397bae | 4147 | lock_chunks(root); |
53e489bc FM |
4148 | |
4149 | /* | |
4150 | * We checked in the above loop all device extents that were already in | |
4151 | * the device tree. However before we have updated the device's | |
4152 | * total_bytes to the new size, we might have had chunk allocations that | |
4153 | * have not complete yet (new block groups attached to transaction | |
4154 | * handles), and therefore their device extents were not yet in the | |
4155 | * device tree and we missed them in the loop above. So if we have any | |
4156 | * pending chunk using a device extent that overlaps the device range | |
4157 | * that we can not use anymore, commit the current transaction and | |
4158 | * repeat the search on the device tree - this way we guarantee we will | |
4159 | * not have chunks using device extents that end beyond 'new_size'. | |
4160 | */ | |
4161 | if (!checked_pending_chunks) { | |
4162 | u64 start = new_size; | |
4163 | u64 len = old_size - new_size; | |
4164 | ||
4165 | if (contains_pending_extent(trans, device, &start, len)) { | |
4166 | unlock_chunks(root); | |
4167 | checked_pending_chunks = true; | |
4168 | failed = 0; | |
4169 | retried = false; | |
4170 | ret = btrfs_commit_transaction(trans, root); | |
4171 | if (ret) | |
4172 | goto done; | |
4173 | goto again; | |
4174 | } | |
4175 | } | |
4176 | ||
7cc8e58d | 4177 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4178 | if (list_empty(&device->resized_list)) |
4179 | list_add_tail(&device->resized_list, | |
4180 | &root->fs_info->fs_devices->resized_devices); | |
d6397bae | 4181 | |
d6397bae CB |
4182 | WARN_ON(diff > old_total); |
4183 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
4184 | unlock_chunks(root); | |
2196d6e8 MX |
4185 | |
4186 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4187 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4188 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4189 | done: |
4190 | btrfs_free_path(path); | |
53e489bc FM |
4191 | if (ret) { |
4192 | lock_chunks(root); | |
4193 | btrfs_device_set_total_bytes(device, old_size); | |
4194 | if (device->writeable) | |
4195 | device->fs_devices->total_rw_bytes += diff; | |
4196 | spin_lock(&root->fs_info->free_chunk_lock); | |
4197 | root->fs_info->free_chunk_space += diff; | |
4198 | spin_unlock(&root->fs_info->free_chunk_lock); | |
4199 | unlock_chunks(root); | |
4200 | } | |
8f18cf13 CM |
4201 | return ret; |
4202 | } | |
4203 | ||
125ccb0a | 4204 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4205 | struct btrfs_key *key, |
4206 | struct btrfs_chunk *chunk, int item_size) | |
4207 | { | |
6c41761f | 4208 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
0b86a832 CM |
4209 | struct btrfs_disk_key disk_key; |
4210 | u32 array_size; | |
4211 | u8 *ptr; | |
4212 | ||
fe48a5c0 | 4213 | lock_chunks(root); |
0b86a832 | 4214 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4215 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 MX |
4216 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
4217 | unlock_chunks(root); | |
0b86a832 | 4218 | return -EFBIG; |
fe48a5c0 | 4219 | } |
0b86a832 CM |
4220 | |
4221 | ptr = super_copy->sys_chunk_array + array_size; | |
4222 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4223 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4224 | ptr += sizeof(disk_key); | |
4225 | memcpy(ptr, chunk, item_size); | |
4226 | item_size += sizeof(disk_key); | |
4227 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
fe48a5c0 MX |
4228 | unlock_chunks(root); |
4229 | ||
0b86a832 CM |
4230 | return 0; |
4231 | } | |
4232 | ||
73c5de00 AJ |
4233 | /* |
4234 | * sort the devices in descending order by max_avail, total_avail | |
4235 | */ | |
4236 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4237 | { |
73c5de00 AJ |
4238 | const struct btrfs_device_info *di_a = a; |
4239 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4240 | |
73c5de00 | 4241 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4242 | return -1; |
73c5de00 | 4243 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4244 | return 1; |
73c5de00 AJ |
4245 | if (di_a->total_avail > di_b->total_avail) |
4246 | return -1; | |
4247 | if (di_a->total_avail < di_b->total_avail) | |
4248 | return 1; | |
4249 | return 0; | |
b2117a39 | 4250 | } |
0b86a832 | 4251 | |
e8c9f186 | 4252 | static const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
e6ec716f MX |
4253 | [BTRFS_RAID_RAID10] = { |
4254 | .sub_stripes = 2, | |
4255 | .dev_stripes = 1, | |
4256 | .devs_max = 0, /* 0 == as many as possible */ | |
4257 | .devs_min = 4, | |
4258 | .devs_increment = 2, | |
4259 | .ncopies = 2, | |
4260 | }, | |
4261 | [BTRFS_RAID_RAID1] = { | |
4262 | .sub_stripes = 1, | |
4263 | .dev_stripes = 1, | |
4264 | .devs_max = 2, | |
4265 | .devs_min = 2, | |
4266 | .devs_increment = 2, | |
4267 | .ncopies = 2, | |
4268 | }, | |
4269 | [BTRFS_RAID_DUP] = { | |
4270 | .sub_stripes = 1, | |
4271 | .dev_stripes = 2, | |
4272 | .devs_max = 1, | |
4273 | .devs_min = 1, | |
4274 | .devs_increment = 1, | |
4275 | .ncopies = 2, | |
4276 | }, | |
4277 | [BTRFS_RAID_RAID0] = { | |
4278 | .sub_stripes = 1, | |
4279 | .dev_stripes = 1, | |
4280 | .devs_max = 0, | |
4281 | .devs_min = 2, | |
4282 | .devs_increment = 1, | |
4283 | .ncopies = 1, | |
4284 | }, | |
4285 | [BTRFS_RAID_SINGLE] = { | |
4286 | .sub_stripes = 1, | |
4287 | .dev_stripes = 1, | |
4288 | .devs_max = 1, | |
4289 | .devs_min = 1, | |
4290 | .devs_increment = 1, | |
4291 | .ncopies = 1, | |
4292 | }, | |
e942f883 CM |
4293 | [BTRFS_RAID_RAID5] = { |
4294 | .sub_stripes = 1, | |
4295 | .dev_stripes = 1, | |
4296 | .devs_max = 0, | |
4297 | .devs_min = 2, | |
4298 | .devs_increment = 1, | |
4299 | .ncopies = 2, | |
4300 | }, | |
4301 | [BTRFS_RAID_RAID6] = { | |
4302 | .sub_stripes = 1, | |
4303 | .dev_stripes = 1, | |
4304 | .devs_max = 0, | |
4305 | .devs_min = 3, | |
4306 | .devs_increment = 1, | |
4307 | .ncopies = 3, | |
4308 | }, | |
31e50229 LB |
4309 | }; |
4310 | ||
53b381b3 DW |
4311 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4312 | { | |
4313 | /* TODO allow them to set a preferred stripe size */ | |
4314 | return 64 * 1024; | |
4315 | } | |
4316 | ||
4317 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4318 | { | |
ffe2d203 | 4319 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4320 | return; |
4321 | ||
ceda0864 | 4322 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4323 | } |
4324 | ||
23f8f9b7 GH |
4325 | #define BTRFS_MAX_DEVS(r) ((BTRFS_LEAF_DATA_SIZE(r) \ |
4326 | - sizeof(struct btrfs_item) \ | |
4327 | - sizeof(struct btrfs_chunk)) \ | |
4328 | / sizeof(struct btrfs_stripe) + 1) | |
4329 | ||
4330 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4331 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4332 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4333 | / sizeof(struct btrfs_stripe) + 1) | |
4334 | ||
73c5de00 | 4335 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4336 | struct btrfs_root *extent_root, u64 start, |
4337 | u64 type) | |
b2117a39 | 4338 | { |
73c5de00 AJ |
4339 | struct btrfs_fs_info *info = extent_root->fs_info; |
4340 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4341 | struct list_head *cur; | |
4342 | struct map_lookup *map = NULL; | |
4343 | struct extent_map_tree *em_tree; | |
4344 | struct extent_map *em; | |
4345 | struct btrfs_device_info *devices_info = NULL; | |
4346 | u64 total_avail; | |
4347 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4348 | int data_stripes; /* number of stripes that count for |
4349 | block group size */ | |
73c5de00 AJ |
4350 | int sub_stripes; /* sub_stripes info for map */ |
4351 | int dev_stripes; /* stripes per dev */ | |
4352 | int devs_max; /* max devs to use */ | |
4353 | int devs_min; /* min devs needed */ | |
4354 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4355 | int ncopies; /* how many copies to data has */ | |
4356 | int ret; | |
4357 | u64 max_stripe_size; | |
4358 | u64 max_chunk_size; | |
4359 | u64 stripe_size; | |
4360 | u64 num_bytes; | |
53b381b3 | 4361 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4362 | int ndevs; |
4363 | int i; | |
4364 | int j; | |
31e50229 | 4365 | int index; |
593060d7 | 4366 | |
0c460c0d | 4367 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4368 | |
73c5de00 AJ |
4369 | if (list_empty(&fs_devices->alloc_list)) |
4370 | return -ENOSPC; | |
b2117a39 | 4371 | |
31e50229 | 4372 | index = __get_raid_index(type); |
73c5de00 | 4373 | |
31e50229 LB |
4374 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4375 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4376 | devs_max = btrfs_raid_array[index].devs_max; | |
4377 | devs_min = btrfs_raid_array[index].devs_min; | |
4378 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4379 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4380 | |
9b3f68b9 | 4381 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
73c5de00 AJ |
4382 | max_stripe_size = 1024 * 1024 * 1024; |
4383 | max_chunk_size = 10 * max_stripe_size; | |
23f8f9b7 GH |
4384 | if (!devs_max) |
4385 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4386 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f CM |
4387 | /* for larger filesystems, use larger metadata chunks */ |
4388 | if (fs_devices->total_rw_bytes > 50ULL * 1024 * 1024 * 1024) | |
4389 | max_stripe_size = 1024 * 1024 * 1024; | |
4390 | else | |
4391 | max_stripe_size = 256 * 1024 * 1024; | |
73c5de00 | 4392 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4393 | if (!devs_max) |
4394 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4395 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
96bdc7dc | 4396 | max_stripe_size = 32 * 1024 * 1024; |
73c5de00 | 4397 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4398 | if (!devs_max) |
4399 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4400 | } else { |
351fd353 | 4401 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4402 | type); |
4403 | BUG_ON(1); | |
9b3f68b9 CM |
4404 | } |
4405 | ||
2b82032c YZ |
4406 | /* we don't want a chunk larger than 10% of writeable space */ |
4407 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4408 | max_chunk_size); | |
9b3f68b9 | 4409 | |
31e818fe | 4410 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4411 | GFP_NOFS); |
4412 | if (!devices_info) | |
4413 | return -ENOMEM; | |
0cad8a11 | 4414 | |
73c5de00 | 4415 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4416 | |
9f680ce0 | 4417 | /* |
73c5de00 AJ |
4418 | * in the first pass through the devices list, we gather information |
4419 | * about the available holes on each device. | |
9f680ce0 | 4420 | */ |
73c5de00 AJ |
4421 | ndevs = 0; |
4422 | while (cur != &fs_devices->alloc_list) { | |
4423 | struct btrfs_device *device; | |
4424 | u64 max_avail; | |
4425 | u64 dev_offset; | |
b2117a39 | 4426 | |
73c5de00 | 4427 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4428 | |
73c5de00 | 4429 | cur = cur->next; |
b2117a39 | 4430 | |
73c5de00 | 4431 | if (!device->writeable) { |
31b1a2bd | 4432 | WARN(1, KERN_ERR |
efe120a0 | 4433 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4434 | continue; |
4435 | } | |
b2117a39 | 4436 | |
63a212ab SB |
4437 | if (!device->in_fs_metadata || |
4438 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4439 | continue; |
b2117a39 | 4440 | |
73c5de00 AJ |
4441 | if (device->total_bytes > device->bytes_used) |
4442 | total_avail = device->total_bytes - device->bytes_used; | |
4443 | else | |
4444 | total_avail = 0; | |
38c01b96 | 4445 | |
4446 | /* If there is no space on this device, skip it. */ | |
4447 | if (total_avail == 0) | |
4448 | continue; | |
b2117a39 | 4449 | |
6df9a95e | 4450 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4451 | max_stripe_size * dev_stripes, |
4452 | &dev_offset, &max_avail); | |
4453 | if (ret && ret != -ENOSPC) | |
4454 | goto error; | |
b2117a39 | 4455 | |
73c5de00 AJ |
4456 | if (ret == 0) |
4457 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4458 | |
73c5de00 AJ |
4459 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4460 | continue; | |
b2117a39 | 4461 | |
063d006f ES |
4462 | if (ndevs == fs_devices->rw_devices) { |
4463 | WARN(1, "%s: found more than %llu devices\n", | |
4464 | __func__, fs_devices->rw_devices); | |
4465 | break; | |
4466 | } | |
73c5de00 AJ |
4467 | devices_info[ndevs].dev_offset = dev_offset; |
4468 | devices_info[ndevs].max_avail = max_avail; | |
4469 | devices_info[ndevs].total_avail = total_avail; | |
4470 | devices_info[ndevs].dev = device; | |
4471 | ++ndevs; | |
4472 | } | |
b2117a39 | 4473 | |
73c5de00 AJ |
4474 | /* |
4475 | * now sort the devices by hole size / available space | |
4476 | */ | |
4477 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4478 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4479 | |
73c5de00 AJ |
4480 | /* round down to number of usable stripes */ |
4481 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4482 | |
73c5de00 AJ |
4483 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4484 | ret = -ENOSPC; | |
4485 | goto error; | |
b2117a39 | 4486 | } |
9f680ce0 | 4487 | |
73c5de00 AJ |
4488 | if (devs_max && ndevs > devs_max) |
4489 | ndevs = devs_max; | |
4490 | /* | |
4491 | * the primary goal is to maximize the number of stripes, so use as many | |
4492 | * devices as possible, even if the stripes are not maximum sized. | |
4493 | */ | |
4494 | stripe_size = devices_info[ndevs-1].max_avail; | |
4495 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4496 | |
53b381b3 DW |
4497 | /* |
4498 | * this will have to be fixed for RAID1 and RAID10 over | |
4499 | * more drives | |
4500 | */ | |
4501 | data_stripes = num_stripes / ncopies; | |
4502 | ||
53b381b3 DW |
4503 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4504 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
4505 | btrfs_super_stripesize(info->super_copy)); | |
4506 | data_stripes = num_stripes - 1; | |
4507 | } | |
4508 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4509 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
4510 | btrfs_super_stripesize(info->super_copy)); | |
4511 | data_stripes = num_stripes - 2; | |
4512 | } | |
86db2578 CM |
4513 | |
4514 | /* | |
4515 | * Use the number of data stripes to figure out how big this chunk | |
4516 | * is really going to be in terms of logical address space, | |
4517 | * and compare that answer with the max chunk size | |
4518 | */ | |
4519 | if (stripe_size * data_stripes > max_chunk_size) { | |
4520 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4521 | |
4522 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4523 | |
4524 | /* bump the answer up to a 16MB boundary */ | |
4525 | stripe_size = (stripe_size + mask) & ~mask; | |
4526 | ||
4527 | /* but don't go higher than the limits we found | |
4528 | * while searching for free extents | |
4529 | */ | |
4530 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4531 | stripe_size = devices_info[ndevs-1].max_avail; | |
4532 | } | |
4533 | ||
b8b93add | 4534 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4535 | |
4536 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4537 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4538 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4539 | |
4540 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4541 | if (!map) { | |
4542 | ret = -ENOMEM; | |
4543 | goto error; | |
4544 | } | |
4545 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4546 | |
73c5de00 AJ |
4547 | for (i = 0; i < ndevs; ++i) { |
4548 | for (j = 0; j < dev_stripes; ++j) { | |
4549 | int s = i * dev_stripes + j; | |
4550 | map->stripes[s].dev = devices_info[i].dev; | |
4551 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4552 | j * stripe_size; | |
6324fbf3 | 4553 | } |
6324fbf3 | 4554 | } |
2b82032c | 4555 | map->sector_size = extent_root->sectorsize; |
53b381b3 DW |
4556 | map->stripe_len = raid_stripe_len; |
4557 | map->io_align = raid_stripe_len; | |
4558 | map->io_width = raid_stripe_len; | |
2b82032c | 4559 | map->type = type; |
2b82032c | 4560 | map->sub_stripes = sub_stripes; |
0b86a832 | 4561 | |
53b381b3 | 4562 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4563 | |
73c5de00 | 4564 | trace_btrfs_chunk_alloc(info->chunk_root, map, start, num_bytes); |
1abe9b8a | 4565 | |
172ddd60 | 4566 | em = alloc_extent_map(); |
2b82032c | 4567 | if (!em) { |
298a8f9c | 4568 | kfree(map); |
b2117a39 MX |
4569 | ret = -ENOMEM; |
4570 | goto error; | |
593060d7 | 4571 | } |
298a8f9c | 4572 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
2b82032c YZ |
4573 | em->bdev = (struct block_device *)map; |
4574 | em->start = start; | |
73c5de00 | 4575 | em->len = num_bytes; |
2b82032c YZ |
4576 | em->block_start = 0; |
4577 | em->block_len = em->len; | |
6df9a95e | 4578 | em->orig_block_len = stripe_size; |
593060d7 | 4579 | |
2b82032c | 4580 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
890871be | 4581 | write_lock(&em_tree->lock); |
09a2a8f9 | 4582 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4583 | if (!ret) { |
4584 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4585 | atomic_inc(&em->refs); | |
4586 | } | |
890871be | 4587 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4588 | if (ret) { |
4589 | free_extent_map(em); | |
1dd4602f | 4590 | goto error; |
0f5d42b2 | 4591 | } |
0b86a832 | 4592 | |
04487488 JB |
4593 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4594 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4595 | start, num_bytes); | |
6df9a95e JB |
4596 | if (ret) |
4597 | goto error_del_extent; | |
2b82032c | 4598 | |
7cc8e58d MX |
4599 | for (i = 0; i < map->num_stripes; i++) { |
4600 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4601 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4602 | } | |
43530c46 | 4603 | |
1c116187 MX |
4604 | spin_lock(&extent_root->fs_info->free_chunk_lock); |
4605 | extent_root->fs_info->free_chunk_space -= (stripe_size * | |
4606 | map->num_stripes); | |
4607 | spin_unlock(&extent_root->fs_info->free_chunk_lock); | |
4608 | ||
0f5d42b2 | 4609 | free_extent_map(em); |
53b381b3 DW |
4610 | check_raid56_incompat_flag(extent_root->fs_info, type); |
4611 | ||
b2117a39 | 4612 | kfree(devices_info); |
2b82032c | 4613 | return 0; |
b2117a39 | 4614 | |
6df9a95e | 4615 | error_del_extent: |
0f5d42b2 JB |
4616 | write_lock(&em_tree->lock); |
4617 | remove_extent_mapping(em_tree, em); | |
4618 | write_unlock(&em_tree->lock); | |
4619 | ||
4620 | /* One for our allocation */ | |
4621 | free_extent_map(em); | |
4622 | /* One for the tree reference */ | |
4623 | free_extent_map(em); | |
495e64f4 FM |
4624 | /* One for the pending_chunks list reference */ |
4625 | free_extent_map(em); | |
b2117a39 | 4626 | error: |
b2117a39 MX |
4627 | kfree(devices_info); |
4628 | return ret; | |
2b82032c YZ |
4629 | } |
4630 | ||
6df9a95e | 4631 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
2b82032c | 4632 | struct btrfs_root *extent_root, |
6df9a95e | 4633 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4634 | { |
2b82032c YZ |
4635 | struct btrfs_key key; |
4636 | struct btrfs_root *chunk_root = extent_root->fs_info->chunk_root; | |
4637 | struct btrfs_device *device; | |
4638 | struct btrfs_chunk *chunk; | |
4639 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4640 | struct extent_map_tree *em_tree; |
4641 | struct extent_map *em; | |
4642 | struct map_lookup *map; | |
4643 | size_t item_size; | |
4644 | u64 dev_offset; | |
4645 | u64 stripe_size; | |
4646 | int i = 0; | |
2b82032c YZ |
4647 | int ret; |
4648 | ||
6df9a95e JB |
4649 | em_tree = &extent_root->fs_info->mapping_tree.map_tree; |
4650 | read_lock(&em_tree->lock); | |
4651 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4652 | read_unlock(&em_tree->lock); | |
4653 | ||
4654 | if (!em) { | |
4655 | btrfs_crit(extent_root->fs_info, "unable to find logical " | |
4656 | "%Lu len %Lu", chunk_offset, chunk_size); | |
4657 | return -EINVAL; | |
4658 | } | |
4659 | ||
4660 | if (em->start != chunk_offset || em->len != chunk_size) { | |
4661 | btrfs_crit(extent_root->fs_info, "found a bad mapping, wanted" | |
351fd353 | 4662 | " %Lu-%Lu, found %Lu-%Lu", chunk_offset, |
6df9a95e JB |
4663 | chunk_size, em->start, em->len); |
4664 | free_extent_map(em); | |
4665 | return -EINVAL; | |
4666 | } | |
4667 | ||
4668 | map = (struct map_lookup *)em->bdev; | |
4669 | item_size = btrfs_chunk_item_size(map->num_stripes); | |
4670 | stripe_size = em->orig_block_len; | |
4671 | ||
2b82032c | 4672 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4673 | if (!chunk) { |
4674 | ret = -ENOMEM; | |
4675 | goto out; | |
4676 | } | |
4677 | ||
4678 | for (i = 0; i < map->num_stripes; i++) { | |
4679 | device = map->stripes[i].dev; | |
4680 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4681 | |
0b86a832 | 4682 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4683 | if (ret) |
6df9a95e JB |
4684 | goto out; |
4685 | ret = btrfs_alloc_dev_extent(trans, device, | |
4686 | chunk_root->root_key.objectid, | |
4687 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4688 | chunk_offset, dev_offset, | |
4689 | stripe_size); | |
4690 | if (ret) | |
4691 | goto out; | |
2b82032c YZ |
4692 | } |
4693 | ||
2b82032c | 4694 | stripe = &chunk->stripe; |
6df9a95e JB |
4695 | for (i = 0; i < map->num_stripes; i++) { |
4696 | device = map->stripes[i].dev; | |
4697 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4698 | |
e17cade2 CM |
4699 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4700 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4701 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4702 | stripe++; |
0b86a832 CM |
4703 | } |
4704 | ||
2b82032c | 4705 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4706 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4707 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4708 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4709 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4710 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4711 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b86a832 | 4712 | btrfs_set_stack_chunk_sector_size(chunk, extent_root->sectorsize); |
2b82032c | 4713 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4714 | |
2b82032c YZ |
4715 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4716 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4717 | key.offset = chunk_offset; | |
0b86a832 | 4718 | |
2b82032c | 4719 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4720 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4721 | /* | |
4722 | * TODO: Cleanup of inserted chunk root in case of | |
4723 | * failure. | |
4724 | */ | |
125ccb0a | 4725 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 4726 | item_size); |
8f18cf13 | 4727 | } |
1abe9b8a | 4728 | |
6df9a95e | 4729 | out: |
0b86a832 | 4730 | kfree(chunk); |
6df9a95e | 4731 | free_extent_map(em); |
4ed1d16e | 4732 | return ret; |
2b82032c | 4733 | } |
0b86a832 | 4734 | |
2b82032c YZ |
4735 | /* |
4736 | * Chunk allocation falls into two parts. The first part does works | |
4737 | * that make the new allocated chunk useable, but not do any operation | |
4738 | * that modifies the chunk tree. The second part does the works that | |
4739 | * require modifying the chunk tree. This division is important for the | |
4740 | * bootstrap process of adding storage to a seed btrfs. | |
4741 | */ | |
4742 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
4743 | struct btrfs_root *extent_root, u64 type) | |
4744 | { | |
4745 | u64 chunk_offset; | |
2b82032c | 4746 | |
a9629596 | 4747 | ASSERT(mutex_is_locked(&extent_root->fs_info->chunk_mutex)); |
6df9a95e JB |
4748 | chunk_offset = find_next_chunk(extent_root->fs_info); |
4749 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); | |
2b82032c YZ |
4750 | } |
4751 | ||
d397712b | 4752 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
4753 | struct btrfs_root *root, |
4754 | struct btrfs_device *device) | |
4755 | { | |
4756 | u64 chunk_offset; | |
4757 | u64 sys_chunk_offset; | |
2b82032c | 4758 | u64 alloc_profile; |
2b82032c YZ |
4759 | struct btrfs_fs_info *fs_info = root->fs_info; |
4760 | struct btrfs_root *extent_root = fs_info->extent_root; | |
4761 | int ret; | |
4762 | ||
6df9a95e | 4763 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 4764 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
4765 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
4766 | alloc_profile); | |
79787eaa JM |
4767 | if (ret) |
4768 | return ret; | |
2b82032c | 4769 | |
6df9a95e | 4770 | sys_chunk_offset = find_next_chunk(root->fs_info); |
de98ced9 | 4771 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
4772 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
4773 | alloc_profile); | |
79787eaa | 4774 | return ret; |
2b82032c YZ |
4775 | } |
4776 | ||
d20983b4 MX |
4777 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
4778 | { | |
4779 | int max_errors; | |
4780 | ||
4781 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
4782 | BTRFS_BLOCK_GROUP_RAID10 | | |
4783 | BTRFS_BLOCK_GROUP_RAID5 | | |
4784 | BTRFS_BLOCK_GROUP_DUP)) { | |
4785 | max_errors = 1; | |
4786 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
4787 | max_errors = 2; | |
4788 | } else { | |
4789 | max_errors = 0; | |
005d6427 | 4790 | } |
2b82032c | 4791 | |
d20983b4 | 4792 | return max_errors; |
2b82032c YZ |
4793 | } |
4794 | ||
4795 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
4796 | { | |
4797 | struct extent_map *em; | |
4798 | struct map_lookup *map; | |
4799 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
4800 | int readonly = 0; | |
d20983b4 | 4801 | int miss_ndevs = 0; |
2b82032c YZ |
4802 | int i; |
4803 | ||
890871be | 4804 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 4805 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 4806 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
4807 | if (!em) |
4808 | return 1; | |
4809 | ||
4810 | map = (struct map_lookup *)em->bdev; | |
4811 | for (i = 0; i < map->num_stripes; i++) { | |
d20983b4 MX |
4812 | if (map->stripes[i].dev->missing) { |
4813 | miss_ndevs++; | |
4814 | continue; | |
4815 | } | |
4816 | ||
2b82032c YZ |
4817 | if (!map->stripes[i].dev->writeable) { |
4818 | readonly = 1; | |
d20983b4 | 4819 | goto end; |
2b82032c YZ |
4820 | } |
4821 | } | |
d20983b4 MX |
4822 | |
4823 | /* | |
4824 | * If the number of missing devices is larger than max errors, | |
4825 | * we can not write the data into that chunk successfully, so | |
4826 | * set it readonly. | |
4827 | */ | |
4828 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
4829 | readonly = 1; | |
4830 | end: | |
0b86a832 | 4831 | free_extent_map(em); |
2b82032c | 4832 | return readonly; |
0b86a832 CM |
4833 | } |
4834 | ||
4835 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
4836 | { | |
a8067e02 | 4837 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
4838 | } |
4839 | ||
4840 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
4841 | { | |
4842 | struct extent_map *em; | |
4843 | ||
d397712b | 4844 | while (1) { |
890871be | 4845 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
4846 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
4847 | if (em) | |
4848 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 4849 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
4850 | if (!em) |
4851 | break; | |
0b86a832 CM |
4852 | /* once for us */ |
4853 | free_extent_map(em); | |
4854 | /* once for the tree */ | |
4855 | free_extent_map(em); | |
4856 | } | |
4857 | } | |
4858 | ||
5d964051 | 4859 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 4860 | { |
5d964051 | 4861 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
4862 | struct extent_map *em; |
4863 | struct map_lookup *map; | |
4864 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4865 | int ret; | |
4866 | ||
890871be | 4867 | read_lock(&em_tree->lock); |
f188591e | 4868 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 4869 | read_unlock(&em_tree->lock); |
f188591e | 4870 | |
fb7669b5 JB |
4871 | /* |
4872 | * We could return errors for these cases, but that could get ugly and | |
4873 | * we'd probably do the same thing which is just not do anything else | |
4874 | * and exit, so return 1 so the callers don't try to use other copies. | |
4875 | */ | |
4876 | if (!em) { | |
351fd353 | 4877 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
4878 | logical+len); |
4879 | return 1; | |
4880 | } | |
4881 | ||
4882 | if (em->start > logical || em->start + em->len < logical) { | |
ccf39f92 | 4883 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got " |
351fd353 | 4884 | "%Lu-%Lu", logical, logical+len, em->start, |
fb7669b5 | 4885 | em->start + em->len); |
7d3d1744 | 4886 | free_extent_map(em); |
fb7669b5 JB |
4887 | return 1; |
4888 | } | |
4889 | ||
f188591e CM |
4890 | map = (struct map_lookup *)em->bdev; |
4891 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) | |
4892 | ret = map->num_stripes; | |
321aecc6 CM |
4893 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
4894 | ret = map->sub_stripes; | |
53b381b3 DW |
4895 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
4896 | ret = 2; | |
4897 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
4898 | ret = 3; | |
f188591e CM |
4899 | else |
4900 | ret = 1; | |
4901 | free_extent_map(em); | |
ad6d620e SB |
4902 | |
4903 | btrfs_dev_replace_lock(&fs_info->dev_replace); | |
4904 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) | |
4905 | ret++; | |
4906 | btrfs_dev_replace_unlock(&fs_info->dev_replace); | |
4907 | ||
f188591e CM |
4908 | return ret; |
4909 | } | |
4910 | ||
53b381b3 DW |
4911 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
4912 | struct btrfs_mapping_tree *map_tree, | |
4913 | u64 logical) | |
4914 | { | |
4915 | struct extent_map *em; | |
4916 | struct map_lookup *map; | |
4917 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4918 | unsigned long len = root->sectorsize; | |
4919 | ||
4920 | read_lock(&em_tree->lock); | |
4921 | em = lookup_extent_mapping(em_tree, logical, len); | |
4922 | read_unlock(&em_tree->lock); | |
4923 | BUG_ON(!em); | |
4924 | ||
4925 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4926 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4927 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 4928 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
4929 | free_extent_map(em); |
4930 | return len; | |
4931 | } | |
4932 | ||
4933 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
4934 | u64 logical, u64 len, int mirror_num) | |
4935 | { | |
4936 | struct extent_map *em; | |
4937 | struct map_lookup *map; | |
4938 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
4939 | int ret = 0; | |
4940 | ||
4941 | read_lock(&em_tree->lock); | |
4942 | em = lookup_extent_mapping(em_tree, logical, len); | |
4943 | read_unlock(&em_tree->lock); | |
4944 | BUG_ON(!em); | |
4945 | ||
4946 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
4947 | map = (struct map_lookup *)em->bdev; | |
ffe2d203 | 4948 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
4949 | ret = 1; |
4950 | free_extent_map(em); | |
4951 | return ret; | |
4952 | } | |
4953 | ||
30d9861f SB |
4954 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
4955 | struct map_lookup *map, int first, int num, | |
4956 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
4957 | { |
4958 | int i; | |
30d9861f SB |
4959 | int tolerance; |
4960 | struct btrfs_device *srcdev; | |
4961 | ||
4962 | if (dev_replace_is_ongoing && | |
4963 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
4964 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
4965 | srcdev = fs_info->dev_replace.srcdev; | |
4966 | else | |
4967 | srcdev = NULL; | |
4968 | ||
4969 | /* | |
4970 | * try to avoid the drive that is the source drive for a | |
4971 | * dev-replace procedure, only choose it if no other non-missing | |
4972 | * mirror is available | |
4973 | */ | |
4974 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
4975 | if (map->stripes[optimal].dev->bdev && | |
4976 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
4977 | return optimal; | |
4978 | for (i = first; i < first + num; i++) { | |
4979 | if (map->stripes[i].dev->bdev && | |
4980 | (tolerance || map->stripes[i].dev != srcdev)) | |
4981 | return i; | |
4982 | } | |
dfe25020 | 4983 | } |
30d9861f | 4984 | |
dfe25020 CM |
4985 | /* we couldn't find one that doesn't fail. Just return something |
4986 | * and the io error handling code will clean up eventually | |
4987 | */ | |
4988 | return optimal; | |
4989 | } | |
4990 | ||
53b381b3 DW |
4991 | static inline int parity_smaller(u64 a, u64 b) |
4992 | { | |
4993 | return a > b; | |
4994 | } | |
4995 | ||
4996 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 4997 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
4998 | { |
4999 | struct btrfs_bio_stripe s; | |
5000 | int i; | |
5001 | u64 l; | |
5002 | int again = 1; | |
5003 | ||
5004 | while (again) { | |
5005 | again = 0; | |
cc7539ed | 5006 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5007 | if (parity_smaller(bbio->raid_map[i], |
5008 | bbio->raid_map[i+1])) { | |
53b381b3 | 5009 | s = bbio->stripes[i]; |
8e5cfb55 | 5010 | l = bbio->raid_map[i]; |
53b381b3 | 5011 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5012 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5013 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5014 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5015 | |
53b381b3 DW |
5016 | again = 1; |
5017 | } | |
5018 | } | |
5019 | } | |
5020 | } | |
5021 | ||
6e9606d2 ZL |
5022 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5023 | { | |
5024 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5025 | /* the size of the btrfs_bio */ |
6e9606d2 | 5026 | sizeof(struct btrfs_bio) + |
e57cf21e | 5027 | /* plus the variable array for the stripes */ |
6e9606d2 | 5028 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5029 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5030 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5031 | /* |
5032 | * plus the raid_map, which includes both the tgt dev | |
5033 | * and the stripes | |
5034 | */ | |
5035 | sizeof(u64) * (total_stripes), | |
6e9606d2 ZL |
5036 | GFP_NOFS); |
5037 | if (!bbio) | |
5038 | return NULL; | |
5039 | ||
5040 | atomic_set(&bbio->error, 0); | |
5041 | atomic_set(&bbio->refs, 1); | |
5042 | ||
5043 | return bbio; | |
5044 | } | |
5045 | ||
5046 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5047 | { | |
5048 | WARN_ON(!atomic_read(&bbio->refs)); | |
5049 | atomic_inc(&bbio->refs); | |
5050 | } | |
5051 | ||
5052 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5053 | { | |
5054 | if (!bbio) | |
5055 | return; | |
5056 | if (atomic_dec_and_test(&bbio->refs)) | |
5057 | kfree(bbio); | |
5058 | } | |
5059 | ||
3ec706c8 | 5060 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5061 | u64 logical, u64 *length, |
a1d3c478 | 5062 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5063 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5064 | { |
5065 | struct extent_map *em; | |
5066 | struct map_lookup *map; | |
3ec706c8 | 5067 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5068 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5069 | u64 offset; | |
593060d7 | 5070 | u64 stripe_offset; |
fce3bb9a | 5071 | u64 stripe_end_offset; |
593060d7 | 5072 | u64 stripe_nr; |
fce3bb9a LD |
5073 | u64 stripe_nr_orig; |
5074 | u64 stripe_nr_end; | |
53b381b3 | 5075 | u64 stripe_len; |
9d644a62 | 5076 | u32 stripe_index; |
cea9e445 | 5077 | int i; |
de11cc12 | 5078 | int ret = 0; |
f2d8d74d | 5079 | int num_stripes; |
a236aed1 | 5080 | int max_errors = 0; |
2c8cdd6e | 5081 | int tgtdev_indexes = 0; |
a1d3c478 | 5082 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5083 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5084 | int dev_replace_is_ongoing = 0; | |
5085 | int num_alloc_stripes; | |
ad6d620e SB |
5086 | int patch_the_first_stripe_for_dev_replace = 0; |
5087 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5088 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5089 | |
890871be | 5090 | read_lock(&em_tree->lock); |
0b86a832 | 5091 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5092 | read_unlock(&em_tree->lock); |
f2d8d74d | 5093 | |
3b951516 | 5094 | if (!em) { |
c2cf52eb | 5095 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5096 | logical, *length); |
9bb91873 JB |
5097 | return -EINVAL; |
5098 | } | |
5099 | ||
5100 | if (em->start > logical || em->start + em->len < logical) { | |
5101 | btrfs_crit(fs_info, "found a bad mapping, wanted %Lu, " | |
351fd353 | 5102 | "found %Lu-%Lu", logical, em->start, |
9bb91873 | 5103 | em->start + em->len); |
7d3d1744 | 5104 | free_extent_map(em); |
9bb91873 | 5105 | return -EINVAL; |
3b951516 | 5106 | } |
0b86a832 | 5107 | |
0b86a832 CM |
5108 | map = (struct map_lookup *)em->bdev; |
5109 | offset = logical - em->start; | |
593060d7 | 5110 | |
53b381b3 | 5111 | stripe_len = map->stripe_len; |
593060d7 CM |
5112 | stripe_nr = offset; |
5113 | /* | |
5114 | * stripe_nr counts the total number of stripes we have to stride | |
5115 | * to get to this block | |
5116 | */ | |
47c5713f | 5117 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5118 | |
53b381b3 | 5119 | stripe_offset = stripe_nr * stripe_len; |
593060d7 CM |
5120 | BUG_ON(offset < stripe_offset); |
5121 | ||
5122 | /* stripe_offset is the offset of this block in its stripe*/ | |
5123 | stripe_offset = offset - stripe_offset; | |
5124 | ||
53b381b3 | 5125 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5126 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5127 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5128 | raid56_full_stripe_start = offset; | |
5129 | ||
5130 | /* allow a write of a full stripe, but make sure we don't | |
5131 | * allow straddling of stripes | |
5132 | */ | |
47c5713f DS |
5133 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5134 | full_stripe_len); | |
53b381b3 DW |
5135 | raid56_full_stripe_start *= full_stripe_len; |
5136 | } | |
5137 | ||
5138 | if (rw & REQ_DISCARD) { | |
5139 | /* we don't discard raid56 yet */ | |
ffe2d203 | 5140 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5141 | ret = -EOPNOTSUPP; |
5142 | goto out; | |
5143 | } | |
fce3bb9a | 5144 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5145 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5146 | u64 max_len; | |
5147 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5148 | For other RAID types and for RAID[56] reads, just allow a single | |
5149 | stripe (on a single disk). */ | |
ffe2d203 | 5150 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
53b381b3 DW |
5151 | (rw & REQ_WRITE)) { |
5152 | max_len = stripe_len * nr_data_stripes(map) - | |
5153 | (offset - raid56_full_stripe_start); | |
5154 | } else { | |
5155 | /* we limit the length of each bio to what fits in a stripe */ | |
5156 | max_len = stripe_len - stripe_offset; | |
5157 | } | |
5158 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5159 | } else { |
5160 | *length = em->len - offset; | |
5161 | } | |
f2d8d74d | 5162 | |
53b381b3 DW |
5163 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5164 | it cares about is the length */ | |
a1d3c478 | 5165 | if (!bbio_ret) |
cea9e445 CM |
5166 | goto out; |
5167 | ||
472262f3 SB |
5168 | btrfs_dev_replace_lock(dev_replace); |
5169 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); | |
5170 | if (!dev_replace_is_ongoing) | |
5171 | btrfs_dev_replace_unlock(dev_replace); | |
5172 | ||
ad6d620e SB |
5173 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
5174 | !(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) && | |
5175 | dev_replace->tgtdev != NULL) { | |
5176 | /* | |
5177 | * in dev-replace case, for repair case (that's the only | |
5178 | * case where the mirror is selected explicitly when | |
5179 | * calling btrfs_map_block), blocks left of the left cursor | |
5180 | * can also be read from the target drive. | |
5181 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5182 | * the last one to the array of stripes. For READ, it also | |
5183 | * needs to be supported using the same mirror number. | |
5184 | * If the requested block is not left of the left cursor, | |
5185 | * EIO is returned. This can happen because btrfs_num_copies() | |
5186 | * returns one more in the dev-replace case. | |
5187 | */ | |
5188 | u64 tmp_length = *length; | |
5189 | struct btrfs_bio *tmp_bbio = NULL; | |
5190 | int tmp_num_stripes; | |
5191 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5192 | int index_srcdev = 0; | |
5193 | int found = 0; | |
5194 | u64 physical_of_found = 0; | |
5195 | ||
5196 | ret = __btrfs_map_block(fs_info, REQ_GET_READ_MIRRORS, | |
8e5cfb55 | 5197 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5198 | if (ret) { |
5199 | WARN_ON(tmp_bbio != NULL); | |
5200 | goto out; | |
5201 | } | |
5202 | ||
5203 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5204 | if (mirror_num > tmp_num_stripes) { | |
5205 | /* | |
5206 | * REQ_GET_READ_MIRRORS does not contain this | |
5207 | * mirror, that means that the requested area | |
5208 | * is not left of the left cursor | |
5209 | */ | |
5210 | ret = -EIO; | |
6e9606d2 | 5211 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5212 | goto out; |
5213 | } | |
5214 | ||
5215 | /* | |
5216 | * process the rest of the function using the mirror_num | |
5217 | * of the source drive. Therefore look it up first. | |
5218 | * At the end, patch the device pointer to the one of the | |
5219 | * target drive. | |
5220 | */ | |
5221 | for (i = 0; i < tmp_num_stripes; i++) { | |
5222 | if (tmp_bbio->stripes[i].dev->devid == srcdev_devid) { | |
5223 | /* | |
5224 | * In case of DUP, in order to keep it | |
5225 | * simple, only add the mirror with the | |
5226 | * lowest physical address | |
5227 | */ | |
5228 | if (found && | |
5229 | physical_of_found <= | |
5230 | tmp_bbio->stripes[i].physical) | |
5231 | continue; | |
5232 | index_srcdev = i; | |
5233 | found = 1; | |
5234 | physical_of_found = | |
5235 | tmp_bbio->stripes[i].physical; | |
5236 | } | |
5237 | } | |
5238 | ||
5239 | if (found) { | |
5240 | mirror_num = index_srcdev + 1; | |
5241 | patch_the_first_stripe_for_dev_replace = 1; | |
5242 | physical_to_patch_in_first_stripe = physical_of_found; | |
5243 | } else { | |
5244 | WARN_ON(1); | |
5245 | ret = -EIO; | |
6e9606d2 | 5246 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5247 | goto out; |
5248 | } | |
5249 | ||
6e9606d2 | 5250 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5251 | } else if (mirror_num > map->num_stripes) { |
5252 | mirror_num = 0; | |
5253 | } | |
5254 | ||
f2d8d74d | 5255 | num_stripes = 1; |
cea9e445 | 5256 | stripe_index = 0; |
fce3bb9a | 5257 | stripe_nr_orig = stripe_nr; |
fda2832f | 5258 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5259 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5260 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5261 | (offset + *length); | |
53b381b3 | 5262 | |
fce3bb9a LD |
5263 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5264 | if (rw & REQ_DISCARD) | |
5265 | num_stripes = min_t(u64, map->num_stripes, | |
5266 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5267 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5268 | &stripe_index); | |
28e1cc7d MX |
5269 | if (!(rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS))) |
5270 | mirror_num = 1; | |
fce3bb9a | 5271 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
29a8d9a0 | 5272 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5273 | num_stripes = map->num_stripes; |
2fff734f | 5274 | else if (mirror_num) |
f188591e | 5275 | stripe_index = mirror_num - 1; |
dfe25020 | 5276 | else { |
30d9861f | 5277 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5278 | map->num_stripes, |
30d9861f SB |
5279 | current->pid % map->num_stripes, |
5280 | dev_replace_is_ongoing); | |
a1d3c478 | 5281 | mirror_num = stripe_index + 1; |
dfe25020 | 5282 | } |
2fff734f | 5283 | |
611f0e00 | 5284 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
29a8d9a0 | 5285 | if (rw & (REQ_WRITE | REQ_DISCARD | REQ_GET_READ_MIRRORS)) { |
f2d8d74d | 5286 | num_stripes = map->num_stripes; |
a1d3c478 | 5287 | } else if (mirror_num) { |
f188591e | 5288 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5289 | } else { |
5290 | mirror_num = 1; | |
5291 | } | |
2fff734f | 5292 | |
321aecc6 | 5293 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5294 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5295 | |
47c5713f | 5296 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5297 | stripe_index *= map->sub_stripes; |
5298 | ||
29a8d9a0 | 5299 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
f2d8d74d | 5300 | num_stripes = map->sub_stripes; |
fce3bb9a LD |
5301 | else if (rw & REQ_DISCARD) |
5302 | num_stripes = min_t(u64, map->sub_stripes * | |
5303 | (stripe_nr_end - stripe_nr_orig), | |
5304 | map->num_stripes); | |
321aecc6 CM |
5305 | else if (mirror_num) |
5306 | stripe_index += mirror_num - 1; | |
dfe25020 | 5307 | else { |
3e74317a | 5308 | int old_stripe_index = stripe_index; |
30d9861f SB |
5309 | stripe_index = find_live_mirror(fs_info, map, |
5310 | stripe_index, | |
dfe25020 | 5311 | map->sub_stripes, stripe_index + |
30d9861f SB |
5312 | current->pid % map->sub_stripes, |
5313 | dev_replace_is_ongoing); | |
3e74317a | 5314 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5315 | } |
53b381b3 | 5316 | |
ffe2d203 | 5317 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5318 | if (need_raid_map && |
af8e2d1d MX |
5319 | ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5320 | mirror_num > 1)) { | |
53b381b3 | 5321 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5322 | stripe_nr = div_u64(raid56_full_stripe_start, |
5323 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5324 | |
5325 | /* RAID[56] write or recovery. Return all stripes */ | |
5326 | num_stripes = map->num_stripes; | |
5327 | max_errors = nr_parity_stripes(map); | |
5328 | ||
53b381b3 DW |
5329 | *length = map->stripe_len; |
5330 | stripe_index = 0; | |
5331 | stripe_offset = 0; | |
5332 | } else { | |
5333 | /* | |
5334 | * Mirror #0 or #1 means the original data block. | |
5335 | * Mirror #2 is RAID5 parity block. | |
5336 | * Mirror #3 is RAID6 Q block. | |
5337 | */ | |
47c5713f DS |
5338 | stripe_nr = div_u64_rem(stripe_nr, |
5339 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5340 | if (mirror_num > 1) |
5341 | stripe_index = nr_data_stripes(map) + | |
5342 | mirror_num - 2; | |
5343 | ||
5344 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5345 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5346 | &stripe_index); | |
28e1cc7d MX |
5347 | if (!(rw & (REQ_WRITE | REQ_DISCARD | |
5348 | REQ_GET_READ_MIRRORS)) && mirror_num <= 1) | |
5349 | mirror_num = 1; | |
53b381b3 | 5350 | } |
8790d502 CM |
5351 | } else { |
5352 | /* | |
47c5713f DS |
5353 | * after this, stripe_nr is the number of stripes on this |
5354 | * device we have to walk to find the data, and stripe_index is | |
5355 | * the number of our device in the stripe array | |
8790d502 | 5356 | */ |
47c5713f DS |
5357 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5358 | &stripe_index); | |
a1d3c478 | 5359 | mirror_num = stripe_index + 1; |
8790d502 | 5360 | } |
593060d7 | 5361 | BUG_ON(stripe_index >= map->num_stripes); |
cea9e445 | 5362 | |
472262f3 | 5363 | num_alloc_stripes = num_stripes; |
ad6d620e SB |
5364 | if (dev_replace_is_ongoing) { |
5365 | if (rw & (REQ_WRITE | REQ_DISCARD)) | |
5366 | num_alloc_stripes <<= 1; | |
5367 | if (rw & REQ_GET_READ_MIRRORS) | |
5368 | num_alloc_stripes++; | |
2c8cdd6e | 5369 | tgtdev_indexes = num_stripes; |
ad6d620e | 5370 | } |
2c8cdd6e | 5371 | |
6e9606d2 | 5372 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5373 | if (!bbio) { |
5374 | ret = -ENOMEM; | |
5375 | goto out; | |
5376 | } | |
2c8cdd6e MX |
5377 | if (dev_replace_is_ongoing) |
5378 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5379 | |
8e5cfb55 | 5380 | /* build raid_map */ |
ffe2d203 | 5381 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
8e5cfb55 ZL |
5382 | need_raid_map && ((rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) || |
5383 | mirror_num > 1)) { | |
5384 | u64 tmp; | |
9d644a62 | 5385 | unsigned rot; |
8e5cfb55 ZL |
5386 | |
5387 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5388 | sizeof(struct btrfs_bio_stripe) * | |
5389 | num_alloc_stripes + | |
5390 | sizeof(int) * tgtdev_indexes); | |
5391 | ||
5392 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5393 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5394 | |
5395 | /* Fill in the logical address of each stripe */ | |
5396 | tmp = stripe_nr * nr_data_stripes(map); | |
5397 | for (i = 0; i < nr_data_stripes(map); i++) | |
5398 | bbio->raid_map[(i+rot) % num_stripes] = | |
5399 | em->start + (tmp + i) * map->stripe_len; | |
5400 | ||
5401 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5402 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5403 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5404 | RAID6_Q_STRIPE; | |
5405 | } | |
5406 | ||
fce3bb9a | 5407 | if (rw & REQ_DISCARD) { |
9d644a62 DS |
5408 | u32 factor = 0; |
5409 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5410 | u64 stripes_per_dev = 0; |
5411 | u32 remaining_stripes = 0; | |
b89203f7 | 5412 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5413 | |
5414 | if (map->type & | |
5415 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5416 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5417 | sub_stripes = 1; | |
5418 | else | |
5419 | sub_stripes = map->sub_stripes; | |
5420 | ||
5421 | factor = map->num_stripes / sub_stripes; | |
5422 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5423 | stripe_nr_orig, | |
5424 | factor, | |
5425 | &remaining_stripes); | |
b89203f7 LB |
5426 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5427 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5428 | } |
5429 | ||
fce3bb9a | 5430 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5431 | bbio->stripes[i].physical = |
f2d8d74d CM |
5432 | map->stripes[stripe_index].physical + |
5433 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5434 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5435 | |
ec9ef7a1 LZ |
5436 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5437 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5438 | bbio->stripes[i].length = stripes_per_dev * | |
5439 | map->stripe_len; | |
b89203f7 | 5440 | |
ec9ef7a1 LZ |
5441 | if (i / sub_stripes < remaining_stripes) |
5442 | bbio->stripes[i].length += | |
5443 | map->stripe_len; | |
b89203f7 LB |
5444 | |
5445 | /* | |
5446 | * Special for the first stripe and | |
5447 | * the last stripe: | |
5448 | * | |
5449 | * |-------|...|-------| | |
5450 | * |----------| | |
5451 | * off end_off | |
5452 | */ | |
ec9ef7a1 | 5453 | if (i < sub_stripes) |
a1d3c478 | 5454 | bbio->stripes[i].length -= |
fce3bb9a | 5455 | stripe_offset; |
b89203f7 LB |
5456 | |
5457 | if (stripe_index >= last_stripe && | |
5458 | stripe_index <= (last_stripe + | |
5459 | sub_stripes - 1)) | |
a1d3c478 | 5460 | bbio->stripes[i].length -= |
fce3bb9a | 5461 | stripe_end_offset; |
b89203f7 | 5462 | |
ec9ef7a1 LZ |
5463 | if (i == sub_stripes - 1) |
5464 | stripe_offset = 0; | |
fce3bb9a | 5465 | } else |
a1d3c478 | 5466 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5467 | |
5468 | stripe_index++; | |
5469 | if (stripe_index == map->num_stripes) { | |
5470 | /* This could only happen for RAID0/10 */ | |
5471 | stripe_index = 0; | |
5472 | stripe_nr++; | |
5473 | } | |
5474 | } | |
5475 | } else { | |
5476 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5477 | bbio->stripes[i].physical = |
212a17ab LT |
5478 | map->stripes[stripe_index].physical + |
5479 | stripe_offset + | |
5480 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5481 | bbio->stripes[i].dev = |
212a17ab | 5482 | map->stripes[stripe_index].dev; |
fce3bb9a | 5483 | stripe_index++; |
f2d8d74d | 5484 | } |
593060d7 | 5485 | } |
de11cc12 | 5486 | |
d20983b4 MX |
5487 | if (rw & (REQ_WRITE | REQ_GET_READ_MIRRORS)) |
5488 | max_errors = btrfs_chunk_max_errors(map); | |
de11cc12 | 5489 | |
8e5cfb55 ZL |
5490 | if (bbio->raid_map) |
5491 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5492 | |
2c8cdd6e | 5493 | tgtdev_indexes = 0; |
472262f3 SB |
5494 | if (dev_replace_is_ongoing && (rw & (REQ_WRITE | REQ_DISCARD)) && |
5495 | dev_replace->tgtdev != NULL) { | |
5496 | int index_where_to_add; | |
5497 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5498 | ||
5499 | /* | |
5500 | * duplicate the write operations while the dev replace | |
5501 | * procedure is running. Since the copying of the old disk | |
5502 | * to the new disk takes place at run time while the | |
5503 | * filesystem is mounted writable, the regular write | |
5504 | * operations to the old disk have to be duplicated to go | |
5505 | * to the new disk as well. | |
5506 | * Note that device->missing is handled by the caller, and | |
5507 | * that the write to the old disk is already set up in the | |
5508 | * stripes array. | |
5509 | */ | |
5510 | index_where_to_add = num_stripes; | |
5511 | for (i = 0; i < num_stripes; i++) { | |
5512 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5513 | /* write to new disk, too */ | |
5514 | struct btrfs_bio_stripe *new = | |
5515 | bbio->stripes + index_where_to_add; | |
5516 | struct btrfs_bio_stripe *old = | |
5517 | bbio->stripes + i; | |
5518 | ||
5519 | new->physical = old->physical; | |
5520 | new->length = old->length; | |
5521 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5522 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5523 | index_where_to_add++; |
5524 | max_errors++; | |
2c8cdd6e | 5525 | tgtdev_indexes++; |
472262f3 SB |
5526 | } |
5527 | } | |
5528 | num_stripes = index_where_to_add; | |
ad6d620e SB |
5529 | } else if (dev_replace_is_ongoing && (rw & REQ_GET_READ_MIRRORS) && |
5530 | dev_replace->tgtdev != NULL) { | |
5531 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5532 | int index_srcdev = 0; | |
5533 | int found = 0; | |
5534 | u64 physical_of_found = 0; | |
5535 | ||
5536 | /* | |
5537 | * During the dev-replace procedure, the target drive can | |
5538 | * also be used to read data in case it is needed to repair | |
5539 | * a corrupt block elsewhere. This is possible if the | |
5540 | * requested area is left of the left cursor. In this area, | |
5541 | * the target drive is a full copy of the source drive. | |
5542 | */ | |
5543 | for (i = 0; i < num_stripes; i++) { | |
5544 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5545 | /* | |
5546 | * In case of DUP, in order to keep it | |
5547 | * simple, only add the mirror with the | |
5548 | * lowest physical address | |
5549 | */ | |
5550 | if (found && | |
5551 | physical_of_found <= | |
5552 | bbio->stripes[i].physical) | |
5553 | continue; | |
5554 | index_srcdev = i; | |
5555 | found = 1; | |
5556 | physical_of_found = bbio->stripes[i].physical; | |
5557 | } | |
5558 | } | |
5559 | if (found) { | |
258ece02 | 5560 | if (physical_of_found + map->stripe_len <= |
ad6d620e SB |
5561 | dev_replace->cursor_left) { |
5562 | struct btrfs_bio_stripe *tgtdev_stripe = | |
5563 | bbio->stripes + num_stripes; | |
5564 | ||
5565 | tgtdev_stripe->physical = physical_of_found; | |
5566 | tgtdev_stripe->length = | |
5567 | bbio->stripes[index_srcdev].length; | |
5568 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5569 | bbio->tgtdev_map[index_srcdev] = num_stripes; |
ad6d620e | 5570 | |
2c8cdd6e | 5571 | tgtdev_indexes++; |
ad6d620e SB |
5572 | num_stripes++; |
5573 | } | |
5574 | } | |
472262f3 SB |
5575 | } |
5576 | ||
de11cc12 | 5577 | *bbio_ret = bbio; |
10f11900 | 5578 | bbio->map_type = map->type; |
de11cc12 LZ |
5579 | bbio->num_stripes = num_stripes; |
5580 | bbio->max_errors = max_errors; | |
5581 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5582 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5583 | |
5584 | /* | |
5585 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5586 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5587 | * available as a mirror | |
5588 | */ | |
5589 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5590 | WARN_ON(num_stripes > 1); | |
5591 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5592 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5593 | bbio->mirror_num = map->num_stripes + 1; | |
5594 | } | |
cea9e445 | 5595 | out: |
472262f3 SB |
5596 | if (dev_replace_is_ongoing) |
5597 | btrfs_dev_replace_unlock(dev_replace); | |
0b86a832 | 5598 | free_extent_map(em); |
de11cc12 | 5599 | return ret; |
0b86a832 CM |
5600 | } |
5601 | ||
3ec706c8 | 5602 | int btrfs_map_block(struct btrfs_fs_info *fs_info, int rw, |
f2d8d74d | 5603 | u64 logical, u64 *length, |
a1d3c478 | 5604 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5605 | { |
3ec706c8 | 5606 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, |
8e5cfb55 | 5607 | mirror_num, 0); |
f2d8d74d CM |
5608 | } |
5609 | ||
af8e2d1d MX |
5610 | /* For Scrub/replace */ |
5611 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, int rw, | |
5612 | u64 logical, u64 *length, | |
5613 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5614 | int need_raid_map) |
af8e2d1d MX |
5615 | { |
5616 | return __btrfs_map_block(fs_info, rw, logical, length, bbio_ret, | |
8e5cfb55 | 5617 | mirror_num, need_raid_map); |
af8e2d1d MX |
5618 | } |
5619 | ||
a512bbf8 YZ |
5620 | int btrfs_rmap_block(struct btrfs_mapping_tree *map_tree, |
5621 | u64 chunk_start, u64 physical, u64 devid, | |
5622 | u64 **logical, int *naddrs, int *stripe_len) | |
5623 | { | |
5624 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5625 | struct extent_map *em; | |
5626 | struct map_lookup *map; | |
5627 | u64 *buf; | |
5628 | u64 bytenr; | |
5629 | u64 length; | |
5630 | u64 stripe_nr; | |
53b381b3 | 5631 | u64 rmap_len; |
a512bbf8 YZ |
5632 | int i, j, nr = 0; |
5633 | ||
890871be | 5634 | read_lock(&em_tree->lock); |
a512bbf8 | 5635 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5636 | read_unlock(&em_tree->lock); |
a512bbf8 | 5637 | |
835d974f | 5638 | if (!em) { |
efe120a0 | 5639 | printk(KERN_ERR "BTRFS: couldn't find em for chunk %Lu\n", |
835d974f JB |
5640 | chunk_start); |
5641 | return -EIO; | |
5642 | } | |
5643 | ||
5644 | if (em->start != chunk_start) { | |
efe120a0 | 5645 | printk(KERN_ERR "BTRFS: bad chunk start, em=%Lu, wanted=%Lu\n", |
835d974f JB |
5646 | em->start, chunk_start); |
5647 | free_extent_map(em); | |
5648 | return -EIO; | |
5649 | } | |
a512bbf8 YZ |
5650 | map = (struct map_lookup *)em->bdev; |
5651 | ||
5652 | length = em->len; | |
53b381b3 DW |
5653 | rmap_len = map->stripe_len; |
5654 | ||
a512bbf8 | 5655 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5656 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5657 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5658 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5659 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5660 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5661 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5662 | } | |
a512bbf8 | 5663 | |
31e818fe | 5664 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5665 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5666 | |
5667 | for (i = 0; i < map->num_stripes; i++) { | |
5668 | if (devid && map->stripes[i].dev->devid != devid) | |
5669 | continue; | |
5670 | if (map->stripes[i].physical > physical || | |
5671 | map->stripes[i].physical + length <= physical) | |
5672 | continue; | |
5673 | ||
5674 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5675 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5676 | |
5677 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5678 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5679 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5680 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5681 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5682 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5683 | * Alternatively, just use rmap_len below instead of | |
5684 | * map->stripe_len */ | |
5685 | ||
5686 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5687 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5688 | for (j = 0; j < nr; j++) { |
5689 | if (buf[j] == bytenr) | |
5690 | break; | |
5691 | } | |
934d375b CM |
5692 | if (j == nr) { |
5693 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5694 | buf[nr++] = bytenr; |
934d375b | 5695 | } |
a512bbf8 YZ |
5696 | } |
5697 | ||
a512bbf8 YZ |
5698 | *logical = buf; |
5699 | *naddrs = nr; | |
53b381b3 | 5700 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5701 | |
5702 | free_extent_map(em); | |
5703 | return 0; | |
f2d8d74d CM |
5704 | } |
5705 | ||
8408c716 MX |
5706 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio, int err) |
5707 | { | |
326e1dbb MS |
5708 | bio->bi_private = bbio->private; |
5709 | bio->bi_end_io = bbio->end_io; | |
5710 | bio_endio(bio, err); | |
5711 | ||
6e9606d2 | 5712 | btrfs_put_bbio(bbio); |
8408c716 MX |
5713 | } |
5714 | ||
a1d3c478 | 5715 | static void btrfs_end_bio(struct bio *bio, int err) |
8790d502 | 5716 | { |
9be3395b | 5717 | struct btrfs_bio *bbio = bio->bi_private; |
c404e0dc | 5718 | struct btrfs_device *dev = bbio->stripes[0].dev; |
7d2b4daa | 5719 | int is_orig_bio = 0; |
8790d502 | 5720 | |
442a4f63 | 5721 | if (err) { |
a1d3c478 | 5722 | atomic_inc(&bbio->error); |
442a4f63 SB |
5723 | if (err == -EIO || err == -EREMOTEIO) { |
5724 | unsigned int stripe_index = | |
9be3395b | 5725 | btrfs_io_bio(bio)->stripe_index; |
442a4f63 SB |
5726 | |
5727 | BUG_ON(stripe_index >= bbio->num_stripes); | |
5728 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa SB |
5729 | if (dev->bdev) { |
5730 | if (bio->bi_rw & WRITE) | |
5731 | btrfs_dev_stat_inc(dev, | |
5732 | BTRFS_DEV_STAT_WRITE_ERRS); | |
5733 | else | |
5734 | btrfs_dev_stat_inc(dev, | |
5735 | BTRFS_DEV_STAT_READ_ERRS); | |
5736 | if ((bio->bi_rw & WRITE_FLUSH) == WRITE_FLUSH) | |
5737 | btrfs_dev_stat_inc(dev, | |
5738 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
5739 | btrfs_dev_stat_print_on_error(dev); | |
5740 | } | |
442a4f63 SB |
5741 | } |
5742 | } | |
8790d502 | 5743 | |
a1d3c478 | 5744 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
5745 | is_orig_bio = 1; |
5746 | ||
c404e0dc MX |
5747 | btrfs_bio_counter_dec(bbio->fs_info); |
5748 | ||
a1d3c478 | 5749 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
5750 | if (!is_orig_bio) { |
5751 | bio_put(bio); | |
a1d3c478 | 5752 | bio = bbio->orig_bio; |
7d2b4daa | 5753 | } |
c7b22bb1 | 5754 | |
9be3395b | 5755 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 5756 | /* only send an error to the higher layers if it is |
53b381b3 | 5757 | * beyond the tolerance of the btrfs bio |
a236aed1 | 5758 | */ |
a1d3c478 | 5759 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
a236aed1 | 5760 | err = -EIO; |
5dbc8fca | 5761 | } else { |
1259ab75 CM |
5762 | /* |
5763 | * this bio is actually up to date, we didn't | |
5764 | * go over the max number of errors | |
5765 | */ | |
5766 | set_bit(BIO_UPTODATE, &bio->bi_flags); | |
a236aed1 | 5767 | err = 0; |
1259ab75 | 5768 | } |
c55f1396 | 5769 | |
8408c716 | 5770 | btrfs_end_bbio(bbio, bio, err); |
7d2b4daa | 5771 | } else if (!is_orig_bio) { |
8790d502 CM |
5772 | bio_put(bio); |
5773 | } | |
8790d502 CM |
5774 | } |
5775 | ||
8b712842 CM |
5776 | /* |
5777 | * see run_scheduled_bios for a description of why bios are collected for | |
5778 | * async submit. | |
5779 | * | |
5780 | * This will add one bio to the pending list for a device and make sure | |
5781 | * the work struct is scheduled. | |
5782 | */ | |
48a3b636 ES |
5783 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
5784 | struct btrfs_device *device, | |
5785 | int rw, struct bio *bio) | |
8b712842 CM |
5786 | { |
5787 | int should_queue = 1; | |
ffbd517d | 5788 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 5789 | |
53b381b3 DW |
5790 | if (device->missing || !device->bdev) { |
5791 | bio_endio(bio, -EIO); | |
5792 | return; | |
5793 | } | |
5794 | ||
8b712842 | 5795 | /* don't bother with additional async steps for reads, right now */ |
7b6d91da | 5796 | if (!(rw & REQ_WRITE)) { |
492bb6de | 5797 | bio_get(bio); |
21adbd5c | 5798 | btrfsic_submit_bio(rw, bio); |
492bb6de | 5799 | bio_put(bio); |
143bede5 | 5800 | return; |
8b712842 CM |
5801 | } |
5802 | ||
5803 | /* | |
0986fe9e | 5804 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
5805 | * higher layers. Otherwise, the async bio makes it appear we have |
5806 | * made progress against dirty pages when we've really just put it | |
5807 | * on a queue for later | |
5808 | */ | |
0986fe9e | 5809 | atomic_inc(&root->fs_info->nr_async_bios); |
492bb6de | 5810 | WARN_ON(bio->bi_next); |
8b712842 CM |
5811 | bio->bi_next = NULL; |
5812 | bio->bi_rw |= rw; | |
5813 | ||
5814 | spin_lock(&device->io_lock); | |
7b6d91da | 5815 | if (bio->bi_rw & REQ_SYNC) |
ffbd517d CM |
5816 | pending_bios = &device->pending_sync_bios; |
5817 | else | |
5818 | pending_bios = &device->pending_bios; | |
8b712842 | 5819 | |
ffbd517d CM |
5820 | if (pending_bios->tail) |
5821 | pending_bios->tail->bi_next = bio; | |
8b712842 | 5822 | |
ffbd517d CM |
5823 | pending_bios->tail = bio; |
5824 | if (!pending_bios->head) | |
5825 | pending_bios->head = bio; | |
8b712842 CM |
5826 | if (device->running_pending) |
5827 | should_queue = 0; | |
5828 | ||
5829 | spin_unlock(&device->io_lock); | |
5830 | ||
5831 | if (should_queue) | |
a8c93d4e QW |
5832 | btrfs_queue_work(root->fs_info->submit_workers, |
5833 | &device->work); | |
8b712842 CM |
5834 | } |
5835 | ||
de1ee92a JB |
5836 | static int bio_size_ok(struct block_device *bdev, struct bio *bio, |
5837 | sector_t sector) | |
5838 | { | |
5839 | struct bio_vec *prev; | |
5840 | struct request_queue *q = bdev_get_queue(bdev); | |
475bf36f | 5841 | unsigned int max_sectors = queue_max_sectors(q); |
de1ee92a JB |
5842 | struct bvec_merge_data bvm = { |
5843 | .bi_bdev = bdev, | |
5844 | .bi_sector = sector, | |
5845 | .bi_rw = bio->bi_rw, | |
5846 | }; | |
5847 | ||
fae7f21c | 5848 | if (WARN_ON(bio->bi_vcnt == 0)) |
de1ee92a | 5849 | return 1; |
de1ee92a JB |
5850 | |
5851 | prev = &bio->bi_io_vec[bio->bi_vcnt - 1]; | |
aa8b57aa | 5852 | if (bio_sectors(bio) > max_sectors) |
de1ee92a JB |
5853 | return 0; |
5854 | ||
5855 | if (!q->merge_bvec_fn) | |
5856 | return 1; | |
5857 | ||
4f024f37 | 5858 | bvm.bi_size = bio->bi_iter.bi_size - prev->bv_len; |
de1ee92a JB |
5859 | if (q->merge_bvec_fn(q, &bvm, prev) < prev->bv_len) |
5860 | return 0; | |
5861 | return 1; | |
5862 | } | |
5863 | ||
5864 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5865 | struct bio *bio, u64 physical, int dev_nr, | |
5866 | int rw, int async) | |
5867 | { | |
5868 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
5869 | ||
5870 | bio->bi_private = bbio; | |
9be3395b | 5871 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 5872 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 5873 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
5874 | #ifdef DEBUG |
5875 | { | |
5876 | struct rcu_string *name; | |
5877 | ||
5878 | rcu_read_lock(); | |
5879 | name = rcu_dereference(dev->name); | |
d1423248 | 5880 | pr_debug("btrfs_map_bio: rw %d, sector=%llu, dev=%lu " |
de1ee92a | 5881 | "(%s id %llu), size=%u\n", rw, |
1b6e4469 FF |
5882 | (u64)bio->bi_iter.bi_sector, (u_long)dev->bdev->bd_dev, |
5883 | name->str, dev->devid, bio->bi_iter.bi_size); | |
de1ee92a JB |
5884 | rcu_read_unlock(); |
5885 | } | |
5886 | #endif | |
5887 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
5888 | |
5889 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
5890 | ||
de1ee92a | 5891 | if (async) |
53b381b3 | 5892 | btrfs_schedule_bio(root, dev, rw, bio); |
de1ee92a JB |
5893 | else |
5894 | btrfsic_submit_bio(rw, bio); | |
5895 | } | |
5896 | ||
5897 | static int breakup_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, | |
5898 | struct bio *first_bio, struct btrfs_device *dev, | |
5899 | int dev_nr, int rw, int async) | |
5900 | { | |
5901 | struct bio_vec *bvec = first_bio->bi_io_vec; | |
5902 | struct bio *bio; | |
5903 | int nr_vecs = bio_get_nr_vecs(dev->bdev); | |
5904 | u64 physical = bbio->stripes[dev_nr].physical; | |
5905 | ||
5906 | again: | |
5907 | bio = btrfs_bio_alloc(dev->bdev, physical >> 9, nr_vecs, GFP_NOFS); | |
5908 | if (!bio) | |
5909 | return -ENOMEM; | |
5910 | ||
5911 | while (bvec <= (first_bio->bi_io_vec + first_bio->bi_vcnt - 1)) { | |
5912 | if (bio_add_page(bio, bvec->bv_page, bvec->bv_len, | |
5913 | bvec->bv_offset) < bvec->bv_len) { | |
4f024f37 | 5914 | u64 len = bio->bi_iter.bi_size; |
de1ee92a JB |
5915 | |
5916 | atomic_inc(&bbio->stripes_pending); | |
5917 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, | |
5918 | rw, async); | |
5919 | physical += len; | |
5920 | goto again; | |
5921 | } | |
5922 | bvec++; | |
5923 | } | |
5924 | ||
5925 | submit_stripe_bio(root, bbio, bio, physical, dev_nr, rw, async); | |
5926 | return 0; | |
5927 | } | |
5928 | ||
5929 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) | |
5930 | { | |
5931 | atomic_inc(&bbio->error); | |
5932 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
8408c716 MX |
5933 | /* Shoud be the original bio. */ |
5934 | WARN_ON(bio != bbio->orig_bio); | |
5935 | ||
9be3395b | 5936 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 5937 | bio->bi_iter.bi_sector = logical >> 9; |
8408c716 MX |
5938 | |
5939 | btrfs_end_bbio(bbio, bio, -EIO); | |
de1ee92a JB |
5940 | } |
5941 | } | |
5942 | ||
f188591e | 5943 | int btrfs_map_bio(struct btrfs_root *root, int rw, struct bio *bio, |
8b712842 | 5944 | int mirror_num, int async_submit) |
0b86a832 | 5945 | { |
0b86a832 | 5946 | struct btrfs_device *dev; |
8790d502 | 5947 | struct bio *first_bio = bio; |
4f024f37 | 5948 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
5949 | u64 length = 0; |
5950 | u64 map_length; | |
0b86a832 | 5951 | int ret; |
08da757d ZL |
5952 | int dev_nr; |
5953 | int total_devs; | |
a1d3c478 | 5954 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 5955 | |
4f024f37 | 5956 | length = bio->bi_iter.bi_size; |
0b86a832 | 5957 | map_length = length; |
cea9e445 | 5958 | |
c404e0dc | 5959 | btrfs_bio_counter_inc_blocked(root->fs_info); |
53b381b3 | 5960 | ret = __btrfs_map_block(root->fs_info, rw, logical, &map_length, &bbio, |
8e5cfb55 | 5961 | mirror_num, 1); |
c404e0dc MX |
5962 | if (ret) { |
5963 | btrfs_bio_counter_dec(root->fs_info); | |
79787eaa | 5964 | return ret; |
c404e0dc | 5965 | } |
cea9e445 | 5966 | |
a1d3c478 | 5967 | total_devs = bbio->num_stripes; |
53b381b3 DW |
5968 | bbio->orig_bio = first_bio; |
5969 | bbio->private = first_bio->bi_private; | |
5970 | bbio->end_io = first_bio->bi_end_io; | |
c404e0dc | 5971 | bbio->fs_info = root->fs_info; |
53b381b3 DW |
5972 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
5973 | ||
8e5cfb55 | 5974 | if (bbio->raid_map) { |
53b381b3 DW |
5975 | /* In this case, map_length has been set to the length of |
5976 | a single stripe; not the whole write */ | |
5977 | if (rw & WRITE) { | |
8e5cfb55 | 5978 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 5979 | } else { |
8e5cfb55 | 5980 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 5981 | mirror_num, 1); |
53b381b3 | 5982 | } |
4245215d | 5983 | |
c404e0dc MX |
5984 | btrfs_bio_counter_dec(root->fs_info); |
5985 | return ret; | |
53b381b3 DW |
5986 | } |
5987 | ||
cea9e445 | 5988 | if (map_length < length) { |
c2cf52eb | 5989 | btrfs_crit(root->fs_info, "mapping failed logical %llu bio len %llu len %llu", |
c1c9ff7c | 5990 | logical, length, map_length); |
cea9e445 CM |
5991 | BUG(); |
5992 | } | |
a1d3c478 | 5993 | |
08da757d | 5994 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a JB |
5995 | dev = bbio->stripes[dev_nr].dev; |
5996 | if (!dev || !dev->bdev || (rw & WRITE && !dev->writeable)) { | |
5997 | bbio_error(bbio, first_bio, logical); | |
de1ee92a JB |
5998 | continue; |
5999 | } | |
6000 | ||
6001 | /* | |
6002 | * Check and see if we're ok with this bio based on it's size | |
6003 | * and offset with the given device. | |
6004 | */ | |
6005 | if (!bio_size_ok(dev->bdev, first_bio, | |
6006 | bbio->stripes[dev_nr].physical >> 9)) { | |
6007 | ret = breakup_stripe_bio(root, bbio, first_bio, dev, | |
6008 | dev_nr, rw, async_submit); | |
6009 | BUG_ON(ret); | |
de1ee92a JB |
6010 | continue; |
6011 | } | |
6012 | ||
a1d3c478 | 6013 | if (dev_nr < total_devs - 1) { |
9be3395b | 6014 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6015 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6016 | } else |
a1d3c478 | 6017 | bio = first_bio; |
de1ee92a JB |
6018 | |
6019 | submit_stripe_bio(root, bbio, bio, | |
6020 | bbio->stripes[dev_nr].physical, dev_nr, rw, | |
6021 | async_submit); | |
8790d502 | 6022 | } |
c404e0dc | 6023 | btrfs_bio_counter_dec(root->fs_info); |
0b86a832 CM |
6024 | return 0; |
6025 | } | |
6026 | ||
aa1b8cd4 | 6027 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6028 | u8 *uuid, u8 *fsid) |
0b86a832 | 6029 | { |
2b82032c YZ |
6030 | struct btrfs_device *device; |
6031 | struct btrfs_fs_devices *cur_devices; | |
6032 | ||
aa1b8cd4 | 6033 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6034 | while (cur_devices) { |
6035 | if (!fsid || | |
6036 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6037 | device = __find_device(&cur_devices->devices, | |
6038 | devid, uuid); | |
6039 | if (device) | |
6040 | return device; | |
6041 | } | |
6042 | cur_devices = cur_devices->seed; | |
6043 | } | |
6044 | return NULL; | |
0b86a832 CM |
6045 | } |
6046 | ||
dfe25020 | 6047 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6048 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6049 | u64 devid, u8 *dev_uuid) |
6050 | { | |
6051 | struct btrfs_device *device; | |
dfe25020 | 6052 | |
12bd2fc0 ID |
6053 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6054 | if (IS_ERR(device)) | |
7cbd8a83 | 6055 | return NULL; |
12bd2fc0 ID |
6056 | |
6057 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6058 | device->fs_devices = fs_devices; |
dfe25020 | 6059 | fs_devices->num_devices++; |
12bd2fc0 ID |
6060 | |
6061 | device->missing = 1; | |
cd02dca5 | 6062 | fs_devices->missing_devices++; |
12bd2fc0 | 6063 | |
dfe25020 CM |
6064 | return device; |
6065 | } | |
6066 | ||
12bd2fc0 ID |
6067 | /** |
6068 | * btrfs_alloc_device - allocate struct btrfs_device | |
6069 | * @fs_info: used only for generating a new devid, can be NULL if | |
6070 | * devid is provided (i.e. @devid != NULL). | |
6071 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6072 | * is generated. | |
6073 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6074 | * is generated. | |
6075 | * | |
6076 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6077 | * on error. Returned struct is not linked onto any lists and can be | |
6078 | * destroyed with kfree() right away. | |
6079 | */ | |
6080 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6081 | const u64 *devid, | |
6082 | const u8 *uuid) | |
6083 | { | |
6084 | struct btrfs_device *dev; | |
6085 | u64 tmp; | |
6086 | ||
fae7f21c | 6087 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6088 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6089 | |
6090 | dev = __alloc_device(); | |
6091 | if (IS_ERR(dev)) | |
6092 | return dev; | |
6093 | ||
6094 | if (devid) | |
6095 | tmp = *devid; | |
6096 | else { | |
6097 | int ret; | |
6098 | ||
6099 | ret = find_next_devid(fs_info, &tmp); | |
6100 | if (ret) { | |
6101 | kfree(dev); | |
6102 | return ERR_PTR(ret); | |
6103 | } | |
6104 | } | |
6105 | dev->devid = tmp; | |
6106 | ||
6107 | if (uuid) | |
6108 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6109 | else | |
6110 | generate_random_uuid(dev->uuid); | |
6111 | ||
9e0af237 LB |
6112 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6113 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6114 | |
6115 | return dev; | |
6116 | } | |
6117 | ||
0b86a832 CM |
6118 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, |
6119 | struct extent_buffer *leaf, | |
6120 | struct btrfs_chunk *chunk) | |
6121 | { | |
6122 | struct btrfs_mapping_tree *map_tree = &root->fs_info->mapping_tree; | |
6123 | struct map_lookup *map; | |
6124 | struct extent_map *em; | |
6125 | u64 logical; | |
6126 | u64 length; | |
6127 | u64 devid; | |
a443755f | 6128 | u8 uuid[BTRFS_UUID_SIZE]; |
593060d7 | 6129 | int num_stripes; |
0b86a832 | 6130 | int ret; |
593060d7 | 6131 | int i; |
0b86a832 | 6132 | |
e17cade2 CM |
6133 | logical = key->offset; |
6134 | length = btrfs_chunk_length(leaf, chunk); | |
a061fc8d | 6135 | |
890871be | 6136 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6137 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6138 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6139 | |
6140 | /* already mapped? */ | |
6141 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6142 | free_extent_map(em); | |
0b86a832 CM |
6143 | return 0; |
6144 | } else if (em) { | |
6145 | free_extent_map(em); | |
6146 | } | |
0b86a832 | 6147 | |
172ddd60 | 6148 | em = alloc_extent_map(); |
0b86a832 CM |
6149 | if (!em) |
6150 | return -ENOMEM; | |
593060d7 CM |
6151 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); |
6152 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
0b86a832 CM |
6153 | if (!map) { |
6154 | free_extent_map(em); | |
6155 | return -ENOMEM; | |
6156 | } | |
6157 | ||
298a8f9c | 6158 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
0b86a832 CM |
6159 | em->bdev = (struct block_device *)map; |
6160 | em->start = logical; | |
6161 | em->len = length; | |
70c8a91c | 6162 | em->orig_start = 0; |
0b86a832 | 6163 | em->block_start = 0; |
c8b97818 | 6164 | em->block_len = em->len; |
0b86a832 | 6165 | |
593060d7 CM |
6166 | map->num_stripes = num_stripes; |
6167 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6168 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6169 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6170 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6171 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6172 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6173 | for (i = 0; i < num_stripes; i++) { |
6174 | map->stripes[i].physical = | |
6175 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6176 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6177 | read_extent_buffer(leaf, uuid, (unsigned long) |
6178 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6179 | BTRFS_UUID_SIZE); | |
aa1b8cd4 SB |
6180 | map->stripes[i].dev = btrfs_find_device(root->fs_info, devid, |
6181 | uuid, NULL); | |
dfe25020 | 6182 | if (!map->stripes[i].dev && !btrfs_test_opt(root, DEGRADED)) { |
593060d7 CM |
6183 | free_extent_map(em); |
6184 | return -EIO; | |
6185 | } | |
dfe25020 CM |
6186 | if (!map->stripes[i].dev) { |
6187 | map->stripes[i].dev = | |
5f375835 MX |
6188 | add_missing_dev(root, root->fs_info->fs_devices, |
6189 | devid, uuid); | |
dfe25020 | 6190 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6191 | free_extent_map(em); |
6192 | return -EIO; | |
6193 | } | |
816fcebe AJ |
6194 | btrfs_warn(root->fs_info, "devid %llu uuid %pU is missing", |
6195 | devid, uuid); | |
dfe25020 CM |
6196 | } |
6197 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6198 | } |
6199 | ||
890871be | 6200 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6201 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6202 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6203 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6204 | free_extent_map(em); |
6205 | ||
6206 | return 0; | |
6207 | } | |
6208 | ||
143bede5 | 6209 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6210 | struct btrfs_dev_item *dev_item, |
6211 | struct btrfs_device *device) | |
6212 | { | |
6213 | unsigned long ptr; | |
0b86a832 CM |
6214 | |
6215 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6216 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6217 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6218 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6219 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6220 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6221 | device->type = btrfs_device_type(leaf, dev_item); |
6222 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6223 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6224 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6225 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6226 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6227 | |
410ba3a2 | 6228 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6229 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6230 | } |
6231 | ||
5f375835 MX |
6232 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6233 | u8 *fsid) | |
2b82032c YZ |
6234 | { |
6235 | struct btrfs_fs_devices *fs_devices; | |
6236 | int ret; | |
6237 | ||
b367e47f | 6238 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c YZ |
6239 | |
6240 | fs_devices = root->fs_info->fs_devices->seed; | |
6241 | while (fs_devices) { | |
5f375835 MX |
6242 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6243 | return fs_devices; | |
6244 | ||
2b82032c YZ |
6245 | fs_devices = fs_devices->seed; |
6246 | } | |
6247 | ||
6248 | fs_devices = find_fsid(fsid); | |
6249 | if (!fs_devices) { | |
5f375835 MX |
6250 | if (!btrfs_test_opt(root, DEGRADED)) |
6251 | return ERR_PTR(-ENOENT); | |
6252 | ||
6253 | fs_devices = alloc_fs_devices(fsid); | |
6254 | if (IS_ERR(fs_devices)) | |
6255 | return fs_devices; | |
6256 | ||
6257 | fs_devices->seeding = 1; | |
6258 | fs_devices->opened = 1; | |
6259 | return fs_devices; | |
2b82032c | 6260 | } |
e4404d6e YZ |
6261 | |
6262 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6263 | if (IS_ERR(fs_devices)) |
6264 | return fs_devices; | |
2b82032c | 6265 | |
97288f2c | 6266 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
15916de8 | 6267 | root->fs_info->bdev_holder); |
48d28232 JL |
6268 | if (ret) { |
6269 | free_fs_devices(fs_devices); | |
5f375835 | 6270 | fs_devices = ERR_PTR(ret); |
2b82032c | 6271 | goto out; |
48d28232 | 6272 | } |
2b82032c YZ |
6273 | |
6274 | if (!fs_devices->seeding) { | |
6275 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6276 | free_fs_devices(fs_devices); |
5f375835 | 6277 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6278 | goto out; |
6279 | } | |
6280 | ||
6281 | fs_devices->seed = root->fs_info->fs_devices->seed; | |
6282 | root->fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6283 | out: |
5f375835 | 6284 | return fs_devices; |
2b82032c YZ |
6285 | } |
6286 | ||
0d81ba5d | 6287 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6288 | struct extent_buffer *leaf, |
6289 | struct btrfs_dev_item *dev_item) | |
6290 | { | |
5f375835 | 6291 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; |
0b86a832 CM |
6292 | struct btrfs_device *device; |
6293 | u64 devid; | |
6294 | int ret; | |
2b82032c | 6295 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6296 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6297 | ||
0b86a832 | 6298 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6299 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6300 | BTRFS_UUID_SIZE); |
1473b24e | 6301 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6302 | BTRFS_UUID_SIZE); |
6303 | ||
6304 | if (memcmp(fs_uuid, root->fs_info->fsid, BTRFS_UUID_SIZE)) { | |
5f375835 MX |
6305 | fs_devices = open_seed_devices(root, fs_uuid); |
6306 | if (IS_ERR(fs_devices)) | |
6307 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6308 | } |
6309 | ||
aa1b8cd4 | 6310 | device = btrfs_find_device(root->fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6311 | if (!device) { |
e4404d6e | 6312 | if (!btrfs_test_opt(root, DEGRADED)) |
2b82032c YZ |
6313 | return -EIO; |
6314 | ||
5f375835 MX |
6315 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6316 | if (!device) | |
6317 | return -ENOMEM; | |
33b97e43 AJ |
6318 | btrfs_warn(root->fs_info, "devid %llu uuid %pU missing", |
6319 | devid, dev_uuid); | |
5f375835 MX |
6320 | } else { |
6321 | if (!device->bdev && !btrfs_test_opt(root, DEGRADED)) | |
6322 | return -EIO; | |
6323 | ||
6324 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6325 | /* |
6326 | * this happens when a device that was properly setup | |
6327 | * in the device info lists suddenly goes bad. | |
6328 | * device->bdev is NULL, and so we have to set | |
6329 | * device->missing to one here | |
6330 | */ | |
5f375835 | 6331 | device->fs_devices->missing_devices++; |
cd02dca5 | 6332 | device->missing = 1; |
2b82032c | 6333 | } |
5f375835 MX |
6334 | |
6335 | /* Move the device to its own fs_devices */ | |
6336 | if (device->fs_devices != fs_devices) { | |
6337 | ASSERT(device->missing); | |
6338 | ||
6339 | list_move(&device->dev_list, &fs_devices->devices); | |
6340 | device->fs_devices->num_devices--; | |
6341 | fs_devices->num_devices++; | |
6342 | ||
6343 | device->fs_devices->missing_devices--; | |
6344 | fs_devices->missing_devices++; | |
6345 | ||
6346 | device->fs_devices = fs_devices; | |
6347 | } | |
2b82032c YZ |
6348 | } |
6349 | ||
6350 | if (device->fs_devices != root->fs_info->fs_devices) { | |
6351 | BUG_ON(device->writeable); | |
6352 | if (device->generation != | |
6353 | btrfs_device_generation(leaf, dev_item)) | |
6354 | return -EINVAL; | |
6324fbf3 | 6355 | } |
0b86a832 CM |
6356 | |
6357 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6358 | device->in_fs_metadata = 1; |
63a212ab | 6359 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6360 | device->fs_devices->total_rw_bytes += device->total_bytes; |
2bf64758 JB |
6361 | spin_lock(&root->fs_info->free_chunk_lock); |
6362 | root->fs_info->free_chunk_space += device->total_bytes - | |
6363 | device->bytes_used; | |
6364 | spin_unlock(&root->fs_info->free_chunk_lock); | |
6365 | } | |
0b86a832 | 6366 | ret = 0; |
0b86a832 CM |
6367 | return ret; |
6368 | } | |
6369 | ||
e4404d6e | 6370 | int btrfs_read_sys_array(struct btrfs_root *root) |
0b86a832 | 6371 | { |
6c41761f | 6372 | struct btrfs_super_block *super_copy = root->fs_info->super_copy; |
a061fc8d | 6373 | struct extent_buffer *sb; |
0b86a832 | 6374 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6375 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6376 | u8 *array_ptr; |
6377 | unsigned long sb_array_offset; | |
84eed90f | 6378 | int ret = 0; |
0b86a832 CM |
6379 | u32 num_stripes; |
6380 | u32 array_size; | |
6381 | u32 len = 0; | |
1ffb22cf | 6382 | u32 cur_offset; |
84eed90f | 6383 | struct btrfs_key key; |
0b86a832 | 6384 | |
a83fffb7 DS |
6385 | ASSERT(BTRFS_SUPER_INFO_SIZE <= root->nodesize); |
6386 | /* | |
6387 | * This will create extent buffer of nodesize, superblock size is | |
6388 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6389 | * overallocate but we can keep it as-is, only the first page is used. | |
6390 | */ | |
6391 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
a061fc8d CM |
6392 | if (!sb) |
6393 | return -ENOMEM; | |
6394 | btrfs_set_buffer_uptodate(sb); | |
85d4e461 | 6395 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 DS |
6396 | /* |
6397 | * The sb extent buffer is artifical and just used to read the system array. | |
6398 | * btrfs_set_buffer_uptodate() call does not properly mark all it's | |
6399 | * pages up-to-date when the page is larger: extent does not cover the | |
6400 | * whole page and consequently check_page_uptodate does not find all | |
6401 | * the page's extents up-to-date (the hole beyond sb), | |
6402 | * write_extent_buffer then triggers a WARN_ON. | |
6403 | * | |
6404 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6405 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6406 | * to silence the warning eg. on PowerPC 64. | |
6407 | */ | |
6408 | if (PAGE_CACHE_SIZE > BTRFS_SUPER_INFO_SIZE) | |
727011e0 | 6409 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6410 | |
a061fc8d | 6411 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6412 | array_size = btrfs_super_sys_array_size(super_copy); |
6413 | ||
1ffb22cf DS |
6414 | array_ptr = super_copy->sys_chunk_array; |
6415 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6416 | cur_offset = 0; | |
0b86a832 | 6417 | |
1ffb22cf DS |
6418 | while (cur_offset < array_size) { |
6419 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6420 | len = sizeof(*disk_key); |
6421 | if (cur_offset + len > array_size) | |
6422 | goto out_short_read; | |
6423 | ||
0b86a832 CM |
6424 | btrfs_disk_key_to_cpu(&key, disk_key); |
6425 | ||
1ffb22cf DS |
6426 | array_ptr += len; |
6427 | sb_array_offset += len; | |
6428 | cur_offset += len; | |
0b86a832 | 6429 | |
0d81ba5d | 6430 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6431 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6432 | /* |
6433 | * At least one btrfs_chunk with one stripe must be | |
6434 | * present, exact stripe count check comes afterwards | |
6435 | */ | |
6436 | len = btrfs_chunk_item_size(1); | |
6437 | if (cur_offset + len > array_size) | |
6438 | goto out_short_read; | |
6439 | ||
6440 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
6441 | len = btrfs_chunk_item_size(num_stripes); | |
6442 | if (cur_offset + len > array_size) | |
6443 | goto out_short_read; | |
6444 | ||
0d81ba5d | 6445 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6446 | if (ret) |
6447 | break; | |
0b86a832 | 6448 | } else { |
84eed90f CM |
6449 | ret = -EIO; |
6450 | break; | |
0b86a832 | 6451 | } |
1ffb22cf DS |
6452 | array_ptr += len; |
6453 | sb_array_offset += len; | |
6454 | cur_offset += len; | |
0b86a832 | 6455 | } |
a061fc8d | 6456 | free_extent_buffer(sb); |
84eed90f | 6457 | return ret; |
e3540eab DS |
6458 | |
6459 | out_short_read: | |
6460 | printk(KERN_ERR "BTRFS: sys_array too short to read %u bytes at offset %u\n", | |
6461 | len, cur_offset); | |
6462 | free_extent_buffer(sb); | |
6463 | return -EIO; | |
0b86a832 CM |
6464 | } |
6465 | ||
6466 | int btrfs_read_chunk_tree(struct btrfs_root *root) | |
6467 | { | |
6468 | struct btrfs_path *path; | |
6469 | struct extent_buffer *leaf; | |
6470 | struct btrfs_key key; | |
6471 | struct btrfs_key found_key; | |
6472 | int ret; | |
6473 | int slot; | |
6474 | ||
6475 | root = root->fs_info->chunk_root; | |
6476 | ||
6477 | path = btrfs_alloc_path(); | |
6478 | if (!path) | |
6479 | return -ENOMEM; | |
6480 | ||
b367e47f LZ |
6481 | mutex_lock(&uuid_mutex); |
6482 | lock_chunks(root); | |
6483 | ||
395927a9 FDBM |
6484 | /* |
6485 | * Read all device items, and then all the chunk items. All | |
6486 | * device items are found before any chunk item (their object id | |
6487 | * is smaller than the lowest possible object id for a chunk | |
6488 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6489 | */ |
6490 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6491 | key.offset = 0; | |
6492 | key.type = 0; | |
0b86a832 | 6493 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6494 | if (ret < 0) |
6495 | goto error; | |
d397712b | 6496 | while (1) { |
0b86a832 CM |
6497 | leaf = path->nodes[0]; |
6498 | slot = path->slots[0]; | |
6499 | if (slot >= btrfs_header_nritems(leaf)) { | |
6500 | ret = btrfs_next_leaf(root, path); | |
6501 | if (ret == 0) | |
6502 | continue; | |
6503 | if (ret < 0) | |
6504 | goto error; | |
6505 | break; | |
6506 | } | |
6507 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6508 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6509 | struct btrfs_dev_item *dev_item; | |
6510 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6511 | struct btrfs_dev_item); |
395927a9 FDBM |
6512 | ret = read_one_dev(root, leaf, dev_item); |
6513 | if (ret) | |
6514 | goto error; | |
0b86a832 CM |
6515 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6516 | struct btrfs_chunk *chunk; | |
6517 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6518 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6519 | if (ret) |
6520 | goto error; | |
0b86a832 CM |
6521 | } |
6522 | path->slots[0]++; | |
6523 | } | |
0b86a832 CM |
6524 | ret = 0; |
6525 | error: | |
b367e47f LZ |
6526 | unlock_chunks(root); |
6527 | mutex_unlock(&uuid_mutex); | |
6528 | ||
2b82032c | 6529 | btrfs_free_path(path); |
0b86a832 CM |
6530 | return ret; |
6531 | } | |
442a4f63 | 6532 | |
cb517eab MX |
6533 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6534 | { | |
6535 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6536 | struct btrfs_device *device; | |
6537 | ||
29cc83f6 LB |
6538 | while (fs_devices) { |
6539 | mutex_lock(&fs_devices->device_list_mutex); | |
6540 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
6541 | device->dev_root = fs_info->dev_root; | |
6542 | mutex_unlock(&fs_devices->device_list_mutex); | |
6543 | ||
6544 | fs_devices = fs_devices->seed; | |
6545 | } | |
cb517eab MX |
6546 | } |
6547 | ||
733f4fbb SB |
6548 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6549 | { | |
6550 | int i; | |
6551 | ||
6552 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6553 | btrfs_dev_stat_reset(dev, i); | |
6554 | } | |
6555 | ||
6556 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6557 | { | |
6558 | struct btrfs_key key; | |
6559 | struct btrfs_key found_key; | |
6560 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6561 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6562 | struct extent_buffer *eb; | |
6563 | int slot; | |
6564 | int ret = 0; | |
6565 | struct btrfs_device *device; | |
6566 | struct btrfs_path *path = NULL; | |
6567 | int i; | |
6568 | ||
6569 | path = btrfs_alloc_path(); | |
6570 | if (!path) { | |
6571 | ret = -ENOMEM; | |
6572 | goto out; | |
6573 | } | |
6574 | ||
6575 | mutex_lock(&fs_devices->device_list_mutex); | |
6576 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6577 | int item_size; | |
6578 | struct btrfs_dev_stats_item *ptr; | |
6579 | ||
6580 | key.objectid = 0; | |
6581 | key.type = BTRFS_DEV_STATS_KEY; | |
6582 | key.offset = device->devid; | |
6583 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6584 | if (ret) { | |
733f4fbb SB |
6585 | __btrfs_reset_dev_stats(device); |
6586 | device->dev_stats_valid = 1; | |
6587 | btrfs_release_path(path); | |
6588 | continue; | |
6589 | } | |
6590 | slot = path->slots[0]; | |
6591 | eb = path->nodes[0]; | |
6592 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6593 | item_size = btrfs_item_size_nr(eb, slot); | |
6594 | ||
6595 | ptr = btrfs_item_ptr(eb, slot, | |
6596 | struct btrfs_dev_stats_item); | |
6597 | ||
6598 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6599 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6600 | btrfs_dev_stat_set(device, i, | |
6601 | btrfs_dev_stats_value(eb, ptr, i)); | |
6602 | else | |
6603 | btrfs_dev_stat_reset(device, i); | |
6604 | } | |
6605 | ||
6606 | device->dev_stats_valid = 1; | |
6607 | btrfs_dev_stat_print_on_load(device); | |
6608 | btrfs_release_path(path); | |
6609 | } | |
6610 | mutex_unlock(&fs_devices->device_list_mutex); | |
6611 | ||
6612 | out: | |
6613 | btrfs_free_path(path); | |
6614 | return ret < 0 ? ret : 0; | |
6615 | } | |
6616 | ||
6617 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6618 | struct btrfs_root *dev_root, | |
6619 | struct btrfs_device *device) | |
6620 | { | |
6621 | struct btrfs_path *path; | |
6622 | struct btrfs_key key; | |
6623 | struct extent_buffer *eb; | |
6624 | struct btrfs_dev_stats_item *ptr; | |
6625 | int ret; | |
6626 | int i; | |
6627 | ||
6628 | key.objectid = 0; | |
6629 | key.type = BTRFS_DEV_STATS_KEY; | |
6630 | key.offset = device->devid; | |
6631 | ||
6632 | path = btrfs_alloc_path(); | |
6633 | BUG_ON(!path); | |
6634 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6635 | if (ret < 0) { | |
efe120a0 FH |
6636 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6637 | "error %d while searching for dev_stats item for device %s!\n", | |
606686ee | 6638 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6639 | goto out; |
6640 | } | |
6641 | ||
6642 | if (ret == 0 && | |
6643 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6644 | /* need to delete old one and insert a new one */ | |
6645 | ret = btrfs_del_item(trans, dev_root, path); | |
6646 | if (ret != 0) { | |
efe120a0 FH |
6647 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6648 | "delete too small dev_stats item for device %s failed %d!\n", | |
606686ee | 6649 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6650 | goto out; |
6651 | } | |
6652 | ret = 1; | |
6653 | } | |
6654 | ||
6655 | if (ret == 1) { | |
6656 | /* need to insert a new item */ | |
6657 | btrfs_release_path(path); | |
6658 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
6659 | &key, sizeof(*ptr)); | |
6660 | if (ret < 0) { | |
efe120a0 FH |
6661 | printk_in_rcu(KERN_WARNING "BTRFS: " |
6662 | "insert dev_stats item for device %s failed %d!\n", | |
606686ee | 6663 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
6664 | goto out; |
6665 | } | |
6666 | } | |
6667 | ||
6668 | eb = path->nodes[0]; | |
6669 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
6670 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6671 | btrfs_set_dev_stats_value(eb, ptr, i, | |
6672 | btrfs_dev_stat_read(device, i)); | |
6673 | btrfs_mark_buffer_dirty(eb); | |
6674 | ||
6675 | out: | |
6676 | btrfs_free_path(path); | |
6677 | return ret; | |
6678 | } | |
6679 | ||
6680 | /* | |
6681 | * called from commit_transaction. Writes all changed device stats to disk. | |
6682 | */ | |
6683 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
6684 | struct btrfs_fs_info *fs_info) | |
6685 | { | |
6686 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6687 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6688 | struct btrfs_device *device; | |
addc3fa7 | 6689 | int stats_cnt; |
733f4fbb SB |
6690 | int ret = 0; |
6691 | ||
6692 | mutex_lock(&fs_devices->device_list_mutex); | |
6693 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 6694 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
6695 | continue; |
6696 | ||
addc3fa7 | 6697 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
733f4fbb SB |
6698 | ret = update_dev_stat_item(trans, dev_root, device); |
6699 | if (!ret) | |
addc3fa7 | 6700 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
6701 | } |
6702 | mutex_unlock(&fs_devices->device_list_mutex); | |
6703 | ||
6704 | return ret; | |
6705 | } | |
6706 | ||
442a4f63 SB |
6707 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
6708 | { | |
6709 | btrfs_dev_stat_inc(dev, index); | |
6710 | btrfs_dev_stat_print_on_error(dev); | |
6711 | } | |
6712 | ||
48a3b636 | 6713 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 6714 | { |
733f4fbb SB |
6715 | if (!dev->dev_stats_valid) |
6716 | return; | |
efe120a0 FH |
6717 | printk_ratelimited_in_rcu(KERN_ERR "BTRFS: " |
6718 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6719 | rcu_str_deref(dev->name), |
442a4f63 SB |
6720 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6721 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6722 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
6723 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
6724 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 6725 | } |
c11d2c23 | 6726 | |
733f4fbb SB |
6727 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
6728 | { | |
a98cdb85 SB |
6729 | int i; |
6730 | ||
6731 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6732 | if (btrfs_dev_stat_read(dev, i) != 0) | |
6733 | break; | |
6734 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
6735 | return; /* all values == 0, suppress message */ | |
6736 | ||
efe120a0 FH |
6737 | printk_in_rcu(KERN_INFO "BTRFS: " |
6738 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u\n", | |
606686ee | 6739 | rcu_str_deref(dev->name), |
733f4fbb SB |
6740 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
6741 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
6742 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
6743 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
6744 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
6745 | } | |
6746 | ||
c11d2c23 | 6747 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 6748 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 SB |
6749 | { |
6750 | struct btrfs_device *dev; | |
6751 | struct btrfs_fs_devices *fs_devices = root->fs_info->fs_devices; | |
6752 | int i; | |
6753 | ||
6754 | mutex_lock(&fs_devices->device_list_mutex); | |
aa1b8cd4 | 6755 | dev = btrfs_find_device(root->fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
6756 | mutex_unlock(&fs_devices->device_list_mutex); |
6757 | ||
6758 | if (!dev) { | |
efe120a0 | 6759 | btrfs_warn(root->fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 6760 | return -ENODEV; |
733f4fbb | 6761 | } else if (!dev->dev_stats_valid) { |
efe120a0 | 6762 | btrfs_warn(root->fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 6763 | return -ENODEV; |
b27f7c0c | 6764 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
6765 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
6766 | if (stats->nr_items > i) | |
6767 | stats->values[i] = | |
6768 | btrfs_dev_stat_read_and_reset(dev, i); | |
6769 | else | |
6770 | btrfs_dev_stat_reset(dev, i); | |
6771 | } | |
6772 | } else { | |
6773 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6774 | if (stats->nr_items > i) | |
6775 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
6776 | } | |
6777 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
6778 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
6779 | return 0; | |
6780 | } | |
a8a6dab7 SB |
6781 | |
6782 | int btrfs_scratch_superblock(struct btrfs_device *device) | |
6783 | { | |
6784 | struct buffer_head *bh; | |
6785 | struct btrfs_super_block *disk_super; | |
6786 | ||
6787 | bh = btrfs_read_dev_super(device->bdev); | |
6788 | if (!bh) | |
6789 | return -EINVAL; | |
6790 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
6791 | ||
6792 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
6793 | set_buffer_dirty(bh); | |
6794 | sync_dirty_buffer(bh); | |
6795 | brelse(bh); | |
6796 | ||
6797 | return 0; | |
6798 | } | |
935e5cc9 MX |
6799 | |
6800 | /* | |
6801 | * Update the size of all devices, which is used for writing out the | |
6802 | * super blocks. | |
6803 | */ | |
6804 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
6805 | { | |
6806 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6807 | struct btrfs_device *curr, *next; | |
6808 | ||
6809 | if (list_empty(&fs_devices->resized_devices)) | |
6810 | return; | |
6811 | ||
6812 | mutex_lock(&fs_devices->device_list_mutex); | |
6813 | lock_chunks(fs_info->dev_root); | |
6814 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, | |
6815 | resized_list) { | |
6816 | list_del_init(&curr->resized_list); | |
6817 | curr->commit_total_bytes = curr->disk_total_bytes; | |
6818 | } | |
6819 | unlock_chunks(fs_info->dev_root); | |
6820 | mutex_unlock(&fs_devices->device_list_mutex); | |
6821 | } | |
ce7213c7 MX |
6822 | |
6823 | /* Must be invoked during the transaction commit */ | |
6824 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
6825 | struct btrfs_transaction *transaction) | |
6826 | { | |
6827 | struct extent_map *em; | |
6828 | struct map_lookup *map; | |
6829 | struct btrfs_device *dev; | |
6830 | int i; | |
6831 | ||
6832 | if (list_empty(&transaction->pending_chunks)) | |
6833 | return; | |
6834 | ||
6835 | /* In order to kick the device replace finish process */ | |
6836 | lock_chunks(root); | |
6837 | list_for_each_entry(em, &transaction->pending_chunks, list) { | |
6838 | map = (struct map_lookup *)em->bdev; | |
6839 | ||
6840 | for (i = 0; i < map->num_stripes; i++) { | |
6841 | dev = map->stripes[i].dev; | |
6842 | dev->commit_bytes_used = dev->bytes_used; | |
6843 | } | |
6844 | } | |
6845 | unlock_chunks(root); | |
6846 | } | |
5a13f430 AJ |
6847 | |
6848 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6849 | { | |
6850 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6851 | while (fs_devices) { | |
6852 | fs_devices->fs_info = fs_info; | |
6853 | fs_devices = fs_devices->seed; | |
6854 | } | |
6855 | } | |
6856 | ||
6857 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
6858 | { | |
6859 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6860 | while (fs_devices) { | |
6861 | fs_devices->fs_info = NULL; | |
6862 | fs_devices = fs_devices->seed; | |
6863 | } | |
6864 | } |