]>
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> |
b765ead5 | 23 | #include <linux/iocontext.h> |
6f88a440 | 24 | #include <linux/capability.h> |
442a4f63 | 25 | #include <linux/ratelimit.h> |
59641015 | 26 | #include <linux/kthread.h> |
53b381b3 | 27 | #include <linux/raid/pq.h> |
803b2f54 | 28 | #include <linux/semaphore.h> |
8da4b8c4 | 29 | #include <linux/uuid.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 | |
af902047 ZL |
45 | const struct btrfs_raid_attr btrfs_raid_array[BTRFS_NR_RAID_TYPES] = { |
46 | [BTRFS_RAID_RAID10] = { | |
47 | .sub_stripes = 2, | |
48 | .dev_stripes = 1, | |
49 | .devs_max = 0, /* 0 == as many as possible */ | |
50 | .devs_min = 4, | |
8789f4fe | 51 | .tolerated_failures = 1, |
af902047 ZL |
52 | .devs_increment = 2, |
53 | .ncopies = 2, | |
54 | }, | |
55 | [BTRFS_RAID_RAID1] = { | |
56 | .sub_stripes = 1, | |
57 | .dev_stripes = 1, | |
58 | .devs_max = 2, | |
59 | .devs_min = 2, | |
8789f4fe | 60 | .tolerated_failures = 1, |
af902047 ZL |
61 | .devs_increment = 2, |
62 | .ncopies = 2, | |
63 | }, | |
64 | [BTRFS_RAID_DUP] = { | |
65 | .sub_stripes = 1, | |
66 | .dev_stripes = 2, | |
67 | .devs_max = 1, | |
68 | .devs_min = 1, | |
8789f4fe | 69 | .tolerated_failures = 0, |
af902047 ZL |
70 | .devs_increment = 1, |
71 | .ncopies = 2, | |
72 | }, | |
73 | [BTRFS_RAID_RAID0] = { | |
74 | .sub_stripes = 1, | |
75 | .dev_stripes = 1, | |
76 | .devs_max = 0, | |
77 | .devs_min = 2, | |
8789f4fe | 78 | .tolerated_failures = 0, |
af902047 ZL |
79 | .devs_increment = 1, |
80 | .ncopies = 1, | |
81 | }, | |
82 | [BTRFS_RAID_SINGLE] = { | |
83 | .sub_stripes = 1, | |
84 | .dev_stripes = 1, | |
85 | .devs_max = 1, | |
86 | .devs_min = 1, | |
8789f4fe | 87 | .tolerated_failures = 0, |
af902047 ZL |
88 | .devs_increment = 1, |
89 | .ncopies = 1, | |
90 | }, | |
91 | [BTRFS_RAID_RAID5] = { | |
92 | .sub_stripes = 1, | |
93 | .dev_stripes = 1, | |
94 | .devs_max = 0, | |
95 | .devs_min = 2, | |
8789f4fe | 96 | .tolerated_failures = 1, |
af902047 ZL |
97 | .devs_increment = 1, |
98 | .ncopies = 2, | |
99 | }, | |
100 | [BTRFS_RAID_RAID6] = { | |
101 | .sub_stripes = 1, | |
102 | .dev_stripes = 1, | |
103 | .devs_max = 0, | |
104 | .devs_min = 3, | |
8789f4fe | 105 | .tolerated_failures = 2, |
af902047 ZL |
106 | .devs_increment = 1, |
107 | .ncopies = 3, | |
108 | }, | |
109 | }; | |
110 | ||
fb75d857 | 111 | const u64 btrfs_raid_group[BTRFS_NR_RAID_TYPES] = { |
af902047 ZL |
112 | [BTRFS_RAID_RAID10] = BTRFS_BLOCK_GROUP_RAID10, |
113 | [BTRFS_RAID_RAID1] = BTRFS_BLOCK_GROUP_RAID1, | |
114 | [BTRFS_RAID_DUP] = BTRFS_BLOCK_GROUP_DUP, | |
115 | [BTRFS_RAID_RAID0] = BTRFS_BLOCK_GROUP_RAID0, | |
116 | [BTRFS_RAID_SINGLE] = 0, | |
117 | [BTRFS_RAID_RAID5] = BTRFS_BLOCK_GROUP_RAID5, | |
118 | [BTRFS_RAID_RAID6] = BTRFS_BLOCK_GROUP_RAID6, | |
119 | }; | |
120 | ||
621292ba DS |
121 | /* |
122 | * Table to convert BTRFS_RAID_* to the error code if minimum number of devices | |
123 | * condition is not met. Zero means there's no corresponding | |
124 | * BTRFS_ERROR_DEV_*_NOT_MET value. | |
125 | */ | |
126 | const int btrfs_raid_mindev_error[BTRFS_NR_RAID_TYPES] = { | |
127 | [BTRFS_RAID_RAID10] = BTRFS_ERROR_DEV_RAID10_MIN_NOT_MET, | |
128 | [BTRFS_RAID_RAID1] = BTRFS_ERROR_DEV_RAID1_MIN_NOT_MET, | |
129 | [BTRFS_RAID_DUP] = 0, | |
130 | [BTRFS_RAID_RAID0] = 0, | |
131 | [BTRFS_RAID_SINGLE] = 0, | |
132 | [BTRFS_RAID_RAID5] = BTRFS_ERROR_DEV_RAID5_MIN_NOT_MET, | |
133 | [BTRFS_RAID_RAID6] = BTRFS_ERROR_DEV_RAID6_MIN_NOT_MET, | |
134 | }; | |
135 | ||
2b82032c YZ |
136 | static int init_first_rw_device(struct btrfs_trans_handle *trans, |
137 | struct btrfs_root *root, | |
138 | struct btrfs_device *device); | |
139 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root); | |
733f4fbb | 140 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev); |
48a3b636 | 141 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev); |
733f4fbb | 142 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *device); |
2b82032c | 143 | |
67a2c45e | 144 | DEFINE_MUTEX(uuid_mutex); |
8a4b83cc | 145 | static LIST_HEAD(fs_uuids); |
c73eccf7 AJ |
146 | struct list_head *btrfs_get_fs_uuids(void) |
147 | { | |
148 | return &fs_uuids; | |
149 | } | |
8a4b83cc | 150 | |
2208a378 ID |
151 | static struct btrfs_fs_devices *__alloc_fs_devices(void) |
152 | { | |
153 | struct btrfs_fs_devices *fs_devs; | |
154 | ||
78f2c9e6 | 155 | fs_devs = kzalloc(sizeof(*fs_devs), GFP_KERNEL); |
2208a378 ID |
156 | if (!fs_devs) |
157 | return ERR_PTR(-ENOMEM); | |
158 | ||
159 | mutex_init(&fs_devs->device_list_mutex); | |
160 | ||
161 | INIT_LIST_HEAD(&fs_devs->devices); | |
935e5cc9 | 162 | INIT_LIST_HEAD(&fs_devs->resized_devices); |
2208a378 ID |
163 | INIT_LIST_HEAD(&fs_devs->alloc_list); |
164 | INIT_LIST_HEAD(&fs_devs->list); | |
165 | ||
166 | return fs_devs; | |
167 | } | |
168 | ||
169 | /** | |
170 | * alloc_fs_devices - allocate struct btrfs_fs_devices | |
171 | * @fsid: a pointer to UUID for this FS. If NULL a new UUID is | |
172 | * generated. | |
173 | * | |
174 | * Return: a pointer to a new &struct btrfs_fs_devices on success; | |
175 | * ERR_PTR() on error. Returned struct is not linked onto any lists and | |
176 | * can be destroyed with kfree() right away. | |
177 | */ | |
178 | static struct btrfs_fs_devices *alloc_fs_devices(const u8 *fsid) | |
179 | { | |
180 | struct btrfs_fs_devices *fs_devs; | |
181 | ||
182 | fs_devs = __alloc_fs_devices(); | |
183 | if (IS_ERR(fs_devs)) | |
184 | return fs_devs; | |
185 | ||
186 | if (fsid) | |
187 | memcpy(fs_devs->fsid, fsid, BTRFS_FSID_SIZE); | |
188 | else | |
189 | generate_random_uuid(fs_devs->fsid); | |
190 | ||
191 | return fs_devs; | |
192 | } | |
193 | ||
e4404d6e YZ |
194 | static void free_fs_devices(struct btrfs_fs_devices *fs_devices) |
195 | { | |
196 | struct btrfs_device *device; | |
197 | WARN_ON(fs_devices->opened); | |
198 | while (!list_empty(&fs_devices->devices)) { | |
199 | device = list_entry(fs_devices->devices.next, | |
200 | struct btrfs_device, dev_list); | |
201 | list_del(&device->dev_list); | |
606686ee | 202 | rcu_string_free(device->name); |
e4404d6e YZ |
203 | kfree(device); |
204 | } | |
205 | kfree(fs_devices); | |
206 | } | |
207 | ||
b8b8ff59 LC |
208 | static void btrfs_kobject_uevent(struct block_device *bdev, |
209 | enum kobject_action action) | |
210 | { | |
211 | int ret; | |
212 | ||
213 | ret = kobject_uevent(&disk_to_dev(bdev->bd_disk)->kobj, action); | |
214 | if (ret) | |
efe120a0 | 215 | pr_warn("BTRFS: Sending event '%d' to kobject: '%s' (%p): failed\n", |
b8b8ff59 LC |
216 | action, |
217 | kobject_name(&disk_to_dev(bdev->bd_disk)->kobj), | |
218 | &disk_to_dev(bdev->bd_disk)->kobj); | |
219 | } | |
220 | ||
143bede5 | 221 | void btrfs_cleanup_fs_uuids(void) |
8a4b83cc CM |
222 | { |
223 | struct btrfs_fs_devices *fs_devices; | |
8a4b83cc | 224 | |
2b82032c YZ |
225 | while (!list_empty(&fs_uuids)) { |
226 | fs_devices = list_entry(fs_uuids.next, | |
227 | struct btrfs_fs_devices, list); | |
228 | list_del(&fs_devices->list); | |
e4404d6e | 229 | free_fs_devices(fs_devices); |
8a4b83cc | 230 | } |
8a4b83cc CM |
231 | } |
232 | ||
12bd2fc0 ID |
233 | static struct btrfs_device *__alloc_device(void) |
234 | { | |
235 | struct btrfs_device *dev; | |
236 | ||
78f2c9e6 | 237 | dev = kzalloc(sizeof(*dev), GFP_KERNEL); |
12bd2fc0 ID |
238 | if (!dev) |
239 | return ERR_PTR(-ENOMEM); | |
240 | ||
241 | INIT_LIST_HEAD(&dev->dev_list); | |
242 | INIT_LIST_HEAD(&dev->dev_alloc_list); | |
935e5cc9 | 243 | INIT_LIST_HEAD(&dev->resized_list); |
12bd2fc0 ID |
244 | |
245 | spin_lock_init(&dev->io_lock); | |
246 | ||
247 | spin_lock_init(&dev->reada_lock); | |
248 | atomic_set(&dev->reada_in_flight, 0); | |
addc3fa7 | 249 | atomic_set(&dev->dev_stats_ccnt, 0); |
546bed63 | 250 | btrfs_device_data_ordered_init(dev); |
d0164adc MG |
251 | INIT_RADIX_TREE(&dev->reada_zones, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); |
252 | INIT_RADIX_TREE(&dev->reada_extents, GFP_NOFS & ~__GFP_DIRECT_RECLAIM); | |
12bd2fc0 ID |
253 | |
254 | return dev; | |
255 | } | |
256 | ||
a1b32a59 CM |
257 | static noinline struct btrfs_device *__find_device(struct list_head *head, |
258 | u64 devid, u8 *uuid) | |
8a4b83cc CM |
259 | { |
260 | struct btrfs_device *dev; | |
8a4b83cc | 261 | |
c6e30871 | 262 | list_for_each_entry(dev, head, dev_list) { |
a443755f | 263 | if (dev->devid == devid && |
8f18cf13 | 264 | (!uuid || !memcmp(dev->uuid, uuid, BTRFS_UUID_SIZE))) { |
8a4b83cc | 265 | return dev; |
a443755f | 266 | } |
8a4b83cc CM |
267 | } |
268 | return NULL; | |
269 | } | |
270 | ||
a1b32a59 | 271 | static noinline struct btrfs_fs_devices *find_fsid(u8 *fsid) |
8a4b83cc | 272 | { |
8a4b83cc CM |
273 | struct btrfs_fs_devices *fs_devices; |
274 | ||
c6e30871 | 275 | list_for_each_entry(fs_devices, &fs_uuids, list) { |
8a4b83cc CM |
276 | if (memcmp(fsid, fs_devices->fsid, BTRFS_FSID_SIZE) == 0) |
277 | return fs_devices; | |
278 | } | |
279 | return NULL; | |
280 | } | |
281 | ||
beaf8ab3 SB |
282 | static int |
283 | btrfs_get_bdev_and_sb(const char *device_path, fmode_t flags, void *holder, | |
284 | int flush, struct block_device **bdev, | |
285 | struct buffer_head **bh) | |
286 | { | |
287 | int ret; | |
288 | ||
289 | *bdev = blkdev_get_by_path(device_path, flags, holder); | |
290 | ||
291 | if (IS_ERR(*bdev)) { | |
292 | ret = PTR_ERR(*bdev); | |
beaf8ab3 SB |
293 | goto error; |
294 | } | |
295 | ||
296 | if (flush) | |
297 | filemap_write_and_wait((*bdev)->bd_inode->i_mapping); | |
298 | ret = set_blocksize(*bdev, 4096); | |
299 | if (ret) { | |
300 | blkdev_put(*bdev, flags); | |
301 | goto error; | |
302 | } | |
303 | invalidate_bdev(*bdev); | |
304 | *bh = btrfs_read_dev_super(*bdev); | |
92fc03fb AJ |
305 | if (IS_ERR(*bh)) { |
306 | ret = PTR_ERR(*bh); | |
beaf8ab3 SB |
307 | blkdev_put(*bdev, flags); |
308 | goto error; | |
309 | } | |
310 | ||
311 | return 0; | |
312 | ||
313 | error: | |
314 | *bdev = NULL; | |
315 | *bh = NULL; | |
316 | return ret; | |
317 | } | |
318 | ||
ffbd517d CM |
319 | static void requeue_list(struct btrfs_pending_bios *pending_bios, |
320 | struct bio *head, struct bio *tail) | |
321 | { | |
322 | ||
323 | struct bio *old_head; | |
324 | ||
325 | old_head = pending_bios->head; | |
326 | pending_bios->head = head; | |
327 | if (pending_bios->tail) | |
328 | tail->bi_next = old_head; | |
329 | else | |
330 | pending_bios->tail = tail; | |
331 | } | |
332 | ||
8b712842 CM |
333 | /* |
334 | * we try to collect pending bios for a device so we don't get a large | |
335 | * number of procs sending bios down to the same device. This greatly | |
336 | * improves the schedulers ability to collect and merge the bios. | |
337 | * | |
338 | * But, it also turns into a long list of bios to process and that is sure | |
339 | * to eventually make the worker thread block. The solution here is to | |
340 | * make some progress and then put this work struct back at the end of | |
341 | * the list if the block device is congested. This way, multiple devices | |
342 | * can make progress from a single worker thread. | |
343 | */ | |
143bede5 | 344 | static noinline void run_scheduled_bios(struct btrfs_device *device) |
8b712842 | 345 | { |
0b246afa | 346 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 CM |
347 | struct bio *pending; |
348 | struct backing_dev_info *bdi; | |
ffbd517d | 349 | struct btrfs_pending_bios *pending_bios; |
8b712842 CM |
350 | struct bio *tail; |
351 | struct bio *cur; | |
352 | int again = 0; | |
ffbd517d | 353 | unsigned long num_run; |
d644d8a1 | 354 | unsigned long batch_run = 0; |
b64a2851 | 355 | unsigned long limit; |
b765ead5 | 356 | unsigned long last_waited = 0; |
d84275c9 | 357 | int force_reg = 0; |
0e588859 | 358 | int sync_pending = 0; |
211588ad CM |
359 | struct blk_plug plug; |
360 | ||
361 | /* | |
362 | * this function runs all the bios we've collected for | |
363 | * a particular device. We don't want to wander off to | |
364 | * another device without first sending all of these down. | |
365 | * So, setup a plug here and finish it off before we return | |
366 | */ | |
367 | blk_start_plug(&plug); | |
8b712842 | 368 | |
bedf762b | 369 | bdi = blk_get_backing_dev_info(device->bdev); |
b64a2851 CM |
370 | limit = btrfs_async_submit_limit(fs_info); |
371 | limit = limit * 2 / 3; | |
372 | ||
8b712842 CM |
373 | loop: |
374 | spin_lock(&device->io_lock); | |
375 | ||
a6837051 | 376 | loop_lock: |
d84275c9 | 377 | num_run = 0; |
ffbd517d | 378 | |
8b712842 CM |
379 | /* take all the bios off the list at once and process them |
380 | * later on (without the lock held). But, remember the | |
381 | * tail and other pointers so the bios can be properly reinserted | |
382 | * into the list if we hit congestion | |
383 | */ | |
d84275c9 | 384 | if (!force_reg && device->pending_sync_bios.head) { |
ffbd517d | 385 | pending_bios = &device->pending_sync_bios; |
d84275c9 CM |
386 | force_reg = 1; |
387 | } else { | |
ffbd517d | 388 | pending_bios = &device->pending_bios; |
d84275c9 CM |
389 | force_reg = 0; |
390 | } | |
ffbd517d CM |
391 | |
392 | pending = pending_bios->head; | |
393 | tail = pending_bios->tail; | |
8b712842 | 394 | WARN_ON(pending && !tail); |
8b712842 CM |
395 | |
396 | /* | |
397 | * if pending was null this time around, no bios need processing | |
398 | * at all and we can stop. Otherwise it'll loop back up again | |
399 | * and do an additional check so no bios are missed. | |
400 | * | |
401 | * device->running_pending is used to synchronize with the | |
402 | * schedule_bio code. | |
403 | */ | |
ffbd517d CM |
404 | if (device->pending_sync_bios.head == NULL && |
405 | device->pending_bios.head == NULL) { | |
8b712842 CM |
406 | again = 0; |
407 | device->running_pending = 0; | |
ffbd517d CM |
408 | } else { |
409 | again = 1; | |
410 | device->running_pending = 1; | |
8b712842 | 411 | } |
ffbd517d CM |
412 | |
413 | pending_bios->head = NULL; | |
414 | pending_bios->tail = NULL; | |
415 | ||
8b712842 CM |
416 | spin_unlock(&device->io_lock); |
417 | ||
d397712b | 418 | while (pending) { |
ffbd517d CM |
419 | |
420 | rmb(); | |
d84275c9 CM |
421 | /* we want to work on both lists, but do more bios on the |
422 | * sync list than the regular list | |
423 | */ | |
424 | if ((num_run > 32 && | |
425 | pending_bios != &device->pending_sync_bios && | |
426 | device->pending_sync_bios.head) || | |
427 | (num_run > 64 && pending_bios == &device->pending_sync_bios && | |
428 | device->pending_bios.head)) { | |
ffbd517d CM |
429 | spin_lock(&device->io_lock); |
430 | requeue_list(pending_bios, pending, tail); | |
431 | goto loop_lock; | |
432 | } | |
433 | ||
8b712842 CM |
434 | cur = pending; |
435 | pending = pending->bi_next; | |
436 | cur->bi_next = NULL; | |
b64a2851 | 437 | |
ee863954 DS |
438 | /* |
439 | * atomic_dec_return implies a barrier for waitqueue_active | |
440 | */ | |
66657b31 | 441 | if (atomic_dec_return(&fs_info->nr_async_bios) < limit && |
b64a2851 CM |
442 | waitqueue_active(&fs_info->async_submit_wait)) |
443 | wake_up(&fs_info->async_submit_wait); | |
492bb6de | 444 | |
dac56212 | 445 | BUG_ON(atomic_read(&cur->__bi_cnt) == 0); |
d644d8a1 | 446 | |
2ab1ba68 CM |
447 | /* |
448 | * if we're doing the sync list, record that our | |
449 | * plug has some sync requests on it | |
450 | * | |
451 | * If we're doing the regular list and there are | |
452 | * sync requests sitting around, unplug before | |
453 | * we add more | |
454 | */ | |
455 | if (pending_bios == &device->pending_sync_bios) { | |
456 | sync_pending = 1; | |
457 | } else if (sync_pending) { | |
458 | blk_finish_plug(&plug); | |
459 | blk_start_plug(&plug); | |
460 | sync_pending = 0; | |
461 | } | |
462 | ||
4e49ea4a | 463 | btrfsic_submit_bio(cur); |
5ff7ba3a CM |
464 | num_run++; |
465 | batch_run++; | |
853d8ec4 DS |
466 | |
467 | cond_resched(); | |
8b712842 CM |
468 | |
469 | /* | |
470 | * we made progress, there is more work to do and the bdi | |
471 | * is now congested. Back off and let other work structs | |
472 | * run instead | |
473 | */ | |
57fd5a5f | 474 | if (pending && bdi_write_congested(bdi) && batch_run > 8 && |
5f2cc086 | 475 | fs_info->fs_devices->open_devices > 1) { |
b765ead5 | 476 | struct io_context *ioc; |
8b712842 | 477 | |
b765ead5 CM |
478 | ioc = current->io_context; |
479 | ||
480 | /* | |
481 | * the main goal here is that we don't want to | |
482 | * block if we're going to be able to submit | |
483 | * more requests without blocking. | |
484 | * | |
485 | * This code does two great things, it pokes into | |
486 | * the elevator code from a filesystem _and_ | |
487 | * it makes assumptions about how batching works. | |
488 | */ | |
489 | if (ioc && ioc->nr_batch_requests > 0 && | |
490 | time_before(jiffies, ioc->last_waited + HZ/50UL) && | |
491 | (last_waited == 0 || | |
492 | ioc->last_waited == last_waited)) { | |
493 | /* | |
494 | * we want to go through our batch of | |
495 | * requests and stop. So, we copy out | |
496 | * the ioc->last_waited time and test | |
497 | * against it before looping | |
498 | */ | |
499 | last_waited = ioc->last_waited; | |
853d8ec4 | 500 | cond_resched(); |
b765ead5 CM |
501 | continue; |
502 | } | |
8b712842 | 503 | spin_lock(&device->io_lock); |
ffbd517d | 504 | requeue_list(pending_bios, pending, tail); |
a6837051 | 505 | device->running_pending = 1; |
8b712842 CM |
506 | |
507 | spin_unlock(&device->io_lock); | |
a8c93d4e QW |
508 | btrfs_queue_work(fs_info->submit_workers, |
509 | &device->work); | |
8b712842 CM |
510 | goto done; |
511 | } | |
d85c8a6f CM |
512 | /* unplug every 64 requests just for good measure */ |
513 | if (batch_run % 64 == 0) { | |
514 | blk_finish_plug(&plug); | |
515 | blk_start_plug(&plug); | |
516 | sync_pending = 0; | |
517 | } | |
8b712842 | 518 | } |
ffbd517d | 519 | |
51684082 CM |
520 | cond_resched(); |
521 | if (again) | |
522 | goto loop; | |
523 | ||
524 | spin_lock(&device->io_lock); | |
525 | if (device->pending_bios.head || device->pending_sync_bios.head) | |
526 | goto loop_lock; | |
527 | spin_unlock(&device->io_lock); | |
528 | ||
8b712842 | 529 | done: |
211588ad | 530 | blk_finish_plug(&plug); |
8b712842 CM |
531 | } |
532 | ||
b2950863 | 533 | static void pending_bios_fn(struct btrfs_work *work) |
8b712842 CM |
534 | { |
535 | struct btrfs_device *device; | |
536 | ||
537 | device = container_of(work, struct btrfs_device, work); | |
538 | run_scheduled_bios(device); | |
539 | } | |
540 | ||
4fde46f0 AJ |
541 | |
542 | void btrfs_free_stale_device(struct btrfs_device *cur_dev) | |
543 | { | |
544 | struct btrfs_fs_devices *fs_devs; | |
545 | struct btrfs_device *dev; | |
546 | ||
547 | if (!cur_dev->name) | |
548 | return; | |
549 | ||
550 | list_for_each_entry(fs_devs, &fs_uuids, list) { | |
551 | int del = 1; | |
552 | ||
553 | if (fs_devs->opened) | |
554 | continue; | |
555 | if (fs_devs->seeding) | |
556 | continue; | |
557 | ||
558 | list_for_each_entry(dev, &fs_devs->devices, dev_list) { | |
559 | ||
560 | if (dev == cur_dev) | |
561 | continue; | |
562 | if (!dev->name) | |
563 | continue; | |
564 | ||
565 | /* | |
566 | * Todo: This won't be enough. What if the same device | |
567 | * comes back (with new uuid and) with its mapper path? | |
568 | * But for now, this does help as mostly an admin will | |
569 | * either use mapper or non mapper path throughout. | |
570 | */ | |
571 | rcu_read_lock(); | |
572 | del = strcmp(rcu_str_deref(dev->name), | |
573 | rcu_str_deref(cur_dev->name)); | |
574 | rcu_read_unlock(); | |
575 | if (!del) | |
576 | break; | |
577 | } | |
578 | ||
579 | if (!del) { | |
580 | /* delete the stale device */ | |
581 | if (fs_devs->num_devices == 1) { | |
582 | btrfs_sysfs_remove_fsid(fs_devs); | |
583 | list_del(&fs_devs->list); | |
584 | free_fs_devices(fs_devs); | |
585 | } else { | |
586 | fs_devs->num_devices--; | |
587 | list_del(&dev->dev_list); | |
588 | rcu_string_free(dev->name); | |
589 | kfree(dev); | |
590 | } | |
591 | break; | |
592 | } | |
593 | } | |
594 | } | |
595 | ||
60999ca4 DS |
596 | /* |
597 | * Add new device to list of registered devices | |
598 | * | |
599 | * Returns: | |
600 | * 1 - first time device is seen | |
601 | * 0 - device already known | |
602 | * < 0 - error | |
603 | */ | |
a1b32a59 | 604 | static noinline int device_list_add(const char *path, |
8a4b83cc CM |
605 | struct btrfs_super_block *disk_super, |
606 | u64 devid, struct btrfs_fs_devices **fs_devices_ret) | |
607 | { | |
608 | struct btrfs_device *device; | |
609 | struct btrfs_fs_devices *fs_devices; | |
606686ee | 610 | struct rcu_string *name; |
60999ca4 | 611 | int ret = 0; |
8a4b83cc CM |
612 | u64 found_transid = btrfs_super_generation(disk_super); |
613 | ||
614 | fs_devices = find_fsid(disk_super->fsid); | |
615 | if (!fs_devices) { | |
2208a378 ID |
616 | fs_devices = alloc_fs_devices(disk_super->fsid); |
617 | if (IS_ERR(fs_devices)) | |
618 | return PTR_ERR(fs_devices); | |
619 | ||
8a4b83cc | 620 | list_add(&fs_devices->list, &fs_uuids); |
2208a378 | 621 | |
8a4b83cc CM |
622 | device = NULL; |
623 | } else { | |
a443755f CM |
624 | device = __find_device(&fs_devices->devices, devid, |
625 | disk_super->dev_item.uuid); | |
8a4b83cc | 626 | } |
443f24fe | 627 | |
8a4b83cc | 628 | if (!device) { |
2b82032c YZ |
629 | if (fs_devices->opened) |
630 | return -EBUSY; | |
631 | ||
12bd2fc0 ID |
632 | device = btrfs_alloc_device(NULL, &devid, |
633 | disk_super->dev_item.uuid); | |
634 | if (IS_ERR(device)) { | |
8a4b83cc | 635 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 636 | return PTR_ERR(device); |
8a4b83cc | 637 | } |
606686ee JB |
638 | |
639 | name = rcu_string_strdup(path, GFP_NOFS); | |
640 | if (!name) { | |
8a4b83cc CM |
641 | kfree(device); |
642 | return -ENOMEM; | |
643 | } | |
606686ee | 644 | rcu_assign_pointer(device->name, name); |
90519d66 | 645 | |
e5e9a520 | 646 | mutex_lock(&fs_devices->device_list_mutex); |
1f78160c | 647 | list_add_rcu(&device->dev_list, &fs_devices->devices); |
f7171750 | 648 | fs_devices->num_devices++; |
e5e9a520 CM |
649 | mutex_unlock(&fs_devices->device_list_mutex); |
650 | ||
60999ca4 | 651 | ret = 1; |
2b82032c | 652 | device->fs_devices = fs_devices; |
606686ee | 653 | } else if (!device->name || strcmp(device->name->str, path)) { |
b96de000 AJ |
654 | /* |
655 | * When FS is already mounted. | |
656 | * 1. If you are here and if the device->name is NULL that | |
657 | * means this device was missing at time of FS mount. | |
658 | * 2. If you are here and if the device->name is different | |
659 | * from 'path' that means either | |
660 | * a. The same device disappeared and reappeared with | |
661 | * different name. or | |
662 | * b. The missing-disk-which-was-replaced, has | |
663 | * reappeared now. | |
664 | * | |
665 | * We must allow 1 and 2a above. But 2b would be a spurious | |
666 | * and unintentional. | |
667 | * | |
668 | * Further in case of 1 and 2a above, the disk at 'path' | |
669 | * would have missed some transaction when it was away and | |
670 | * in case of 2a the stale bdev has to be updated as well. | |
671 | * 2b must not be allowed at all time. | |
672 | */ | |
673 | ||
674 | /* | |
0f23ae74 CM |
675 | * For now, we do allow update to btrfs_fs_device through the |
676 | * btrfs dev scan cli after FS has been mounted. We're still | |
677 | * tracking a problem where systems fail mount by subvolume id | |
678 | * when we reject replacement on a mounted FS. | |
b96de000 | 679 | */ |
0f23ae74 | 680 | if (!fs_devices->opened && found_transid < device->generation) { |
77bdae4d AJ |
681 | /* |
682 | * That is if the FS is _not_ mounted and if you | |
683 | * are here, that means there is more than one | |
684 | * disk with same uuid and devid.We keep the one | |
685 | * with larger generation number or the last-in if | |
686 | * generation are equal. | |
687 | */ | |
0f23ae74 | 688 | return -EEXIST; |
77bdae4d | 689 | } |
b96de000 | 690 | |
606686ee | 691 | name = rcu_string_strdup(path, GFP_NOFS); |
3a0524dc TH |
692 | if (!name) |
693 | return -ENOMEM; | |
606686ee JB |
694 | rcu_string_free(device->name); |
695 | rcu_assign_pointer(device->name, name); | |
cd02dca5 CM |
696 | if (device->missing) { |
697 | fs_devices->missing_devices--; | |
698 | device->missing = 0; | |
699 | } | |
8a4b83cc CM |
700 | } |
701 | ||
77bdae4d AJ |
702 | /* |
703 | * Unmount does not free the btrfs_device struct but would zero | |
704 | * generation along with most of the other members. So just update | |
705 | * it back. We need it to pick the disk with largest generation | |
706 | * (as above). | |
707 | */ | |
708 | if (!fs_devices->opened) | |
709 | device->generation = found_transid; | |
710 | ||
4fde46f0 AJ |
711 | /* |
712 | * if there is new btrfs on an already registered device, | |
713 | * then remove the stale device entry. | |
714 | */ | |
02feae3c AJ |
715 | if (ret > 0) |
716 | btrfs_free_stale_device(device); | |
4fde46f0 | 717 | |
8a4b83cc | 718 | *fs_devices_ret = fs_devices; |
60999ca4 DS |
719 | |
720 | return ret; | |
8a4b83cc CM |
721 | } |
722 | ||
e4404d6e YZ |
723 | static struct btrfs_fs_devices *clone_fs_devices(struct btrfs_fs_devices *orig) |
724 | { | |
725 | struct btrfs_fs_devices *fs_devices; | |
726 | struct btrfs_device *device; | |
727 | struct btrfs_device *orig_dev; | |
728 | ||
2208a378 ID |
729 | fs_devices = alloc_fs_devices(orig->fsid); |
730 | if (IS_ERR(fs_devices)) | |
731 | return fs_devices; | |
e4404d6e | 732 | |
adbbb863 | 733 | mutex_lock(&orig->device_list_mutex); |
02db0844 | 734 | fs_devices->total_devices = orig->total_devices; |
e4404d6e | 735 | |
46224705 | 736 | /* We have held the volume lock, it is safe to get the devices. */ |
e4404d6e | 737 | list_for_each_entry(orig_dev, &orig->devices, dev_list) { |
606686ee JB |
738 | struct rcu_string *name; |
739 | ||
12bd2fc0 ID |
740 | device = btrfs_alloc_device(NULL, &orig_dev->devid, |
741 | orig_dev->uuid); | |
742 | if (IS_ERR(device)) | |
e4404d6e YZ |
743 | goto error; |
744 | ||
606686ee JB |
745 | /* |
746 | * This is ok to do without rcu read locked because we hold the | |
747 | * uuid mutex so nothing we touch in here is going to disappear. | |
748 | */ | |
e755f780 | 749 | if (orig_dev->name) { |
78f2c9e6 DS |
750 | name = rcu_string_strdup(orig_dev->name->str, |
751 | GFP_KERNEL); | |
e755f780 AJ |
752 | if (!name) { |
753 | kfree(device); | |
754 | goto error; | |
755 | } | |
756 | rcu_assign_pointer(device->name, name); | |
fd2696f3 | 757 | } |
e4404d6e | 758 | |
e4404d6e YZ |
759 | list_add(&device->dev_list, &fs_devices->devices); |
760 | device->fs_devices = fs_devices; | |
761 | fs_devices->num_devices++; | |
762 | } | |
adbbb863 | 763 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
764 | return fs_devices; |
765 | error: | |
adbbb863 | 766 | mutex_unlock(&orig->device_list_mutex); |
e4404d6e YZ |
767 | free_fs_devices(fs_devices); |
768 | return ERR_PTR(-ENOMEM); | |
769 | } | |
770 | ||
9eaed21e | 771 | void btrfs_close_extra_devices(struct btrfs_fs_devices *fs_devices, int step) |
dfe25020 | 772 | { |
c6e30871 | 773 | struct btrfs_device *device, *next; |
443f24fe | 774 | struct btrfs_device *latest_dev = NULL; |
a6b0d5c8 | 775 | |
dfe25020 CM |
776 | mutex_lock(&uuid_mutex); |
777 | again: | |
46224705 | 778 | /* This is the initialized path, it is safe to release the devices. */ |
c6e30871 | 779 | list_for_each_entry_safe(device, next, &fs_devices->devices, dev_list) { |
a6b0d5c8 | 780 | if (device->in_fs_metadata) { |
63a212ab | 781 | if (!device->is_tgtdev_for_dev_replace && |
443f24fe MX |
782 | (!latest_dev || |
783 | device->generation > latest_dev->generation)) { | |
784 | latest_dev = device; | |
a6b0d5c8 | 785 | } |
2b82032c | 786 | continue; |
a6b0d5c8 | 787 | } |
2b82032c | 788 | |
8dabb742 SB |
789 | if (device->devid == BTRFS_DEV_REPLACE_DEVID) { |
790 | /* | |
791 | * In the first step, keep the device which has | |
792 | * the correct fsid and the devid that is used | |
793 | * for the dev_replace procedure. | |
794 | * In the second step, the dev_replace state is | |
795 | * read from the device tree and it is known | |
796 | * whether the procedure is really active or | |
797 | * not, which means whether this device is | |
798 | * used or whether it should be removed. | |
799 | */ | |
800 | if (step == 0 || device->is_tgtdev_for_dev_replace) { | |
801 | continue; | |
802 | } | |
803 | } | |
2b82032c | 804 | if (device->bdev) { |
d4d77629 | 805 | blkdev_put(device->bdev, device->mode); |
2b82032c YZ |
806 | device->bdev = NULL; |
807 | fs_devices->open_devices--; | |
808 | } | |
809 | if (device->writeable) { | |
810 | list_del_init(&device->dev_alloc_list); | |
811 | device->writeable = 0; | |
8dabb742 SB |
812 | if (!device->is_tgtdev_for_dev_replace) |
813 | fs_devices->rw_devices--; | |
2b82032c | 814 | } |
e4404d6e YZ |
815 | list_del_init(&device->dev_list); |
816 | fs_devices->num_devices--; | |
606686ee | 817 | rcu_string_free(device->name); |
e4404d6e | 818 | kfree(device); |
dfe25020 | 819 | } |
2b82032c YZ |
820 | |
821 | if (fs_devices->seed) { | |
822 | fs_devices = fs_devices->seed; | |
2b82032c YZ |
823 | goto again; |
824 | } | |
825 | ||
443f24fe | 826 | fs_devices->latest_bdev = latest_dev->bdev; |
a6b0d5c8 | 827 | |
dfe25020 | 828 | mutex_unlock(&uuid_mutex); |
dfe25020 | 829 | } |
a0af469b | 830 | |
1f78160c XG |
831 | static void __free_device(struct work_struct *work) |
832 | { | |
833 | struct btrfs_device *device; | |
834 | ||
835 | device = container_of(work, struct btrfs_device, rcu_work); | |
606686ee | 836 | rcu_string_free(device->name); |
1f78160c XG |
837 | kfree(device); |
838 | } | |
839 | ||
840 | static void free_device(struct rcu_head *head) | |
841 | { | |
842 | struct btrfs_device *device; | |
843 | ||
844 | device = container_of(head, struct btrfs_device, rcu); | |
845 | ||
846 | INIT_WORK(&device->rcu_work, __free_device); | |
847 | schedule_work(&device->rcu_work); | |
848 | } | |
849 | ||
14238819 AJ |
850 | static void btrfs_close_bdev(struct btrfs_device *device) |
851 | { | |
852 | if (device->bdev && device->writeable) { | |
853 | sync_blockdev(device->bdev); | |
854 | invalidate_bdev(device->bdev); | |
855 | } | |
856 | ||
857 | if (device->bdev) | |
858 | blkdev_put(device->bdev, device->mode); | |
859 | } | |
860 | ||
0ccd0528 | 861 | static void btrfs_prepare_close_one_device(struct btrfs_device *device) |
f448341a AJ |
862 | { |
863 | struct btrfs_fs_devices *fs_devices = device->fs_devices; | |
864 | struct btrfs_device *new_device; | |
865 | struct rcu_string *name; | |
866 | ||
867 | if (device->bdev) | |
868 | fs_devices->open_devices--; | |
869 | ||
870 | if (device->writeable && | |
871 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
872 | list_del_init(&device->dev_alloc_list); | |
873 | fs_devices->rw_devices--; | |
874 | } | |
875 | ||
876 | if (device->missing) | |
877 | fs_devices->missing_devices--; | |
878 | ||
879 | new_device = btrfs_alloc_device(NULL, &device->devid, | |
880 | device->uuid); | |
881 | BUG_ON(IS_ERR(new_device)); /* -ENOMEM */ | |
882 | ||
883 | /* Safe because we are under uuid_mutex */ | |
884 | if (device->name) { | |
885 | name = rcu_string_strdup(device->name->str, GFP_NOFS); | |
886 | BUG_ON(!name); /* -ENOMEM */ | |
887 | rcu_assign_pointer(new_device->name, name); | |
888 | } | |
889 | ||
890 | list_replace_rcu(&device->dev_list, &new_device->dev_list); | |
891 | new_device->fs_devices = device->fs_devices; | |
f448341a AJ |
892 | } |
893 | ||
2b82032c | 894 | static int __btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
8a4b83cc | 895 | { |
2037a093 | 896 | struct btrfs_device *device, *tmp; |
0ccd0528 AJ |
897 | struct list_head pending_put; |
898 | ||
899 | INIT_LIST_HEAD(&pending_put); | |
e4404d6e | 900 | |
2b82032c YZ |
901 | if (--fs_devices->opened > 0) |
902 | return 0; | |
8a4b83cc | 903 | |
c9513edb | 904 | mutex_lock(&fs_devices->device_list_mutex); |
2037a093 | 905 | list_for_each_entry_safe(device, tmp, &fs_devices->devices, dev_list) { |
0ccd0528 AJ |
906 | btrfs_prepare_close_one_device(device); |
907 | list_add(&device->dev_list, &pending_put); | |
8a4b83cc | 908 | } |
c9513edb XG |
909 | mutex_unlock(&fs_devices->device_list_mutex); |
910 | ||
0ccd0528 AJ |
911 | /* |
912 | * btrfs_show_devname() is using the device_list_mutex, | |
913 | * sometimes call to blkdev_put() leads vfs calling | |
914 | * into this func. So do put outside of device_list_mutex, | |
915 | * as of now. | |
916 | */ | |
917 | while (!list_empty(&pending_put)) { | |
918 | device = list_first_entry(&pending_put, | |
919 | struct btrfs_device, dev_list); | |
920 | list_del(&device->dev_list); | |
921 | btrfs_close_bdev(device); | |
922 | call_rcu(&device->rcu, free_device); | |
923 | } | |
924 | ||
e4404d6e YZ |
925 | WARN_ON(fs_devices->open_devices); |
926 | WARN_ON(fs_devices->rw_devices); | |
2b82032c YZ |
927 | fs_devices->opened = 0; |
928 | fs_devices->seeding = 0; | |
2b82032c | 929 | |
8a4b83cc CM |
930 | return 0; |
931 | } | |
932 | ||
2b82032c YZ |
933 | int btrfs_close_devices(struct btrfs_fs_devices *fs_devices) |
934 | { | |
e4404d6e | 935 | struct btrfs_fs_devices *seed_devices = NULL; |
2b82032c YZ |
936 | int ret; |
937 | ||
938 | mutex_lock(&uuid_mutex); | |
939 | ret = __btrfs_close_devices(fs_devices); | |
e4404d6e YZ |
940 | if (!fs_devices->opened) { |
941 | seed_devices = fs_devices->seed; | |
942 | fs_devices->seed = NULL; | |
943 | } | |
2b82032c | 944 | mutex_unlock(&uuid_mutex); |
e4404d6e YZ |
945 | |
946 | while (seed_devices) { | |
947 | fs_devices = seed_devices; | |
948 | seed_devices = fs_devices->seed; | |
949 | __btrfs_close_devices(fs_devices); | |
950 | free_fs_devices(fs_devices); | |
951 | } | |
bc178622 ES |
952 | /* |
953 | * Wait for rcu kworkers under __btrfs_close_devices | |
954 | * to finish all blkdev_puts so device is really | |
955 | * free when umount is done. | |
956 | */ | |
957 | rcu_barrier(); | |
2b82032c YZ |
958 | return ret; |
959 | } | |
960 | ||
e4404d6e YZ |
961 | static int __btrfs_open_devices(struct btrfs_fs_devices *fs_devices, |
962 | fmode_t flags, void *holder) | |
8a4b83cc | 963 | { |
d5e2003c | 964 | struct request_queue *q; |
8a4b83cc CM |
965 | struct block_device *bdev; |
966 | struct list_head *head = &fs_devices->devices; | |
8a4b83cc | 967 | struct btrfs_device *device; |
443f24fe | 968 | struct btrfs_device *latest_dev = NULL; |
a0af469b CM |
969 | struct buffer_head *bh; |
970 | struct btrfs_super_block *disk_super; | |
a0af469b | 971 | u64 devid; |
2b82032c | 972 | int seeding = 1; |
a0af469b | 973 | int ret = 0; |
8a4b83cc | 974 | |
d4d77629 TH |
975 | flags |= FMODE_EXCL; |
976 | ||
c6e30871 | 977 | list_for_each_entry(device, head, dev_list) { |
c1c4d91c CM |
978 | if (device->bdev) |
979 | continue; | |
dfe25020 CM |
980 | if (!device->name) |
981 | continue; | |
982 | ||
f63e0cca ES |
983 | /* Just open everything we can; ignore failures here */ |
984 | if (btrfs_get_bdev_and_sb(device->name->str, flags, holder, 1, | |
985 | &bdev, &bh)) | |
beaf8ab3 | 986 | continue; |
a0af469b CM |
987 | |
988 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
a343832f | 989 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
a0af469b CM |
990 | if (devid != device->devid) |
991 | goto error_brelse; | |
992 | ||
2b82032c YZ |
993 | if (memcmp(device->uuid, disk_super->dev_item.uuid, |
994 | BTRFS_UUID_SIZE)) | |
995 | goto error_brelse; | |
996 | ||
997 | device->generation = btrfs_super_generation(disk_super); | |
443f24fe MX |
998 | if (!latest_dev || |
999 | device->generation > latest_dev->generation) | |
1000 | latest_dev = device; | |
a0af469b | 1001 | |
2b82032c YZ |
1002 | if (btrfs_super_flags(disk_super) & BTRFS_SUPER_FLAG_SEEDING) { |
1003 | device->writeable = 0; | |
1004 | } else { | |
1005 | device->writeable = !bdev_read_only(bdev); | |
1006 | seeding = 0; | |
1007 | } | |
1008 | ||
d5e2003c | 1009 | q = bdev_get_queue(bdev); |
90180da4 | 1010 | if (blk_queue_discard(q)) |
d5e2003c | 1011 | device->can_discard = 1; |
d5e2003c | 1012 | |
8a4b83cc | 1013 | device->bdev = bdev; |
dfe25020 | 1014 | device->in_fs_metadata = 0; |
15916de8 CM |
1015 | device->mode = flags; |
1016 | ||
c289811c CM |
1017 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
1018 | fs_devices->rotating = 1; | |
1019 | ||
a0af469b | 1020 | fs_devices->open_devices++; |
55e50e45 ID |
1021 | if (device->writeable && |
1022 | device->devid != BTRFS_DEV_REPLACE_DEVID) { | |
2b82032c YZ |
1023 | fs_devices->rw_devices++; |
1024 | list_add(&device->dev_alloc_list, | |
1025 | &fs_devices->alloc_list); | |
1026 | } | |
4f6c9328 | 1027 | brelse(bh); |
a0af469b | 1028 | continue; |
a061fc8d | 1029 | |
a0af469b CM |
1030 | error_brelse: |
1031 | brelse(bh); | |
d4d77629 | 1032 | blkdev_put(bdev, flags); |
a0af469b | 1033 | continue; |
8a4b83cc | 1034 | } |
a0af469b | 1035 | if (fs_devices->open_devices == 0) { |
20bcd649 | 1036 | ret = -EINVAL; |
a0af469b CM |
1037 | goto out; |
1038 | } | |
2b82032c YZ |
1039 | fs_devices->seeding = seeding; |
1040 | fs_devices->opened = 1; | |
443f24fe | 1041 | fs_devices->latest_bdev = latest_dev->bdev; |
2b82032c | 1042 | fs_devices->total_rw_bytes = 0; |
a0af469b | 1043 | out: |
2b82032c YZ |
1044 | return ret; |
1045 | } | |
1046 | ||
1047 | int btrfs_open_devices(struct btrfs_fs_devices *fs_devices, | |
97288f2c | 1048 | fmode_t flags, void *holder) |
2b82032c YZ |
1049 | { |
1050 | int ret; | |
1051 | ||
1052 | mutex_lock(&uuid_mutex); | |
1053 | if (fs_devices->opened) { | |
e4404d6e YZ |
1054 | fs_devices->opened++; |
1055 | ret = 0; | |
2b82032c | 1056 | } else { |
15916de8 | 1057 | ret = __btrfs_open_devices(fs_devices, flags, holder); |
2b82032c | 1058 | } |
8a4b83cc | 1059 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1060 | return ret; |
1061 | } | |
1062 | ||
6cf86a00 AJ |
1063 | void btrfs_release_disk_super(struct page *page) |
1064 | { | |
1065 | kunmap(page); | |
1066 | put_page(page); | |
1067 | } | |
1068 | ||
1069 | int btrfs_read_disk_super(struct block_device *bdev, u64 bytenr, | |
1070 | struct page **page, struct btrfs_super_block **disk_super) | |
1071 | { | |
1072 | void *p; | |
1073 | pgoff_t index; | |
1074 | ||
1075 | /* make sure our super fits in the device */ | |
1076 | if (bytenr + PAGE_SIZE >= i_size_read(bdev->bd_inode)) | |
1077 | return 1; | |
1078 | ||
1079 | /* make sure our super fits in the page */ | |
1080 | if (sizeof(**disk_super) > PAGE_SIZE) | |
1081 | return 1; | |
1082 | ||
1083 | /* make sure our super doesn't straddle pages on disk */ | |
1084 | index = bytenr >> PAGE_SHIFT; | |
1085 | if ((bytenr + sizeof(**disk_super) - 1) >> PAGE_SHIFT != index) | |
1086 | return 1; | |
1087 | ||
1088 | /* pull in the page with our super */ | |
1089 | *page = read_cache_page_gfp(bdev->bd_inode->i_mapping, | |
1090 | index, GFP_KERNEL); | |
1091 | ||
1092 | if (IS_ERR_OR_NULL(*page)) | |
1093 | return 1; | |
1094 | ||
1095 | p = kmap(*page); | |
1096 | ||
1097 | /* align our pointer to the offset of the super block */ | |
1098 | *disk_super = p + (bytenr & ~PAGE_MASK); | |
1099 | ||
1100 | if (btrfs_super_bytenr(*disk_super) != bytenr || | |
1101 | btrfs_super_magic(*disk_super) != BTRFS_MAGIC) { | |
1102 | btrfs_release_disk_super(*page); | |
1103 | return 1; | |
1104 | } | |
1105 | ||
1106 | if ((*disk_super)->label[0] && | |
1107 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1]) | |
1108 | (*disk_super)->label[BTRFS_LABEL_SIZE - 1] = '\0'; | |
1109 | ||
1110 | return 0; | |
1111 | } | |
1112 | ||
6f60cbd3 DS |
1113 | /* |
1114 | * Look for a btrfs signature on a device. This may be called out of the mount path | |
1115 | * and we are not allowed to call set_blocksize during the scan. The superblock | |
1116 | * is read via pagecache | |
1117 | */ | |
97288f2c | 1118 | int btrfs_scan_one_device(const char *path, fmode_t flags, void *holder, |
8a4b83cc CM |
1119 | struct btrfs_fs_devices **fs_devices_ret) |
1120 | { | |
1121 | struct btrfs_super_block *disk_super; | |
1122 | struct block_device *bdev; | |
6f60cbd3 | 1123 | struct page *page; |
6f60cbd3 | 1124 | int ret = -EINVAL; |
8a4b83cc | 1125 | u64 devid; |
f2984462 | 1126 | u64 transid; |
02db0844 | 1127 | u64 total_devices; |
6f60cbd3 | 1128 | u64 bytenr; |
8a4b83cc | 1129 | |
6f60cbd3 DS |
1130 | /* |
1131 | * we would like to check all the supers, but that would make | |
1132 | * a btrfs mount succeed after a mkfs from a different FS. | |
1133 | * So, we need to add a special mount option to scan for | |
1134 | * later supers, using BTRFS_SUPER_MIRROR_MAX instead | |
1135 | */ | |
1136 | bytenr = btrfs_sb_offset(0); | |
d4d77629 | 1137 | flags |= FMODE_EXCL; |
10f6327b | 1138 | mutex_lock(&uuid_mutex); |
6f60cbd3 DS |
1139 | |
1140 | bdev = blkdev_get_by_path(path, flags, holder); | |
6f60cbd3 DS |
1141 | if (IS_ERR(bdev)) { |
1142 | ret = PTR_ERR(bdev); | |
beaf8ab3 | 1143 | goto error; |
6f60cbd3 DS |
1144 | } |
1145 | ||
6cf86a00 | 1146 | if (btrfs_read_disk_super(bdev, bytenr, &page, &disk_super)) |
6f60cbd3 DS |
1147 | goto error_bdev_put; |
1148 | ||
a343832f | 1149 | devid = btrfs_stack_device_id(&disk_super->dev_item); |
f2984462 | 1150 | transid = btrfs_super_generation(disk_super); |
02db0844 | 1151 | total_devices = btrfs_super_num_devices(disk_super); |
6f60cbd3 | 1152 | |
8a4b83cc | 1153 | ret = device_list_add(path, disk_super, devid, fs_devices_ret); |
60999ca4 DS |
1154 | if (ret > 0) { |
1155 | if (disk_super->label[0]) { | |
62e85577 | 1156 | pr_info("BTRFS: device label %s ", disk_super->label); |
60999ca4 | 1157 | } else { |
62e85577 | 1158 | pr_info("BTRFS: device fsid %pU ", disk_super->fsid); |
60999ca4 DS |
1159 | } |
1160 | ||
62e85577 | 1161 | pr_cont("devid %llu transid %llu %s\n", devid, transid, path); |
60999ca4 DS |
1162 | ret = 0; |
1163 | } | |
02db0844 JB |
1164 | if (!ret && fs_devices_ret) |
1165 | (*fs_devices_ret)->total_devices = total_devices; | |
6f60cbd3 | 1166 | |
6cf86a00 | 1167 | btrfs_release_disk_super(page); |
6f60cbd3 DS |
1168 | |
1169 | error_bdev_put: | |
d4d77629 | 1170 | blkdev_put(bdev, flags); |
8a4b83cc | 1171 | error: |
beaf8ab3 | 1172 | mutex_unlock(&uuid_mutex); |
8a4b83cc CM |
1173 | return ret; |
1174 | } | |
0b86a832 | 1175 | |
6d07bcec MX |
1176 | /* helper to account the used device space in the range */ |
1177 | int btrfs_account_dev_extents_size(struct btrfs_device *device, u64 start, | |
1178 | u64 end, u64 *length) | |
1179 | { | |
1180 | struct btrfs_key key; | |
fb456252 | 1181 | struct btrfs_root *root = device->fs_info->dev_root; |
6d07bcec MX |
1182 | struct btrfs_dev_extent *dev_extent; |
1183 | struct btrfs_path *path; | |
1184 | u64 extent_end; | |
1185 | int ret; | |
1186 | int slot; | |
1187 | struct extent_buffer *l; | |
1188 | ||
1189 | *length = 0; | |
1190 | ||
63a212ab | 1191 | if (start >= device->total_bytes || device->is_tgtdev_for_dev_replace) |
6d07bcec MX |
1192 | return 0; |
1193 | ||
1194 | path = btrfs_alloc_path(); | |
1195 | if (!path) | |
1196 | return -ENOMEM; | |
e4058b54 | 1197 | path->reada = READA_FORWARD; |
6d07bcec MX |
1198 | |
1199 | key.objectid = device->devid; | |
1200 | key.offset = start; | |
1201 | key.type = BTRFS_DEV_EXTENT_KEY; | |
1202 | ||
1203 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); | |
1204 | if (ret < 0) | |
1205 | goto out; | |
1206 | if (ret > 0) { | |
1207 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1208 | if (ret < 0) | |
1209 | goto out; | |
1210 | } | |
1211 | ||
1212 | while (1) { | |
1213 | l = path->nodes[0]; | |
1214 | slot = path->slots[0]; | |
1215 | if (slot >= btrfs_header_nritems(l)) { | |
1216 | ret = btrfs_next_leaf(root, path); | |
1217 | if (ret == 0) | |
1218 | continue; | |
1219 | if (ret < 0) | |
1220 | goto out; | |
1221 | ||
1222 | break; | |
1223 | } | |
1224 | btrfs_item_key_to_cpu(l, &key, slot); | |
1225 | ||
1226 | if (key.objectid < device->devid) | |
1227 | goto next; | |
1228 | ||
1229 | if (key.objectid > device->devid) | |
1230 | break; | |
1231 | ||
962a298f | 1232 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
6d07bcec MX |
1233 | goto next; |
1234 | ||
1235 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
1236 | extent_end = key.offset + btrfs_dev_extent_length(l, | |
1237 | dev_extent); | |
1238 | if (key.offset <= start && extent_end > end) { | |
1239 | *length = end - start + 1; | |
1240 | break; | |
1241 | } else if (key.offset <= start && extent_end > start) | |
1242 | *length += extent_end - start; | |
1243 | else if (key.offset > start && extent_end <= end) | |
1244 | *length += extent_end - key.offset; | |
1245 | else if (key.offset > start && key.offset <= end) { | |
1246 | *length += end - key.offset + 1; | |
1247 | break; | |
1248 | } else if (key.offset > end) | |
1249 | break; | |
1250 | ||
1251 | next: | |
1252 | path->slots[0]++; | |
1253 | } | |
1254 | ret = 0; | |
1255 | out: | |
1256 | btrfs_free_path(path); | |
1257 | return ret; | |
1258 | } | |
1259 | ||
499f377f | 1260 | static int contains_pending_extent(struct btrfs_transaction *transaction, |
6df9a95e JB |
1261 | struct btrfs_device *device, |
1262 | u64 *start, u64 len) | |
1263 | { | |
fb456252 | 1264 | struct btrfs_fs_info *fs_info = device->fs_info; |
6df9a95e | 1265 | struct extent_map *em; |
499f377f | 1266 | struct list_head *search_list = &fs_info->pinned_chunks; |
6df9a95e | 1267 | int ret = 0; |
1b984508 | 1268 | u64 physical_start = *start; |
6df9a95e | 1269 | |
499f377f JM |
1270 | if (transaction) |
1271 | search_list = &transaction->pending_chunks; | |
04216820 FM |
1272 | again: |
1273 | list_for_each_entry(em, search_list, list) { | |
6df9a95e JB |
1274 | struct map_lookup *map; |
1275 | int i; | |
1276 | ||
95617d69 | 1277 | map = em->map_lookup; |
6df9a95e | 1278 | for (i = 0; i < map->num_stripes; i++) { |
c152b63e FM |
1279 | u64 end; |
1280 | ||
6df9a95e JB |
1281 | if (map->stripes[i].dev != device) |
1282 | continue; | |
1b984508 | 1283 | if (map->stripes[i].physical >= physical_start + len || |
6df9a95e | 1284 | map->stripes[i].physical + em->orig_block_len <= |
1b984508 | 1285 | physical_start) |
6df9a95e | 1286 | continue; |
c152b63e FM |
1287 | /* |
1288 | * Make sure that while processing the pinned list we do | |
1289 | * not override our *start with a lower value, because | |
1290 | * we can have pinned chunks that fall within this | |
1291 | * device hole and that have lower physical addresses | |
1292 | * than the pending chunks we processed before. If we | |
1293 | * do not take this special care we can end up getting | |
1294 | * 2 pending chunks that start at the same physical | |
1295 | * device offsets because the end offset of a pinned | |
1296 | * chunk can be equal to the start offset of some | |
1297 | * pending chunk. | |
1298 | */ | |
1299 | end = map->stripes[i].physical + em->orig_block_len; | |
1300 | if (end > *start) { | |
1301 | *start = end; | |
1302 | ret = 1; | |
1303 | } | |
6df9a95e JB |
1304 | } |
1305 | } | |
499f377f JM |
1306 | if (search_list != &fs_info->pinned_chunks) { |
1307 | search_list = &fs_info->pinned_chunks; | |
04216820 FM |
1308 | goto again; |
1309 | } | |
6df9a95e JB |
1310 | |
1311 | return ret; | |
1312 | } | |
1313 | ||
1314 | ||
0b86a832 | 1315 | /* |
499f377f JM |
1316 | * find_free_dev_extent_start - find free space in the specified device |
1317 | * @device: the device which we search the free space in | |
1318 | * @num_bytes: the size of the free space that we need | |
1319 | * @search_start: the position from which to begin the search | |
1320 | * @start: store the start of the free space. | |
1321 | * @len: the size of the free space. that we find, or the size | |
1322 | * of the max free space if we don't find suitable free space | |
7bfc837d | 1323 | * |
0b86a832 CM |
1324 | * this uses a pretty simple search, the expectation is that it is |
1325 | * called very infrequently and that a given device has a small number | |
1326 | * of extents | |
7bfc837d MX |
1327 | * |
1328 | * @start is used to store the start of the free space if we find. But if we | |
1329 | * don't find suitable free space, it will be used to store the start position | |
1330 | * of the max free space. | |
1331 | * | |
1332 | * @len is used to store the size of the free space that we find. | |
1333 | * But if we don't find suitable free space, it is used to store the size of | |
1334 | * the max free space. | |
0b86a832 | 1335 | */ |
499f377f JM |
1336 | int find_free_dev_extent_start(struct btrfs_transaction *transaction, |
1337 | struct btrfs_device *device, u64 num_bytes, | |
1338 | u64 search_start, u64 *start, u64 *len) | |
0b86a832 | 1339 | { |
0b246afa JM |
1340 | struct btrfs_fs_info *fs_info = device->fs_info; |
1341 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 | 1342 | struct btrfs_key key; |
7bfc837d | 1343 | struct btrfs_dev_extent *dev_extent; |
2b82032c | 1344 | struct btrfs_path *path; |
7bfc837d MX |
1345 | u64 hole_size; |
1346 | u64 max_hole_start; | |
1347 | u64 max_hole_size; | |
1348 | u64 extent_end; | |
0b86a832 CM |
1349 | u64 search_end = device->total_bytes; |
1350 | int ret; | |
7bfc837d | 1351 | int slot; |
0b86a832 | 1352 | struct extent_buffer *l; |
8cdc7c5b FM |
1353 | u64 min_search_start; |
1354 | ||
1355 | /* | |
1356 | * We don't want to overwrite the superblock on the drive nor any area | |
1357 | * used by the boot loader (grub for example), so we make sure to start | |
1358 | * at an offset of at least 1MB. | |
1359 | */ | |
0b246afa | 1360 | min_search_start = max(fs_info->alloc_start, 1024ull * 1024); |
8cdc7c5b | 1361 | search_start = max(search_start, min_search_start); |
0b86a832 | 1362 | |
6df9a95e JB |
1363 | path = btrfs_alloc_path(); |
1364 | if (!path) | |
1365 | return -ENOMEM; | |
f2ab7618 | 1366 | |
7bfc837d MX |
1367 | max_hole_start = search_start; |
1368 | max_hole_size = 0; | |
1369 | ||
f2ab7618 | 1370 | again: |
63a212ab | 1371 | if (search_start >= search_end || device->is_tgtdev_for_dev_replace) { |
7bfc837d | 1372 | ret = -ENOSPC; |
6df9a95e | 1373 | goto out; |
7bfc837d MX |
1374 | } |
1375 | ||
e4058b54 | 1376 | path->reada = READA_FORWARD; |
6df9a95e JB |
1377 | path->search_commit_root = 1; |
1378 | path->skip_locking = 1; | |
7bfc837d | 1379 | |
0b86a832 CM |
1380 | key.objectid = device->devid; |
1381 | key.offset = search_start; | |
1382 | key.type = BTRFS_DEV_EXTENT_KEY; | |
7bfc837d | 1383 | |
125ccb0a | 1384 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
0b86a832 | 1385 | if (ret < 0) |
7bfc837d | 1386 | goto out; |
1fcbac58 YZ |
1387 | if (ret > 0) { |
1388 | ret = btrfs_previous_item(root, path, key.objectid, key.type); | |
1389 | if (ret < 0) | |
7bfc837d | 1390 | goto out; |
1fcbac58 | 1391 | } |
7bfc837d | 1392 | |
0b86a832 CM |
1393 | while (1) { |
1394 | l = path->nodes[0]; | |
1395 | slot = path->slots[0]; | |
1396 | if (slot >= btrfs_header_nritems(l)) { | |
1397 | ret = btrfs_next_leaf(root, path); | |
1398 | if (ret == 0) | |
1399 | continue; | |
1400 | if (ret < 0) | |
7bfc837d MX |
1401 | goto out; |
1402 | ||
1403 | break; | |
0b86a832 CM |
1404 | } |
1405 | btrfs_item_key_to_cpu(l, &key, slot); | |
1406 | ||
1407 | if (key.objectid < device->devid) | |
1408 | goto next; | |
1409 | ||
1410 | if (key.objectid > device->devid) | |
7bfc837d | 1411 | break; |
0b86a832 | 1412 | |
962a298f | 1413 | if (key.type != BTRFS_DEV_EXTENT_KEY) |
7bfc837d | 1414 | goto next; |
9779b72f | 1415 | |
7bfc837d MX |
1416 | if (key.offset > search_start) { |
1417 | hole_size = key.offset - search_start; | |
9779b72f | 1418 | |
6df9a95e JB |
1419 | /* |
1420 | * Have to check before we set max_hole_start, otherwise | |
1421 | * we could end up sending back this offset anyway. | |
1422 | */ | |
499f377f | 1423 | if (contains_pending_extent(transaction, device, |
6df9a95e | 1424 | &search_start, |
1b984508 FL |
1425 | hole_size)) { |
1426 | if (key.offset >= search_start) { | |
1427 | hole_size = key.offset - search_start; | |
1428 | } else { | |
1429 | WARN_ON_ONCE(1); | |
1430 | hole_size = 0; | |
1431 | } | |
1432 | } | |
6df9a95e | 1433 | |
7bfc837d MX |
1434 | if (hole_size > max_hole_size) { |
1435 | max_hole_start = search_start; | |
1436 | max_hole_size = hole_size; | |
1437 | } | |
9779b72f | 1438 | |
7bfc837d MX |
1439 | /* |
1440 | * If this free space is greater than which we need, | |
1441 | * it must be the max free space that we have found | |
1442 | * until now, so max_hole_start must point to the start | |
1443 | * of this free space and the length of this free space | |
1444 | * is stored in max_hole_size. Thus, we return | |
1445 | * max_hole_start and max_hole_size and go back to the | |
1446 | * caller. | |
1447 | */ | |
1448 | if (hole_size >= num_bytes) { | |
1449 | ret = 0; | |
1450 | goto out; | |
0b86a832 CM |
1451 | } |
1452 | } | |
0b86a832 | 1453 | |
0b86a832 | 1454 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); |
7bfc837d MX |
1455 | extent_end = key.offset + btrfs_dev_extent_length(l, |
1456 | dev_extent); | |
1457 | if (extent_end > search_start) | |
1458 | search_start = extent_end; | |
0b86a832 CM |
1459 | next: |
1460 | path->slots[0]++; | |
1461 | cond_resched(); | |
1462 | } | |
0b86a832 | 1463 | |
38c01b96 | 1464 | /* |
1465 | * At this point, search_start should be the end of | |
1466 | * allocated dev extents, and when shrinking the device, | |
1467 | * search_end may be smaller than search_start. | |
1468 | */ | |
f2ab7618 | 1469 | if (search_end > search_start) { |
38c01b96 | 1470 | hole_size = search_end - search_start; |
1471 | ||
499f377f | 1472 | if (contains_pending_extent(transaction, device, &search_start, |
f2ab7618 ZL |
1473 | hole_size)) { |
1474 | btrfs_release_path(path); | |
1475 | goto again; | |
1476 | } | |
0b86a832 | 1477 | |
f2ab7618 ZL |
1478 | if (hole_size > max_hole_size) { |
1479 | max_hole_start = search_start; | |
1480 | max_hole_size = hole_size; | |
1481 | } | |
6df9a95e JB |
1482 | } |
1483 | ||
7bfc837d | 1484 | /* See above. */ |
f2ab7618 | 1485 | if (max_hole_size < num_bytes) |
7bfc837d MX |
1486 | ret = -ENOSPC; |
1487 | else | |
1488 | ret = 0; | |
1489 | ||
1490 | out: | |
2b82032c | 1491 | btrfs_free_path(path); |
7bfc837d | 1492 | *start = max_hole_start; |
b2117a39 | 1493 | if (len) |
7bfc837d | 1494 | *len = max_hole_size; |
0b86a832 CM |
1495 | return ret; |
1496 | } | |
1497 | ||
499f377f JM |
1498 | int find_free_dev_extent(struct btrfs_trans_handle *trans, |
1499 | struct btrfs_device *device, u64 num_bytes, | |
1500 | u64 *start, u64 *len) | |
1501 | { | |
499f377f | 1502 | /* FIXME use last free of some kind */ |
499f377f | 1503 | return find_free_dev_extent_start(trans->transaction, device, |
8cdc7c5b | 1504 | num_bytes, 0, start, len); |
499f377f JM |
1505 | } |
1506 | ||
b2950863 | 1507 | static int btrfs_free_dev_extent(struct btrfs_trans_handle *trans, |
8f18cf13 | 1508 | struct btrfs_device *device, |
2196d6e8 | 1509 | u64 start, u64 *dev_extent_len) |
8f18cf13 | 1510 | { |
0b246afa JM |
1511 | struct btrfs_fs_info *fs_info = device->fs_info; |
1512 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 CM |
1513 | int ret; |
1514 | struct btrfs_path *path; | |
8f18cf13 | 1515 | struct btrfs_key key; |
a061fc8d CM |
1516 | struct btrfs_key found_key; |
1517 | struct extent_buffer *leaf = NULL; | |
1518 | struct btrfs_dev_extent *extent = NULL; | |
8f18cf13 CM |
1519 | |
1520 | path = btrfs_alloc_path(); | |
1521 | if (!path) | |
1522 | return -ENOMEM; | |
1523 | ||
1524 | key.objectid = device->devid; | |
1525 | key.offset = start; | |
1526 | key.type = BTRFS_DEV_EXTENT_KEY; | |
924cd8fb | 1527 | again: |
8f18cf13 | 1528 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); |
a061fc8d CM |
1529 | if (ret > 0) { |
1530 | ret = btrfs_previous_item(root, path, key.objectid, | |
1531 | BTRFS_DEV_EXTENT_KEY); | |
b0b802d7 TI |
1532 | if (ret) |
1533 | goto out; | |
a061fc8d CM |
1534 | leaf = path->nodes[0]; |
1535 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1536 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1537 | struct btrfs_dev_extent); | |
1538 | BUG_ON(found_key.offset > start || found_key.offset + | |
1539 | btrfs_dev_extent_length(leaf, extent) < start); | |
924cd8fb MX |
1540 | key = found_key; |
1541 | btrfs_release_path(path); | |
1542 | goto again; | |
a061fc8d CM |
1543 | } else if (ret == 0) { |
1544 | leaf = path->nodes[0]; | |
1545 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1546 | struct btrfs_dev_extent); | |
79787eaa | 1547 | } else { |
0b246afa | 1548 | btrfs_handle_fs_error(fs_info, ret, "Slot search failed"); |
79787eaa | 1549 | goto out; |
a061fc8d | 1550 | } |
8f18cf13 | 1551 | |
2196d6e8 MX |
1552 | *dev_extent_len = btrfs_dev_extent_length(leaf, extent); |
1553 | ||
8f18cf13 | 1554 | ret = btrfs_del_item(trans, root, path); |
79787eaa | 1555 | if (ret) { |
0b246afa JM |
1556 | btrfs_handle_fs_error(fs_info, ret, |
1557 | "Failed to remove dev extent item"); | |
13212b54 | 1558 | } else { |
3204d33c | 1559 | set_bit(BTRFS_TRANS_HAVE_FREE_BGS, &trans->transaction->flags); |
79787eaa | 1560 | } |
b0b802d7 | 1561 | out: |
8f18cf13 CM |
1562 | btrfs_free_path(path); |
1563 | return ret; | |
1564 | } | |
1565 | ||
48a3b636 ES |
1566 | static int btrfs_alloc_dev_extent(struct btrfs_trans_handle *trans, |
1567 | struct btrfs_device *device, | |
1568 | u64 chunk_tree, u64 chunk_objectid, | |
1569 | u64 chunk_offset, u64 start, u64 num_bytes) | |
0b86a832 CM |
1570 | { |
1571 | int ret; | |
1572 | struct btrfs_path *path; | |
0b246afa JM |
1573 | struct btrfs_fs_info *fs_info = device->fs_info; |
1574 | struct btrfs_root *root = fs_info->dev_root; | |
0b86a832 CM |
1575 | struct btrfs_dev_extent *extent; |
1576 | struct extent_buffer *leaf; | |
1577 | struct btrfs_key key; | |
1578 | ||
dfe25020 | 1579 | WARN_ON(!device->in_fs_metadata); |
63a212ab | 1580 | WARN_ON(device->is_tgtdev_for_dev_replace); |
0b86a832 CM |
1581 | path = btrfs_alloc_path(); |
1582 | if (!path) | |
1583 | return -ENOMEM; | |
1584 | ||
0b86a832 | 1585 | key.objectid = device->devid; |
2b82032c | 1586 | key.offset = start; |
0b86a832 CM |
1587 | key.type = BTRFS_DEV_EXTENT_KEY; |
1588 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
1589 | sizeof(*extent)); | |
2cdcecbc MF |
1590 | if (ret) |
1591 | goto out; | |
0b86a832 CM |
1592 | |
1593 | leaf = path->nodes[0]; | |
1594 | extent = btrfs_item_ptr(leaf, path->slots[0], | |
1595 | struct btrfs_dev_extent); | |
e17cade2 CM |
1596 | btrfs_set_dev_extent_chunk_tree(leaf, extent, chunk_tree); |
1597 | btrfs_set_dev_extent_chunk_objectid(leaf, extent, chunk_objectid); | |
1598 | btrfs_set_dev_extent_chunk_offset(leaf, extent, chunk_offset); | |
1599 | ||
0b246afa | 1600 | write_extent_buffer_chunk_tree_uuid(leaf, fs_info->chunk_tree_uuid); |
e17cade2 | 1601 | |
0b86a832 CM |
1602 | btrfs_set_dev_extent_length(leaf, extent, num_bytes); |
1603 | btrfs_mark_buffer_dirty(leaf); | |
2cdcecbc | 1604 | out: |
0b86a832 CM |
1605 | btrfs_free_path(path); |
1606 | return ret; | |
1607 | } | |
1608 | ||
6df9a95e | 1609 | static u64 find_next_chunk(struct btrfs_fs_info *fs_info) |
0b86a832 | 1610 | { |
6df9a95e JB |
1611 | struct extent_map_tree *em_tree; |
1612 | struct extent_map *em; | |
1613 | struct rb_node *n; | |
1614 | u64 ret = 0; | |
0b86a832 | 1615 | |
6df9a95e JB |
1616 | em_tree = &fs_info->mapping_tree.map_tree; |
1617 | read_lock(&em_tree->lock); | |
1618 | n = rb_last(&em_tree->map); | |
1619 | if (n) { | |
1620 | em = rb_entry(n, struct extent_map, rb_node); | |
1621 | ret = em->start + em->len; | |
0b86a832 | 1622 | } |
6df9a95e JB |
1623 | read_unlock(&em_tree->lock); |
1624 | ||
0b86a832 CM |
1625 | return ret; |
1626 | } | |
1627 | ||
53f10659 ID |
1628 | static noinline int find_next_devid(struct btrfs_fs_info *fs_info, |
1629 | u64 *devid_ret) | |
0b86a832 CM |
1630 | { |
1631 | int ret; | |
1632 | struct btrfs_key key; | |
1633 | struct btrfs_key found_key; | |
2b82032c YZ |
1634 | struct btrfs_path *path; |
1635 | ||
2b82032c YZ |
1636 | path = btrfs_alloc_path(); |
1637 | if (!path) | |
1638 | return -ENOMEM; | |
0b86a832 CM |
1639 | |
1640 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
1641 | key.type = BTRFS_DEV_ITEM_KEY; | |
1642 | key.offset = (u64)-1; | |
1643 | ||
53f10659 | 1644 | ret = btrfs_search_slot(NULL, fs_info->chunk_root, &key, path, 0, 0); |
0b86a832 CM |
1645 | if (ret < 0) |
1646 | goto error; | |
1647 | ||
79787eaa | 1648 | BUG_ON(ret == 0); /* Corruption */ |
0b86a832 | 1649 | |
53f10659 ID |
1650 | ret = btrfs_previous_item(fs_info->chunk_root, path, |
1651 | BTRFS_DEV_ITEMS_OBJECTID, | |
0b86a832 CM |
1652 | BTRFS_DEV_ITEM_KEY); |
1653 | if (ret) { | |
53f10659 | 1654 | *devid_ret = 1; |
0b86a832 CM |
1655 | } else { |
1656 | btrfs_item_key_to_cpu(path->nodes[0], &found_key, | |
1657 | path->slots[0]); | |
53f10659 | 1658 | *devid_ret = found_key.offset + 1; |
0b86a832 CM |
1659 | } |
1660 | ret = 0; | |
1661 | error: | |
2b82032c | 1662 | btrfs_free_path(path); |
0b86a832 CM |
1663 | return ret; |
1664 | } | |
1665 | ||
1666 | /* | |
1667 | * the device information is stored in the chunk root | |
1668 | * the btrfs_device struct should be fully filled in | |
1669 | */ | |
48a3b636 | 1670 | static int btrfs_add_device(struct btrfs_trans_handle *trans, |
5b4aacef | 1671 | struct btrfs_fs_info *fs_info, |
48a3b636 | 1672 | struct btrfs_device *device) |
0b86a832 | 1673 | { |
5b4aacef | 1674 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
1675 | int ret; |
1676 | struct btrfs_path *path; | |
1677 | struct btrfs_dev_item *dev_item; | |
1678 | struct extent_buffer *leaf; | |
1679 | struct btrfs_key key; | |
1680 | unsigned long ptr; | |
0b86a832 | 1681 | |
0b86a832 CM |
1682 | path = btrfs_alloc_path(); |
1683 | if (!path) | |
1684 | return -ENOMEM; | |
1685 | ||
0b86a832 CM |
1686 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1687 | key.type = BTRFS_DEV_ITEM_KEY; | |
2b82032c | 1688 | key.offset = device->devid; |
0b86a832 CM |
1689 | |
1690 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
0d81ba5d | 1691 | sizeof(*dev_item)); |
0b86a832 CM |
1692 | if (ret) |
1693 | goto out; | |
1694 | ||
1695 | leaf = path->nodes[0]; | |
1696 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
1697 | ||
1698 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2b82032c | 1699 | btrfs_set_device_generation(leaf, dev_item, 0); |
0b86a832 CM |
1700 | btrfs_set_device_type(leaf, dev_item, device->type); |
1701 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
1702 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
1703 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
1704 | btrfs_set_device_total_bytes(leaf, dev_item, |
1705 | btrfs_device_get_disk_total_bytes(device)); | |
1706 | btrfs_set_device_bytes_used(leaf, dev_item, | |
1707 | btrfs_device_get_bytes_used(device)); | |
e17cade2 CM |
1708 | btrfs_set_device_group(leaf, dev_item, 0); |
1709 | btrfs_set_device_seek_speed(leaf, dev_item, 0); | |
1710 | btrfs_set_device_bandwidth(leaf, dev_item, 0); | |
c3027eb5 | 1711 | btrfs_set_device_start_offset(leaf, dev_item, 0); |
0b86a832 | 1712 | |
410ba3a2 | 1713 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 1714 | write_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
1473b24e | 1715 | ptr = btrfs_device_fsid(dev_item); |
0b246afa | 1716 | write_extent_buffer(leaf, fs_info->fsid, ptr, BTRFS_UUID_SIZE); |
0b86a832 | 1717 | btrfs_mark_buffer_dirty(leaf); |
0b86a832 | 1718 | |
2b82032c | 1719 | ret = 0; |
0b86a832 CM |
1720 | out: |
1721 | btrfs_free_path(path); | |
1722 | return ret; | |
1723 | } | |
8f18cf13 | 1724 | |
5a1972bd QW |
1725 | /* |
1726 | * Function to update ctime/mtime for a given device path. | |
1727 | * Mainly used for ctime/mtime based probe like libblkid. | |
1728 | */ | |
1729 | static void update_dev_time(char *path_name) | |
1730 | { | |
1731 | struct file *filp; | |
1732 | ||
1733 | filp = filp_open(path_name, O_RDWR, 0); | |
98af592f | 1734 | if (IS_ERR(filp)) |
5a1972bd QW |
1735 | return; |
1736 | file_update_time(filp); | |
1737 | filp_close(filp, NULL); | |
5a1972bd QW |
1738 | } |
1739 | ||
5b4aacef | 1740 | static int btrfs_rm_dev_item(struct btrfs_fs_info *fs_info, |
a061fc8d CM |
1741 | struct btrfs_device *device) |
1742 | { | |
5b4aacef | 1743 | struct btrfs_root *root = fs_info->chunk_root; |
a061fc8d CM |
1744 | int ret; |
1745 | struct btrfs_path *path; | |
a061fc8d | 1746 | struct btrfs_key key; |
a061fc8d CM |
1747 | struct btrfs_trans_handle *trans; |
1748 | ||
a061fc8d CM |
1749 | path = btrfs_alloc_path(); |
1750 | if (!path) | |
1751 | return -ENOMEM; | |
1752 | ||
a22285a6 | 1753 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
1754 | if (IS_ERR(trans)) { |
1755 | btrfs_free_path(path); | |
1756 | return PTR_ERR(trans); | |
1757 | } | |
a061fc8d CM |
1758 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
1759 | key.type = BTRFS_DEV_ITEM_KEY; | |
1760 | key.offset = device->devid; | |
1761 | ||
1762 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
1763 | if (ret < 0) | |
1764 | goto out; | |
1765 | ||
1766 | if (ret > 0) { | |
1767 | ret = -ENOENT; | |
1768 | goto out; | |
1769 | } | |
1770 | ||
1771 | ret = btrfs_del_item(trans, root, path); | |
1772 | if (ret) | |
1773 | goto out; | |
a061fc8d CM |
1774 | out: |
1775 | btrfs_free_path(path); | |
1776 | btrfs_commit_transaction(trans, root); | |
1777 | return ret; | |
1778 | } | |
1779 | ||
3cc31a0d DS |
1780 | /* |
1781 | * Verify that @num_devices satisfies the RAID profile constraints in the whole | |
1782 | * filesystem. It's up to the caller to adjust that number regarding eg. device | |
1783 | * replace. | |
1784 | */ | |
1785 | static int btrfs_check_raid_min_devices(struct btrfs_fs_info *fs_info, | |
1786 | u64 num_devices) | |
a061fc8d | 1787 | { |
a061fc8d | 1788 | u64 all_avail; |
de98ced9 | 1789 | unsigned seq; |
418775a2 | 1790 | int i; |
a061fc8d | 1791 | |
de98ced9 | 1792 | do { |
bd45ffbc | 1793 | seq = read_seqbegin(&fs_info->profiles_lock); |
de98ced9 | 1794 | |
bd45ffbc AJ |
1795 | all_avail = fs_info->avail_data_alloc_bits | |
1796 | fs_info->avail_system_alloc_bits | | |
1797 | fs_info->avail_metadata_alloc_bits; | |
1798 | } while (read_seqretry(&fs_info->profiles_lock, seq)); | |
a061fc8d | 1799 | |
418775a2 DS |
1800 | for (i = 0; i < BTRFS_NR_RAID_TYPES; i++) { |
1801 | if (!(all_avail & btrfs_raid_group[i])) | |
1802 | continue; | |
a061fc8d | 1803 | |
418775a2 DS |
1804 | if (num_devices < btrfs_raid_array[i].devs_min) { |
1805 | int ret = btrfs_raid_mindev_error[i]; | |
bd45ffbc | 1806 | |
418775a2 DS |
1807 | if (ret) |
1808 | return ret; | |
1809 | } | |
53b381b3 DW |
1810 | } |
1811 | ||
bd45ffbc | 1812 | return 0; |
f1fa7f26 AJ |
1813 | } |
1814 | ||
88acff64 AJ |
1815 | struct btrfs_device *btrfs_find_next_active_device(struct btrfs_fs_devices *fs_devs, |
1816 | struct btrfs_device *device) | |
a061fc8d | 1817 | { |
2b82032c | 1818 | struct btrfs_device *next_device; |
88acff64 AJ |
1819 | |
1820 | list_for_each_entry(next_device, &fs_devs->devices, dev_list) { | |
1821 | if (next_device != device && | |
1822 | !next_device->missing && next_device->bdev) | |
1823 | return next_device; | |
1824 | } | |
1825 | ||
1826 | return NULL; | |
1827 | } | |
1828 | ||
1829 | /* | |
1830 | * Helper function to check if the given device is part of s_bdev / latest_bdev | |
1831 | * and replace it with the provided or the next active device, in the context | |
1832 | * where this function called, there should be always be another device (or | |
1833 | * this_dev) which is active. | |
1834 | */ | |
1835 | void btrfs_assign_next_active_device(struct btrfs_fs_info *fs_info, | |
1836 | struct btrfs_device *device, struct btrfs_device *this_dev) | |
1837 | { | |
1838 | struct btrfs_device *next_device; | |
1839 | ||
1840 | if (this_dev) | |
1841 | next_device = this_dev; | |
1842 | else | |
1843 | next_device = btrfs_find_next_active_device(fs_info->fs_devices, | |
1844 | device); | |
1845 | ASSERT(next_device); | |
1846 | ||
1847 | if (fs_info->sb->s_bdev && | |
1848 | (fs_info->sb->s_bdev == device->bdev)) | |
1849 | fs_info->sb->s_bdev = next_device->bdev; | |
1850 | ||
1851 | if (fs_info->fs_devices->latest_bdev == device->bdev) | |
1852 | fs_info->fs_devices->latest_bdev = next_device->bdev; | |
1853 | } | |
1854 | ||
6b526ed7 | 1855 | int btrfs_rm_device(struct btrfs_root *root, char *device_path, u64 devid) |
f1fa7f26 | 1856 | { |
0b246afa | 1857 | struct btrfs_fs_info *fs_info = root->fs_info; |
f1fa7f26 | 1858 | struct btrfs_device *device; |
1f78160c | 1859 | struct btrfs_fs_devices *cur_devices; |
2b82032c | 1860 | u64 num_devices; |
a061fc8d | 1861 | int ret = 0; |
1f78160c | 1862 | bool clear_super = false; |
a061fc8d | 1863 | |
a061fc8d CM |
1864 | mutex_lock(&uuid_mutex); |
1865 | ||
0b246afa JM |
1866 | num_devices = fs_info->fs_devices->num_devices; |
1867 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); | |
1868 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { | |
8dabb742 SB |
1869 | WARN_ON(num_devices < 1); |
1870 | num_devices--; | |
1871 | } | |
0b246afa | 1872 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
8dabb742 | 1873 | |
0b246afa | 1874 | ret = btrfs_check_raid_min_devices(fs_info, num_devices - 1); |
f1fa7f26 | 1875 | if (ret) |
a061fc8d | 1876 | goto out; |
a061fc8d | 1877 | |
5c5c0df0 | 1878 | ret = btrfs_find_device_by_devspec(root, devid, device_path, |
24fc572f AJ |
1879 | &device); |
1880 | if (ret) | |
53b381b3 | 1881 | goto out; |
dfe25020 | 1882 | |
63a212ab | 1883 | if (device->is_tgtdev_for_dev_replace) { |
183860f6 | 1884 | ret = BTRFS_ERROR_DEV_TGT_REPLACE; |
24fc572f | 1885 | goto out; |
63a212ab SB |
1886 | } |
1887 | ||
0b246afa | 1888 | if (device->writeable && fs_info->fs_devices->rw_devices == 1) { |
183860f6 | 1889 | ret = BTRFS_ERROR_DEV_ONLY_WRITABLE; |
24fc572f | 1890 | goto out; |
2b82032c YZ |
1891 | } |
1892 | ||
1893 | if (device->writeable) { | |
0b246afa | 1894 | lock_chunks(fs_info); |
2b82032c | 1895 | list_del_init(&device->dev_alloc_list); |
c3929c36 | 1896 | device->fs_devices->rw_devices--; |
0b246afa | 1897 | unlock_chunks(fs_info); |
1f78160c | 1898 | clear_super = true; |
dfe25020 | 1899 | } |
a061fc8d | 1900 | |
d7901554 | 1901 | mutex_unlock(&uuid_mutex); |
a061fc8d | 1902 | ret = btrfs_shrink_device(device, 0); |
d7901554 | 1903 | mutex_lock(&uuid_mutex); |
a061fc8d | 1904 | if (ret) |
9b3517e9 | 1905 | goto error_undo; |
a061fc8d | 1906 | |
63a212ab SB |
1907 | /* |
1908 | * TODO: the superblock still includes this device in its num_devices | |
1909 | * counter although write_all_supers() is not locked out. This | |
1910 | * could give a filesystem state which requires a degraded mount. | |
1911 | */ | |
0b246afa | 1912 | ret = btrfs_rm_dev_item(fs_info, device); |
a061fc8d | 1913 | if (ret) |
9b3517e9 | 1914 | goto error_undo; |
a061fc8d | 1915 | |
2b82032c | 1916 | device->in_fs_metadata = 0; |
0b246afa | 1917 | btrfs_scrub_cancel_dev(fs_info, device); |
e5e9a520 CM |
1918 | |
1919 | /* | |
1920 | * the device list mutex makes sure that we don't change | |
1921 | * the device list while someone else is writing out all | |
d7306801 FDBM |
1922 | * the device supers. Whoever is writing all supers, should |
1923 | * lock the device list mutex before getting the number of | |
1924 | * devices in the super block (super_copy). Conversely, | |
1925 | * whoever updates the number of devices in the super block | |
1926 | * (super_copy) should hold the device list mutex. | |
e5e9a520 | 1927 | */ |
1f78160c XG |
1928 | |
1929 | cur_devices = device->fs_devices; | |
0b246afa | 1930 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c | 1931 | list_del_rcu(&device->dev_list); |
e5e9a520 | 1932 | |
e4404d6e | 1933 | device->fs_devices->num_devices--; |
02db0844 | 1934 | device->fs_devices->total_devices--; |
2b82032c | 1935 | |
cd02dca5 | 1936 | if (device->missing) |
3a7d55c8 | 1937 | device->fs_devices->missing_devices--; |
cd02dca5 | 1938 | |
0b246afa | 1939 | btrfs_assign_next_active_device(fs_info, device, NULL); |
2b82032c | 1940 | |
0bfaa9c5 | 1941 | if (device->bdev) { |
e4404d6e | 1942 | device->fs_devices->open_devices--; |
0bfaa9c5 | 1943 | /* remove sysfs entry */ |
0b246afa | 1944 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
0bfaa9c5 | 1945 | } |
99994cde | 1946 | |
0b246afa JM |
1947 | num_devices = btrfs_super_num_devices(fs_info->super_copy) - 1; |
1948 | btrfs_set_super_num_devices(fs_info->super_copy, num_devices); | |
1949 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
2b82032c | 1950 | |
cea67ab9 JM |
1951 | /* |
1952 | * at this point, the device is zero sized and detached from | |
1953 | * the devices list. All that's left is to zero out the old | |
1954 | * supers and free the device. | |
1955 | */ | |
1956 | if (device->writeable) | |
1957 | btrfs_scratch_superblocks(device->bdev, device->name->str); | |
1958 | ||
1959 | btrfs_close_bdev(device); | |
1960 | call_rcu(&device->rcu, free_device); | |
1961 | ||
1f78160c | 1962 | if (cur_devices->open_devices == 0) { |
e4404d6e | 1963 | struct btrfs_fs_devices *fs_devices; |
0b246afa | 1964 | fs_devices = fs_info->fs_devices; |
e4404d6e | 1965 | while (fs_devices) { |
8321cf25 RS |
1966 | if (fs_devices->seed == cur_devices) { |
1967 | fs_devices->seed = cur_devices->seed; | |
e4404d6e | 1968 | break; |
8321cf25 | 1969 | } |
e4404d6e | 1970 | fs_devices = fs_devices->seed; |
2b82032c | 1971 | } |
1f78160c | 1972 | cur_devices->seed = NULL; |
1f78160c | 1973 | __btrfs_close_devices(cur_devices); |
1f78160c | 1974 | free_fs_devices(cur_devices); |
2b82032c YZ |
1975 | } |
1976 | ||
0b246afa JM |
1977 | fs_info->num_tolerated_disk_barrier_failures = |
1978 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
5af3e8cc | 1979 | |
a061fc8d CM |
1980 | out: |
1981 | mutex_unlock(&uuid_mutex); | |
a061fc8d | 1982 | return ret; |
24fc572f | 1983 | |
9b3517e9 ID |
1984 | error_undo: |
1985 | if (device->writeable) { | |
0b246afa | 1986 | lock_chunks(fs_info); |
9b3517e9 | 1987 | list_add(&device->dev_alloc_list, |
0b246afa | 1988 | &fs_info->fs_devices->alloc_list); |
c3929c36 | 1989 | device->fs_devices->rw_devices++; |
0b246afa | 1990 | unlock_chunks(fs_info); |
9b3517e9 | 1991 | } |
24fc572f | 1992 | goto out; |
a061fc8d CM |
1993 | } |
1994 | ||
084b6e7c QW |
1995 | void btrfs_rm_dev_replace_remove_srcdev(struct btrfs_fs_info *fs_info, |
1996 | struct btrfs_device *srcdev) | |
e93c89c1 | 1997 | { |
d51908ce AJ |
1998 | struct btrfs_fs_devices *fs_devices; |
1999 | ||
e93c89c1 | 2000 | WARN_ON(!mutex_is_locked(&fs_info->fs_devices->device_list_mutex)); |
1357272f | 2001 | |
25e8e911 AJ |
2002 | /* |
2003 | * in case of fs with no seed, srcdev->fs_devices will point | |
2004 | * to fs_devices of fs_info. However when the dev being replaced is | |
2005 | * a seed dev it will point to the seed's local fs_devices. In short | |
2006 | * srcdev will have its correct fs_devices in both the cases. | |
2007 | */ | |
2008 | fs_devices = srcdev->fs_devices; | |
d51908ce | 2009 | |
e93c89c1 SB |
2010 | list_del_rcu(&srcdev->dev_list); |
2011 | list_del_rcu(&srcdev->dev_alloc_list); | |
d51908ce | 2012 | fs_devices->num_devices--; |
82372bc8 | 2013 | if (srcdev->missing) |
d51908ce | 2014 | fs_devices->missing_devices--; |
e93c89c1 | 2015 | |
48b3b9d4 | 2016 | if (srcdev->writeable) |
82372bc8 | 2017 | fs_devices->rw_devices--; |
1357272f | 2018 | |
82372bc8 | 2019 | if (srcdev->bdev) |
d51908ce | 2020 | fs_devices->open_devices--; |
084b6e7c QW |
2021 | } |
2022 | ||
2023 | void btrfs_rm_dev_replace_free_srcdev(struct btrfs_fs_info *fs_info, | |
2024 | struct btrfs_device *srcdev) | |
2025 | { | |
2026 | struct btrfs_fs_devices *fs_devices = srcdev->fs_devices; | |
e93c89c1 | 2027 | |
48b3b9d4 AJ |
2028 | if (srcdev->writeable) { |
2029 | /* zero out the old super if it is writable */ | |
2030 | btrfs_scratch_superblocks(srcdev->bdev, srcdev->name->str); | |
2031 | } | |
14238819 AJ |
2032 | |
2033 | btrfs_close_bdev(srcdev); | |
2034 | ||
e93c89c1 | 2035 | call_rcu(&srcdev->rcu, free_device); |
94d5f0c2 AJ |
2036 | |
2037 | /* | |
2038 | * unless fs_devices is seed fs, num_devices shouldn't go | |
2039 | * zero | |
2040 | */ | |
2041 | BUG_ON(!fs_devices->num_devices && !fs_devices->seeding); | |
2042 | ||
2043 | /* if this is no devs we rather delete the fs_devices */ | |
2044 | if (!fs_devices->num_devices) { | |
2045 | struct btrfs_fs_devices *tmp_fs_devices; | |
2046 | ||
2047 | tmp_fs_devices = fs_info->fs_devices; | |
2048 | while (tmp_fs_devices) { | |
2049 | if (tmp_fs_devices->seed == fs_devices) { | |
2050 | tmp_fs_devices->seed = fs_devices->seed; | |
2051 | break; | |
2052 | } | |
2053 | tmp_fs_devices = tmp_fs_devices->seed; | |
2054 | } | |
2055 | fs_devices->seed = NULL; | |
8bef8401 AJ |
2056 | __btrfs_close_devices(fs_devices); |
2057 | free_fs_devices(fs_devices); | |
94d5f0c2 | 2058 | } |
e93c89c1 SB |
2059 | } |
2060 | ||
2061 | void btrfs_destroy_dev_replace_tgtdev(struct btrfs_fs_info *fs_info, | |
2062 | struct btrfs_device *tgtdev) | |
2063 | { | |
67a2c45e | 2064 | mutex_lock(&uuid_mutex); |
e93c89c1 SB |
2065 | WARN_ON(!tgtdev); |
2066 | mutex_lock(&fs_info->fs_devices->device_list_mutex); | |
d2ff1b20 | 2067 | |
32576040 | 2068 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, tgtdev); |
d2ff1b20 | 2069 | |
779bf3fe | 2070 | if (tgtdev->bdev) |
e93c89c1 | 2071 | fs_info->fs_devices->open_devices--; |
779bf3fe | 2072 | |
e93c89c1 | 2073 | fs_info->fs_devices->num_devices--; |
e93c89c1 | 2074 | |
88acff64 | 2075 | btrfs_assign_next_active_device(fs_info, tgtdev, NULL); |
e93c89c1 | 2076 | |
e93c89c1 | 2077 | list_del_rcu(&tgtdev->dev_list); |
e93c89c1 SB |
2078 | |
2079 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
67a2c45e | 2080 | mutex_unlock(&uuid_mutex); |
779bf3fe AJ |
2081 | |
2082 | /* | |
2083 | * The update_dev_time() with in btrfs_scratch_superblocks() | |
2084 | * may lead to a call to btrfs_show_devname() which will try | |
2085 | * to hold device_list_mutex. And here this device | |
2086 | * is already out of device list, so we don't have to hold | |
2087 | * the device_list_mutex lock. | |
2088 | */ | |
2089 | btrfs_scratch_superblocks(tgtdev->bdev, tgtdev->name->str); | |
14238819 AJ |
2090 | |
2091 | btrfs_close_bdev(tgtdev); | |
779bf3fe | 2092 | call_rcu(&tgtdev->rcu, free_device); |
e93c89c1 SB |
2093 | } |
2094 | ||
48a3b636 ES |
2095 | static int btrfs_find_device_by_path(struct btrfs_root *root, char *device_path, |
2096 | struct btrfs_device **device) | |
7ba15b7d | 2097 | { |
0b246afa | 2098 | struct btrfs_fs_info *fs_info = root->fs_info; |
7ba15b7d SB |
2099 | int ret = 0; |
2100 | struct btrfs_super_block *disk_super; | |
2101 | u64 devid; | |
2102 | u8 *dev_uuid; | |
2103 | struct block_device *bdev; | |
2104 | struct buffer_head *bh; | |
2105 | ||
2106 | *device = NULL; | |
2107 | ret = btrfs_get_bdev_and_sb(device_path, FMODE_READ, | |
0b246afa | 2108 | fs_info->bdev_holder, 0, &bdev, &bh); |
7ba15b7d SB |
2109 | if (ret) |
2110 | return ret; | |
2111 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
2112 | devid = btrfs_stack_device_id(&disk_super->dev_item); | |
2113 | dev_uuid = disk_super->dev_item.uuid; | |
0b246afa | 2114 | *device = btrfs_find_device(fs_info, devid, dev_uuid, disk_super->fsid); |
7ba15b7d SB |
2115 | brelse(bh); |
2116 | if (!*device) | |
2117 | ret = -ENOENT; | |
2118 | blkdev_put(bdev, FMODE_READ); | |
2119 | return ret; | |
2120 | } | |
2121 | ||
2122 | int btrfs_find_device_missing_or_by_path(struct btrfs_root *root, | |
2123 | char *device_path, | |
2124 | struct btrfs_device **device) | |
2125 | { | |
0b246afa JM |
2126 | struct btrfs_fs_info *fs_info = root->fs_info; |
2127 | ||
7ba15b7d SB |
2128 | *device = NULL; |
2129 | if (strcmp(device_path, "missing") == 0) { | |
2130 | struct list_head *devices; | |
2131 | struct btrfs_device *tmp; | |
2132 | ||
0b246afa | 2133 | devices = &fs_info->fs_devices->devices; |
7ba15b7d SB |
2134 | /* |
2135 | * It is safe to read the devices since the volume_mutex | |
2136 | * is held by the caller. | |
2137 | */ | |
2138 | list_for_each_entry(tmp, devices, dev_list) { | |
2139 | if (tmp->in_fs_metadata && !tmp->bdev) { | |
2140 | *device = tmp; | |
2141 | break; | |
2142 | } | |
2143 | } | |
2144 | ||
d74a6259 AJ |
2145 | if (!*device) |
2146 | return BTRFS_ERROR_DEV_MISSING_NOT_FOUND; | |
7ba15b7d SB |
2147 | |
2148 | return 0; | |
2149 | } else { | |
2150 | return btrfs_find_device_by_path(root, device_path, device); | |
2151 | } | |
2152 | } | |
2153 | ||
5c5c0df0 DS |
2154 | /* |
2155 | * Lookup a device given by device id, or the path if the id is 0. | |
2156 | */ | |
2157 | int btrfs_find_device_by_devspec(struct btrfs_root *root, u64 devid, | |
2158 | char *devpath, | |
24e0474b AJ |
2159 | struct btrfs_device **device) |
2160 | { | |
0b246afa | 2161 | struct btrfs_fs_info *fs_info = root->fs_info; |
24e0474b AJ |
2162 | int ret; |
2163 | ||
5c5c0df0 | 2164 | if (devid) { |
24e0474b | 2165 | ret = 0; |
0b246afa | 2166 | *device = btrfs_find_device(fs_info, devid, NULL, NULL); |
24e0474b AJ |
2167 | if (!*device) |
2168 | ret = -ENOENT; | |
2169 | } else { | |
5c5c0df0 | 2170 | if (!devpath || !devpath[0]) |
b3d1b153 AJ |
2171 | return -EINVAL; |
2172 | ||
5c5c0df0 | 2173 | ret = btrfs_find_device_missing_or_by_path(root, devpath, |
24e0474b AJ |
2174 | device); |
2175 | } | |
2176 | return ret; | |
2177 | } | |
2178 | ||
2b82032c YZ |
2179 | /* |
2180 | * does all the dirty work required for changing file system's UUID. | |
2181 | */ | |
125ccb0a | 2182 | static int btrfs_prepare_sprout(struct btrfs_root *root) |
2b82032c | 2183 | { |
0b246afa JM |
2184 | struct btrfs_fs_info *fs_info = root->fs_info; |
2185 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
2b82032c | 2186 | struct btrfs_fs_devices *old_devices; |
e4404d6e | 2187 | struct btrfs_fs_devices *seed_devices; |
0b246afa | 2188 | struct btrfs_super_block *disk_super = fs_info->super_copy; |
2b82032c YZ |
2189 | struct btrfs_device *device; |
2190 | u64 super_flags; | |
2191 | ||
2192 | BUG_ON(!mutex_is_locked(&uuid_mutex)); | |
e4404d6e | 2193 | if (!fs_devices->seeding) |
2b82032c YZ |
2194 | return -EINVAL; |
2195 | ||
2208a378 ID |
2196 | seed_devices = __alloc_fs_devices(); |
2197 | if (IS_ERR(seed_devices)) | |
2198 | return PTR_ERR(seed_devices); | |
2b82032c | 2199 | |
e4404d6e YZ |
2200 | old_devices = clone_fs_devices(fs_devices); |
2201 | if (IS_ERR(old_devices)) { | |
2202 | kfree(seed_devices); | |
2203 | return PTR_ERR(old_devices); | |
2b82032c | 2204 | } |
e4404d6e | 2205 | |
2b82032c YZ |
2206 | list_add(&old_devices->list, &fs_uuids); |
2207 | ||
e4404d6e YZ |
2208 | memcpy(seed_devices, fs_devices, sizeof(*seed_devices)); |
2209 | seed_devices->opened = 1; | |
2210 | INIT_LIST_HEAD(&seed_devices->devices); | |
2211 | INIT_LIST_HEAD(&seed_devices->alloc_list); | |
e5e9a520 | 2212 | mutex_init(&seed_devices->device_list_mutex); |
c9513edb | 2213 | |
0b246afa | 2214 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
1f78160c XG |
2215 | list_splice_init_rcu(&fs_devices->devices, &seed_devices->devices, |
2216 | synchronize_rcu); | |
2196d6e8 MX |
2217 | list_for_each_entry(device, &seed_devices->devices, dev_list) |
2218 | device->fs_devices = seed_devices; | |
c9513edb | 2219 | |
0b246afa | 2220 | lock_chunks(fs_info); |
e4404d6e | 2221 | list_splice_init(&fs_devices->alloc_list, &seed_devices->alloc_list); |
0b246afa | 2222 | unlock_chunks(fs_info); |
e4404d6e | 2223 | |
2b82032c YZ |
2224 | fs_devices->seeding = 0; |
2225 | fs_devices->num_devices = 0; | |
2226 | fs_devices->open_devices = 0; | |
69611ac8 | 2227 | fs_devices->missing_devices = 0; |
69611ac8 | 2228 | fs_devices->rotating = 0; |
e4404d6e | 2229 | fs_devices->seed = seed_devices; |
2b82032c YZ |
2230 | |
2231 | generate_random_uuid(fs_devices->fsid); | |
0b246afa | 2232 | memcpy(fs_info->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
2b82032c | 2233 | memcpy(disk_super->fsid, fs_devices->fsid, BTRFS_FSID_SIZE); |
0b246afa | 2234 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
f7171750 | 2235 | |
2b82032c YZ |
2236 | super_flags = btrfs_super_flags(disk_super) & |
2237 | ~BTRFS_SUPER_FLAG_SEEDING; | |
2238 | btrfs_set_super_flags(disk_super, super_flags); | |
2239 | ||
2240 | return 0; | |
2241 | } | |
2242 | ||
2243 | /* | |
01327610 | 2244 | * Store the expected generation for seed devices in device items. |
2b82032c YZ |
2245 | */ |
2246 | static int btrfs_finish_sprout(struct btrfs_trans_handle *trans, | |
5b4aacef | 2247 | struct btrfs_fs_info *fs_info) |
2b82032c | 2248 | { |
5b4aacef | 2249 | struct btrfs_root *root = fs_info->chunk_root; |
2b82032c YZ |
2250 | struct btrfs_path *path; |
2251 | struct extent_buffer *leaf; | |
2252 | struct btrfs_dev_item *dev_item; | |
2253 | struct btrfs_device *device; | |
2254 | struct btrfs_key key; | |
2255 | u8 fs_uuid[BTRFS_UUID_SIZE]; | |
2256 | u8 dev_uuid[BTRFS_UUID_SIZE]; | |
2257 | u64 devid; | |
2258 | int ret; | |
2259 | ||
2260 | path = btrfs_alloc_path(); | |
2261 | if (!path) | |
2262 | return -ENOMEM; | |
2263 | ||
2b82032c YZ |
2264 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; |
2265 | key.offset = 0; | |
2266 | key.type = BTRFS_DEV_ITEM_KEY; | |
2267 | ||
2268 | while (1) { | |
2269 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2270 | if (ret < 0) | |
2271 | goto error; | |
2272 | ||
2273 | leaf = path->nodes[0]; | |
2274 | next_slot: | |
2275 | if (path->slots[0] >= btrfs_header_nritems(leaf)) { | |
2276 | ret = btrfs_next_leaf(root, path); | |
2277 | if (ret > 0) | |
2278 | break; | |
2279 | if (ret < 0) | |
2280 | goto error; | |
2281 | leaf = path->nodes[0]; | |
2282 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
b3b4aa74 | 2283 | btrfs_release_path(path); |
2b82032c YZ |
2284 | continue; |
2285 | } | |
2286 | ||
2287 | btrfs_item_key_to_cpu(leaf, &key, path->slots[0]); | |
2288 | if (key.objectid != BTRFS_DEV_ITEMS_OBJECTID || | |
2289 | key.type != BTRFS_DEV_ITEM_KEY) | |
2290 | break; | |
2291 | ||
2292 | dev_item = btrfs_item_ptr(leaf, path->slots[0], | |
2293 | struct btrfs_dev_item); | |
2294 | devid = btrfs_device_id(leaf, dev_item); | |
410ba3a2 | 2295 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
2b82032c | 2296 | BTRFS_UUID_SIZE); |
1473b24e | 2297 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c | 2298 | BTRFS_UUID_SIZE); |
0b246afa | 2299 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
79787eaa | 2300 | BUG_ON(!device); /* Logic error */ |
2b82032c YZ |
2301 | |
2302 | if (device->fs_devices->seeding) { | |
2303 | btrfs_set_device_generation(leaf, dev_item, | |
2304 | device->generation); | |
2305 | btrfs_mark_buffer_dirty(leaf); | |
2306 | } | |
2307 | ||
2308 | path->slots[0]++; | |
2309 | goto next_slot; | |
2310 | } | |
2311 | ret = 0; | |
2312 | error: | |
2313 | btrfs_free_path(path); | |
2314 | return ret; | |
2315 | } | |
2316 | ||
5112febb | 2317 | int btrfs_init_new_device(struct btrfs_fs_info *fs_info, char *device_path) |
788f20eb | 2318 | { |
5112febb | 2319 | struct btrfs_root *root = fs_info->dev_root; |
d5e2003c | 2320 | struct request_queue *q; |
788f20eb CM |
2321 | struct btrfs_trans_handle *trans; |
2322 | struct btrfs_device *device; | |
2323 | struct block_device *bdev; | |
788f20eb | 2324 | struct list_head *devices; |
0b246afa | 2325 | struct super_block *sb = fs_info->sb; |
606686ee | 2326 | struct rcu_string *name; |
3c1dbdf5 | 2327 | u64 tmp; |
2b82032c | 2328 | int seeding_dev = 0; |
788f20eb CM |
2329 | int ret = 0; |
2330 | ||
0b246afa | 2331 | if ((sb->s_flags & MS_RDONLY) && !fs_info->fs_devices->seeding) |
f8c5d0b4 | 2332 | return -EROFS; |
788f20eb | 2333 | |
a5d16333 | 2334 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, |
0b246afa | 2335 | fs_info->bdev_holder); |
7f59203a JB |
2336 | if (IS_ERR(bdev)) |
2337 | return PTR_ERR(bdev); | |
a2135011 | 2338 | |
0b246afa | 2339 | if (fs_info->fs_devices->seeding) { |
2b82032c YZ |
2340 | seeding_dev = 1; |
2341 | down_write(&sb->s_umount); | |
2342 | mutex_lock(&uuid_mutex); | |
2343 | } | |
2344 | ||
8c8bee1d | 2345 | filemap_write_and_wait(bdev->bd_inode->i_mapping); |
a2135011 | 2346 | |
0b246afa | 2347 | devices = &fs_info->fs_devices->devices; |
d25628bd | 2348 | |
0b246afa | 2349 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
c6e30871 | 2350 | list_for_each_entry(device, devices, dev_list) { |
788f20eb CM |
2351 | if (device->bdev == bdev) { |
2352 | ret = -EEXIST; | |
d25628bd | 2353 | mutex_unlock( |
0b246afa | 2354 | &fs_info->fs_devices->device_list_mutex); |
2b82032c | 2355 | goto error; |
788f20eb CM |
2356 | } |
2357 | } | |
0b246afa | 2358 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
788f20eb | 2359 | |
0b246afa | 2360 | device = btrfs_alloc_device(fs_info, NULL, NULL); |
12bd2fc0 | 2361 | if (IS_ERR(device)) { |
788f20eb | 2362 | /* we can safely leave the fs_devices entry around */ |
12bd2fc0 | 2363 | ret = PTR_ERR(device); |
2b82032c | 2364 | goto error; |
788f20eb CM |
2365 | } |
2366 | ||
78f2c9e6 | 2367 | name = rcu_string_strdup(device_path, GFP_KERNEL); |
606686ee | 2368 | if (!name) { |
788f20eb | 2369 | kfree(device); |
2b82032c YZ |
2370 | ret = -ENOMEM; |
2371 | goto error; | |
788f20eb | 2372 | } |
606686ee | 2373 | rcu_assign_pointer(device->name, name); |
2b82032c | 2374 | |
a22285a6 | 2375 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 | 2376 | if (IS_ERR(trans)) { |
606686ee | 2377 | rcu_string_free(device->name); |
98d5dc13 TI |
2378 | kfree(device); |
2379 | ret = PTR_ERR(trans); | |
2380 | goto error; | |
2381 | } | |
2382 | ||
d5e2003c JB |
2383 | q = bdev_get_queue(bdev); |
2384 | if (blk_queue_discard(q)) | |
2385 | device->can_discard = 1; | |
2b82032c | 2386 | device->writeable = 1; |
2b82032c | 2387 | device->generation = trans->transid; |
0b246afa JM |
2388 | device->io_width = fs_info->sectorsize; |
2389 | device->io_align = fs_info->sectorsize; | |
2390 | device->sector_size = fs_info->sectorsize; | |
788f20eb | 2391 | device->total_bytes = i_size_read(bdev->bd_inode); |
2cc3c559 | 2392 | device->disk_total_bytes = device->total_bytes; |
935e5cc9 | 2393 | device->commit_total_bytes = device->total_bytes; |
fb456252 | 2394 | device->fs_info = fs_info; |
788f20eb | 2395 | device->bdev = bdev; |
dfe25020 | 2396 | device->in_fs_metadata = 1; |
63a212ab | 2397 | device->is_tgtdev_for_dev_replace = 0; |
fb01aa85 | 2398 | device->mode = FMODE_EXCL; |
27087f37 | 2399 | device->dev_stats_valid = 1; |
2b82032c | 2400 | set_blocksize(device->bdev, 4096); |
788f20eb | 2401 | |
2b82032c YZ |
2402 | if (seeding_dev) { |
2403 | sb->s_flags &= ~MS_RDONLY; | |
125ccb0a | 2404 | ret = btrfs_prepare_sprout(root); |
79787eaa | 2405 | BUG_ON(ret); /* -ENOMEM */ |
2b82032c | 2406 | } |
788f20eb | 2407 | |
0b246afa | 2408 | device->fs_devices = fs_info->fs_devices; |
e5e9a520 | 2409 | |
0b246afa JM |
2410 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
2411 | lock_chunks(fs_info); | |
2412 | list_add_rcu(&device->dev_list, &fs_info->fs_devices->devices); | |
2b82032c | 2413 | list_add(&device->dev_alloc_list, |
0b246afa JM |
2414 | &fs_info->fs_devices->alloc_list); |
2415 | fs_info->fs_devices->num_devices++; | |
2416 | fs_info->fs_devices->open_devices++; | |
2417 | fs_info->fs_devices->rw_devices++; | |
2418 | fs_info->fs_devices->total_devices++; | |
2419 | fs_info->fs_devices->total_rw_bytes += device->total_bytes; | |
325cd4ba | 2420 | |
0b246afa JM |
2421 | spin_lock(&fs_info->free_chunk_lock); |
2422 | fs_info->free_chunk_space += device->total_bytes; | |
2423 | spin_unlock(&fs_info->free_chunk_lock); | |
2bf64758 | 2424 | |
c289811c | 2425 | if (!blk_queue_nonrot(bdev_get_queue(bdev))) |
0b246afa | 2426 | fs_info->fs_devices->rotating = 1; |
c289811c | 2427 | |
0b246afa JM |
2428 | tmp = btrfs_super_total_bytes(fs_info->super_copy); |
2429 | btrfs_set_super_total_bytes(fs_info->super_copy, | |
3c1dbdf5 | 2430 | tmp + device->total_bytes); |
788f20eb | 2431 | |
0b246afa JM |
2432 | tmp = btrfs_super_num_devices(fs_info->super_copy); |
2433 | btrfs_set_super_num_devices(fs_info->super_copy, tmp + 1); | |
0d39376a AJ |
2434 | |
2435 | /* add sysfs device entry */ | |
0b246afa | 2436 | btrfs_sysfs_add_device_link(fs_info->fs_devices, device); |
0d39376a | 2437 | |
2196d6e8 MX |
2438 | /* |
2439 | * we've got more storage, clear any full flags on the space | |
2440 | * infos | |
2441 | */ | |
0b246afa | 2442 | btrfs_clear_space_info_full(fs_info); |
2196d6e8 | 2443 | |
0b246afa JM |
2444 | unlock_chunks(fs_info); |
2445 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); | |
788f20eb | 2446 | |
2b82032c | 2447 | if (seeding_dev) { |
0b246afa | 2448 | lock_chunks(fs_info); |
2b82032c | 2449 | ret = init_first_rw_device(trans, root, device); |
0b246afa | 2450 | unlock_chunks(fs_info); |
005d6427 | 2451 | if (ret) { |
66642832 | 2452 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2453 | goto error_trans; |
005d6427 | 2454 | } |
2196d6e8 MX |
2455 | } |
2456 | ||
0b246afa | 2457 | ret = btrfs_add_device(trans, fs_info, device); |
2196d6e8 | 2458 | if (ret) { |
66642832 | 2459 | btrfs_abort_transaction(trans, ret); |
2196d6e8 MX |
2460 | goto error_trans; |
2461 | } | |
2462 | ||
2463 | if (seeding_dev) { | |
2464 | char fsid_buf[BTRFS_UUID_UNPARSED_SIZE]; | |
2465 | ||
0b246afa | 2466 | ret = btrfs_finish_sprout(trans, fs_info); |
005d6427 | 2467 | if (ret) { |
66642832 | 2468 | btrfs_abort_transaction(trans, ret); |
79787eaa | 2469 | goto error_trans; |
005d6427 | 2470 | } |
b2373f25 AJ |
2471 | |
2472 | /* Sprouting would change fsid of the mounted root, | |
2473 | * so rename the fsid on the sysfs | |
2474 | */ | |
2475 | snprintf(fsid_buf, BTRFS_UUID_UNPARSED_SIZE, "%pU", | |
0b246afa JM |
2476 | fs_info->fsid); |
2477 | if (kobject_rename(&fs_info->fs_devices->fsid_kobj, fsid_buf)) | |
2478 | btrfs_warn(fs_info, | |
2479 | "sysfs: failed to create fsid for sprout"); | |
2b82032c YZ |
2480 | } |
2481 | ||
0b246afa JM |
2482 | fs_info->num_tolerated_disk_barrier_failures = |
2483 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
79787eaa | 2484 | ret = btrfs_commit_transaction(trans, root); |
a2135011 | 2485 | |
2b82032c YZ |
2486 | if (seeding_dev) { |
2487 | mutex_unlock(&uuid_mutex); | |
2488 | up_write(&sb->s_umount); | |
788f20eb | 2489 | |
79787eaa JM |
2490 | if (ret) /* transaction commit */ |
2491 | return ret; | |
2492 | ||
2b82032c | 2493 | ret = btrfs_relocate_sys_chunks(root); |
79787eaa | 2494 | if (ret < 0) |
0b246afa | 2495 | btrfs_handle_fs_error(fs_info, ret, |
5d163e0e | 2496 | "Failed to relocate sys chunks after device initialization. This can be fixed using the \"btrfs balance\" command."); |
671415b7 MX |
2497 | trans = btrfs_attach_transaction(root); |
2498 | if (IS_ERR(trans)) { | |
2499 | if (PTR_ERR(trans) == -ENOENT) | |
2500 | return 0; | |
2501 | return PTR_ERR(trans); | |
2502 | } | |
2503 | ret = btrfs_commit_transaction(trans, root); | |
2b82032c | 2504 | } |
c9e9f97b | 2505 | |
5a1972bd QW |
2506 | /* Update ctime/mtime for libblkid */ |
2507 | update_dev_time(device_path); | |
2b82032c | 2508 | return ret; |
79787eaa JM |
2509 | |
2510 | error_trans: | |
79787eaa | 2511 | btrfs_end_transaction(trans, root); |
606686ee | 2512 | rcu_string_free(device->name); |
0b246afa | 2513 | btrfs_sysfs_rm_device_link(fs_info->fs_devices, device); |
79787eaa | 2514 | kfree(device); |
2b82032c | 2515 | error: |
e525fd89 | 2516 | blkdev_put(bdev, FMODE_EXCL); |
2b82032c YZ |
2517 | if (seeding_dev) { |
2518 | mutex_unlock(&uuid_mutex); | |
2519 | up_write(&sb->s_umount); | |
2520 | } | |
c9e9f97b | 2521 | return ret; |
788f20eb CM |
2522 | } |
2523 | ||
e93c89c1 | 2524 | int btrfs_init_dev_replace_tgtdev(struct btrfs_root *root, char *device_path, |
1c43366d | 2525 | struct btrfs_device *srcdev, |
e93c89c1 SB |
2526 | struct btrfs_device **device_out) |
2527 | { | |
2528 | struct request_queue *q; | |
2529 | struct btrfs_device *device; | |
2530 | struct block_device *bdev; | |
2531 | struct btrfs_fs_info *fs_info = root->fs_info; | |
2532 | struct list_head *devices; | |
2533 | struct rcu_string *name; | |
12bd2fc0 | 2534 | u64 devid = BTRFS_DEV_REPLACE_DEVID; |
e93c89c1 SB |
2535 | int ret = 0; |
2536 | ||
2537 | *device_out = NULL; | |
1c43366d MX |
2538 | if (fs_info->fs_devices->seeding) { |
2539 | btrfs_err(fs_info, "the filesystem is a seed filesystem!"); | |
e93c89c1 | 2540 | return -EINVAL; |
1c43366d | 2541 | } |
e93c89c1 SB |
2542 | |
2543 | bdev = blkdev_get_by_path(device_path, FMODE_WRITE | FMODE_EXCL, | |
2544 | fs_info->bdev_holder); | |
1c43366d MX |
2545 | if (IS_ERR(bdev)) { |
2546 | btrfs_err(fs_info, "target device %s is invalid!", device_path); | |
e93c89c1 | 2547 | return PTR_ERR(bdev); |
1c43366d | 2548 | } |
e93c89c1 SB |
2549 | |
2550 | filemap_write_and_wait(bdev->bd_inode->i_mapping); | |
2551 | ||
2552 | devices = &fs_info->fs_devices->devices; | |
2553 | list_for_each_entry(device, devices, dev_list) { | |
2554 | if (device->bdev == bdev) { | |
5d163e0e JM |
2555 | btrfs_err(fs_info, |
2556 | "target device is in the filesystem!"); | |
e93c89c1 SB |
2557 | ret = -EEXIST; |
2558 | goto error; | |
2559 | } | |
2560 | } | |
2561 | ||
1c43366d | 2562 | |
7cc8e58d MX |
2563 | if (i_size_read(bdev->bd_inode) < |
2564 | btrfs_device_get_total_bytes(srcdev)) { | |
5d163e0e JM |
2565 | btrfs_err(fs_info, |
2566 | "target device is smaller than source device!"); | |
1c43366d MX |
2567 | ret = -EINVAL; |
2568 | goto error; | |
2569 | } | |
2570 | ||
2571 | ||
12bd2fc0 ID |
2572 | device = btrfs_alloc_device(NULL, &devid, NULL); |
2573 | if (IS_ERR(device)) { | |
2574 | ret = PTR_ERR(device); | |
e93c89c1 SB |
2575 | goto error; |
2576 | } | |
2577 | ||
2578 | name = rcu_string_strdup(device_path, GFP_NOFS); | |
2579 | if (!name) { | |
2580 | kfree(device); | |
2581 | ret = -ENOMEM; | |
2582 | goto error; | |
2583 | } | |
2584 | rcu_assign_pointer(device->name, name); | |
2585 | ||
2586 | q = bdev_get_queue(bdev); | |
2587 | if (blk_queue_discard(q)) | |
2588 | device->can_discard = 1; | |
0b246afa | 2589 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 | 2590 | device->writeable = 1; |
e93c89c1 | 2591 | device->generation = 0; |
0b246afa JM |
2592 | device->io_width = fs_info->sectorsize; |
2593 | device->io_align = fs_info->sectorsize; | |
2594 | device->sector_size = fs_info->sectorsize; | |
7cc8e58d MX |
2595 | device->total_bytes = btrfs_device_get_total_bytes(srcdev); |
2596 | device->disk_total_bytes = btrfs_device_get_disk_total_bytes(srcdev); | |
2597 | device->bytes_used = btrfs_device_get_bytes_used(srcdev); | |
935e5cc9 MX |
2598 | ASSERT(list_empty(&srcdev->resized_list)); |
2599 | device->commit_total_bytes = srcdev->commit_total_bytes; | |
ce7213c7 | 2600 | device->commit_bytes_used = device->bytes_used; |
fb456252 | 2601 | device->fs_info = fs_info; |
e93c89c1 SB |
2602 | device->bdev = bdev; |
2603 | device->in_fs_metadata = 1; | |
2604 | device->is_tgtdev_for_dev_replace = 1; | |
2605 | device->mode = FMODE_EXCL; | |
27087f37 | 2606 | device->dev_stats_valid = 1; |
e93c89c1 SB |
2607 | set_blocksize(device->bdev, 4096); |
2608 | device->fs_devices = fs_info->fs_devices; | |
2609 | list_add(&device->dev_list, &fs_info->fs_devices->devices); | |
2610 | fs_info->fs_devices->num_devices++; | |
2611 | fs_info->fs_devices->open_devices++; | |
0b246afa | 2612 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
e93c89c1 SB |
2613 | |
2614 | *device_out = device; | |
2615 | return ret; | |
2616 | ||
2617 | error: | |
2618 | blkdev_put(bdev, FMODE_EXCL); | |
2619 | return ret; | |
2620 | } | |
2621 | ||
2622 | void btrfs_init_dev_replace_tgtdev_for_resume(struct btrfs_fs_info *fs_info, | |
2623 | struct btrfs_device *tgtdev) | |
2624 | { | |
da17066c JM |
2625 | u32 sectorsize = fs_info->sectorsize; |
2626 | ||
e93c89c1 | 2627 | WARN_ON(fs_info->fs_devices->rw_devices == 0); |
da17066c JM |
2628 | tgtdev->io_width = sectorsize; |
2629 | tgtdev->io_align = sectorsize; | |
2630 | tgtdev->sector_size = sectorsize; | |
fb456252 | 2631 | tgtdev->fs_info = fs_info; |
e93c89c1 SB |
2632 | tgtdev->in_fs_metadata = 1; |
2633 | } | |
2634 | ||
d397712b CM |
2635 | static noinline int btrfs_update_device(struct btrfs_trans_handle *trans, |
2636 | struct btrfs_device *device) | |
0b86a832 CM |
2637 | { |
2638 | int ret; | |
2639 | struct btrfs_path *path; | |
0b246afa | 2640 | struct btrfs_root *root = device->fs_info->chunk_root; |
0b86a832 CM |
2641 | struct btrfs_dev_item *dev_item; |
2642 | struct extent_buffer *leaf; | |
2643 | struct btrfs_key key; | |
2644 | ||
0b86a832 CM |
2645 | path = btrfs_alloc_path(); |
2646 | if (!path) | |
2647 | return -ENOMEM; | |
2648 | ||
2649 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
2650 | key.type = BTRFS_DEV_ITEM_KEY; | |
2651 | key.offset = device->devid; | |
2652 | ||
2653 | ret = btrfs_search_slot(trans, root, &key, path, 0, 1); | |
2654 | if (ret < 0) | |
2655 | goto out; | |
2656 | ||
2657 | if (ret > 0) { | |
2658 | ret = -ENOENT; | |
2659 | goto out; | |
2660 | } | |
2661 | ||
2662 | leaf = path->nodes[0]; | |
2663 | dev_item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_dev_item); | |
2664 | ||
2665 | btrfs_set_device_id(leaf, dev_item, device->devid); | |
2666 | btrfs_set_device_type(leaf, dev_item, device->type); | |
2667 | btrfs_set_device_io_align(leaf, dev_item, device->io_align); | |
2668 | btrfs_set_device_io_width(leaf, dev_item, device->io_width); | |
2669 | btrfs_set_device_sector_size(leaf, dev_item, device->sector_size); | |
7cc8e58d MX |
2670 | btrfs_set_device_total_bytes(leaf, dev_item, |
2671 | btrfs_device_get_disk_total_bytes(device)); | |
2672 | btrfs_set_device_bytes_used(leaf, dev_item, | |
2673 | btrfs_device_get_bytes_used(device)); | |
0b86a832 CM |
2674 | btrfs_mark_buffer_dirty(leaf); |
2675 | ||
2676 | out: | |
2677 | btrfs_free_path(path); | |
2678 | return ret; | |
2679 | } | |
2680 | ||
2196d6e8 | 2681 | int btrfs_grow_device(struct btrfs_trans_handle *trans, |
8f18cf13 CM |
2682 | struct btrfs_device *device, u64 new_size) |
2683 | { | |
0b246afa JM |
2684 | struct btrfs_fs_info *fs_info = device->fs_info; |
2685 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
935e5cc9 | 2686 | struct btrfs_fs_devices *fs_devices; |
2196d6e8 MX |
2687 | u64 old_total; |
2688 | u64 diff; | |
8f18cf13 | 2689 | |
2b82032c YZ |
2690 | if (!device->writeable) |
2691 | return -EACCES; | |
2196d6e8 | 2692 | |
0b246afa | 2693 | lock_chunks(fs_info); |
2196d6e8 MX |
2694 | old_total = btrfs_super_total_bytes(super_copy); |
2695 | diff = new_size - device->total_bytes; | |
2696 | ||
63a212ab | 2697 | if (new_size <= device->total_bytes || |
2196d6e8 | 2698 | device->is_tgtdev_for_dev_replace) { |
0b246afa | 2699 | unlock_chunks(fs_info); |
2b82032c | 2700 | return -EINVAL; |
2196d6e8 | 2701 | } |
2b82032c | 2702 | |
0b246afa | 2703 | fs_devices = fs_info->fs_devices; |
2b82032c | 2704 | |
8f18cf13 | 2705 | btrfs_set_super_total_bytes(super_copy, old_total + diff); |
2b82032c YZ |
2706 | device->fs_devices->total_rw_bytes += diff; |
2707 | ||
7cc8e58d MX |
2708 | btrfs_device_set_total_bytes(device, new_size); |
2709 | btrfs_device_set_disk_total_bytes(device, new_size); | |
fb456252 | 2710 | btrfs_clear_space_info_full(device->fs_info); |
935e5cc9 MX |
2711 | if (list_empty(&device->resized_list)) |
2712 | list_add_tail(&device->resized_list, | |
2713 | &fs_devices->resized_devices); | |
0b246afa | 2714 | unlock_chunks(fs_info); |
4184ea7f | 2715 | |
8f18cf13 CM |
2716 | return btrfs_update_device(trans, device); |
2717 | } | |
2718 | ||
2719 | static int btrfs_free_chunk(struct btrfs_trans_handle *trans, | |
5b4aacef | 2720 | struct btrfs_fs_info *fs_info, u64 chunk_objectid, |
8f18cf13 CM |
2721 | u64 chunk_offset) |
2722 | { | |
5b4aacef | 2723 | struct btrfs_root *root = fs_info->chunk_root; |
8f18cf13 CM |
2724 | int ret; |
2725 | struct btrfs_path *path; | |
2726 | struct btrfs_key key; | |
2727 | ||
8f18cf13 CM |
2728 | path = btrfs_alloc_path(); |
2729 | if (!path) | |
2730 | return -ENOMEM; | |
2731 | ||
2732 | key.objectid = chunk_objectid; | |
2733 | key.offset = chunk_offset; | |
2734 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2735 | ||
2736 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
79787eaa JM |
2737 | if (ret < 0) |
2738 | goto out; | |
2739 | else if (ret > 0) { /* Logic error or corruption */ | |
0b246afa JM |
2740 | btrfs_handle_fs_error(fs_info, -ENOENT, |
2741 | "Failed lookup while freeing chunk."); | |
79787eaa JM |
2742 | ret = -ENOENT; |
2743 | goto out; | |
2744 | } | |
8f18cf13 CM |
2745 | |
2746 | ret = btrfs_del_item(trans, root, path); | |
79787eaa | 2747 | if (ret < 0) |
0b246afa JM |
2748 | btrfs_handle_fs_error(fs_info, ret, |
2749 | "Failed to delete chunk item."); | |
79787eaa | 2750 | out: |
8f18cf13 | 2751 | btrfs_free_path(path); |
65a246c5 | 2752 | return ret; |
8f18cf13 CM |
2753 | } |
2754 | ||
6bccf3ab JM |
2755 | static int btrfs_del_sys_chunk(struct btrfs_fs_info *fs_info, |
2756 | u64 chunk_objectid, u64 chunk_offset) | |
8f18cf13 | 2757 | { |
0b246afa | 2758 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 CM |
2759 | struct btrfs_disk_key *disk_key; |
2760 | struct btrfs_chunk *chunk; | |
2761 | u8 *ptr; | |
2762 | int ret = 0; | |
2763 | u32 num_stripes; | |
2764 | u32 array_size; | |
2765 | u32 len = 0; | |
2766 | u32 cur; | |
2767 | struct btrfs_key key; | |
2768 | ||
0b246afa | 2769 | lock_chunks(fs_info); |
8f18cf13 CM |
2770 | array_size = btrfs_super_sys_array_size(super_copy); |
2771 | ||
2772 | ptr = super_copy->sys_chunk_array; | |
2773 | cur = 0; | |
2774 | ||
2775 | while (cur < array_size) { | |
2776 | disk_key = (struct btrfs_disk_key *)ptr; | |
2777 | btrfs_disk_key_to_cpu(&key, disk_key); | |
2778 | ||
2779 | len = sizeof(*disk_key); | |
2780 | ||
2781 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { | |
2782 | chunk = (struct btrfs_chunk *)(ptr + len); | |
2783 | num_stripes = btrfs_stack_chunk_num_stripes(chunk); | |
2784 | len += btrfs_chunk_item_size(num_stripes); | |
2785 | } else { | |
2786 | ret = -EIO; | |
2787 | break; | |
2788 | } | |
2789 | if (key.objectid == chunk_objectid && | |
2790 | key.offset == chunk_offset) { | |
2791 | memmove(ptr, ptr + len, array_size - (cur + len)); | |
2792 | array_size -= len; | |
2793 | btrfs_set_super_sys_array_size(super_copy, array_size); | |
2794 | } else { | |
2795 | ptr += len; | |
2796 | cur += len; | |
2797 | } | |
2798 | } | |
0b246afa | 2799 | unlock_chunks(fs_info); |
8f18cf13 CM |
2800 | return ret; |
2801 | } | |
2802 | ||
47ab2a6c | 2803 | int btrfs_remove_chunk(struct btrfs_trans_handle *trans, |
5b4aacef | 2804 | struct btrfs_fs_info *fs_info, u64 chunk_offset) |
8f18cf13 CM |
2805 | { |
2806 | struct extent_map_tree *em_tree; | |
8f18cf13 | 2807 | struct extent_map *em; |
0b246afa | 2808 | struct btrfs_root *extent_root = fs_info->extent_root; |
8f18cf13 | 2809 | struct map_lookup *map; |
2196d6e8 | 2810 | u64 dev_extent_len = 0; |
47ab2a6c | 2811 | u64 chunk_objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
47ab2a6c | 2812 | int i, ret = 0; |
0b246afa | 2813 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
8f18cf13 | 2814 | |
5b4aacef | 2815 | em_tree = &fs_info->mapping_tree.map_tree; |
8f18cf13 | 2816 | |
890871be | 2817 | read_lock(&em_tree->lock); |
8f18cf13 | 2818 | em = lookup_extent_mapping(em_tree, chunk_offset, 1); |
890871be | 2819 | read_unlock(&em_tree->lock); |
8f18cf13 | 2820 | |
47ab2a6c JB |
2821 | if (!em || em->start > chunk_offset || |
2822 | em->start + em->len < chunk_offset) { | |
2823 | /* | |
2824 | * This is a logic error, but we don't want to just rely on the | |
bb7ab3b9 | 2825 | * user having built with ASSERT enabled, so if ASSERT doesn't |
47ab2a6c JB |
2826 | * do anything we still error out. |
2827 | */ | |
2828 | ASSERT(0); | |
2829 | if (em) | |
2830 | free_extent_map(em); | |
2831 | return -EINVAL; | |
2832 | } | |
95617d69 | 2833 | map = em->map_lookup; |
3796d335 | 2834 | lock_chunks(fs_info); |
4617ea3a | 2835 | check_system_chunk(trans, extent_root, map->type); |
3796d335 | 2836 | unlock_chunks(fs_info); |
8f18cf13 | 2837 | |
57ba4cb8 FM |
2838 | /* |
2839 | * Take the device list mutex to prevent races with the final phase of | |
2840 | * a device replace operation that replaces the device object associated | |
2841 | * with map stripes (dev-replace.c:btrfs_dev_replace_finishing()). | |
2842 | */ | |
2843 | mutex_lock(&fs_devices->device_list_mutex); | |
8f18cf13 | 2844 | for (i = 0; i < map->num_stripes; i++) { |
47ab2a6c | 2845 | struct btrfs_device *device = map->stripes[i].dev; |
2196d6e8 MX |
2846 | ret = btrfs_free_dev_extent(trans, device, |
2847 | map->stripes[i].physical, | |
2848 | &dev_extent_len); | |
47ab2a6c | 2849 | if (ret) { |
57ba4cb8 | 2850 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2851 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2852 | goto out; |
2853 | } | |
a061fc8d | 2854 | |
2196d6e8 | 2855 | if (device->bytes_used > 0) { |
0b246afa | 2856 | lock_chunks(fs_info); |
2196d6e8 MX |
2857 | btrfs_device_set_bytes_used(device, |
2858 | device->bytes_used - dev_extent_len); | |
0b246afa JM |
2859 | spin_lock(&fs_info->free_chunk_lock); |
2860 | fs_info->free_chunk_space += dev_extent_len; | |
2861 | spin_unlock(&fs_info->free_chunk_lock); | |
2862 | btrfs_clear_space_info_full(fs_info); | |
2863 | unlock_chunks(fs_info); | |
2196d6e8 | 2864 | } |
a061fc8d | 2865 | |
dfe25020 CM |
2866 | if (map->stripes[i].dev) { |
2867 | ret = btrfs_update_device(trans, map->stripes[i].dev); | |
47ab2a6c | 2868 | if (ret) { |
57ba4cb8 | 2869 | mutex_unlock(&fs_devices->device_list_mutex); |
66642832 | 2870 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2871 | goto out; |
2872 | } | |
dfe25020 | 2873 | } |
8f18cf13 | 2874 | } |
57ba4cb8 FM |
2875 | mutex_unlock(&fs_devices->device_list_mutex); |
2876 | ||
0b246afa | 2877 | ret = btrfs_free_chunk(trans, fs_info, chunk_objectid, chunk_offset); |
47ab2a6c | 2878 | if (ret) { |
66642832 | 2879 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2880 | goto out; |
2881 | } | |
8f18cf13 | 2882 | |
6bccf3ab | 2883 | trace_btrfs_chunk_free(fs_info, map, chunk_offset, em->len); |
1abe9b8a | 2884 | |
8f18cf13 | 2885 | if (map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
6bccf3ab JM |
2886 | ret = btrfs_del_sys_chunk(fs_info, chunk_objectid, |
2887 | chunk_offset); | |
47ab2a6c | 2888 | if (ret) { |
66642832 | 2889 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2890 | goto out; |
2891 | } | |
8f18cf13 CM |
2892 | } |
2893 | ||
6bccf3ab | 2894 | ret = btrfs_remove_block_group(trans, fs_info, chunk_offset, em); |
47ab2a6c | 2895 | if (ret) { |
66642832 | 2896 | btrfs_abort_transaction(trans, ret); |
47ab2a6c JB |
2897 | goto out; |
2898 | } | |
2b82032c | 2899 | |
47ab2a6c | 2900 | out: |
2b82032c YZ |
2901 | /* once for us */ |
2902 | free_extent_map(em); | |
47ab2a6c JB |
2903 | return ret; |
2904 | } | |
2b82032c | 2905 | |
5b4aacef | 2906 | static int btrfs_relocate_chunk(struct btrfs_fs_info *fs_info, u64 chunk_offset) |
47ab2a6c | 2907 | { |
5b4aacef JM |
2908 | struct btrfs_root *root = fs_info->chunk_root; |
2909 | struct btrfs_root *extent_root = fs_info->extent_root; | |
19c4d2f9 | 2910 | struct btrfs_trans_handle *trans; |
47ab2a6c | 2911 | int ret; |
2b82032c | 2912 | |
67c5e7d4 FM |
2913 | /* |
2914 | * Prevent races with automatic removal of unused block groups. | |
2915 | * After we relocate and before we remove the chunk with offset | |
2916 | * chunk_offset, automatic removal of the block group can kick in, | |
2917 | * resulting in a failure when calling btrfs_remove_chunk() below. | |
2918 | * | |
2919 | * Make sure to acquire this mutex before doing a tree search (dev | |
2920 | * or chunk trees) to find chunks. Otherwise the cleaner kthread might | |
2921 | * call btrfs_remove_chunk() (through btrfs_delete_unused_bgs()) after | |
2922 | * we release the path used to search the chunk/dev tree and before | |
2923 | * the current task acquires this mutex and calls us. | |
2924 | */ | |
0b246afa | 2925 | ASSERT(mutex_is_locked(&fs_info->delete_unused_bgs_mutex)); |
67c5e7d4 | 2926 | |
0b246afa | 2927 | ret = btrfs_can_relocate(fs_info, chunk_offset); |
47ab2a6c JB |
2928 | if (ret) |
2929 | return -ENOSPC; | |
2930 | ||
2931 | /* step one, relocate all the extents inside this chunk */ | |
55e3a601 | 2932 | btrfs_scrub_pause(root); |
0b246afa | 2933 | ret = btrfs_relocate_block_group(fs_info, chunk_offset); |
55e3a601 | 2934 | btrfs_scrub_continue(root); |
47ab2a6c JB |
2935 | if (ret) |
2936 | return ret; | |
2937 | ||
19c4d2f9 CM |
2938 | trans = btrfs_start_trans_remove_block_group(root->fs_info, |
2939 | chunk_offset); | |
2940 | if (IS_ERR(trans)) { | |
2941 | ret = PTR_ERR(trans); | |
2942 | btrfs_handle_fs_error(root->fs_info, ret, NULL); | |
2943 | return ret; | |
2944 | } | |
2945 | ||
47ab2a6c | 2946 | /* |
19c4d2f9 CM |
2947 | * step two, delete the device extents and the |
2948 | * chunk tree entries | |
47ab2a6c | 2949 | */ |
5b4aacef | 2950 | ret = btrfs_remove_chunk(trans, fs_info, chunk_offset); |
19c4d2f9 CM |
2951 | btrfs_end_transaction(trans, extent_root); |
2952 | return ret; | |
2b82032c YZ |
2953 | } |
2954 | ||
2955 | static int btrfs_relocate_sys_chunks(struct btrfs_root *root) | |
2956 | { | |
0b246afa JM |
2957 | struct btrfs_fs_info *fs_info = root->fs_info; |
2958 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c YZ |
2959 | struct btrfs_path *path; |
2960 | struct extent_buffer *leaf; | |
2961 | struct btrfs_chunk *chunk; | |
2962 | struct btrfs_key key; | |
2963 | struct btrfs_key found_key; | |
2b82032c | 2964 | u64 chunk_type; |
ba1bf481 JB |
2965 | bool retried = false; |
2966 | int failed = 0; | |
2b82032c YZ |
2967 | int ret; |
2968 | ||
2969 | path = btrfs_alloc_path(); | |
2970 | if (!path) | |
2971 | return -ENOMEM; | |
2972 | ||
ba1bf481 | 2973 | again: |
2b82032c YZ |
2974 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
2975 | key.offset = (u64)-1; | |
2976 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
2977 | ||
2978 | while (1) { | |
0b246afa | 2979 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2980 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 | 2981 | if (ret < 0) { |
0b246afa | 2982 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c | 2983 | goto error; |
67c5e7d4 | 2984 | } |
79787eaa | 2985 | BUG_ON(ret == 0); /* Corruption */ |
2b82032c YZ |
2986 | |
2987 | ret = btrfs_previous_item(chunk_root, path, key.objectid, | |
2988 | key.type); | |
67c5e7d4 | 2989 | if (ret) |
0b246afa | 2990 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
2b82032c YZ |
2991 | if (ret < 0) |
2992 | goto error; | |
2993 | if (ret > 0) | |
2994 | break; | |
1a40e23b | 2995 | |
2b82032c YZ |
2996 | leaf = path->nodes[0]; |
2997 | btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]); | |
1a40e23b | 2998 | |
2b82032c YZ |
2999 | chunk = btrfs_item_ptr(leaf, path->slots[0], |
3000 | struct btrfs_chunk); | |
3001 | chunk_type = btrfs_chunk_type(leaf, chunk); | |
b3b4aa74 | 3002 | btrfs_release_path(path); |
8f18cf13 | 3003 | |
2b82032c | 3004 | if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) { |
0b246afa | 3005 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
ba1bf481 JB |
3006 | if (ret == -ENOSPC) |
3007 | failed++; | |
14586651 HS |
3008 | else |
3009 | BUG_ON(ret); | |
2b82032c | 3010 | } |
0b246afa | 3011 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 3012 | |
2b82032c YZ |
3013 | if (found_key.offset == 0) |
3014 | break; | |
3015 | key.offset = found_key.offset - 1; | |
3016 | } | |
3017 | ret = 0; | |
ba1bf481 JB |
3018 | if (failed && !retried) { |
3019 | failed = 0; | |
3020 | retried = true; | |
3021 | goto again; | |
fae7f21c | 3022 | } else if (WARN_ON(failed && retried)) { |
ba1bf481 JB |
3023 | ret = -ENOSPC; |
3024 | } | |
2b82032c YZ |
3025 | error: |
3026 | btrfs_free_path(path); | |
3027 | return ret; | |
8f18cf13 CM |
3028 | } |
3029 | ||
6bccf3ab | 3030 | static int insert_balance_item(struct btrfs_fs_info *fs_info, |
0940ebf6 ID |
3031 | struct btrfs_balance_control *bctl) |
3032 | { | |
6bccf3ab | 3033 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3034 | struct btrfs_trans_handle *trans; |
3035 | struct btrfs_balance_item *item; | |
3036 | struct btrfs_disk_balance_args disk_bargs; | |
3037 | struct btrfs_path *path; | |
3038 | struct extent_buffer *leaf; | |
3039 | struct btrfs_key key; | |
3040 | int ret, err; | |
3041 | ||
3042 | path = btrfs_alloc_path(); | |
3043 | if (!path) | |
3044 | return -ENOMEM; | |
3045 | ||
3046 | trans = btrfs_start_transaction(root, 0); | |
3047 | if (IS_ERR(trans)) { | |
3048 | btrfs_free_path(path); | |
3049 | return PTR_ERR(trans); | |
3050 | } | |
3051 | ||
3052 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3053 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3054 | key.offset = 0; |
3055 | ||
3056 | ret = btrfs_insert_empty_item(trans, root, path, &key, | |
3057 | sizeof(*item)); | |
3058 | if (ret) | |
3059 | goto out; | |
3060 | ||
3061 | leaf = path->nodes[0]; | |
3062 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
3063 | ||
b159fa28 | 3064 | memzero_extent_buffer(leaf, (unsigned long)item, sizeof(*item)); |
0940ebf6 ID |
3065 | |
3066 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->data); | |
3067 | btrfs_set_balance_data(leaf, item, &disk_bargs); | |
3068 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->meta); | |
3069 | btrfs_set_balance_meta(leaf, item, &disk_bargs); | |
3070 | btrfs_cpu_balance_args_to_disk(&disk_bargs, &bctl->sys); | |
3071 | btrfs_set_balance_sys(leaf, item, &disk_bargs); | |
3072 | ||
3073 | btrfs_set_balance_flags(leaf, item, bctl->flags); | |
3074 | ||
3075 | btrfs_mark_buffer_dirty(leaf); | |
3076 | out: | |
3077 | btrfs_free_path(path); | |
3078 | err = btrfs_commit_transaction(trans, root); | |
3079 | if (err && !ret) | |
3080 | ret = err; | |
3081 | return ret; | |
3082 | } | |
3083 | ||
6bccf3ab | 3084 | static int del_balance_item(struct btrfs_fs_info *fs_info) |
0940ebf6 | 3085 | { |
6bccf3ab | 3086 | struct btrfs_root *root = fs_info->tree_root; |
0940ebf6 ID |
3087 | struct btrfs_trans_handle *trans; |
3088 | struct btrfs_path *path; | |
3089 | struct btrfs_key key; | |
3090 | int ret, err; | |
3091 | ||
3092 | path = btrfs_alloc_path(); | |
3093 | if (!path) | |
3094 | return -ENOMEM; | |
3095 | ||
3096 | trans = btrfs_start_transaction(root, 0); | |
3097 | if (IS_ERR(trans)) { | |
3098 | btrfs_free_path(path); | |
3099 | return PTR_ERR(trans); | |
3100 | } | |
3101 | ||
3102 | key.objectid = BTRFS_BALANCE_OBJECTID; | |
c479cb4f | 3103 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
0940ebf6 ID |
3104 | key.offset = 0; |
3105 | ||
3106 | ret = btrfs_search_slot(trans, root, &key, path, -1, 1); | |
3107 | if (ret < 0) | |
3108 | goto out; | |
3109 | if (ret > 0) { | |
3110 | ret = -ENOENT; | |
3111 | goto out; | |
3112 | } | |
3113 | ||
3114 | ret = btrfs_del_item(trans, root, path); | |
3115 | out: | |
3116 | btrfs_free_path(path); | |
3117 | err = btrfs_commit_transaction(trans, root); | |
3118 | if (err && !ret) | |
3119 | ret = err; | |
3120 | return ret; | |
3121 | } | |
3122 | ||
59641015 ID |
3123 | /* |
3124 | * This is a heuristic used to reduce the number of chunks balanced on | |
3125 | * resume after balance was interrupted. | |
3126 | */ | |
3127 | static void update_balance_args(struct btrfs_balance_control *bctl) | |
3128 | { | |
3129 | /* | |
3130 | * Turn on soft mode for chunk types that were being converted. | |
3131 | */ | |
3132 | if (bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3133 | bctl->data.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3134 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3135 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3136 | if (bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) | |
3137 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_SOFT; | |
3138 | ||
3139 | /* | |
3140 | * Turn on usage filter if is not already used. The idea is | |
3141 | * that chunks that we have already balanced should be | |
3142 | * reasonably full. Don't do it for chunks that are being | |
3143 | * converted - that will keep us from relocating unconverted | |
3144 | * (albeit full) chunks. | |
3145 | */ | |
3146 | if (!(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3147 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3148 | !(bctl->data.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3149 | bctl->data.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3150 | bctl->data.usage = 90; | |
3151 | } | |
3152 | if (!(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3153 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3154 | !(bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3155 | bctl->sys.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3156 | bctl->sys.usage = 90; | |
3157 | } | |
3158 | if (!(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE) && | |
bc309467 | 3159 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
59641015 ID |
3160 | !(bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT)) { |
3161 | bctl->meta.flags |= BTRFS_BALANCE_ARGS_USAGE; | |
3162 | bctl->meta.usage = 90; | |
3163 | } | |
3164 | } | |
3165 | ||
c9e9f97b ID |
3166 | /* |
3167 | * Should be called with both balance and volume mutexes held to | |
3168 | * serialize other volume operations (add_dev/rm_dev/resize) with | |
3169 | * restriper. Same goes for unset_balance_control. | |
3170 | */ | |
3171 | static void set_balance_control(struct btrfs_balance_control *bctl) | |
3172 | { | |
3173 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
3174 | ||
3175 | BUG_ON(fs_info->balance_ctl); | |
3176 | ||
3177 | spin_lock(&fs_info->balance_lock); | |
3178 | fs_info->balance_ctl = bctl; | |
3179 | spin_unlock(&fs_info->balance_lock); | |
3180 | } | |
3181 | ||
3182 | static void unset_balance_control(struct btrfs_fs_info *fs_info) | |
3183 | { | |
3184 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
3185 | ||
3186 | BUG_ON(!fs_info->balance_ctl); | |
3187 | ||
3188 | spin_lock(&fs_info->balance_lock); | |
3189 | fs_info->balance_ctl = NULL; | |
3190 | spin_unlock(&fs_info->balance_lock); | |
3191 | ||
3192 | kfree(bctl); | |
3193 | } | |
3194 | ||
ed25e9b2 ID |
3195 | /* |
3196 | * Balance filters. Return 1 if chunk should be filtered out | |
3197 | * (should not be balanced). | |
3198 | */ | |
899c81ea | 3199 | static int chunk_profiles_filter(u64 chunk_type, |
ed25e9b2 ID |
3200 | struct btrfs_balance_args *bargs) |
3201 | { | |
899c81ea ID |
3202 | chunk_type = chunk_to_extended(chunk_type) & |
3203 | BTRFS_EXTENDED_PROFILE_MASK; | |
ed25e9b2 | 3204 | |
899c81ea | 3205 | if (bargs->profiles & chunk_type) |
ed25e9b2 ID |
3206 | return 0; |
3207 | ||
3208 | return 1; | |
3209 | } | |
3210 | ||
dba72cb3 | 3211 | static int chunk_usage_range_filter(struct btrfs_fs_info *fs_info, u64 chunk_offset, |
5ce5b3c0 | 3212 | struct btrfs_balance_args *bargs) |
bc309467 DS |
3213 | { |
3214 | struct btrfs_block_group_cache *cache; | |
3215 | u64 chunk_used; | |
3216 | u64 user_thresh_min; | |
3217 | u64 user_thresh_max; | |
3218 | int ret = 1; | |
3219 | ||
3220 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3221 | chunk_used = btrfs_block_group_used(&cache->item); | |
3222 | ||
3223 | if (bargs->usage_min == 0) | |
3224 | user_thresh_min = 0; | |
3225 | else | |
3226 | user_thresh_min = div_factor_fine(cache->key.offset, | |
3227 | bargs->usage_min); | |
3228 | ||
3229 | if (bargs->usage_max == 0) | |
3230 | user_thresh_max = 1; | |
3231 | else if (bargs->usage_max > 100) | |
3232 | user_thresh_max = cache->key.offset; | |
3233 | else | |
3234 | user_thresh_max = div_factor_fine(cache->key.offset, | |
3235 | bargs->usage_max); | |
3236 | ||
3237 | if (user_thresh_min <= chunk_used && chunk_used < user_thresh_max) | |
3238 | ret = 0; | |
3239 | ||
3240 | btrfs_put_block_group(cache); | |
3241 | return ret; | |
3242 | } | |
3243 | ||
dba72cb3 | 3244 | static int chunk_usage_filter(struct btrfs_fs_info *fs_info, |
bc309467 | 3245 | u64 chunk_offset, struct btrfs_balance_args *bargs) |
5ce5b3c0 ID |
3246 | { |
3247 | struct btrfs_block_group_cache *cache; | |
3248 | u64 chunk_used, user_thresh; | |
3249 | int ret = 1; | |
3250 | ||
3251 | cache = btrfs_lookup_block_group(fs_info, chunk_offset); | |
3252 | chunk_used = btrfs_block_group_used(&cache->item); | |
3253 | ||
bc309467 | 3254 | if (bargs->usage_min == 0) |
3e39cea6 | 3255 | user_thresh = 1; |
a105bb88 ID |
3256 | else if (bargs->usage > 100) |
3257 | user_thresh = cache->key.offset; | |
3258 | else | |
3259 | user_thresh = div_factor_fine(cache->key.offset, | |
3260 | bargs->usage); | |
3261 | ||
5ce5b3c0 ID |
3262 | if (chunk_used < user_thresh) |
3263 | ret = 0; | |
3264 | ||
3265 | btrfs_put_block_group(cache); | |
3266 | return ret; | |
3267 | } | |
3268 | ||
409d404b ID |
3269 | static int chunk_devid_filter(struct extent_buffer *leaf, |
3270 | struct btrfs_chunk *chunk, | |
3271 | struct btrfs_balance_args *bargs) | |
3272 | { | |
3273 | struct btrfs_stripe *stripe; | |
3274 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3275 | int i; | |
3276 | ||
3277 | for (i = 0; i < num_stripes; i++) { | |
3278 | stripe = btrfs_stripe_nr(chunk, i); | |
3279 | if (btrfs_stripe_devid(leaf, stripe) == bargs->devid) | |
3280 | return 0; | |
3281 | } | |
3282 | ||
3283 | return 1; | |
3284 | } | |
3285 | ||
94e60d5a ID |
3286 | /* [pstart, pend) */ |
3287 | static int chunk_drange_filter(struct extent_buffer *leaf, | |
3288 | struct btrfs_chunk *chunk, | |
3289 | u64 chunk_offset, | |
3290 | struct btrfs_balance_args *bargs) | |
3291 | { | |
3292 | struct btrfs_stripe *stripe; | |
3293 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3294 | u64 stripe_offset; | |
3295 | u64 stripe_length; | |
3296 | int factor; | |
3297 | int i; | |
3298 | ||
3299 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_DEVID)) | |
3300 | return 0; | |
3301 | ||
3302 | if (btrfs_chunk_type(leaf, chunk) & (BTRFS_BLOCK_GROUP_DUP | | |
53b381b3 DW |
3303 | BTRFS_BLOCK_GROUP_RAID1 | BTRFS_BLOCK_GROUP_RAID10)) { |
3304 | factor = num_stripes / 2; | |
3305 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID5) { | |
3306 | factor = num_stripes - 1; | |
3307 | } else if (btrfs_chunk_type(leaf, chunk) & BTRFS_BLOCK_GROUP_RAID6) { | |
3308 | factor = num_stripes - 2; | |
3309 | } else { | |
3310 | factor = num_stripes; | |
3311 | } | |
94e60d5a ID |
3312 | |
3313 | for (i = 0; i < num_stripes; i++) { | |
3314 | stripe = btrfs_stripe_nr(chunk, i); | |
3315 | if (btrfs_stripe_devid(leaf, stripe) != bargs->devid) | |
3316 | continue; | |
3317 | ||
3318 | stripe_offset = btrfs_stripe_offset(leaf, stripe); | |
3319 | stripe_length = btrfs_chunk_length(leaf, chunk); | |
b8b93add | 3320 | stripe_length = div_u64(stripe_length, factor); |
94e60d5a ID |
3321 | |
3322 | if (stripe_offset < bargs->pend && | |
3323 | stripe_offset + stripe_length > bargs->pstart) | |
3324 | return 0; | |
3325 | } | |
3326 | ||
3327 | return 1; | |
3328 | } | |
3329 | ||
ea67176a ID |
3330 | /* [vstart, vend) */ |
3331 | static int chunk_vrange_filter(struct extent_buffer *leaf, | |
3332 | struct btrfs_chunk *chunk, | |
3333 | u64 chunk_offset, | |
3334 | struct btrfs_balance_args *bargs) | |
3335 | { | |
3336 | if (chunk_offset < bargs->vend && | |
3337 | chunk_offset + btrfs_chunk_length(leaf, chunk) > bargs->vstart) | |
3338 | /* at least part of the chunk is inside this vrange */ | |
3339 | return 0; | |
3340 | ||
3341 | return 1; | |
3342 | } | |
3343 | ||
dee32d0a GAP |
3344 | static int chunk_stripes_range_filter(struct extent_buffer *leaf, |
3345 | struct btrfs_chunk *chunk, | |
3346 | struct btrfs_balance_args *bargs) | |
3347 | { | |
3348 | int num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
3349 | ||
3350 | if (bargs->stripes_min <= num_stripes | |
3351 | && num_stripes <= bargs->stripes_max) | |
3352 | return 0; | |
3353 | ||
3354 | return 1; | |
3355 | } | |
3356 | ||
899c81ea | 3357 | static int chunk_soft_convert_filter(u64 chunk_type, |
cfa4c961 ID |
3358 | struct btrfs_balance_args *bargs) |
3359 | { | |
3360 | if (!(bargs->flags & BTRFS_BALANCE_ARGS_CONVERT)) | |
3361 | return 0; | |
3362 | ||
899c81ea ID |
3363 | chunk_type = chunk_to_extended(chunk_type) & |
3364 | BTRFS_EXTENDED_PROFILE_MASK; | |
cfa4c961 | 3365 | |
899c81ea | 3366 | if (bargs->target == chunk_type) |
cfa4c961 ID |
3367 | return 1; |
3368 | ||
3369 | return 0; | |
3370 | } | |
3371 | ||
f43ffb60 ID |
3372 | static int should_balance_chunk(struct btrfs_root *root, |
3373 | struct extent_buffer *leaf, | |
3374 | struct btrfs_chunk *chunk, u64 chunk_offset) | |
3375 | { | |
0b246afa JM |
3376 | struct btrfs_fs_info *fs_info = root->fs_info; |
3377 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; | |
f43ffb60 ID |
3378 | struct btrfs_balance_args *bargs = NULL; |
3379 | u64 chunk_type = btrfs_chunk_type(leaf, chunk); | |
3380 | ||
3381 | /* type filter */ | |
3382 | if (!((chunk_type & BTRFS_BLOCK_GROUP_TYPE_MASK) & | |
3383 | (bctl->flags & BTRFS_BALANCE_TYPE_MASK))) { | |
3384 | return 0; | |
3385 | } | |
3386 | ||
3387 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3388 | bargs = &bctl->data; | |
3389 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3390 | bargs = &bctl->sys; | |
3391 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3392 | bargs = &bctl->meta; | |
3393 | ||
ed25e9b2 ID |
3394 | /* profiles filter */ |
3395 | if ((bargs->flags & BTRFS_BALANCE_ARGS_PROFILES) && | |
3396 | chunk_profiles_filter(chunk_type, bargs)) { | |
3397 | return 0; | |
5ce5b3c0 ID |
3398 | } |
3399 | ||
3400 | /* usage filter */ | |
3401 | if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE) && | |
0b246afa | 3402 | chunk_usage_filter(fs_info, chunk_offset, bargs)) { |
5ce5b3c0 | 3403 | return 0; |
bc309467 | 3404 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_USAGE_RANGE) && |
0b246afa | 3405 | chunk_usage_range_filter(fs_info, chunk_offset, bargs)) { |
bc309467 | 3406 | return 0; |
409d404b ID |
3407 | } |
3408 | ||
3409 | /* devid filter */ | |
3410 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DEVID) && | |
3411 | chunk_devid_filter(leaf, chunk, bargs)) { | |
3412 | return 0; | |
94e60d5a ID |
3413 | } |
3414 | ||
3415 | /* drange filter, makes sense only with devid filter */ | |
3416 | if ((bargs->flags & BTRFS_BALANCE_ARGS_DRANGE) && | |
3417 | chunk_drange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3418 | return 0; | |
ea67176a ID |
3419 | } |
3420 | ||
3421 | /* vrange filter */ | |
3422 | if ((bargs->flags & BTRFS_BALANCE_ARGS_VRANGE) && | |
3423 | chunk_vrange_filter(leaf, chunk, chunk_offset, bargs)) { | |
3424 | return 0; | |
ed25e9b2 ID |
3425 | } |
3426 | ||
dee32d0a GAP |
3427 | /* stripes filter */ |
3428 | if ((bargs->flags & BTRFS_BALANCE_ARGS_STRIPES_RANGE) && | |
3429 | chunk_stripes_range_filter(leaf, chunk, bargs)) { | |
3430 | return 0; | |
3431 | } | |
3432 | ||
cfa4c961 ID |
3433 | /* soft profile changing mode */ |
3434 | if ((bargs->flags & BTRFS_BALANCE_ARGS_SOFT) && | |
3435 | chunk_soft_convert_filter(chunk_type, bargs)) { | |
3436 | return 0; | |
3437 | } | |
3438 | ||
7d824b6f DS |
3439 | /* |
3440 | * limited by count, must be the last filter | |
3441 | */ | |
3442 | if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT)) { | |
3443 | if (bargs->limit == 0) | |
3444 | return 0; | |
3445 | else | |
3446 | bargs->limit--; | |
12907fc7 DS |
3447 | } else if ((bargs->flags & BTRFS_BALANCE_ARGS_LIMIT_RANGE)) { |
3448 | /* | |
3449 | * Same logic as the 'limit' filter; the minimum cannot be | |
01327610 | 3450 | * determined here because we do not have the global information |
12907fc7 DS |
3451 | * about the count of all chunks that satisfy the filters. |
3452 | */ | |
3453 | if (bargs->limit_max == 0) | |
3454 | return 0; | |
3455 | else | |
3456 | bargs->limit_max--; | |
7d824b6f DS |
3457 | } |
3458 | ||
f43ffb60 ID |
3459 | return 1; |
3460 | } | |
3461 | ||
c9e9f97b | 3462 | static int __btrfs_balance(struct btrfs_fs_info *fs_info) |
ec44a35c | 3463 | { |
19a39dce | 3464 | struct btrfs_balance_control *bctl = fs_info->balance_ctl; |
c9e9f97b ID |
3465 | struct btrfs_root *chunk_root = fs_info->chunk_root; |
3466 | struct btrfs_root *dev_root = fs_info->dev_root; | |
3467 | struct list_head *devices; | |
ec44a35c CM |
3468 | struct btrfs_device *device; |
3469 | u64 old_size; | |
3470 | u64 size_to_free; | |
12907fc7 | 3471 | u64 chunk_type; |
f43ffb60 | 3472 | struct btrfs_chunk *chunk; |
5a488b9d | 3473 | struct btrfs_path *path = NULL; |
ec44a35c | 3474 | struct btrfs_key key; |
ec44a35c | 3475 | struct btrfs_key found_key; |
c9e9f97b | 3476 | struct btrfs_trans_handle *trans; |
f43ffb60 ID |
3477 | struct extent_buffer *leaf; |
3478 | int slot; | |
c9e9f97b ID |
3479 | int ret; |
3480 | int enospc_errors = 0; | |
19a39dce | 3481 | bool counting = true; |
12907fc7 | 3482 | /* The single value limit and min/max limits use the same bytes in the */ |
7d824b6f DS |
3483 | u64 limit_data = bctl->data.limit; |
3484 | u64 limit_meta = bctl->meta.limit; | |
3485 | u64 limit_sys = bctl->sys.limit; | |
12907fc7 DS |
3486 | u32 count_data = 0; |
3487 | u32 count_meta = 0; | |
3488 | u32 count_sys = 0; | |
2c9fe835 | 3489 | int chunk_reserved = 0; |
cf25ce51 | 3490 | u64 bytes_used = 0; |
ec44a35c | 3491 | |
ec44a35c | 3492 | /* step one make some room on all the devices */ |
c9e9f97b | 3493 | devices = &fs_info->fs_devices->devices; |
c6e30871 | 3494 | list_for_each_entry(device, devices, dev_list) { |
7cc8e58d | 3495 | old_size = btrfs_device_get_total_bytes(device); |
ec44a35c | 3496 | size_to_free = div_factor(old_size, 1); |
ee22184b | 3497 | size_to_free = min_t(u64, size_to_free, SZ_1M); |
2b82032c | 3498 | if (!device->writeable || |
7cc8e58d MX |
3499 | btrfs_device_get_total_bytes(device) - |
3500 | btrfs_device_get_bytes_used(device) > size_to_free || | |
63a212ab | 3501 | device->is_tgtdev_for_dev_replace) |
ec44a35c CM |
3502 | continue; |
3503 | ||
3504 | ret = btrfs_shrink_device(device, old_size - size_to_free); | |
ba1bf481 JB |
3505 | if (ret == -ENOSPC) |
3506 | break; | |
5a488b9d LB |
3507 | if (ret) { |
3508 | /* btrfs_shrink_device never returns ret > 0 */ | |
3509 | WARN_ON(ret > 0); | |
3510 | goto error; | |
3511 | } | |
ec44a35c | 3512 | |
a22285a6 | 3513 | trans = btrfs_start_transaction(dev_root, 0); |
5a488b9d LB |
3514 | if (IS_ERR(trans)) { |
3515 | ret = PTR_ERR(trans); | |
3516 | btrfs_info_in_rcu(fs_info, | |
3517 | "resize: unable to start transaction after shrinking device %s (error %d), old size %llu, new size %llu", | |
3518 | rcu_str_deref(device->name), ret, | |
3519 | old_size, old_size - size_to_free); | |
3520 | goto error; | |
3521 | } | |
ec44a35c CM |
3522 | |
3523 | ret = btrfs_grow_device(trans, device, old_size); | |
5a488b9d LB |
3524 | if (ret) { |
3525 | btrfs_end_transaction(trans, dev_root); | |
3526 | /* btrfs_grow_device never returns ret > 0 */ | |
3527 | WARN_ON(ret > 0); | |
3528 | btrfs_info_in_rcu(fs_info, | |
3529 | "resize: unable to grow device after shrinking device %s (error %d), old size %llu, new size %llu", | |
3530 | rcu_str_deref(device->name), ret, | |
3531 | old_size, old_size - size_to_free); | |
3532 | goto error; | |
3533 | } | |
ec44a35c CM |
3534 | |
3535 | btrfs_end_transaction(trans, dev_root); | |
3536 | } | |
3537 | ||
3538 | /* step two, relocate all the chunks */ | |
3539 | path = btrfs_alloc_path(); | |
17e9f796 MF |
3540 | if (!path) { |
3541 | ret = -ENOMEM; | |
3542 | goto error; | |
3543 | } | |
19a39dce ID |
3544 | |
3545 | /* zero out stat counters */ | |
3546 | spin_lock(&fs_info->balance_lock); | |
3547 | memset(&bctl->stat, 0, sizeof(bctl->stat)); | |
3548 | spin_unlock(&fs_info->balance_lock); | |
3549 | again: | |
7d824b6f | 3550 | if (!counting) { |
12907fc7 DS |
3551 | /* |
3552 | * The single value limit and min/max limits use the same bytes | |
3553 | * in the | |
3554 | */ | |
7d824b6f DS |
3555 | bctl->data.limit = limit_data; |
3556 | bctl->meta.limit = limit_meta; | |
3557 | bctl->sys.limit = limit_sys; | |
3558 | } | |
ec44a35c CM |
3559 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
3560 | key.offset = (u64)-1; | |
3561 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
3562 | ||
d397712b | 3563 | while (1) { |
19a39dce | 3564 | if ((!counting && atomic_read(&fs_info->balance_pause_req)) || |
a7e99c69 | 3565 | atomic_read(&fs_info->balance_cancel_req)) { |
837d5b6e ID |
3566 | ret = -ECANCELED; |
3567 | goto error; | |
3568 | } | |
3569 | ||
67c5e7d4 | 3570 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
ec44a35c | 3571 | ret = btrfs_search_slot(NULL, chunk_root, &key, path, 0, 0); |
67c5e7d4 FM |
3572 | if (ret < 0) { |
3573 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3574 | goto error; |
67c5e7d4 | 3575 | } |
ec44a35c CM |
3576 | |
3577 | /* | |
3578 | * this shouldn't happen, it means the last relocate | |
3579 | * failed | |
3580 | */ | |
3581 | if (ret == 0) | |
c9e9f97b | 3582 | BUG(); /* FIXME break ? */ |
ec44a35c CM |
3583 | |
3584 | ret = btrfs_previous_item(chunk_root, path, 0, | |
3585 | BTRFS_CHUNK_ITEM_KEY); | |
c9e9f97b | 3586 | if (ret) { |
67c5e7d4 | 3587 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
c9e9f97b | 3588 | ret = 0; |
ec44a35c | 3589 | break; |
c9e9f97b | 3590 | } |
7d9eb12c | 3591 | |
f43ffb60 ID |
3592 | leaf = path->nodes[0]; |
3593 | slot = path->slots[0]; | |
3594 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
7d9eb12c | 3595 | |
67c5e7d4 FM |
3596 | if (found_key.objectid != key.objectid) { |
3597 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ec44a35c | 3598 | break; |
67c5e7d4 | 3599 | } |
7d9eb12c | 3600 | |
f43ffb60 | 3601 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); |
12907fc7 | 3602 | chunk_type = btrfs_chunk_type(leaf, chunk); |
f43ffb60 | 3603 | |
19a39dce ID |
3604 | if (!counting) { |
3605 | spin_lock(&fs_info->balance_lock); | |
3606 | bctl->stat.considered++; | |
3607 | spin_unlock(&fs_info->balance_lock); | |
3608 | } | |
3609 | ||
f43ffb60 ID |
3610 | ret = should_balance_chunk(chunk_root, leaf, chunk, |
3611 | found_key.offset); | |
2c9fe835 | 3612 | |
b3b4aa74 | 3613 | btrfs_release_path(path); |
67c5e7d4 FM |
3614 | if (!ret) { |
3615 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
f43ffb60 | 3616 | goto loop; |
67c5e7d4 | 3617 | } |
f43ffb60 | 3618 | |
19a39dce | 3619 | if (counting) { |
67c5e7d4 | 3620 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
19a39dce ID |
3621 | spin_lock(&fs_info->balance_lock); |
3622 | bctl->stat.expected++; | |
3623 | spin_unlock(&fs_info->balance_lock); | |
12907fc7 DS |
3624 | |
3625 | if (chunk_type & BTRFS_BLOCK_GROUP_DATA) | |
3626 | count_data++; | |
3627 | else if (chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) | |
3628 | count_sys++; | |
3629 | else if (chunk_type & BTRFS_BLOCK_GROUP_METADATA) | |
3630 | count_meta++; | |
3631 | ||
3632 | goto loop; | |
3633 | } | |
3634 | ||
3635 | /* | |
3636 | * Apply limit_min filter, no need to check if the LIMITS | |
3637 | * filter is used, limit_min is 0 by default | |
3638 | */ | |
3639 | if (((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3640 | count_data < bctl->data.limit_min) | |
3641 | || ((chunk_type & BTRFS_BLOCK_GROUP_METADATA) && | |
3642 | count_meta < bctl->meta.limit_min) | |
3643 | || ((chunk_type & BTRFS_BLOCK_GROUP_SYSTEM) && | |
3644 | count_sys < bctl->sys.limit_min)) { | |
3645 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
19a39dce ID |
3646 | goto loop; |
3647 | } | |
3648 | ||
cf25ce51 LB |
3649 | ASSERT(fs_info->data_sinfo); |
3650 | spin_lock(&fs_info->data_sinfo->lock); | |
3651 | bytes_used = fs_info->data_sinfo->bytes_used; | |
3652 | spin_unlock(&fs_info->data_sinfo->lock); | |
3653 | ||
3654 | if ((chunk_type & BTRFS_BLOCK_GROUP_DATA) && | |
3655 | !chunk_reserved && !bytes_used) { | |
2c9fe835 ZL |
3656 | trans = btrfs_start_transaction(chunk_root, 0); |
3657 | if (IS_ERR(trans)) { | |
3658 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3659 | ret = PTR_ERR(trans); | |
3660 | goto error; | |
3661 | } | |
3662 | ||
3663 | ret = btrfs_force_chunk_alloc(trans, chunk_root, | |
3664 | BTRFS_BLOCK_GROUP_DATA); | |
8a7d656f | 3665 | btrfs_end_transaction(trans, chunk_root); |
2c9fe835 ZL |
3666 | if (ret < 0) { |
3667 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
3668 | goto error; | |
3669 | } | |
2c9fe835 ZL |
3670 | chunk_reserved = 1; |
3671 | } | |
3672 | ||
5b4aacef | 3673 | ret = btrfs_relocate_chunk(fs_info, found_key.offset); |
67c5e7d4 | 3674 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
508794eb JB |
3675 | if (ret && ret != -ENOSPC) |
3676 | goto error; | |
19a39dce | 3677 | if (ret == -ENOSPC) { |
c9e9f97b | 3678 | enospc_errors++; |
19a39dce ID |
3679 | } else { |
3680 | spin_lock(&fs_info->balance_lock); | |
3681 | bctl->stat.completed++; | |
3682 | spin_unlock(&fs_info->balance_lock); | |
3683 | } | |
f43ffb60 | 3684 | loop: |
795a3321 ID |
3685 | if (found_key.offset == 0) |
3686 | break; | |
ba1bf481 | 3687 | key.offset = found_key.offset - 1; |
ec44a35c | 3688 | } |
c9e9f97b | 3689 | |
19a39dce ID |
3690 | if (counting) { |
3691 | btrfs_release_path(path); | |
3692 | counting = false; | |
3693 | goto again; | |
3694 | } | |
ec44a35c CM |
3695 | error: |
3696 | btrfs_free_path(path); | |
c9e9f97b | 3697 | if (enospc_errors) { |
efe120a0 | 3698 | btrfs_info(fs_info, "%d enospc errors during balance", |
5d163e0e | 3699 | enospc_errors); |
c9e9f97b ID |
3700 | if (!ret) |
3701 | ret = -ENOSPC; | |
3702 | } | |
3703 | ||
ec44a35c CM |
3704 | return ret; |
3705 | } | |
3706 | ||
0c460c0d ID |
3707 | /** |
3708 | * alloc_profile_is_valid - see if a given profile is valid and reduced | |
3709 | * @flags: profile to validate | |
3710 | * @extended: if true @flags is treated as an extended profile | |
3711 | */ | |
3712 | static int alloc_profile_is_valid(u64 flags, int extended) | |
3713 | { | |
3714 | u64 mask = (extended ? BTRFS_EXTENDED_PROFILE_MASK : | |
3715 | BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
3716 | ||
3717 | flags &= ~BTRFS_BLOCK_GROUP_TYPE_MASK; | |
3718 | ||
3719 | /* 1) check that all other bits are zeroed */ | |
3720 | if (flags & ~mask) | |
3721 | return 0; | |
3722 | ||
3723 | /* 2) see if profile is reduced */ | |
3724 | if (flags == 0) | |
3725 | return !extended; /* "0" is valid for usual profiles */ | |
3726 | ||
3727 | /* true if exactly one bit set */ | |
3728 | return (flags & (flags - 1)) == 0; | |
3729 | } | |
3730 | ||
837d5b6e ID |
3731 | static inline int balance_need_close(struct btrfs_fs_info *fs_info) |
3732 | { | |
a7e99c69 ID |
3733 | /* cancel requested || normal exit path */ |
3734 | return atomic_read(&fs_info->balance_cancel_req) || | |
3735 | (atomic_read(&fs_info->balance_pause_req) == 0 && | |
3736 | atomic_read(&fs_info->balance_cancel_req) == 0); | |
837d5b6e ID |
3737 | } |
3738 | ||
c9e9f97b ID |
3739 | static void __cancel_balance(struct btrfs_fs_info *fs_info) |
3740 | { | |
0940ebf6 ID |
3741 | int ret; |
3742 | ||
c9e9f97b | 3743 | unset_balance_control(fs_info); |
6bccf3ab | 3744 | ret = del_balance_item(fs_info); |
0f788c58 | 3745 | if (ret) |
34d97007 | 3746 | btrfs_handle_fs_error(fs_info, ret, NULL); |
ed0fb78f ID |
3747 | |
3748 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); | |
c9e9f97b ID |
3749 | } |
3750 | ||
bdcd3c97 AM |
3751 | /* Non-zero return value signifies invalidity */ |
3752 | static inline int validate_convert_profile(struct btrfs_balance_args *bctl_arg, | |
3753 | u64 allowed) | |
3754 | { | |
3755 | return ((bctl_arg->flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3756 | (!alloc_profile_is_valid(bctl_arg->target, 1) || | |
3757 | (bctl_arg->target & ~allowed))); | |
3758 | } | |
3759 | ||
c9e9f97b ID |
3760 | /* |
3761 | * Should be called with both balance and volume mutexes held | |
3762 | */ | |
3763 | int btrfs_balance(struct btrfs_balance_control *bctl, | |
3764 | struct btrfs_ioctl_balance_args *bargs) | |
3765 | { | |
3766 | struct btrfs_fs_info *fs_info = bctl->fs_info; | |
f43ffb60 | 3767 | u64 allowed; |
e4837f8f | 3768 | int mixed = 0; |
c9e9f97b | 3769 | int ret; |
8dabb742 | 3770 | u64 num_devices; |
de98ced9 | 3771 | unsigned seq; |
c9e9f97b | 3772 | |
837d5b6e | 3773 | if (btrfs_fs_closing(fs_info) || |
a7e99c69 ID |
3774 | atomic_read(&fs_info->balance_pause_req) || |
3775 | atomic_read(&fs_info->balance_cancel_req)) { | |
c9e9f97b ID |
3776 | ret = -EINVAL; |
3777 | goto out; | |
3778 | } | |
3779 | ||
e4837f8f ID |
3780 | allowed = btrfs_super_incompat_flags(fs_info->super_copy); |
3781 | if (allowed & BTRFS_FEATURE_INCOMPAT_MIXED_GROUPS) | |
3782 | mixed = 1; | |
3783 | ||
f43ffb60 ID |
3784 | /* |
3785 | * In case of mixed groups both data and meta should be picked, | |
3786 | * and identical options should be given for both of them. | |
3787 | */ | |
e4837f8f ID |
3788 | allowed = BTRFS_BALANCE_DATA | BTRFS_BALANCE_METADATA; |
3789 | if (mixed && (bctl->flags & allowed)) { | |
f43ffb60 ID |
3790 | if (!(bctl->flags & BTRFS_BALANCE_DATA) || |
3791 | !(bctl->flags & BTRFS_BALANCE_METADATA) || | |
3792 | memcmp(&bctl->data, &bctl->meta, sizeof(bctl->data))) { | |
5d163e0e JM |
3793 | btrfs_err(fs_info, |
3794 | "with mixed groups data and metadata balance options must be the same"); | |
f43ffb60 ID |
3795 | ret = -EINVAL; |
3796 | goto out; | |
3797 | } | |
3798 | } | |
3799 | ||
8dabb742 | 3800 | num_devices = fs_info->fs_devices->num_devices; |
73beece9 | 3801 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
8dabb742 SB |
3802 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) { |
3803 | BUG_ON(num_devices < 1); | |
3804 | num_devices--; | |
3805 | } | |
73beece9 | 3806 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
88be159c AH |
3807 | allowed = BTRFS_AVAIL_ALLOC_BIT_SINGLE | BTRFS_BLOCK_GROUP_DUP; |
3808 | if (num_devices > 1) | |
e4d8ec0f | 3809 | allowed |= (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID1); |
8250dabe AP |
3810 | if (num_devices > 2) |
3811 | allowed |= BTRFS_BLOCK_GROUP_RAID5; | |
3812 | if (num_devices > 3) | |
3813 | allowed |= (BTRFS_BLOCK_GROUP_RAID10 | | |
3814 | BTRFS_BLOCK_GROUP_RAID6); | |
bdcd3c97 | 3815 | if (validate_convert_profile(&bctl->data, allowed)) { |
5d163e0e JM |
3816 | btrfs_err(fs_info, |
3817 | "unable to start balance with target data profile %llu", | |
3818 | bctl->data.target); | |
e4d8ec0f ID |
3819 | ret = -EINVAL; |
3820 | goto out; | |
3821 | } | |
bdcd3c97 | 3822 | if (validate_convert_profile(&bctl->meta, allowed)) { |
efe120a0 | 3823 | btrfs_err(fs_info, |
5d163e0e JM |
3824 | "unable to start balance with target metadata profile %llu", |
3825 | bctl->meta.target); | |
e4d8ec0f ID |
3826 | ret = -EINVAL; |
3827 | goto out; | |
3828 | } | |
bdcd3c97 | 3829 | if (validate_convert_profile(&bctl->sys, allowed)) { |
efe120a0 | 3830 | btrfs_err(fs_info, |
5d163e0e JM |
3831 | "unable to start balance with target system profile %llu", |
3832 | bctl->sys.target); | |
e4d8ec0f ID |
3833 | ret = -EINVAL; |
3834 | goto out; | |
3835 | } | |
3836 | ||
e4d8ec0f ID |
3837 | /* allow to reduce meta or sys integrity only if force set */ |
3838 | allowed = BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1 | | |
53b381b3 DW |
3839 | BTRFS_BLOCK_GROUP_RAID10 | |
3840 | BTRFS_BLOCK_GROUP_RAID5 | | |
3841 | BTRFS_BLOCK_GROUP_RAID6; | |
de98ced9 MX |
3842 | do { |
3843 | seq = read_seqbegin(&fs_info->profiles_lock); | |
3844 | ||
3845 | if (((bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3846 | (fs_info->avail_system_alloc_bits & allowed) && | |
3847 | !(bctl->sys.target & allowed)) || | |
3848 | ((bctl->meta.flags & BTRFS_BALANCE_ARGS_CONVERT) && | |
3849 | (fs_info->avail_metadata_alloc_bits & allowed) && | |
3850 | !(bctl->meta.target & allowed))) { | |
3851 | if (bctl->flags & BTRFS_BALANCE_FORCE) { | |
5d163e0e JM |
3852 | btrfs_info(fs_info, |
3853 | "force reducing metadata integrity"); | |
de98ced9 | 3854 | } else { |
5d163e0e JM |
3855 | btrfs_err(fs_info, |
3856 | "balance will reduce metadata integrity, use force if you want this"); | |
de98ced9 MX |
3857 | ret = -EINVAL; |
3858 | goto out; | |
3859 | } | |
e4d8ec0f | 3860 | } |
de98ced9 | 3861 | } while (read_seqretry(&fs_info->profiles_lock, seq)); |
e4d8ec0f | 3862 | |
ee592d07 ST |
3863 | if (btrfs_get_num_tolerated_disk_barrier_failures(bctl->meta.target) < |
3864 | btrfs_get_num_tolerated_disk_barrier_failures(bctl->data.target)) { | |
3865 | btrfs_warn(fs_info, | |
5d163e0e JM |
3866 | "metadata profile 0x%llx has lower redundancy than data profile 0x%llx", |
3867 | bctl->meta.target, bctl->data.target); | |
ee592d07 ST |
3868 | } |
3869 | ||
5af3e8cc | 3870 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
943c6e99 ZL |
3871 | fs_info->num_tolerated_disk_barrier_failures = min( |
3872 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info), | |
3873 | btrfs_get_num_tolerated_disk_barrier_failures( | |
3874 | bctl->sys.target)); | |
5af3e8cc SB |
3875 | } |
3876 | ||
6bccf3ab | 3877 | ret = insert_balance_item(fs_info, bctl); |
59641015 | 3878 | if (ret && ret != -EEXIST) |
0940ebf6 ID |
3879 | goto out; |
3880 | ||
59641015 ID |
3881 | if (!(bctl->flags & BTRFS_BALANCE_RESUME)) { |
3882 | BUG_ON(ret == -EEXIST); | |
3883 | set_balance_control(bctl); | |
3884 | } else { | |
3885 | BUG_ON(ret != -EEXIST); | |
3886 | spin_lock(&fs_info->balance_lock); | |
3887 | update_balance_args(bctl); | |
3888 | spin_unlock(&fs_info->balance_lock); | |
3889 | } | |
c9e9f97b | 3890 | |
837d5b6e | 3891 | atomic_inc(&fs_info->balance_running); |
c9e9f97b ID |
3892 | mutex_unlock(&fs_info->balance_mutex); |
3893 | ||
3894 | ret = __btrfs_balance(fs_info); | |
3895 | ||
3896 | mutex_lock(&fs_info->balance_mutex); | |
837d5b6e | 3897 | atomic_dec(&fs_info->balance_running); |
c9e9f97b | 3898 | |
bf023ecf ID |
3899 | if (bctl->sys.flags & BTRFS_BALANCE_ARGS_CONVERT) { |
3900 | fs_info->num_tolerated_disk_barrier_failures = | |
3901 | btrfs_calc_num_tolerated_disk_barrier_failures(fs_info); | |
3902 | } | |
3903 | ||
c9e9f97b ID |
3904 | if (bargs) { |
3905 | memset(bargs, 0, sizeof(*bargs)); | |
19a39dce | 3906 | update_ioctl_balance_args(fs_info, 0, bargs); |
c9e9f97b ID |
3907 | } |
3908 | ||
3a01aa7a ID |
3909 | if ((ret && ret != -ECANCELED && ret != -ENOSPC) || |
3910 | balance_need_close(fs_info)) { | |
3911 | __cancel_balance(fs_info); | |
3912 | } | |
3913 | ||
837d5b6e | 3914 | wake_up(&fs_info->balance_wait_q); |
c9e9f97b ID |
3915 | |
3916 | return ret; | |
3917 | out: | |
59641015 ID |
3918 | if (bctl->flags & BTRFS_BALANCE_RESUME) |
3919 | __cancel_balance(fs_info); | |
ed0fb78f | 3920 | else { |
59641015 | 3921 | kfree(bctl); |
ed0fb78f ID |
3922 | atomic_set(&fs_info->mutually_exclusive_operation_running, 0); |
3923 | } | |
59641015 ID |
3924 | return ret; |
3925 | } | |
3926 | ||
3927 | static int balance_kthread(void *data) | |
3928 | { | |
2b6ba629 | 3929 | struct btrfs_fs_info *fs_info = data; |
9555c6c1 | 3930 | int ret = 0; |
59641015 ID |
3931 | |
3932 | mutex_lock(&fs_info->volume_mutex); | |
3933 | mutex_lock(&fs_info->balance_mutex); | |
3934 | ||
2b6ba629 | 3935 | if (fs_info->balance_ctl) { |
efe120a0 | 3936 | btrfs_info(fs_info, "continuing balance"); |
2b6ba629 | 3937 | ret = btrfs_balance(fs_info->balance_ctl, NULL); |
9555c6c1 | 3938 | } |
59641015 ID |
3939 | |
3940 | mutex_unlock(&fs_info->balance_mutex); | |
3941 | mutex_unlock(&fs_info->volume_mutex); | |
2b6ba629 | 3942 | |
59641015 ID |
3943 | return ret; |
3944 | } | |
3945 | ||
2b6ba629 ID |
3946 | int btrfs_resume_balance_async(struct btrfs_fs_info *fs_info) |
3947 | { | |
3948 | struct task_struct *tsk; | |
3949 | ||
3950 | spin_lock(&fs_info->balance_lock); | |
3951 | if (!fs_info->balance_ctl) { | |
3952 | spin_unlock(&fs_info->balance_lock); | |
3953 | return 0; | |
3954 | } | |
3955 | spin_unlock(&fs_info->balance_lock); | |
3956 | ||
3cdde224 | 3957 | if (btrfs_test_opt(fs_info, SKIP_BALANCE)) { |
efe120a0 | 3958 | btrfs_info(fs_info, "force skipping balance"); |
2b6ba629 ID |
3959 | return 0; |
3960 | } | |
3961 | ||
3962 | tsk = kthread_run(balance_kthread, fs_info, "btrfs-balance"); | |
cd633972 | 3963 | return PTR_ERR_OR_ZERO(tsk); |
2b6ba629 ID |
3964 | } |
3965 | ||
68310a5e | 3966 | int btrfs_recover_balance(struct btrfs_fs_info *fs_info) |
59641015 | 3967 | { |
59641015 ID |
3968 | struct btrfs_balance_control *bctl; |
3969 | struct btrfs_balance_item *item; | |
3970 | struct btrfs_disk_balance_args disk_bargs; | |
3971 | struct btrfs_path *path; | |
3972 | struct extent_buffer *leaf; | |
3973 | struct btrfs_key key; | |
3974 | int ret; | |
3975 | ||
3976 | path = btrfs_alloc_path(); | |
3977 | if (!path) | |
3978 | return -ENOMEM; | |
3979 | ||
59641015 | 3980 | key.objectid = BTRFS_BALANCE_OBJECTID; |
c479cb4f | 3981 | key.type = BTRFS_TEMPORARY_ITEM_KEY; |
59641015 ID |
3982 | key.offset = 0; |
3983 | ||
68310a5e | 3984 | ret = btrfs_search_slot(NULL, fs_info->tree_root, &key, path, 0, 0); |
59641015 | 3985 | if (ret < 0) |
68310a5e | 3986 | goto out; |
59641015 ID |
3987 | if (ret > 0) { /* ret = -ENOENT; */ |
3988 | ret = 0; | |
68310a5e ID |
3989 | goto out; |
3990 | } | |
3991 | ||
3992 | bctl = kzalloc(sizeof(*bctl), GFP_NOFS); | |
3993 | if (!bctl) { | |
3994 | ret = -ENOMEM; | |
3995 | goto out; | |
59641015 ID |
3996 | } |
3997 | ||
3998 | leaf = path->nodes[0]; | |
3999 | item = btrfs_item_ptr(leaf, path->slots[0], struct btrfs_balance_item); | |
4000 | ||
68310a5e ID |
4001 | bctl->fs_info = fs_info; |
4002 | bctl->flags = btrfs_balance_flags(leaf, item); | |
4003 | bctl->flags |= BTRFS_BALANCE_RESUME; | |
59641015 ID |
4004 | |
4005 | btrfs_balance_data(leaf, item, &disk_bargs); | |
4006 | btrfs_disk_balance_args_to_cpu(&bctl->data, &disk_bargs); | |
4007 | btrfs_balance_meta(leaf, item, &disk_bargs); | |
4008 | btrfs_disk_balance_args_to_cpu(&bctl->meta, &disk_bargs); | |
4009 | btrfs_balance_sys(leaf, item, &disk_bargs); | |
4010 | btrfs_disk_balance_args_to_cpu(&bctl->sys, &disk_bargs); | |
4011 | ||
ed0fb78f ID |
4012 | WARN_ON(atomic_xchg(&fs_info->mutually_exclusive_operation_running, 1)); |
4013 | ||
68310a5e ID |
4014 | mutex_lock(&fs_info->volume_mutex); |
4015 | mutex_lock(&fs_info->balance_mutex); | |
59641015 | 4016 | |
68310a5e ID |
4017 | set_balance_control(bctl); |
4018 | ||
4019 | mutex_unlock(&fs_info->balance_mutex); | |
4020 | mutex_unlock(&fs_info->volume_mutex); | |
59641015 ID |
4021 | out: |
4022 | btrfs_free_path(path); | |
ec44a35c CM |
4023 | return ret; |
4024 | } | |
4025 | ||
837d5b6e ID |
4026 | int btrfs_pause_balance(struct btrfs_fs_info *fs_info) |
4027 | { | |
4028 | int ret = 0; | |
4029 | ||
4030 | mutex_lock(&fs_info->balance_mutex); | |
4031 | if (!fs_info->balance_ctl) { | |
4032 | mutex_unlock(&fs_info->balance_mutex); | |
4033 | return -ENOTCONN; | |
4034 | } | |
4035 | ||
4036 | if (atomic_read(&fs_info->balance_running)) { | |
4037 | atomic_inc(&fs_info->balance_pause_req); | |
4038 | mutex_unlock(&fs_info->balance_mutex); | |
4039 | ||
4040 | wait_event(fs_info->balance_wait_q, | |
4041 | atomic_read(&fs_info->balance_running) == 0); | |
4042 | ||
4043 | mutex_lock(&fs_info->balance_mutex); | |
4044 | /* we are good with balance_ctl ripped off from under us */ | |
4045 | BUG_ON(atomic_read(&fs_info->balance_running)); | |
4046 | atomic_dec(&fs_info->balance_pause_req); | |
4047 | } else { | |
4048 | ret = -ENOTCONN; | |
4049 | } | |
4050 | ||
4051 | mutex_unlock(&fs_info->balance_mutex); | |
4052 | return ret; | |
4053 | } | |
4054 | ||
a7e99c69 ID |
4055 | int btrfs_cancel_balance(struct btrfs_fs_info *fs_info) |
4056 | { | |
e649e587 ID |
4057 | if (fs_info->sb->s_flags & MS_RDONLY) |
4058 | return -EROFS; | |
4059 | ||
a7e99c69 ID |
4060 | mutex_lock(&fs_info->balance_mutex); |
4061 | if (!fs_info->balance_ctl) { | |
4062 | mutex_unlock(&fs_info->balance_mutex); | |
4063 | return -ENOTCONN; | |
4064 | } | |
4065 | ||
4066 | atomic_inc(&fs_info->balance_cancel_req); | |
4067 | /* | |
4068 | * if we are running just wait and return, balance item is | |
4069 | * deleted in btrfs_balance in this case | |
4070 | */ | |
4071 | if (atomic_read(&fs_info->balance_running)) { | |
4072 | mutex_unlock(&fs_info->balance_mutex); | |
4073 | wait_event(fs_info->balance_wait_q, | |
4074 | atomic_read(&fs_info->balance_running) == 0); | |
4075 | mutex_lock(&fs_info->balance_mutex); | |
4076 | } else { | |
4077 | /* __cancel_balance needs volume_mutex */ | |
4078 | mutex_unlock(&fs_info->balance_mutex); | |
4079 | mutex_lock(&fs_info->volume_mutex); | |
4080 | mutex_lock(&fs_info->balance_mutex); | |
4081 | ||
4082 | if (fs_info->balance_ctl) | |
4083 | __cancel_balance(fs_info); | |
4084 | ||
4085 | mutex_unlock(&fs_info->volume_mutex); | |
4086 | } | |
4087 | ||
4088 | BUG_ON(fs_info->balance_ctl || atomic_read(&fs_info->balance_running)); | |
4089 | atomic_dec(&fs_info->balance_cancel_req); | |
4090 | mutex_unlock(&fs_info->balance_mutex); | |
4091 | return 0; | |
4092 | } | |
4093 | ||
803b2f54 SB |
4094 | static int btrfs_uuid_scan_kthread(void *data) |
4095 | { | |
4096 | struct btrfs_fs_info *fs_info = data; | |
4097 | struct btrfs_root *root = fs_info->tree_root; | |
4098 | struct btrfs_key key; | |
4099 | struct btrfs_key max_key; | |
4100 | struct btrfs_path *path = NULL; | |
4101 | int ret = 0; | |
4102 | struct extent_buffer *eb; | |
4103 | int slot; | |
4104 | struct btrfs_root_item root_item; | |
4105 | u32 item_size; | |
f45388f3 | 4106 | struct btrfs_trans_handle *trans = NULL; |
803b2f54 SB |
4107 | |
4108 | path = btrfs_alloc_path(); | |
4109 | if (!path) { | |
4110 | ret = -ENOMEM; | |
4111 | goto out; | |
4112 | } | |
4113 | ||
4114 | key.objectid = 0; | |
4115 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4116 | key.offset = 0; | |
4117 | ||
4118 | max_key.objectid = (u64)-1; | |
4119 | max_key.type = BTRFS_ROOT_ITEM_KEY; | |
4120 | max_key.offset = (u64)-1; | |
4121 | ||
803b2f54 | 4122 | while (1) { |
6174d3cb | 4123 | ret = btrfs_search_forward(root, &key, path, 0); |
803b2f54 SB |
4124 | if (ret) { |
4125 | if (ret > 0) | |
4126 | ret = 0; | |
4127 | break; | |
4128 | } | |
4129 | ||
4130 | if (key.type != BTRFS_ROOT_ITEM_KEY || | |
4131 | (key.objectid < BTRFS_FIRST_FREE_OBJECTID && | |
4132 | key.objectid != BTRFS_FS_TREE_OBJECTID) || | |
4133 | key.objectid > BTRFS_LAST_FREE_OBJECTID) | |
4134 | goto skip; | |
4135 | ||
4136 | eb = path->nodes[0]; | |
4137 | slot = path->slots[0]; | |
4138 | item_size = btrfs_item_size_nr(eb, slot); | |
4139 | if (item_size < sizeof(root_item)) | |
4140 | goto skip; | |
4141 | ||
803b2f54 SB |
4142 | read_extent_buffer(eb, &root_item, |
4143 | btrfs_item_ptr_offset(eb, slot), | |
4144 | (int)sizeof(root_item)); | |
4145 | if (btrfs_root_refs(&root_item) == 0) | |
4146 | goto skip; | |
f45388f3 FDBM |
4147 | |
4148 | if (!btrfs_is_empty_uuid(root_item.uuid) || | |
4149 | !btrfs_is_empty_uuid(root_item.received_uuid)) { | |
4150 | if (trans) | |
4151 | goto update_tree; | |
4152 | ||
4153 | btrfs_release_path(path); | |
803b2f54 SB |
4154 | /* |
4155 | * 1 - subvol uuid item | |
4156 | * 1 - received_subvol uuid item | |
4157 | */ | |
4158 | trans = btrfs_start_transaction(fs_info->uuid_root, 2); | |
4159 | if (IS_ERR(trans)) { | |
4160 | ret = PTR_ERR(trans); | |
4161 | break; | |
4162 | } | |
f45388f3 FDBM |
4163 | continue; |
4164 | } else { | |
4165 | goto skip; | |
4166 | } | |
4167 | update_tree: | |
4168 | if (!btrfs_is_empty_uuid(root_item.uuid)) { | |
6bccf3ab | 4169 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4170 | root_item.uuid, |
4171 | BTRFS_UUID_KEY_SUBVOL, | |
4172 | key.objectid); | |
4173 | if (ret < 0) { | |
efe120a0 | 4174 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4175 | ret); |
803b2f54 SB |
4176 | break; |
4177 | } | |
4178 | } | |
4179 | ||
4180 | if (!btrfs_is_empty_uuid(root_item.received_uuid)) { | |
6bccf3ab | 4181 | ret = btrfs_uuid_tree_add(trans, fs_info, |
803b2f54 SB |
4182 | root_item.received_uuid, |
4183 | BTRFS_UUID_KEY_RECEIVED_SUBVOL, | |
4184 | key.objectid); | |
4185 | if (ret < 0) { | |
efe120a0 | 4186 | btrfs_warn(fs_info, "uuid_tree_add failed %d", |
803b2f54 | 4187 | ret); |
803b2f54 SB |
4188 | break; |
4189 | } | |
4190 | } | |
4191 | ||
f45388f3 | 4192 | skip: |
803b2f54 SB |
4193 | if (trans) { |
4194 | ret = btrfs_end_transaction(trans, fs_info->uuid_root); | |
f45388f3 | 4195 | trans = NULL; |
803b2f54 SB |
4196 | if (ret) |
4197 | break; | |
4198 | } | |
4199 | ||
803b2f54 SB |
4200 | btrfs_release_path(path); |
4201 | if (key.offset < (u64)-1) { | |
4202 | key.offset++; | |
4203 | } else if (key.type < BTRFS_ROOT_ITEM_KEY) { | |
4204 | key.offset = 0; | |
4205 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4206 | } else if (key.objectid < (u64)-1) { | |
4207 | key.offset = 0; | |
4208 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4209 | key.objectid++; | |
4210 | } else { | |
4211 | break; | |
4212 | } | |
4213 | cond_resched(); | |
4214 | } | |
4215 | ||
4216 | out: | |
4217 | btrfs_free_path(path); | |
f45388f3 FDBM |
4218 | if (trans && !IS_ERR(trans)) |
4219 | btrfs_end_transaction(trans, fs_info->uuid_root); | |
803b2f54 | 4220 | if (ret) |
efe120a0 | 4221 | btrfs_warn(fs_info, "btrfs_uuid_scan_kthread failed %d", ret); |
70f80175 | 4222 | else |
afcdd129 | 4223 | set_bit(BTRFS_FS_UPDATE_UUID_TREE_GEN, &fs_info->flags); |
803b2f54 SB |
4224 | up(&fs_info->uuid_tree_rescan_sem); |
4225 | return 0; | |
4226 | } | |
4227 | ||
70f80175 SB |
4228 | /* |
4229 | * Callback for btrfs_uuid_tree_iterate(). | |
4230 | * returns: | |
4231 | * 0 check succeeded, the entry is not outdated. | |
bb7ab3b9 | 4232 | * < 0 if an error occurred. |
70f80175 SB |
4233 | * > 0 if the check failed, which means the caller shall remove the entry. |
4234 | */ | |
4235 | static int btrfs_check_uuid_tree_entry(struct btrfs_fs_info *fs_info, | |
4236 | u8 *uuid, u8 type, u64 subid) | |
4237 | { | |
4238 | struct btrfs_key key; | |
4239 | int ret = 0; | |
4240 | struct btrfs_root *subvol_root; | |
4241 | ||
4242 | if (type != BTRFS_UUID_KEY_SUBVOL && | |
4243 | type != BTRFS_UUID_KEY_RECEIVED_SUBVOL) | |
4244 | goto out; | |
4245 | ||
4246 | key.objectid = subid; | |
4247 | key.type = BTRFS_ROOT_ITEM_KEY; | |
4248 | key.offset = (u64)-1; | |
4249 | subvol_root = btrfs_read_fs_root_no_name(fs_info, &key); | |
4250 | if (IS_ERR(subvol_root)) { | |
4251 | ret = PTR_ERR(subvol_root); | |
4252 | if (ret == -ENOENT) | |
4253 | ret = 1; | |
4254 | goto out; | |
4255 | } | |
4256 | ||
4257 | switch (type) { | |
4258 | case BTRFS_UUID_KEY_SUBVOL: | |
4259 | if (memcmp(uuid, subvol_root->root_item.uuid, BTRFS_UUID_SIZE)) | |
4260 | ret = 1; | |
4261 | break; | |
4262 | case BTRFS_UUID_KEY_RECEIVED_SUBVOL: | |
4263 | if (memcmp(uuid, subvol_root->root_item.received_uuid, | |
4264 | BTRFS_UUID_SIZE)) | |
4265 | ret = 1; | |
4266 | break; | |
4267 | } | |
4268 | ||
4269 | out: | |
4270 | return ret; | |
4271 | } | |
4272 | ||
4273 | static int btrfs_uuid_rescan_kthread(void *data) | |
4274 | { | |
4275 | struct btrfs_fs_info *fs_info = (struct btrfs_fs_info *)data; | |
4276 | int ret; | |
4277 | ||
4278 | /* | |
4279 | * 1st step is to iterate through the existing UUID tree and | |
4280 | * to delete all entries that contain outdated data. | |
4281 | * 2nd step is to add all missing entries to the UUID tree. | |
4282 | */ | |
4283 | ret = btrfs_uuid_tree_iterate(fs_info, btrfs_check_uuid_tree_entry); | |
4284 | if (ret < 0) { | |
efe120a0 | 4285 | btrfs_warn(fs_info, "iterating uuid_tree failed %d", ret); |
70f80175 SB |
4286 | up(&fs_info->uuid_tree_rescan_sem); |
4287 | return ret; | |
4288 | } | |
4289 | return btrfs_uuid_scan_kthread(data); | |
4290 | } | |
4291 | ||
f7a81ea4 SB |
4292 | int btrfs_create_uuid_tree(struct btrfs_fs_info *fs_info) |
4293 | { | |
4294 | struct btrfs_trans_handle *trans; | |
4295 | struct btrfs_root *tree_root = fs_info->tree_root; | |
4296 | struct btrfs_root *uuid_root; | |
803b2f54 SB |
4297 | struct task_struct *task; |
4298 | int ret; | |
f7a81ea4 SB |
4299 | |
4300 | /* | |
4301 | * 1 - root node | |
4302 | * 1 - root item | |
4303 | */ | |
4304 | trans = btrfs_start_transaction(tree_root, 2); | |
4305 | if (IS_ERR(trans)) | |
4306 | return PTR_ERR(trans); | |
4307 | ||
4308 | uuid_root = btrfs_create_tree(trans, fs_info, | |
4309 | BTRFS_UUID_TREE_OBJECTID); | |
4310 | if (IS_ERR(uuid_root)) { | |
6d13f549 | 4311 | ret = PTR_ERR(uuid_root); |
66642832 | 4312 | btrfs_abort_transaction(trans, ret); |
65d4f4c1 | 4313 | btrfs_end_transaction(trans, tree_root); |
6d13f549 | 4314 | return ret; |
f7a81ea4 SB |
4315 | } |
4316 | ||
4317 | fs_info->uuid_root = uuid_root; | |
4318 | ||
803b2f54 SB |
4319 | ret = btrfs_commit_transaction(trans, tree_root); |
4320 | if (ret) | |
4321 | return ret; | |
4322 | ||
4323 | down(&fs_info->uuid_tree_rescan_sem); | |
4324 | task = kthread_run(btrfs_uuid_scan_kthread, fs_info, "btrfs-uuid"); | |
4325 | if (IS_ERR(task)) { | |
70f80175 | 4326 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ |
efe120a0 | 4327 | btrfs_warn(fs_info, "failed to start uuid_scan task"); |
803b2f54 SB |
4328 | up(&fs_info->uuid_tree_rescan_sem); |
4329 | return PTR_ERR(task); | |
4330 | } | |
4331 | ||
4332 | return 0; | |
f7a81ea4 | 4333 | } |
803b2f54 | 4334 | |
70f80175 SB |
4335 | int btrfs_check_uuid_tree(struct btrfs_fs_info *fs_info) |
4336 | { | |
4337 | struct task_struct *task; | |
4338 | ||
4339 | down(&fs_info->uuid_tree_rescan_sem); | |
4340 | task = kthread_run(btrfs_uuid_rescan_kthread, fs_info, "btrfs-uuid"); | |
4341 | if (IS_ERR(task)) { | |
4342 | /* fs_info->update_uuid_tree_gen remains 0 in all error case */ | |
efe120a0 | 4343 | btrfs_warn(fs_info, "failed to start uuid_rescan task"); |
70f80175 SB |
4344 | up(&fs_info->uuid_tree_rescan_sem); |
4345 | return PTR_ERR(task); | |
4346 | } | |
4347 | ||
4348 | return 0; | |
4349 | } | |
4350 | ||
8f18cf13 CM |
4351 | /* |
4352 | * shrinking a device means finding all of the device extents past | |
4353 | * the new size, and then following the back refs to the chunks. | |
4354 | * The chunk relocation code actually frees the device extent | |
4355 | */ | |
4356 | int btrfs_shrink_device(struct btrfs_device *device, u64 new_size) | |
4357 | { | |
0b246afa JM |
4358 | struct btrfs_fs_info *fs_info = device->fs_info; |
4359 | struct btrfs_root *root = fs_info->dev_root; | |
8f18cf13 | 4360 | struct btrfs_trans_handle *trans; |
8f18cf13 CM |
4361 | struct btrfs_dev_extent *dev_extent = NULL; |
4362 | struct btrfs_path *path; | |
4363 | u64 length; | |
8f18cf13 CM |
4364 | u64 chunk_offset; |
4365 | int ret; | |
4366 | int slot; | |
ba1bf481 JB |
4367 | int failed = 0; |
4368 | bool retried = false; | |
53e489bc | 4369 | bool checked_pending_chunks = false; |
8f18cf13 CM |
4370 | struct extent_buffer *l; |
4371 | struct btrfs_key key; | |
0b246afa | 4372 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
8f18cf13 | 4373 | u64 old_total = btrfs_super_total_bytes(super_copy); |
7cc8e58d MX |
4374 | u64 old_size = btrfs_device_get_total_bytes(device); |
4375 | u64 diff = old_size - new_size; | |
8f18cf13 | 4376 | |
63a212ab SB |
4377 | if (device->is_tgtdev_for_dev_replace) |
4378 | return -EINVAL; | |
4379 | ||
8f18cf13 CM |
4380 | path = btrfs_alloc_path(); |
4381 | if (!path) | |
4382 | return -ENOMEM; | |
4383 | ||
e4058b54 | 4384 | path->reada = READA_FORWARD; |
8f18cf13 | 4385 | |
0b246afa | 4386 | lock_chunks(fs_info); |
7d9eb12c | 4387 | |
7cc8e58d | 4388 | btrfs_device_set_total_bytes(device, new_size); |
2bf64758 | 4389 | if (device->writeable) { |
2b82032c | 4390 | device->fs_devices->total_rw_bytes -= diff; |
0b246afa JM |
4391 | spin_lock(&fs_info->free_chunk_lock); |
4392 | fs_info->free_chunk_space -= diff; | |
4393 | spin_unlock(&fs_info->free_chunk_lock); | |
2bf64758 | 4394 | } |
0b246afa | 4395 | unlock_chunks(fs_info); |
8f18cf13 | 4396 | |
ba1bf481 | 4397 | again: |
8f18cf13 CM |
4398 | key.objectid = device->devid; |
4399 | key.offset = (u64)-1; | |
4400 | key.type = BTRFS_DEV_EXTENT_KEY; | |
4401 | ||
213e64da | 4402 | do { |
0b246afa | 4403 | mutex_lock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4404 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
67c5e7d4 | 4405 | if (ret < 0) { |
0b246afa | 4406 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 | 4407 | goto done; |
67c5e7d4 | 4408 | } |
8f18cf13 CM |
4409 | |
4410 | ret = btrfs_previous_item(root, path, 0, key.type); | |
67c5e7d4 | 4411 | if (ret) |
0b246afa | 4412 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
8f18cf13 CM |
4413 | if (ret < 0) |
4414 | goto done; | |
4415 | if (ret) { | |
4416 | ret = 0; | |
b3b4aa74 | 4417 | btrfs_release_path(path); |
bf1fb512 | 4418 | break; |
8f18cf13 CM |
4419 | } |
4420 | ||
4421 | l = path->nodes[0]; | |
4422 | slot = path->slots[0]; | |
4423 | btrfs_item_key_to_cpu(l, &key, path->slots[0]); | |
4424 | ||
ba1bf481 | 4425 | if (key.objectid != device->devid) { |
0b246afa | 4426 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4427 | btrfs_release_path(path); |
bf1fb512 | 4428 | break; |
ba1bf481 | 4429 | } |
8f18cf13 CM |
4430 | |
4431 | dev_extent = btrfs_item_ptr(l, slot, struct btrfs_dev_extent); | |
4432 | length = btrfs_dev_extent_length(l, dev_extent); | |
4433 | ||
ba1bf481 | 4434 | if (key.offset + length <= new_size) { |
0b246afa | 4435 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); |
b3b4aa74 | 4436 | btrfs_release_path(path); |
d6397bae | 4437 | break; |
ba1bf481 | 4438 | } |
8f18cf13 | 4439 | |
8f18cf13 | 4440 | chunk_offset = btrfs_dev_extent_chunk_offset(l, dev_extent); |
b3b4aa74 | 4441 | btrfs_release_path(path); |
8f18cf13 | 4442 | |
0b246afa JM |
4443 | ret = btrfs_relocate_chunk(fs_info, chunk_offset); |
4444 | mutex_unlock(&fs_info->delete_unused_bgs_mutex); | |
ba1bf481 | 4445 | if (ret && ret != -ENOSPC) |
8f18cf13 | 4446 | goto done; |
ba1bf481 JB |
4447 | if (ret == -ENOSPC) |
4448 | failed++; | |
213e64da | 4449 | } while (key.offset-- > 0); |
ba1bf481 JB |
4450 | |
4451 | if (failed && !retried) { | |
4452 | failed = 0; | |
4453 | retried = true; | |
4454 | goto again; | |
4455 | } else if (failed && retried) { | |
4456 | ret = -ENOSPC; | |
ba1bf481 | 4457 | goto done; |
8f18cf13 CM |
4458 | } |
4459 | ||
d6397bae | 4460 | /* Shrinking succeeded, else we would be at "done". */ |
a22285a6 | 4461 | trans = btrfs_start_transaction(root, 0); |
98d5dc13 TI |
4462 | if (IS_ERR(trans)) { |
4463 | ret = PTR_ERR(trans); | |
4464 | goto done; | |
4465 | } | |
4466 | ||
0b246afa | 4467 | lock_chunks(fs_info); |
53e489bc FM |
4468 | |
4469 | /* | |
4470 | * We checked in the above loop all device extents that were already in | |
4471 | * the device tree. However before we have updated the device's | |
4472 | * total_bytes to the new size, we might have had chunk allocations that | |
4473 | * have not complete yet (new block groups attached to transaction | |
4474 | * handles), and therefore their device extents were not yet in the | |
4475 | * device tree and we missed them in the loop above. So if we have any | |
4476 | * pending chunk using a device extent that overlaps the device range | |
4477 | * that we can not use anymore, commit the current transaction and | |
4478 | * repeat the search on the device tree - this way we guarantee we will | |
4479 | * not have chunks using device extents that end beyond 'new_size'. | |
4480 | */ | |
4481 | if (!checked_pending_chunks) { | |
4482 | u64 start = new_size; | |
4483 | u64 len = old_size - new_size; | |
4484 | ||
499f377f JM |
4485 | if (contains_pending_extent(trans->transaction, device, |
4486 | &start, len)) { | |
0b246afa | 4487 | unlock_chunks(fs_info); |
53e489bc FM |
4488 | checked_pending_chunks = true; |
4489 | failed = 0; | |
4490 | retried = false; | |
4491 | ret = btrfs_commit_transaction(trans, root); | |
4492 | if (ret) | |
4493 | goto done; | |
4494 | goto again; | |
4495 | } | |
4496 | } | |
4497 | ||
7cc8e58d | 4498 | btrfs_device_set_disk_total_bytes(device, new_size); |
935e5cc9 MX |
4499 | if (list_empty(&device->resized_list)) |
4500 | list_add_tail(&device->resized_list, | |
0b246afa | 4501 | &fs_info->fs_devices->resized_devices); |
d6397bae | 4502 | |
d6397bae CB |
4503 | WARN_ON(diff > old_total); |
4504 | btrfs_set_super_total_bytes(super_copy, old_total - diff); | |
0b246afa | 4505 | unlock_chunks(fs_info); |
2196d6e8 MX |
4506 | |
4507 | /* Now btrfs_update_device() will change the on-disk size. */ | |
4508 | ret = btrfs_update_device(trans, device); | |
d6397bae | 4509 | btrfs_end_transaction(trans, root); |
8f18cf13 CM |
4510 | done: |
4511 | btrfs_free_path(path); | |
53e489bc | 4512 | if (ret) { |
0b246afa | 4513 | lock_chunks(fs_info); |
53e489bc FM |
4514 | btrfs_device_set_total_bytes(device, old_size); |
4515 | if (device->writeable) | |
4516 | device->fs_devices->total_rw_bytes += diff; | |
0b246afa JM |
4517 | spin_lock(&fs_info->free_chunk_lock); |
4518 | fs_info->free_chunk_space += diff; | |
4519 | spin_unlock(&fs_info->free_chunk_lock); | |
4520 | unlock_chunks(fs_info); | |
53e489bc | 4521 | } |
8f18cf13 CM |
4522 | return ret; |
4523 | } | |
4524 | ||
125ccb0a | 4525 | static int btrfs_add_system_chunk(struct btrfs_root *root, |
0b86a832 CM |
4526 | struct btrfs_key *key, |
4527 | struct btrfs_chunk *chunk, int item_size) | |
4528 | { | |
0b246afa JM |
4529 | struct btrfs_fs_info *fs_info = root->fs_info; |
4530 | struct btrfs_super_block *super_copy = fs_info->super_copy; | |
0b86a832 CM |
4531 | struct btrfs_disk_key disk_key; |
4532 | u32 array_size; | |
4533 | u8 *ptr; | |
4534 | ||
0b246afa | 4535 | lock_chunks(fs_info); |
0b86a832 | 4536 | array_size = btrfs_super_sys_array_size(super_copy); |
5f43f86e | 4537 | if (array_size + item_size + sizeof(disk_key) |
fe48a5c0 | 4538 | > BTRFS_SYSTEM_CHUNK_ARRAY_SIZE) { |
0b246afa | 4539 | unlock_chunks(fs_info); |
0b86a832 | 4540 | return -EFBIG; |
fe48a5c0 | 4541 | } |
0b86a832 CM |
4542 | |
4543 | ptr = super_copy->sys_chunk_array + array_size; | |
4544 | btrfs_cpu_key_to_disk(&disk_key, key); | |
4545 | memcpy(ptr, &disk_key, sizeof(disk_key)); | |
4546 | ptr += sizeof(disk_key); | |
4547 | memcpy(ptr, chunk, item_size); | |
4548 | item_size += sizeof(disk_key); | |
4549 | btrfs_set_super_sys_array_size(super_copy, array_size + item_size); | |
0b246afa | 4550 | unlock_chunks(fs_info); |
fe48a5c0 | 4551 | |
0b86a832 CM |
4552 | return 0; |
4553 | } | |
4554 | ||
73c5de00 AJ |
4555 | /* |
4556 | * sort the devices in descending order by max_avail, total_avail | |
4557 | */ | |
4558 | static int btrfs_cmp_device_info(const void *a, const void *b) | |
9b3f68b9 | 4559 | { |
73c5de00 AJ |
4560 | const struct btrfs_device_info *di_a = a; |
4561 | const struct btrfs_device_info *di_b = b; | |
9b3f68b9 | 4562 | |
73c5de00 | 4563 | if (di_a->max_avail > di_b->max_avail) |
b2117a39 | 4564 | return -1; |
73c5de00 | 4565 | if (di_a->max_avail < di_b->max_avail) |
b2117a39 | 4566 | return 1; |
73c5de00 AJ |
4567 | if (di_a->total_avail > di_b->total_avail) |
4568 | return -1; | |
4569 | if (di_a->total_avail < di_b->total_avail) | |
4570 | return 1; | |
4571 | return 0; | |
b2117a39 | 4572 | } |
0b86a832 | 4573 | |
53b381b3 DW |
4574 | static u32 find_raid56_stripe_len(u32 data_devices, u32 dev_stripe_target) |
4575 | { | |
4576 | /* TODO allow them to set a preferred stripe size */ | |
ee22184b | 4577 | return SZ_64K; |
53b381b3 DW |
4578 | } |
4579 | ||
4580 | static void check_raid56_incompat_flag(struct btrfs_fs_info *info, u64 type) | |
4581 | { | |
ffe2d203 | 4582 | if (!(type & BTRFS_BLOCK_GROUP_RAID56_MASK)) |
53b381b3 DW |
4583 | return; |
4584 | ||
ceda0864 | 4585 | btrfs_set_fs_incompat(info, RAID56); |
53b381b3 DW |
4586 | } |
4587 | ||
da17066c | 4588 | #define BTRFS_MAX_DEVS(r) ((BTRFS_MAX_ITEM_SIZE(r->fs_info) \ |
23f8f9b7 GH |
4589 | - sizeof(struct btrfs_chunk)) \ |
4590 | / sizeof(struct btrfs_stripe) + 1) | |
4591 | ||
4592 | #define BTRFS_MAX_DEVS_SYS_CHUNK ((BTRFS_SYSTEM_CHUNK_ARRAY_SIZE \ | |
4593 | - 2 * sizeof(struct btrfs_disk_key) \ | |
4594 | - 2 * sizeof(struct btrfs_chunk)) \ | |
4595 | / sizeof(struct btrfs_stripe) + 1) | |
4596 | ||
73c5de00 | 4597 | static int __btrfs_alloc_chunk(struct btrfs_trans_handle *trans, |
6df9a95e JB |
4598 | struct btrfs_root *extent_root, u64 start, |
4599 | u64 type) | |
b2117a39 | 4600 | { |
73c5de00 AJ |
4601 | struct btrfs_fs_info *info = extent_root->fs_info; |
4602 | struct btrfs_fs_devices *fs_devices = info->fs_devices; | |
4603 | struct list_head *cur; | |
4604 | struct map_lookup *map = NULL; | |
4605 | struct extent_map_tree *em_tree; | |
4606 | struct extent_map *em; | |
4607 | struct btrfs_device_info *devices_info = NULL; | |
4608 | u64 total_avail; | |
4609 | int num_stripes; /* total number of stripes to allocate */ | |
53b381b3 DW |
4610 | int data_stripes; /* number of stripes that count for |
4611 | block group size */ | |
73c5de00 AJ |
4612 | int sub_stripes; /* sub_stripes info for map */ |
4613 | int dev_stripes; /* stripes per dev */ | |
4614 | int devs_max; /* max devs to use */ | |
4615 | int devs_min; /* min devs needed */ | |
4616 | int devs_increment; /* ndevs has to be a multiple of this */ | |
4617 | int ncopies; /* how many copies to data has */ | |
4618 | int ret; | |
4619 | u64 max_stripe_size; | |
4620 | u64 max_chunk_size; | |
4621 | u64 stripe_size; | |
4622 | u64 num_bytes; | |
53b381b3 | 4623 | u64 raid_stripe_len = BTRFS_STRIPE_LEN; |
73c5de00 AJ |
4624 | int ndevs; |
4625 | int i; | |
4626 | int j; | |
31e50229 | 4627 | int index; |
593060d7 | 4628 | |
0c460c0d | 4629 | BUG_ON(!alloc_profile_is_valid(type, 0)); |
9b3f68b9 | 4630 | |
73c5de00 AJ |
4631 | if (list_empty(&fs_devices->alloc_list)) |
4632 | return -ENOSPC; | |
b2117a39 | 4633 | |
31e50229 | 4634 | index = __get_raid_index(type); |
73c5de00 | 4635 | |
31e50229 LB |
4636 | sub_stripes = btrfs_raid_array[index].sub_stripes; |
4637 | dev_stripes = btrfs_raid_array[index].dev_stripes; | |
4638 | devs_max = btrfs_raid_array[index].devs_max; | |
4639 | devs_min = btrfs_raid_array[index].devs_min; | |
4640 | devs_increment = btrfs_raid_array[index].devs_increment; | |
4641 | ncopies = btrfs_raid_array[index].ncopies; | |
b2117a39 | 4642 | |
9b3f68b9 | 4643 | if (type & BTRFS_BLOCK_GROUP_DATA) { |
ee22184b | 4644 | max_stripe_size = SZ_1G; |
73c5de00 | 4645 | max_chunk_size = 10 * max_stripe_size; |
23f8f9b7 GH |
4646 | if (!devs_max) |
4647 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
9b3f68b9 | 4648 | } else if (type & BTRFS_BLOCK_GROUP_METADATA) { |
1100373f | 4649 | /* for larger filesystems, use larger metadata chunks */ |
ee22184b BL |
4650 | if (fs_devices->total_rw_bytes > 50ULL * SZ_1G) |
4651 | max_stripe_size = SZ_1G; | |
1100373f | 4652 | else |
ee22184b | 4653 | max_stripe_size = SZ_256M; |
73c5de00 | 4654 | max_chunk_size = max_stripe_size; |
23f8f9b7 GH |
4655 | if (!devs_max) |
4656 | devs_max = BTRFS_MAX_DEVS(info->chunk_root); | |
a40a90a0 | 4657 | } else if (type & BTRFS_BLOCK_GROUP_SYSTEM) { |
ee22184b | 4658 | max_stripe_size = SZ_32M; |
73c5de00 | 4659 | max_chunk_size = 2 * max_stripe_size; |
23f8f9b7 GH |
4660 | if (!devs_max) |
4661 | devs_max = BTRFS_MAX_DEVS_SYS_CHUNK; | |
73c5de00 | 4662 | } else { |
351fd353 | 4663 | btrfs_err(info, "invalid chunk type 0x%llx requested", |
73c5de00 AJ |
4664 | type); |
4665 | BUG_ON(1); | |
9b3f68b9 CM |
4666 | } |
4667 | ||
2b82032c YZ |
4668 | /* we don't want a chunk larger than 10% of writeable space */ |
4669 | max_chunk_size = min(div_factor(fs_devices->total_rw_bytes, 1), | |
4670 | max_chunk_size); | |
9b3f68b9 | 4671 | |
31e818fe | 4672 | devices_info = kcalloc(fs_devices->rw_devices, sizeof(*devices_info), |
73c5de00 AJ |
4673 | GFP_NOFS); |
4674 | if (!devices_info) | |
4675 | return -ENOMEM; | |
0cad8a11 | 4676 | |
73c5de00 | 4677 | cur = fs_devices->alloc_list.next; |
9b3f68b9 | 4678 | |
9f680ce0 | 4679 | /* |
73c5de00 AJ |
4680 | * in the first pass through the devices list, we gather information |
4681 | * about the available holes on each device. | |
9f680ce0 | 4682 | */ |
73c5de00 AJ |
4683 | ndevs = 0; |
4684 | while (cur != &fs_devices->alloc_list) { | |
4685 | struct btrfs_device *device; | |
4686 | u64 max_avail; | |
4687 | u64 dev_offset; | |
b2117a39 | 4688 | |
73c5de00 | 4689 | device = list_entry(cur, struct btrfs_device, dev_alloc_list); |
9f680ce0 | 4690 | |
73c5de00 | 4691 | cur = cur->next; |
b2117a39 | 4692 | |
73c5de00 | 4693 | if (!device->writeable) { |
31b1a2bd | 4694 | WARN(1, KERN_ERR |
efe120a0 | 4695 | "BTRFS: read-only device in alloc_list\n"); |
73c5de00 AJ |
4696 | continue; |
4697 | } | |
b2117a39 | 4698 | |
63a212ab SB |
4699 | if (!device->in_fs_metadata || |
4700 | device->is_tgtdev_for_dev_replace) | |
73c5de00 | 4701 | continue; |
b2117a39 | 4702 | |
73c5de00 AJ |
4703 | if (device->total_bytes > device->bytes_used) |
4704 | total_avail = device->total_bytes - device->bytes_used; | |
4705 | else | |
4706 | total_avail = 0; | |
38c01b96 | 4707 | |
4708 | /* If there is no space on this device, skip it. */ | |
4709 | if (total_avail == 0) | |
4710 | continue; | |
b2117a39 | 4711 | |
6df9a95e | 4712 | ret = find_free_dev_extent(trans, device, |
73c5de00 AJ |
4713 | max_stripe_size * dev_stripes, |
4714 | &dev_offset, &max_avail); | |
4715 | if (ret && ret != -ENOSPC) | |
4716 | goto error; | |
b2117a39 | 4717 | |
73c5de00 AJ |
4718 | if (ret == 0) |
4719 | max_avail = max_stripe_size * dev_stripes; | |
b2117a39 | 4720 | |
73c5de00 AJ |
4721 | if (max_avail < BTRFS_STRIPE_LEN * dev_stripes) |
4722 | continue; | |
b2117a39 | 4723 | |
063d006f ES |
4724 | if (ndevs == fs_devices->rw_devices) { |
4725 | WARN(1, "%s: found more than %llu devices\n", | |
4726 | __func__, fs_devices->rw_devices); | |
4727 | break; | |
4728 | } | |
73c5de00 AJ |
4729 | devices_info[ndevs].dev_offset = dev_offset; |
4730 | devices_info[ndevs].max_avail = max_avail; | |
4731 | devices_info[ndevs].total_avail = total_avail; | |
4732 | devices_info[ndevs].dev = device; | |
4733 | ++ndevs; | |
4734 | } | |
b2117a39 | 4735 | |
73c5de00 AJ |
4736 | /* |
4737 | * now sort the devices by hole size / available space | |
4738 | */ | |
4739 | sort(devices_info, ndevs, sizeof(struct btrfs_device_info), | |
4740 | btrfs_cmp_device_info, NULL); | |
b2117a39 | 4741 | |
73c5de00 AJ |
4742 | /* round down to number of usable stripes */ |
4743 | ndevs -= ndevs % devs_increment; | |
b2117a39 | 4744 | |
73c5de00 AJ |
4745 | if (ndevs < devs_increment * sub_stripes || ndevs < devs_min) { |
4746 | ret = -ENOSPC; | |
4747 | goto error; | |
b2117a39 | 4748 | } |
9f680ce0 | 4749 | |
73c5de00 AJ |
4750 | if (devs_max && ndevs > devs_max) |
4751 | ndevs = devs_max; | |
4752 | /* | |
4753 | * the primary goal is to maximize the number of stripes, so use as many | |
4754 | * devices as possible, even if the stripes are not maximum sized. | |
4755 | */ | |
4756 | stripe_size = devices_info[ndevs-1].max_avail; | |
4757 | num_stripes = ndevs * dev_stripes; | |
b2117a39 | 4758 | |
53b381b3 DW |
4759 | /* |
4760 | * this will have to be fixed for RAID1 and RAID10 over | |
4761 | * more drives | |
4762 | */ | |
4763 | data_stripes = num_stripes / ncopies; | |
4764 | ||
53b381b3 DW |
4765 | if (type & BTRFS_BLOCK_GROUP_RAID5) { |
4766 | raid_stripe_len = find_raid56_stripe_len(ndevs - 1, | |
da17066c | 4767 | info->stripesize); |
53b381b3 DW |
4768 | data_stripes = num_stripes - 1; |
4769 | } | |
4770 | if (type & BTRFS_BLOCK_GROUP_RAID6) { | |
4771 | raid_stripe_len = find_raid56_stripe_len(ndevs - 2, | |
da17066c | 4772 | info->stripesize); |
53b381b3 DW |
4773 | data_stripes = num_stripes - 2; |
4774 | } | |
86db2578 CM |
4775 | |
4776 | /* | |
4777 | * Use the number of data stripes to figure out how big this chunk | |
4778 | * is really going to be in terms of logical address space, | |
4779 | * and compare that answer with the max chunk size | |
4780 | */ | |
4781 | if (stripe_size * data_stripes > max_chunk_size) { | |
4782 | u64 mask = (1ULL << 24) - 1; | |
b8b93add DS |
4783 | |
4784 | stripe_size = div_u64(max_chunk_size, data_stripes); | |
86db2578 CM |
4785 | |
4786 | /* bump the answer up to a 16MB boundary */ | |
4787 | stripe_size = (stripe_size + mask) & ~mask; | |
4788 | ||
4789 | /* but don't go higher than the limits we found | |
4790 | * while searching for free extents | |
4791 | */ | |
4792 | if (stripe_size > devices_info[ndevs-1].max_avail) | |
4793 | stripe_size = devices_info[ndevs-1].max_avail; | |
4794 | } | |
4795 | ||
b8b93add | 4796 | stripe_size = div_u64(stripe_size, dev_stripes); |
37db63a4 ID |
4797 | |
4798 | /* align to BTRFS_STRIPE_LEN */ | |
b8b93add | 4799 | stripe_size = div_u64(stripe_size, raid_stripe_len); |
53b381b3 | 4800 | stripe_size *= raid_stripe_len; |
b2117a39 MX |
4801 | |
4802 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); | |
4803 | if (!map) { | |
4804 | ret = -ENOMEM; | |
4805 | goto error; | |
4806 | } | |
4807 | map->num_stripes = num_stripes; | |
9b3f68b9 | 4808 | |
73c5de00 AJ |
4809 | for (i = 0; i < ndevs; ++i) { |
4810 | for (j = 0; j < dev_stripes; ++j) { | |
4811 | int s = i * dev_stripes + j; | |
4812 | map->stripes[s].dev = devices_info[i].dev; | |
4813 | map->stripes[s].physical = devices_info[i].dev_offset + | |
4814 | j * stripe_size; | |
6324fbf3 | 4815 | } |
6324fbf3 | 4816 | } |
da17066c | 4817 | map->sector_size = info->sectorsize; |
53b381b3 DW |
4818 | map->stripe_len = raid_stripe_len; |
4819 | map->io_align = raid_stripe_len; | |
4820 | map->io_width = raid_stripe_len; | |
2b82032c | 4821 | map->type = type; |
2b82032c | 4822 | map->sub_stripes = sub_stripes; |
0b86a832 | 4823 | |
53b381b3 | 4824 | num_bytes = stripe_size * data_stripes; |
0b86a832 | 4825 | |
6bccf3ab | 4826 | trace_btrfs_chunk_alloc(info, map, start, num_bytes); |
1abe9b8a | 4827 | |
172ddd60 | 4828 | em = alloc_extent_map(); |
2b82032c | 4829 | if (!em) { |
298a8f9c | 4830 | kfree(map); |
b2117a39 MX |
4831 | ret = -ENOMEM; |
4832 | goto error; | |
593060d7 | 4833 | } |
298a8f9c | 4834 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 4835 | em->map_lookup = map; |
2b82032c | 4836 | em->start = start; |
73c5de00 | 4837 | em->len = num_bytes; |
2b82032c YZ |
4838 | em->block_start = 0; |
4839 | em->block_len = em->len; | |
6df9a95e | 4840 | em->orig_block_len = stripe_size; |
593060d7 | 4841 | |
0b246afa | 4842 | em_tree = &info->mapping_tree.map_tree; |
890871be | 4843 | write_lock(&em_tree->lock); |
09a2a8f9 | 4844 | ret = add_extent_mapping(em_tree, em, 0); |
6df9a95e JB |
4845 | if (!ret) { |
4846 | list_add_tail(&em->list, &trans->transaction->pending_chunks); | |
4847 | atomic_inc(&em->refs); | |
4848 | } | |
890871be | 4849 | write_unlock(&em_tree->lock); |
0f5d42b2 JB |
4850 | if (ret) { |
4851 | free_extent_map(em); | |
1dd4602f | 4852 | goto error; |
0f5d42b2 | 4853 | } |
0b86a832 | 4854 | |
04487488 JB |
4855 | ret = btrfs_make_block_group(trans, extent_root, 0, type, |
4856 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4857 | start, num_bytes); | |
6df9a95e JB |
4858 | if (ret) |
4859 | goto error_del_extent; | |
2b82032c | 4860 | |
7cc8e58d MX |
4861 | for (i = 0; i < map->num_stripes; i++) { |
4862 | num_bytes = map->stripes[i].dev->bytes_used + stripe_size; | |
4863 | btrfs_device_set_bytes_used(map->stripes[i].dev, num_bytes); | |
4864 | } | |
43530c46 | 4865 | |
0b246afa JM |
4866 | spin_lock(&info->free_chunk_lock); |
4867 | info->free_chunk_space -= (stripe_size * map->num_stripes); | |
4868 | spin_unlock(&info->free_chunk_lock); | |
1c116187 | 4869 | |
0f5d42b2 | 4870 | free_extent_map(em); |
0b246afa | 4871 | check_raid56_incompat_flag(info, type); |
53b381b3 | 4872 | |
b2117a39 | 4873 | kfree(devices_info); |
2b82032c | 4874 | return 0; |
b2117a39 | 4875 | |
6df9a95e | 4876 | error_del_extent: |
0f5d42b2 JB |
4877 | write_lock(&em_tree->lock); |
4878 | remove_extent_mapping(em_tree, em); | |
4879 | write_unlock(&em_tree->lock); | |
4880 | ||
4881 | /* One for our allocation */ | |
4882 | free_extent_map(em); | |
4883 | /* One for the tree reference */ | |
4884 | free_extent_map(em); | |
495e64f4 FM |
4885 | /* One for the pending_chunks list reference */ |
4886 | free_extent_map(em); | |
b2117a39 | 4887 | error: |
b2117a39 MX |
4888 | kfree(devices_info); |
4889 | return ret; | |
2b82032c YZ |
4890 | } |
4891 | ||
6df9a95e | 4892 | int btrfs_finish_chunk_alloc(struct btrfs_trans_handle *trans, |
6bccf3ab | 4893 | struct btrfs_fs_info *fs_info, |
6df9a95e | 4894 | u64 chunk_offset, u64 chunk_size) |
2b82032c | 4895 | { |
6bccf3ab JM |
4896 | struct btrfs_root *extent_root = fs_info->extent_root; |
4897 | struct btrfs_root *chunk_root = fs_info->chunk_root; | |
2b82032c | 4898 | struct btrfs_key key; |
2b82032c YZ |
4899 | struct btrfs_device *device; |
4900 | struct btrfs_chunk *chunk; | |
4901 | struct btrfs_stripe *stripe; | |
6df9a95e JB |
4902 | struct extent_map_tree *em_tree; |
4903 | struct extent_map *em; | |
4904 | struct map_lookup *map; | |
4905 | size_t item_size; | |
4906 | u64 dev_offset; | |
4907 | u64 stripe_size; | |
4908 | int i = 0; | |
140e639f | 4909 | int ret = 0; |
2b82032c | 4910 | |
0b246afa | 4911 | em_tree = &fs_info->mapping_tree.map_tree; |
6df9a95e JB |
4912 | read_lock(&em_tree->lock); |
4913 | em = lookup_extent_mapping(em_tree, chunk_offset, chunk_size); | |
4914 | read_unlock(&em_tree->lock); | |
4915 | ||
4916 | if (!em) { | |
0b246afa | 4917 | btrfs_crit(fs_info, "unable to find logical %Lu len %Lu", |
5d163e0e | 4918 | chunk_offset, chunk_size); |
6df9a95e JB |
4919 | return -EINVAL; |
4920 | } | |
4921 | ||
4922 | if (em->start != chunk_offset || em->len != chunk_size) { | |
0b246afa | 4923 | btrfs_crit(fs_info, |
5d163e0e JM |
4924 | "found a bad mapping, wanted %Lu-%Lu, found %Lu-%Lu", |
4925 | chunk_offset, chunk_size, em->start, em->len); | |
6df9a95e JB |
4926 | free_extent_map(em); |
4927 | return -EINVAL; | |
4928 | } | |
4929 | ||
95617d69 | 4930 | map = em->map_lookup; |
6df9a95e JB |
4931 | item_size = btrfs_chunk_item_size(map->num_stripes); |
4932 | stripe_size = em->orig_block_len; | |
4933 | ||
2b82032c | 4934 | chunk = kzalloc(item_size, GFP_NOFS); |
6df9a95e JB |
4935 | if (!chunk) { |
4936 | ret = -ENOMEM; | |
4937 | goto out; | |
4938 | } | |
4939 | ||
50460e37 FM |
4940 | /* |
4941 | * Take the device list mutex to prevent races with the final phase of | |
4942 | * a device replace operation that replaces the device object associated | |
4943 | * with the map's stripes, because the device object's id can change | |
4944 | * at any time during that final phase of the device replace operation | |
4945 | * (dev-replace.c:btrfs_dev_replace_finishing()). | |
4946 | */ | |
0b246afa | 4947 | mutex_lock(&fs_info->fs_devices->device_list_mutex); |
6df9a95e JB |
4948 | for (i = 0; i < map->num_stripes; i++) { |
4949 | device = map->stripes[i].dev; | |
4950 | dev_offset = map->stripes[i].physical; | |
2b82032c | 4951 | |
0b86a832 | 4952 | ret = btrfs_update_device(trans, device); |
3acd3953 | 4953 | if (ret) |
50460e37 | 4954 | break; |
6df9a95e JB |
4955 | ret = btrfs_alloc_dev_extent(trans, device, |
4956 | chunk_root->root_key.objectid, | |
4957 | BTRFS_FIRST_CHUNK_TREE_OBJECTID, | |
4958 | chunk_offset, dev_offset, | |
4959 | stripe_size); | |
4960 | if (ret) | |
50460e37 FM |
4961 | break; |
4962 | } | |
4963 | if (ret) { | |
0b246afa | 4964 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
50460e37 | 4965 | goto out; |
2b82032c YZ |
4966 | } |
4967 | ||
2b82032c | 4968 | stripe = &chunk->stripe; |
6df9a95e JB |
4969 | for (i = 0; i < map->num_stripes; i++) { |
4970 | device = map->stripes[i].dev; | |
4971 | dev_offset = map->stripes[i].physical; | |
0b86a832 | 4972 | |
e17cade2 CM |
4973 | btrfs_set_stack_stripe_devid(stripe, device->devid); |
4974 | btrfs_set_stack_stripe_offset(stripe, dev_offset); | |
4975 | memcpy(stripe->dev_uuid, device->uuid, BTRFS_UUID_SIZE); | |
2b82032c | 4976 | stripe++; |
0b86a832 | 4977 | } |
0b246afa | 4978 | mutex_unlock(&fs_info->fs_devices->device_list_mutex); |
0b86a832 | 4979 | |
2b82032c | 4980 | btrfs_set_stack_chunk_length(chunk, chunk_size); |
0b86a832 | 4981 | btrfs_set_stack_chunk_owner(chunk, extent_root->root_key.objectid); |
2b82032c YZ |
4982 | btrfs_set_stack_chunk_stripe_len(chunk, map->stripe_len); |
4983 | btrfs_set_stack_chunk_type(chunk, map->type); | |
4984 | btrfs_set_stack_chunk_num_stripes(chunk, map->num_stripes); | |
4985 | btrfs_set_stack_chunk_io_align(chunk, map->stripe_len); | |
4986 | btrfs_set_stack_chunk_io_width(chunk, map->stripe_len); | |
0b246afa | 4987 | btrfs_set_stack_chunk_sector_size(chunk, fs_info->sectorsize); |
2b82032c | 4988 | btrfs_set_stack_chunk_sub_stripes(chunk, map->sub_stripes); |
0b86a832 | 4989 | |
2b82032c YZ |
4990 | key.objectid = BTRFS_FIRST_CHUNK_TREE_OBJECTID; |
4991 | key.type = BTRFS_CHUNK_ITEM_KEY; | |
4992 | key.offset = chunk_offset; | |
0b86a832 | 4993 | |
2b82032c | 4994 | ret = btrfs_insert_item(trans, chunk_root, &key, chunk, item_size); |
4ed1d16e MF |
4995 | if (ret == 0 && map->type & BTRFS_BLOCK_GROUP_SYSTEM) { |
4996 | /* | |
4997 | * TODO: Cleanup of inserted chunk root in case of | |
4998 | * failure. | |
4999 | */ | |
125ccb0a | 5000 | ret = btrfs_add_system_chunk(chunk_root, &key, chunk, |
2b82032c | 5001 | item_size); |
8f18cf13 | 5002 | } |
1abe9b8a | 5003 | |
6df9a95e | 5004 | out: |
0b86a832 | 5005 | kfree(chunk); |
6df9a95e | 5006 | free_extent_map(em); |
4ed1d16e | 5007 | return ret; |
2b82032c | 5008 | } |
0b86a832 | 5009 | |
2b82032c YZ |
5010 | /* |
5011 | * Chunk allocation falls into two parts. The first part does works | |
5012 | * that make the new allocated chunk useable, but not do any operation | |
5013 | * that modifies the chunk tree. The second part does the works that | |
5014 | * require modifying the chunk tree. This division is important for the | |
5015 | * bootstrap process of adding storage to a seed btrfs. | |
5016 | */ | |
5017 | int btrfs_alloc_chunk(struct btrfs_trans_handle *trans, | |
5018 | struct btrfs_root *extent_root, u64 type) | |
5019 | { | |
0b246afa | 5020 | struct btrfs_fs_info *fs_info = extent_root->fs_info; |
2b82032c | 5021 | u64 chunk_offset; |
2b82032c | 5022 | |
0b246afa JM |
5023 | ASSERT(mutex_is_locked(&fs_info->chunk_mutex)); |
5024 | chunk_offset = find_next_chunk(fs_info); | |
6df9a95e | 5025 | return __btrfs_alloc_chunk(trans, extent_root, chunk_offset, type); |
2b82032c YZ |
5026 | } |
5027 | ||
d397712b | 5028 | static noinline int init_first_rw_device(struct btrfs_trans_handle *trans, |
2b82032c YZ |
5029 | struct btrfs_root *root, |
5030 | struct btrfs_device *device) | |
5031 | { | |
5032 | u64 chunk_offset; | |
5033 | u64 sys_chunk_offset; | |
2b82032c | 5034 | u64 alloc_profile; |
2b82032c YZ |
5035 | struct btrfs_fs_info *fs_info = root->fs_info; |
5036 | struct btrfs_root *extent_root = fs_info->extent_root; | |
5037 | int ret; | |
5038 | ||
6df9a95e | 5039 | chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 5040 | alloc_profile = btrfs_get_alloc_profile(extent_root, 0); |
6df9a95e JB |
5041 | ret = __btrfs_alloc_chunk(trans, extent_root, chunk_offset, |
5042 | alloc_profile); | |
79787eaa JM |
5043 | if (ret) |
5044 | return ret; | |
2b82032c | 5045 | |
0b246afa | 5046 | sys_chunk_offset = find_next_chunk(fs_info); |
de98ced9 | 5047 | alloc_profile = btrfs_get_alloc_profile(fs_info->chunk_root, 0); |
6df9a95e JB |
5048 | ret = __btrfs_alloc_chunk(trans, extent_root, sys_chunk_offset, |
5049 | alloc_profile); | |
79787eaa | 5050 | return ret; |
2b82032c YZ |
5051 | } |
5052 | ||
d20983b4 MX |
5053 | static inline int btrfs_chunk_max_errors(struct map_lookup *map) |
5054 | { | |
5055 | int max_errors; | |
5056 | ||
5057 | if (map->type & (BTRFS_BLOCK_GROUP_RAID1 | | |
5058 | BTRFS_BLOCK_GROUP_RAID10 | | |
5059 | BTRFS_BLOCK_GROUP_RAID5 | | |
5060 | BTRFS_BLOCK_GROUP_DUP)) { | |
5061 | max_errors = 1; | |
5062 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID6) { | |
5063 | max_errors = 2; | |
5064 | } else { | |
5065 | max_errors = 0; | |
005d6427 | 5066 | } |
2b82032c | 5067 | |
d20983b4 | 5068 | return max_errors; |
2b82032c YZ |
5069 | } |
5070 | ||
5071 | int btrfs_chunk_readonly(struct btrfs_root *root, u64 chunk_offset) | |
5072 | { | |
0b246afa | 5073 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b82032c YZ |
5074 | struct extent_map *em; |
5075 | struct map_lookup *map; | |
0b246afa | 5076 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
2b82032c | 5077 | int readonly = 0; |
d20983b4 | 5078 | int miss_ndevs = 0; |
2b82032c YZ |
5079 | int i; |
5080 | ||
890871be | 5081 | read_lock(&map_tree->map_tree.lock); |
2b82032c | 5082 | em = lookup_extent_mapping(&map_tree->map_tree, chunk_offset, 1); |
890871be | 5083 | read_unlock(&map_tree->map_tree.lock); |
2b82032c YZ |
5084 | if (!em) |
5085 | return 1; | |
5086 | ||
95617d69 | 5087 | map = em->map_lookup; |
2b82032c | 5088 | for (i = 0; i < map->num_stripes; i++) { |
d20983b4 MX |
5089 | if (map->stripes[i].dev->missing) { |
5090 | miss_ndevs++; | |
5091 | continue; | |
5092 | } | |
5093 | ||
2b82032c YZ |
5094 | if (!map->stripes[i].dev->writeable) { |
5095 | readonly = 1; | |
d20983b4 | 5096 | goto end; |
2b82032c YZ |
5097 | } |
5098 | } | |
d20983b4 MX |
5099 | |
5100 | /* | |
5101 | * If the number of missing devices is larger than max errors, | |
5102 | * we can not write the data into that chunk successfully, so | |
5103 | * set it readonly. | |
5104 | */ | |
5105 | if (miss_ndevs > btrfs_chunk_max_errors(map)) | |
5106 | readonly = 1; | |
5107 | end: | |
0b86a832 | 5108 | free_extent_map(em); |
2b82032c | 5109 | return readonly; |
0b86a832 CM |
5110 | } |
5111 | ||
5112 | void btrfs_mapping_init(struct btrfs_mapping_tree *tree) | |
5113 | { | |
a8067e02 | 5114 | extent_map_tree_init(&tree->map_tree); |
0b86a832 CM |
5115 | } |
5116 | ||
5117 | void btrfs_mapping_tree_free(struct btrfs_mapping_tree *tree) | |
5118 | { | |
5119 | struct extent_map *em; | |
5120 | ||
d397712b | 5121 | while (1) { |
890871be | 5122 | write_lock(&tree->map_tree.lock); |
0b86a832 CM |
5123 | em = lookup_extent_mapping(&tree->map_tree, 0, (u64)-1); |
5124 | if (em) | |
5125 | remove_extent_mapping(&tree->map_tree, em); | |
890871be | 5126 | write_unlock(&tree->map_tree.lock); |
0b86a832 CM |
5127 | if (!em) |
5128 | break; | |
0b86a832 CM |
5129 | /* once for us */ |
5130 | free_extent_map(em); | |
5131 | /* once for the tree */ | |
5132 | free_extent_map(em); | |
5133 | } | |
5134 | } | |
5135 | ||
5d964051 | 5136 | int btrfs_num_copies(struct btrfs_fs_info *fs_info, u64 logical, u64 len) |
f188591e | 5137 | { |
5d964051 | 5138 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
f188591e CM |
5139 | struct extent_map *em; |
5140 | struct map_lookup *map; | |
5141 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5142 | int ret; | |
5143 | ||
890871be | 5144 | read_lock(&em_tree->lock); |
f188591e | 5145 | em = lookup_extent_mapping(em_tree, logical, len); |
890871be | 5146 | read_unlock(&em_tree->lock); |
f188591e | 5147 | |
fb7669b5 JB |
5148 | /* |
5149 | * We could return errors for these cases, but that could get ugly and | |
5150 | * we'd probably do the same thing which is just not do anything else | |
5151 | * and exit, so return 1 so the callers don't try to use other copies. | |
5152 | */ | |
5153 | if (!em) { | |
351fd353 | 5154 | btrfs_crit(fs_info, "No mapping for %Lu-%Lu", logical, |
fb7669b5 JB |
5155 | logical+len); |
5156 | return 1; | |
5157 | } | |
5158 | ||
5159 | if (em->start > logical || em->start + em->len < logical) { | |
5d163e0e JM |
5160 | btrfs_crit(fs_info, "Invalid mapping for %Lu-%Lu, got %Lu-%Lu", |
5161 | logical, logical+len, em->start, | |
5162 | em->start + em->len); | |
7d3d1744 | 5163 | free_extent_map(em); |
fb7669b5 JB |
5164 | return 1; |
5165 | } | |
5166 | ||
95617d69 | 5167 | map = em->map_lookup; |
f188591e CM |
5168 | if (map->type & (BTRFS_BLOCK_GROUP_DUP | BTRFS_BLOCK_GROUP_RAID1)) |
5169 | ret = map->num_stripes; | |
321aecc6 CM |
5170 | else if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
5171 | ret = map->sub_stripes; | |
53b381b3 DW |
5172 | else if (map->type & BTRFS_BLOCK_GROUP_RAID5) |
5173 | ret = 2; | |
5174 | else if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5175 | ret = 3; | |
f188591e CM |
5176 | else |
5177 | ret = 1; | |
5178 | free_extent_map(em); | |
ad6d620e | 5179 | |
73beece9 | 5180 | btrfs_dev_replace_lock(&fs_info->dev_replace, 0); |
ad6d620e SB |
5181 | if (btrfs_dev_replace_is_ongoing(&fs_info->dev_replace)) |
5182 | ret++; | |
73beece9 | 5183 | btrfs_dev_replace_unlock(&fs_info->dev_replace, 0); |
ad6d620e | 5184 | |
f188591e CM |
5185 | return ret; |
5186 | } | |
5187 | ||
53b381b3 DW |
5188 | unsigned long btrfs_full_stripe_len(struct btrfs_root *root, |
5189 | struct btrfs_mapping_tree *map_tree, | |
5190 | u64 logical) | |
5191 | { | |
0b246afa | 5192 | struct btrfs_fs_info *fs_info = root->fs_info; |
53b381b3 DW |
5193 | struct extent_map *em; |
5194 | struct map_lookup *map; | |
5195 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
0b246afa | 5196 | unsigned long len = fs_info->sectorsize; |
53b381b3 DW |
5197 | |
5198 | read_lock(&em_tree->lock); | |
5199 | em = lookup_extent_mapping(em_tree, logical, len); | |
5200 | read_unlock(&em_tree->lock); | |
5201 | BUG_ON(!em); | |
5202 | ||
5203 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5204 | map = em->map_lookup; |
ffe2d203 | 5205 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 | 5206 | len = map->stripe_len * nr_data_stripes(map); |
53b381b3 DW |
5207 | free_extent_map(em); |
5208 | return len; | |
5209 | } | |
5210 | ||
5211 | int btrfs_is_parity_mirror(struct btrfs_mapping_tree *map_tree, | |
5212 | u64 logical, u64 len, int mirror_num) | |
5213 | { | |
5214 | struct extent_map *em; | |
5215 | struct map_lookup *map; | |
5216 | struct extent_map_tree *em_tree = &map_tree->map_tree; | |
5217 | int ret = 0; | |
5218 | ||
5219 | read_lock(&em_tree->lock); | |
5220 | em = lookup_extent_mapping(em_tree, logical, len); | |
5221 | read_unlock(&em_tree->lock); | |
5222 | BUG_ON(!em); | |
5223 | ||
5224 | BUG_ON(em->start > logical || em->start + em->len < logical); | |
95617d69 | 5225 | map = em->map_lookup; |
ffe2d203 | 5226 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) |
53b381b3 DW |
5227 | ret = 1; |
5228 | free_extent_map(em); | |
5229 | return ret; | |
5230 | } | |
5231 | ||
30d9861f SB |
5232 | static int find_live_mirror(struct btrfs_fs_info *fs_info, |
5233 | struct map_lookup *map, int first, int num, | |
5234 | int optimal, int dev_replace_is_ongoing) | |
dfe25020 CM |
5235 | { |
5236 | int i; | |
30d9861f SB |
5237 | int tolerance; |
5238 | struct btrfs_device *srcdev; | |
5239 | ||
5240 | if (dev_replace_is_ongoing && | |
5241 | fs_info->dev_replace.cont_reading_from_srcdev_mode == | |
5242 | BTRFS_DEV_REPLACE_ITEM_CONT_READING_FROM_SRCDEV_MODE_AVOID) | |
5243 | srcdev = fs_info->dev_replace.srcdev; | |
5244 | else | |
5245 | srcdev = NULL; | |
5246 | ||
5247 | /* | |
5248 | * try to avoid the drive that is the source drive for a | |
5249 | * dev-replace procedure, only choose it if no other non-missing | |
5250 | * mirror is available | |
5251 | */ | |
5252 | for (tolerance = 0; tolerance < 2; tolerance++) { | |
5253 | if (map->stripes[optimal].dev->bdev && | |
5254 | (tolerance || map->stripes[optimal].dev != srcdev)) | |
5255 | return optimal; | |
5256 | for (i = first; i < first + num; i++) { | |
5257 | if (map->stripes[i].dev->bdev && | |
5258 | (tolerance || map->stripes[i].dev != srcdev)) | |
5259 | return i; | |
5260 | } | |
dfe25020 | 5261 | } |
30d9861f | 5262 | |
dfe25020 CM |
5263 | /* we couldn't find one that doesn't fail. Just return something |
5264 | * and the io error handling code will clean up eventually | |
5265 | */ | |
5266 | return optimal; | |
5267 | } | |
5268 | ||
53b381b3 DW |
5269 | static inline int parity_smaller(u64 a, u64 b) |
5270 | { | |
5271 | return a > b; | |
5272 | } | |
5273 | ||
5274 | /* Bubble-sort the stripe set to put the parity/syndrome stripes last */ | |
8e5cfb55 | 5275 | static void sort_parity_stripes(struct btrfs_bio *bbio, int num_stripes) |
53b381b3 DW |
5276 | { |
5277 | struct btrfs_bio_stripe s; | |
5278 | int i; | |
5279 | u64 l; | |
5280 | int again = 1; | |
5281 | ||
5282 | while (again) { | |
5283 | again = 0; | |
cc7539ed | 5284 | for (i = 0; i < num_stripes - 1; i++) { |
8e5cfb55 ZL |
5285 | if (parity_smaller(bbio->raid_map[i], |
5286 | bbio->raid_map[i+1])) { | |
53b381b3 | 5287 | s = bbio->stripes[i]; |
8e5cfb55 | 5288 | l = bbio->raid_map[i]; |
53b381b3 | 5289 | bbio->stripes[i] = bbio->stripes[i+1]; |
8e5cfb55 | 5290 | bbio->raid_map[i] = bbio->raid_map[i+1]; |
53b381b3 | 5291 | bbio->stripes[i+1] = s; |
8e5cfb55 | 5292 | bbio->raid_map[i+1] = l; |
2c8cdd6e | 5293 | |
53b381b3 DW |
5294 | again = 1; |
5295 | } | |
5296 | } | |
5297 | } | |
5298 | } | |
5299 | ||
6e9606d2 ZL |
5300 | static struct btrfs_bio *alloc_btrfs_bio(int total_stripes, int real_stripes) |
5301 | { | |
5302 | struct btrfs_bio *bbio = kzalloc( | |
e57cf21e | 5303 | /* the size of the btrfs_bio */ |
6e9606d2 | 5304 | sizeof(struct btrfs_bio) + |
e57cf21e | 5305 | /* plus the variable array for the stripes */ |
6e9606d2 | 5306 | sizeof(struct btrfs_bio_stripe) * (total_stripes) + |
e57cf21e | 5307 | /* plus the variable array for the tgt dev */ |
6e9606d2 | 5308 | sizeof(int) * (real_stripes) + |
e57cf21e CM |
5309 | /* |
5310 | * plus the raid_map, which includes both the tgt dev | |
5311 | * and the stripes | |
5312 | */ | |
5313 | sizeof(u64) * (total_stripes), | |
277fb5fc | 5314 | GFP_NOFS|__GFP_NOFAIL); |
6e9606d2 ZL |
5315 | |
5316 | atomic_set(&bbio->error, 0); | |
5317 | atomic_set(&bbio->refs, 1); | |
5318 | ||
5319 | return bbio; | |
5320 | } | |
5321 | ||
5322 | void btrfs_get_bbio(struct btrfs_bio *bbio) | |
5323 | { | |
5324 | WARN_ON(!atomic_read(&bbio->refs)); | |
5325 | atomic_inc(&bbio->refs); | |
5326 | } | |
5327 | ||
5328 | void btrfs_put_bbio(struct btrfs_bio *bbio) | |
5329 | { | |
5330 | if (!bbio) | |
5331 | return; | |
5332 | if (atomic_dec_and_test(&bbio->refs)) | |
5333 | kfree(bbio); | |
5334 | } | |
5335 | ||
cf8cddd3 CH |
5336 | static int __btrfs_map_block(struct btrfs_fs_info *fs_info, |
5337 | enum btrfs_map_op op, | |
f2d8d74d | 5338 | u64 logical, u64 *length, |
a1d3c478 | 5339 | struct btrfs_bio **bbio_ret, |
8e5cfb55 | 5340 | int mirror_num, int need_raid_map) |
0b86a832 CM |
5341 | { |
5342 | struct extent_map *em; | |
5343 | struct map_lookup *map; | |
3ec706c8 | 5344 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
0b86a832 CM |
5345 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5346 | u64 offset; | |
593060d7 | 5347 | u64 stripe_offset; |
fce3bb9a | 5348 | u64 stripe_end_offset; |
593060d7 | 5349 | u64 stripe_nr; |
fce3bb9a LD |
5350 | u64 stripe_nr_orig; |
5351 | u64 stripe_nr_end; | |
53b381b3 | 5352 | u64 stripe_len; |
9d644a62 | 5353 | u32 stripe_index; |
cea9e445 | 5354 | int i; |
de11cc12 | 5355 | int ret = 0; |
f2d8d74d | 5356 | int num_stripes; |
a236aed1 | 5357 | int max_errors = 0; |
2c8cdd6e | 5358 | int tgtdev_indexes = 0; |
a1d3c478 | 5359 | struct btrfs_bio *bbio = NULL; |
472262f3 SB |
5360 | struct btrfs_dev_replace *dev_replace = &fs_info->dev_replace; |
5361 | int dev_replace_is_ongoing = 0; | |
5362 | int num_alloc_stripes; | |
ad6d620e SB |
5363 | int patch_the_first_stripe_for_dev_replace = 0; |
5364 | u64 physical_to_patch_in_first_stripe = 0; | |
53b381b3 | 5365 | u64 raid56_full_stripe_start = (u64)-1; |
0b86a832 | 5366 | |
890871be | 5367 | read_lock(&em_tree->lock); |
0b86a832 | 5368 | em = lookup_extent_mapping(em_tree, logical, *length); |
890871be | 5369 | read_unlock(&em_tree->lock); |
f2d8d74d | 5370 | |
3b951516 | 5371 | if (!em) { |
c2cf52eb | 5372 | btrfs_crit(fs_info, "unable to find logical %llu len %llu", |
c1c9ff7c | 5373 | logical, *length); |
9bb91873 JB |
5374 | return -EINVAL; |
5375 | } | |
5376 | ||
5377 | if (em->start > logical || em->start + em->len < logical) { | |
5d163e0e JM |
5378 | btrfs_crit(fs_info, |
5379 | "found a bad mapping, wanted %Lu, found %Lu-%Lu", | |
5380 | logical, em->start, em->start + em->len); | |
7d3d1744 | 5381 | free_extent_map(em); |
9bb91873 | 5382 | return -EINVAL; |
3b951516 | 5383 | } |
0b86a832 | 5384 | |
95617d69 | 5385 | map = em->map_lookup; |
0b86a832 | 5386 | offset = logical - em->start; |
593060d7 | 5387 | |
53b381b3 | 5388 | stripe_len = map->stripe_len; |
593060d7 CM |
5389 | stripe_nr = offset; |
5390 | /* | |
5391 | * stripe_nr counts the total number of stripes we have to stride | |
5392 | * to get to this block | |
5393 | */ | |
47c5713f | 5394 | stripe_nr = div64_u64(stripe_nr, stripe_len); |
593060d7 | 5395 | |
53b381b3 | 5396 | stripe_offset = stripe_nr * stripe_len; |
e042d1ec | 5397 | if (offset < stripe_offset) { |
5d163e0e JM |
5398 | btrfs_crit(fs_info, |
5399 | "stripe math has gone wrong, stripe_offset=%llu, offset=%llu, start=%llu, logical=%llu, stripe_len=%llu", | |
e042d1ec JB |
5400 | stripe_offset, offset, em->start, logical, |
5401 | stripe_len); | |
5402 | free_extent_map(em); | |
5403 | return -EINVAL; | |
5404 | } | |
593060d7 CM |
5405 | |
5406 | /* stripe_offset is the offset of this block in its stripe*/ | |
5407 | stripe_offset = offset - stripe_offset; | |
5408 | ||
53b381b3 | 5409 | /* if we're here for raid56, we need to know the stripe aligned start */ |
ffe2d203 | 5410 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5411 | unsigned long full_stripe_len = stripe_len * nr_data_stripes(map); |
5412 | raid56_full_stripe_start = offset; | |
5413 | ||
5414 | /* allow a write of a full stripe, but make sure we don't | |
5415 | * allow straddling of stripes | |
5416 | */ | |
47c5713f DS |
5417 | raid56_full_stripe_start = div64_u64(raid56_full_stripe_start, |
5418 | full_stripe_len); | |
53b381b3 DW |
5419 | raid56_full_stripe_start *= full_stripe_len; |
5420 | } | |
5421 | ||
cf8cddd3 | 5422 | if (op == BTRFS_MAP_DISCARD) { |
53b381b3 | 5423 | /* we don't discard raid56 yet */ |
ffe2d203 | 5424 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
53b381b3 DW |
5425 | ret = -EOPNOTSUPP; |
5426 | goto out; | |
5427 | } | |
fce3bb9a | 5428 | *length = min_t(u64, em->len - offset, *length); |
53b381b3 DW |
5429 | } else if (map->type & BTRFS_BLOCK_GROUP_PROFILE_MASK) { |
5430 | u64 max_len; | |
5431 | /* For writes to RAID[56], allow a full stripeset across all disks. | |
5432 | For other RAID types and for RAID[56] reads, just allow a single | |
5433 | stripe (on a single disk). */ | |
ffe2d203 | 5434 | if ((map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
cf8cddd3 | 5435 | (op == BTRFS_MAP_WRITE)) { |
53b381b3 DW |
5436 | max_len = stripe_len * nr_data_stripes(map) - |
5437 | (offset - raid56_full_stripe_start); | |
5438 | } else { | |
5439 | /* we limit the length of each bio to what fits in a stripe */ | |
5440 | max_len = stripe_len - stripe_offset; | |
5441 | } | |
5442 | *length = min_t(u64, em->len - offset, max_len); | |
cea9e445 CM |
5443 | } else { |
5444 | *length = em->len - offset; | |
5445 | } | |
f2d8d74d | 5446 | |
53b381b3 DW |
5447 | /* This is for when we're called from btrfs_merge_bio_hook() and all |
5448 | it cares about is the length */ | |
a1d3c478 | 5449 | if (!bbio_ret) |
cea9e445 CM |
5450 | goto out; |
5451 | ||
73beece9 | 5452 | btrfs_dev_replace_lock(dev_replace, 0); |
472262f3 SB |
5453 | dev_replace_is_ongoing = btrfs_dev_replace_is_ongoing(dev_replace); |
5454 | if (!dev_replace_is_ongoing) | |
73beece9 LB |
5455 | btrfs_dev_replace_unlock(dev_replace, 0); |
5456 | else | |
5457 | btrfs_dev_replace_set_lock_blocking(dev_replace); | |
472262f3 | 5458 | |
ad6d620e | 5459 | if (dev_replace_is_ongoing && mirror_num == map->num_stripes + 1 && |
cf8cddd3 CH |
5460 | op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD && |
5461 | op != BTRFS_MAP_GET_READ_MIRRORS && dev_replace->tgtdev != NULL) { | |
ad6d620e SB |
5462 | /* |
5463 | * in dev-replace case, for repair case (that's the only | |
5464 | * case where the mirror is selected explicitly when | |
5465 | * calling btrfs_map_block), blocks left of the left cursor | |
5466 | * can also be read from the target drive. | |
5467 | * For REQ_GET_READ_MIRRORS, the target drive is added as | |
5468 | * the last one to the array of stripes. For READ, it also | |
5469 | * needs to be supported using the same mirror number. | |
5470 | * If the requested block is not left of the left cursor, | |
5471 | * EIO is returned. This can happen because btrfs_num_copies() | |
5472 | * returns one more in the dev-replace case. | |
5473 | */ | |
5474 | u64 tmp_length = *length; | |
5475 | struct btrfs_bio *tmp_bbio = NULL; | |
5476 | int tmp_num_stripes; | |
5477 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5478 | int index_srcdev = 0; | |
5479 | int found = 0; | |
5480 | u64 physical_of_found = 0; | |
5481 | ||
cf8cddd3 | 5482 | ret = __btrfs_map_block(fs_info, BTRFS_MAP_GET_READ_MIRRORS, |
8e5cfb55 | 5483 | logical, &tmp_length, &tmp_bbio, 0, 0); |
ad6d620e SB |
5484 | if (ret) { |
5485 | WARN_ON(tmp_bbio != NULL); | |
5486 | goto out; | |
5487 | } | |
5488 | ||
5489 | tmp_num_stripes = tmp_bbio->num_stripes; | |
5490 | if (mirror_num > tmp_num_stripes) { | |
5491 | /* | |
cf8cddd3 | 5492 | * BTRFS_MAP_GET_READ_MIRRORS does not contain this |
ad6d620e SB |
5493 | * mirror, that means that the requested area |
5494 | * is not left of the left cursor | |
5495 | */ | |
5496 | ret = -EIO; | |
6e9606d2 | 5497 | btrfs_put_bbio(tmp_bbio); |
ad6d620e SB |
5498 | goto out; |
5499 | } | |
5500 | ||
5501 | /* | |
5502 | * process the rest of the function using the mirror_num | |
5503 | * of the source drive. Therefore look it up first. | |
5504 | * At the end, patch the device pointer to the one of the | |
5505 | * target drive. | |
5506 | */ | |
5507 | for (i = 0; i < tmp_num_stripes; i++) { | |
94a97dfe ZL |
5508 | if (tmp_bbio->stripes[i].dev->devid != srcdev_devid) |
5509 | continue; | |
5510 | ||
5511 | /* | |
5512 | * In case of DUP, in order to keep it simple, only add | |
5513 | * the mirror with the lowest physical address | |
5514 | */ | |
5515 | if (found && | |
5516 | physical_of_found <= tmp_bbio->stripes[i].physical) | |
5517 | continue; | |
5518 | ||
5519 | index_srcdev = i; | |
5520 | found = 1; | |
5521 | physical_of_found = tmp_bbio->stripes[i].physical; | |
ad6d620e SB |
5522 | } |
5523 | ||
94a97dfe ZL |
5524 | btrfs_put_bbio(tmp_bbio); |
5525 | ||
5526 | if (!found) { | |
ad6d620e SB |
5527 | WARN_ON(1); |
5528 | ret = -EIO; | |
ad6d620e SB |
5529 | goto out; |
5530 | } | |
5531 | ||
94a97dfe ZL |
5532 | mirror_num = index_srcdev + 1; |
5533 | patch_the_first_stripe_for_dev_replace = 1; | |
5534 | physical_to_patch_in_first_stripe = physical_of_found; | |
ad6d620e SB |
5535 | } else if (mirror_num > map->num_stripes) { |
5536 | mirror_num = 0; | |
5537 | } | |
5538 | ||
f2d8d74d | 5539 | num_stripes = 1; |
cea9e445 | 5540 | stripe_index = 0; |
fce3bb9a | 5541 | stripe_nr_orig = stripe_nr; |
fda2832f | 5542 | stripe_nr_end = ALIGN(offset + *length, map->stripe_len); |
b8b93add | 5543 | stripe_nr_end = div_u64(stripe_nr_end, map->stripe_len); |
fce3bb9a LD |
5544 | stripe_end_offset = stripe_nr_end * map->stripe_len - |
5545 | (offset + *length); | |
53b381b3 | 5546 | |
fce3bb9a | 5547 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
cf8cddd3 | 5548 | if (op == BTRFS_MAP_DISCARD) |
fce3bb9a LD |
5549 | num_stripes = min_t(u64, map->num_stripes, |
5550 | stripe_nr_end - stripe_nr_orig); | |
47c5713f DS |
5551 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5552 | &stripe_index); | |
cf8cddd3 CH |
5553 | if (op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD && |
5554 | op != BTRFS_MAP_GET_READ_MIRRORS) | |
28e1cc7d | 5555 | mirror_num = 1; |
fce3bb9a | 5556 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID1) { |
cf8cddd3 CH |
5557 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD || |
5558 | op == BTRFS_MAP_GET_READ_MIRRORS) | |
f2d8d74d | 5559 | num_stripes = map->num_stripes; |
2fff734f | 5560 | else if (mirror_num) |
f188591e | 5561 | stripe_index = mirror_num - 1; |
dfe25020 | 5562 | else { |
30d9861f | 5563 | stripe_index = find_live_mirror(fs_info, map, 0, |
dfe25020 | 5564 | map->num_stripes, |
30d9861f SB |
5565 | current->pid % map->num_stripes, |
5566 | dev_replace_is_ongoing); | |
a1d3c478 | 5567 | mirror_num = stripe_index + 1; |
dfe25020 | 5568 | } |
2fff734f | 5569 | |
611f0e00 | 5570 | } else if (map->type & BTRFS_BLOCK_GROUP_DUP) { |
cf8cddd3 CH |
5571 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD || |
5572 | op == BTRFS_MAP_GET_READ_MIRRORS) { | |
f2d8d74d | 5573 | num_stripes = map->num_stripes; |
a1d3c478 | 5574 | } else if (mirror_num) { |
f188591e | 5575 | stripe_index = mirror_num - 1; |
a1d3c478 JS |
5576 | } else { |
5577 | mirror_num = 1; | |
5578 | } | |
2fff734f | 5579 | |
321aecc6 | 5580 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID10) { |
9d644a62 | 5581 | u32 factor = map->num_stripes / map->sub_stripes; |
321aecc6 | 5582 | |
47c5713f | 5583 | stripe_nr = div_u64_rem(stripe_nr, factor, &stripe_index); |
321aecc6 CM |
5584 | stripe_index *= map->sub_stripes; |
5585 | ||
cf8cddd3 | 5586 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
f2d8d74d | 5587 | num_stripes = map->sub_stripes; |
cf8cddd3 | 5588 | else if (op == BTRFS_MAP_DISCARD) |
fce3bb9a LD |
5589 | num_stripes = min_t(u64, map->sub_stripes * |
5590 | (stripe_nr_end - stripe_nr_orig), | |
5591 | map->num_stripes); | |
321aecc6 CM |
5592 | else if (mirror_num) |
5593 | stripe_index += mirror_num - 1; | |
dfe25020 | 5594 | else { |
3e74317a | 5595 | int old_stripe_index = stripe_index; |
30d9861f SB |
5596 | stripe_index = find_live_mirror(fs_info, map, |
5597 | stripe_index, | |
dfe25020 | 5598 | map->sub_stripes, stripe_index + |
30d9861f SB |
5599 | current->pid % map->sub_stripes, |
5600 | dev_replace_is_ongoing); | |
3e74317a | 5601 | mirror_num = stripe_index - old_stripe_index + 1; |
dfe25020 | 5602 | } |
53b381b3 | 5603 | |
ffe2d203 | 5604 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
8e5cfb55 | 5605 | if (need_raid_map && |
cf8cddd3 | 5606 | (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS || |
af8e2d1d | 5607 | mirror_num > 1)) { |
53b381b3 | 5608 | /* push stripe_nr back to the start of the full stripe */ |
b8b93add DS |
5609 | stripe_nr = div_u64(raid56_full_stripe_start, |
5610 | stripe_len * nr_data_stripes(map)); | |
53b381b3 DW |
5611 | |
5612 | /* RAID[56] write or recovery. Return all stripes */ | |
5613 | num_stripes = map->num_stripes; | |
5614 | max_errors = nr_parity_stripes(map); | |
5615 | ||
53b381b3 DW |
5616 | *length = map->stripe_len; |
5617 | stripe_index = 0; | |
5618 | stripe_offset = 0; | |
5619 | } else { | |
5620 | /* | |
5621 | * Mirror #0 or #1 means the original data block. | |
5622 | * Mirror #2 is RAID5 parity block. | |
5623 | * Mirror #3 is RAID6 Q block. | |
5624 | */ | |
47c5713f DS |
5625 | stripe_nr = div_u64_rem(stripe_nr, |
5626 | nr_data_stripes(map), &stripe_index); | |
53b381b3 DW |
5627 | if (mirror_num > 1) |
5628 | stripe_index = nr_data_stripes(map) + | |
5629 | mirror_num - 2; | |
5630 | ||
5631 | /* We distribute the parity blocks across stripes */ | |
47c5713f DS |
5632 | div_u64_rem(stripe_nr + stripe_index, map->num_stripes, |
5633 | &stripe_index); | |
cf8cddd3 CH |
5634 | if ((op != BTRFS_MAP_WRITE && op != BTRFS_MAP_DISCARD && |
5635 | op != BTRFS_MAP_GET_READ_MIRRORS) && mirror_num <= 1) | |
28e1cc7d | 5636 | mirror_num = 1; |
53b381b3 | 5637 | } |
8790d502 CM |
5638 | } else { |
5639 | /* | |
47c5713f DS |
5640 | * after this, stripe_nr is the number of stripes on this |
5641 | * device we have to walk to find the data, and stripe_index is | |
5642 | * the number of our device in the stripe array | |
8790d502 | 5643 | */ |
47c5713f DS |
5644 | stripe_nr = div_u64_rem(stripe_nr, map->num_stripes, |
5645 | &stripe_index); | |
a1d3c478 | 5646 | mirror_num = stripe_index + 1; |
8790d502 | 5647 | } |
e042d1ec | 5648 | if (stripe_index >= map->num_stripes) { |
5d163e0e JM |
5649 | btrfs_crit(fs_info, |
5650 | "stripe index math went horribly wrong, got stripe_index=%u, num_stripes=%u", | |
e042d1ec JB |
5651 | stripe_index, map->num_stripes); |
5652 | ret = -EINVAL; | |
5653 | goto out; | |
5654 | } | |
cea9e445 | 5655 | |
472262f3 | 5656 | num_alloc_stripes = num_stripes; |
ad6d620e | 5657 | if (dev_replace_is_ongoing) { |
cf8cddd3 | 5658 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD) |
ad6d620e | 5659 | num_alloc_stripes <<= 1; |
cf8cddd3 | 5660 | if (op == BTRFS_MAP_GET_READ_MIRRORS) |
ad6d620e | 5661 | num_alloc_stripes++; |
2c8cdd6e | 5662 | tgtdev_indexes = num_stripes; |
ad6d620e | 5663 | } |
2c8cdd6e | 5664 | |
6e9606d2 | 5665 | bbio = alloc_btrfs_bio(num_alloc_stripes, tgtdev_indexes); |
de11cc12 LZ |
5666 | if (!bbio) { |
5667 | ret = -ENOMEM; | |
5668 | goto out; | |
5669 | } | |
2c8cdd6e MX |
5670 | if (dev_replace_is_ongoing) |
5671 | bbio->tgtdev_map = (int *)(bbio->stripes + num_alloc_stripes); | |
de11cc12 | 5672 | |
8e5cfb55 | 5673 | /* build raid_map */ |
ffe2d203 | 5674 | if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK && |
b3d3fa51 | 5675 | need_raid_map && |
cf8cddd3 | 5676 | ((op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) || |
8e5cfb55 ZL |
5677 | mirror_num > 1)) { |
5678 | u64 tmp; | |
9d644a62 | 5679 | unsigned rot; |
8e5cfb55 ZL |
5680 | |
5681 | bbio->raid_map = (u64 *)((void *)bbio->stripes + | |
5682 | sizeof(struct btrfs_bio_stripe) * | |
5683 | num_alloc_stripes + | |
5684 | sizeof(int) * tgtdev_indexes); | |
5685 | ||
5686 | /* Work out the disk rotation on this stripe-set */ | |
47c5713f | 5687 | div_u64_rem(stripe_nr, num_stripes, &rot); |
8e5cfb55 ZL |
5688 | |
5689 | /* Fill in the logical address of each stripe */ | |
5690 | tmp = stripe_nr * nr_data_stripes(map); | |
5691 | for (i = 0; i < nr_data_stripes(map); i++) | |
5692 | bbio->raid_map[(i+rot) % num_stripes] = | |
5693 | em->start + (tmp + i) * map->stripe_len; | |
5694 | ||
5695 | bbio->raid_map[(i+rot) % map->num_stripes] = RAID5_P_STRIPE; | |
5696 | if (map->type & BTRFS_BLOCK_GROUP_RAID6) | |
5697 | bbio->raid_map[(i+rot+1) % num_stripes] = | |
5698 | RAID6_Q_STRIPE; | |
5699 | } | |
5700 | ||
cf8cddd3 | 5701 | if (op == BTRFS_MAP_DISCARD) { |
9d644a62 DS |
5702 | u32 factor = 0; |
5703 | u32 sub_stripes = 0; | |
ec9ef7a1 LZ |
5704 | u64 stripes_per_dev = 0; |
5705 | u32 remaining_stripes = 0; | |
b89203f7 | 5706 | u32 last_stripe = 0; |
ec9ef7a1 LZ |
5707 | |
5708 | if (map->type & | |
5709 | (BTRFS_BLOCK_GROUP_RAID0 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5710 | if (map->type & BTRFS_BLOCK_GROUP_RAID0) | |
5711 | sub_stripes = 1; | |
5712 | else | |
5713 | sub_stripes = map->sub_stripes; | |
5714 | ||
5715 | factor = map->num_stripes / sub_stripes; | |
5716 | stripes_per_dev = div_u64_rem(stripe_nr_end - | |
5717 | stripe_nr_orig, | |
5718 | factor, | |
5719 | &remaining_stripes); | |
b89203f7 LB |
5720 | div_u64_rem(stripe_nr_end - 1, factor, &last_stripe); |
5721 | last_stripe *= sub_stripes; | |
ec9ef7a1 LZ |
5722 | } |
5723 | ||
fce3bb9a | 5724 | for (i = 0; i < num_stripes; i++) { |
a1d3c478 | 5725 | bbio->stripes[i].physical = |
f2d8d74d CM |
5726 | map->stripes[stripe_index].physical + |
5727 | stripe_offset + stripe_nr * map->stripe_len; | |
a1d3c478 | 5728 | bbio->stripes[i].dev = map->stripes[stripe_index].dev; |
fce3bb9a | 5729 | |
ec9ef7a1 LZ |
5730 | if (map->type & (BTRFS_BLOCK_GROUP_RAID0 | |
5731 | BTRFS_BLOCK_GROUP_RAID10)) { | |
5732 | bbio->stripes[i].length = stripes_per_dev * | |
5733 | map->stripe_len; | |
b89203f7 | 5734 | |
ec9ef7a1 LZ |
5735 | if (i / sub_stripes < remaining_stripes) |
5736 | bbio->stripes[i].length += | |
5737 | map->stripe_len; | |
b89203f7 LB |
5738 | |
5739 | /* | |
5740 | * Special for the first stripe and | |
5741 | * the last stripe: | |
5742 | * | |
5743 | * |-------|...|-------| | |
5744 | * |----------| | |
5745 | * off end_off | |
5746 | */ | |
ec9ef7a1 | 5747 | if (i < sub_stripes) |
a1d3c478 | 5748 | bbio->stripes[i].length -= |
fce3bb9a | 5749 | stripe_offset; |
b89203f7 LB |
5750 | |
5751 | if (stripe_index >= last_stripe && | |
5752 | stripe_index <= (last_stripe + | |
5753 | sub_stripes - 1)) | |
a1d3c478 | 5754 | bbio->stripes[i].length -= |
fce3bb9a | 5755 | stripe_end_offset; |
b89203f7 | 5756 | |
ec9ef7a1 LZ |
5757 | if (i == sub_stripes - 1) |
5758 | stripe_offset = 0; | |
fce3bb9a | 5759 | } else |
a1d3c478 | 5760 | bbio->stripes[i].length = *length; |
fce3bb9a LD |
5761 | |
5762 | stripe_index++; | |
5763 | if (stripe_index == map->num_stripes) { | |
5764 | /* This could only happen for RAID0/10 */ | |
5765 | stripe_index = 0; | |
5766 | stripe_nr++; | |
5767 | } | |
5768 | } | |
5769 | } else { | |
5770 | for (i = 0; i < num_stripes; i++) { | |
a1d3c478 | 5771 | bbio->stripes[i].physical = |
212a17ab LT |
5772 | map->stripes[stripe_index].physical + |
5773 | stripe_offset + | |
5774 | stripe_nr * map->stripe_len; | |
a1d3c478 | 5775 | bbio->stripes[i].dev = |
212a17ab | 5776 | map->stripes[stripe_index].dev; |
fce3bb9a | 5777 | stripe_index++; |
f2d8d74d | 5778 | } |
593060d7 | 5779 | } |
de11cc12 | 5780 | |
cf8cddd3 | 5781 | if (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_GET_READ_MIRRORS) |
d20983b4 | 5782 | max_errors = btrfs_chunk_max_errors(map); |
de11cc12 | 5783 | |
8e5cfb55 ZL |
5784 | if (bbio->raid_map) |
5785 | sort_parity_stripes(bbio, num_stripes); | |
cc7539ed | 5786 | |
2c8cdd6e | 5787 | tgtdev_indexes = 0; |
b3d3fa51 | 5788 | if (dev_replace_is_ongoing && |
cf8cddd3 | 5789 | (op == BTRFS_MAP_WRITE || op == BTRFS_MAP_DISCARD) && |
472262f3 SB |
5790 | dev_replace->tgtdev != NULL) { |
5791 | int index_where_to_add; | |
5792 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5793 | ||
5794 | /* | |
5795 | * duplicate the write operations while the dev replace | |
5796 | * procedure is running. Since the copying of the old disk | |
5797 | * to the new disk takes place at run time while the | |
5798 | * filesystem is mounted writable, the regular write | |
5799 | * operations to the old disk have to be duplicated to go | |
5800 | * to the new disk as well. | |
5801 | * Note that device->missing is handled by the caller, and | |
5802 | * that the write to the old disk is already set up in the | |
5803 | * stripes array. | |
5804 | */ | |
5805 | index_where_to_add = num_stripes; | |
5806 | for (i = 0; i < num_stripes; i++) { | |
5807 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5808 | /* write to new disk, too */ | |
5809 | struct btrfs_bio_stripe *new = | |
5810 | bbio->stripes + index_where_to_add; | |
5811 | struct btrfs_bio_stripe *old = | |
5812 | bbio->stripes + i; | |
5813 | ||
5814 | new->physical = old->physical; | |
5815 | new->length = old->length; | |
5816 | new->dev = dev_replace->tgtdev; | |
2c8cdd6e | 5817 | bbio->tgtdev_map[i] = index_where_to_add; |
472262f3 SB |
5818 | index_where_to_add++; |
5819 | max_errors++; | |
2c8cdd6e | 5820 | tgtdev_indexes++; |
472262f3 SB |
5821 | } |
5822 | } | |
5823 | num_stripes = index_where_to_add; | |
cf8cddd3 CH |
5824 | } else if (dev_replace_is_ongoing && |
5825 | op == BTRFS_MAP_GET_READ_MIRRORS && | |
ad6d620e SB |
5826 | dev_replace->tgtdev != NULL) { |
5827 | u64 srcdev_devid = dev_replace->srcdev->devid; | |
5828 | int index_srcdev = 0; | |
5829 | int found = 0; | |
5830 | u64 physical_of_found = 0; | |
5831 | ||
5832 | /* | |
5833 | * During the dev-replace procedure, the target drive can | |
5834 | * also be used to read data in case it is needed to repair | |
5835 | * a corrupt block elsewhere. This is possible if the | |
5836 | * requested area is left of the left cursor. In this area, | |
5837 | * the target drive is a full copy of the source drive. | |
5838 | */ | |
5839 | for (i = 0; i < num_stripes; i++) { | |
5840 | if (bbio->stripes[i].dev->devid == srcdev_devid) { | |
5841 | /* | |
5842 | * In case of DUP, in order to keep it | |
5843 | * simple, only add the mirror with the | |
5844 | * lowest physical address | |
5845 | */ | |
5846 | if (found && | |
5847 | physical_of_found <= | |
5848 | bbio->stripes[i].physical) | |
5849 | continue; | |
5850 | index_srcdev = i; | |
5851 | found = 1; | |
5852 | physical_of_found = bbio->stripes[i].physical; | |
5853 | } | |
5854 | } | |
5855 | if (found) { | |
22ab04e8 FM |
5856 | struct btrfs_bio_stripe *tgtdev_stripe = |
5857 | bbio->stripes + num_stripes; | |
ad6d620e | 5858 | |
22ab04e8 FM |
5859 | tgtdev_stripe->physical = physical_of_found; |
5860 | tgtdev_stripe->length = | |
5861 | bbio->stripes[index_srcdev].length; | |
5862 | tgtdev_stripe->dev = dev_replace->tgtdev; | |
5863 | bbio->tgtdev_map[index_srcdev] = num_stripes; | |
ad6d620e | 5864 | |
22ab04e8 FM |
5865 | tgtdev_indexes++; |
5866 | num_stripes++; | |
ad6d620e | 5867 | } |
472262f3 SB |
5868 | } |
5869 | ||
de11cc12 | 5870 | *bbio_ret = bbio; |
10f11900 | 5871 | bbio->map_type = map->type; |
de11cc12 LZ |
5872 | bbio->num_stripes = num_stripes; |
5873 | bbio->max_errors = max_errors; | |
5874 | bbio->mirror_num = mirror_num; | |
2c8cdd6e | 5875 | bbio->num_tgtdevs = tgtdev_indexes; |
ad6d620e SB |
5876 | |
5877 | /* | |
5878 | * this is the case that REQ_READ && dev_replace_is_ongoing && | |
5879 | * mirror_num == num_stripes + 1 && dev_replace target drive is | |
5880 | * available as a mirror | |
5881 | */ | |
5882 | if (patch_the_first_stripe_for_dev_replace && num_stripes > 0) { | |
5883 | WARN_ON(num_stripes > 1); | |
5884 | bbio->stripes[0].dev = dev_replace->tgtdev; | |
5885 | bbio->stripes[0].physical = physical_to_patch_in_first_stripe; | |
5886 | bbio->mirror_num = map->num_stripes + 1; | |
5887 | } | |
cea9e445 | 5888 | out: |
73beece9 LB |
5889 | if (dev_replace_is_ongoing) { |
5890 | btrfs_dev_replace_clear_lock_blocking(dev_replace); | |
5891 | btrfs_dev_replace_unlock(dev_replace, 0); | |
5892 | } | |
0b86a832 | 5893 | free_extent_map(em); |
de11cc12 | 5894 | return ret; |
0b86a832 CM |
5895 | } |
5896 | ||
cf8cddd3 | 5897 | int btrfs_map_block(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
f2d8d74d | 5898 | u64 logical, u64 *length, |
a1d3c478 | 5899 | struct btrfs_bio **bbio_ret, int mirror_num) |
f2d8d74d | 5900 | { |
b3d3fa51 | 5901 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5902 | mirror_num, 0); |
f2d8d74d CM |
5903 | } |
5904 | ||
af8e2d1d | 5905 | /* For Scrub/replace */ |
cf8cddd3 | 5906 | int btrfs_map_sblock(struct btrfs_fs_info *fs_info, enum btrfs_map_op op, |
af8e2d1d MX |
5907 | u64 logical, u64 *length, |
5908 | struct btrfs_bio **bbio_ret, int mirror_num, | |
8e5cfb55 | 5909 | int need_raid_map) |
af8e2d1d | 5910 | { |
b3d3fa51 | 5911 | return __btrfs_map_block(fs_info, op, logical, length, bbio_ret, |
8e5cfb55 | 5912 | mirror_num, need_raid_map); |
af8e2d1d MX |
5913 | } |
5914 | ||
ab8d0fc4 | 5915 | int btrfs_rmap_block(struct btrfs_fs_info *fs_info, |
a512bbf8 YZ |
5916 | u64 chunk_start, u64 physical, u64 devid, |
5917 | u64 **logical, int *naddrs, int *stripe_len) | |
5918 | { | |
ab8d0fc4 | 5919 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; |
a512bbf8 YZ |
5920 | struct extent_map_tree *em_tree = &map_tree->map_tree; |
5921 | struct extent_map *em; | |
5922 | struct map_lookup *map; | |
5923 | u64 *buf; | |
5924 | u64 bytenr; | |
5925 | u64 length; | |
5926 | u64 stripe_nr; | |
53b381b3 | 5927 | u64 rmap_len; |
a512bbf8 YZ |
5928 | int i, j, nr = 0; |
5929 | ||
890871be | 5930 | read_lock(&em_tree->lock); |
a512bbf8 | 5931 | em = lookup_extent_mapping(em_tree, chunk_start, 1); |
890871be | 5932 | read_unlock(&em_tree->lock); |
a512bbf8 | 5933 | |
835d974f | 5934 | if (!em) { |
ab8d0fc4 JM |
5935 | btrfs_err(fs_info, "couldn't find em for chunk %Lu", |
5936 | chunk_start); | |
835d974f JB |
5937 | return -EIO; |
5938 | } | |
5939 | ||
5940 | if (em->start != chunk_start) { | |
ab8d0fc4 | 5941 | btrfs_err(fs_info, "bad chunk start, em=%Lu, wanted=%Lu", |
835d974f JB |
5942 | em->start, chunk_start); |
5943 | free_extent_map(em); | |
5944 | return -EIO; | |
5945 | } | |
95617d69 | 5946 | map = em->map_lookup; |
a512bbf8 YZ |
5947 | |
5948 | length = em->len; | |
53b381b3 DW |
5949 | rmap_len = map->stripe_len; |
5950 | ||
a512bbf8 | 5951 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) |
b8b93add | 5952 | length = div_u64(length, map->num_stripes / map->sub_stripes); |
a512bbf8 | 5953 | else if (map->type & BTRFS_BLOCK_GROUP_RAID0) |
b8b93add | 5954 | length = div_u64(length, map->num_stripes); |
ffe2d203 | 5955 | else if (map->type & BTRFS_BLOCK_GROUP_RAID56_MASK) { |
b8b93add | 5956 | length = div_u64(length, nr_data_stripes(map)); |
53b381b3 DW |
5957 | rmap_len = map->stripe_len * nr_data_stripes(map); |
5958 | } | |
a512bbf8 | 5959 | |
31e818fe | 5960 | buf = kcalloc(map->num_stripes, sizeof(u64), GFP_NOFS); |
79787eaa | 5961 | BUG_ON(!buf); /* -ENOMEM */ |
a512bbf8 YZ |
5962 | |
5963 | for (i = 0; i < map->num_stripes; i++) { | |
5964 | if (devid && map->stripes[i].dev->devid != devid) | |
5965 | continue; | |
5966 | if (map->stripes[i].physical > physical || | |
5967 | map->stripes[i].physical + length <= physical) | |
5968 | continue; | |
5969 | ||
5970 | stripe_nr = physical - map->stripes[i].physical; | |
b8b93add | 5971 | stripe_nr = div_u64(stripe_nr, map->stripe_len); |
a512bbf8 YZ |
5972 | |
5973 | if (map->type & BTRFS_BLOCK_GROUP_RAID10) { | |
5974 | stripe_nr = stripe_nr * map->num_stripes + i; | |
b8b93add | 5975 | stripe_nr = div_u64(stripe_nr, map->sub_stripes); |
a512bbf8 YZ |
5976 | } else if (map->type & BTRFS_BLOCK_GROUP_RAID0) { |
5977 | stripe_nr = stripe_nr * map->num_stripes + i; | |
53b381b3 DW |
5978 | } /* else if RAID[56], multiply by nr_data_stripes(). |
5979 | * Alternatively, just use rmap_len below instead of | |
5980 | * map->stripe_len */ | |
5981 | ||
5982 | bytenr = chunk_start + stripe_nr * rmap_len; | |
934d375b | 5983 | WARN_ON(nr >= map->num_stripes); |
a512bbf8 YZ |
5984 | for (j = 0; j < nr; j++) { |
5985 | if (buf[j] == bytenr) | |
5986 | break; | |
5987 | } | |
934d375b CM |
5988 | if (j == nr) { |
5989 | WARN_ON(nr >= map->num_stripes); | |
a512bbf8 | 5990 | buf[nr++] = bytenr; |
934d375b | 5991 | } |
a512bbf8 YZ |
5992 | } |
5993 | ||
a512bbf8 YZ |
5994 | *logical = buf; |
5995 | *naddrs = nr; | |
53b381b3 | 5996 | *stripe_len = rmap_len; |
a512bbf8 YZ |
5997 | |
5998 | free_extent_map(em); | |
5999 | return 0; | |
f2d8d74d CM |
6000 | } |
6001 | ||
4246a0b6 | 6002 | static inline void btrfs_end_bbio(struct btrfs_bio *bbio, struct bio *bio) |
8408c716 | 6003 | { |
326e1dbb MS |
6004 | bio->bi_private = bbio->private; |
6005 | bio->bi_end_io = bbio->end_io; | |
4246a0b6 | 6006 | bio_endio(bio); |
326e1dbb | 6007 | |
6e9606d2 | 6008 | btrfs_put_bbio(bbio); |
8408c716 MX |
6009 | } |
6010 | ||
4246a0b6 | 6011 | static void btrfs_end_bio(struct bio *bio) |
8790d502 | 6012 | { |
9be3395b | 6013 | struct btrfs_bio *bbio = bio->bi_private; |
7d2b4daa | 6014 | int is_orig_bio = 0; |
8790d502 | 6015 | |
4246a0b6 | 6016 | if (bio->bi_error) { |
a1d3c478 | 6017 | atomic_inc(&bbio->error); |
4246a0b6 | 6018 | if (bio->bi_error == -EIO || bio->bi_error == -EREMOTEIO) { |
442a4f63 | 6019 | unsigned int stripe_index = |
9be3395b | 6020 | btrfs_io_bio(bio)->stripe_index; |
65f53338 | 6021 | struct btrfs_device *dev; |
442a4f63 SB |
6022 | |
6023 | BUG_ON(stripe_index >= bbio->num_stripes); | |
6024 | dev = bbio->stripes[stripe_index].dev; | |
597a60fa | 6025 | if (dev->bdev) { |
37226b21 | 6026 | if (bio_op(bio) == REQ_OP_WRITE) |
597a60fa SB |
6027 | btrfs_dev_stat_inc(dev, |
6028 | BTRFS_DEV_STAT_WRITE_ERRS); | |
6029 | else | |
6030 | btrfs_dev_stat_inc(dev, | |
6031 | BTRFS_DEV_STAT_READ_ERRS); | |
1eff9d32 | 6032 | if ((bio->bi_opf & WRITE_FLUSH) == WRITE_FLUSH) |
597a60fa SB |
6033 | btrfs_dev_stat_inc(dev, |
6034 | BTRFS_DEV_STAT_FLUSH_ERRS); | |
6035 | btrfs_dev_stat_print_on_error(dev); | |
6036 | } | |
442a4f63 SB |
6037 | } |
6038 | } | |
8790d502 | 6039 | |
a1d3c478 | 6040 | if (bio == bbio->orig_bio) |
7d2b4daa CM |
6041 | is_orig_bio = 1; |
6042 | ||
c404e0dc MX |
6043 | btrfs_bio_counter_dec(bbio->fs_info); |
6044 | ||
a1d3c478 | 6045 | if (atomic_dec_and_test(&bbio->stripes_pending)) { |
7d2b4daa CM |
6046 | if (!is_orig_bio) { |
6047 | bio_put(bio); | |
a1d3c478 | 6048 | bio = bbio->orig_bio; |
7d2b4daa | 6049 | } |
c7b22bb1 | 6050 | |
9be3395b | 6051 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
a236aed1 | 6052 | /* only send an error to the higher layers if it is |
53b381b3 | 6053 | * beyond the tolerance of the btrfs bio |
a236aed1 | 6054 | */ |
a1d3c478 | 6055 | if (atomic_read(&bbio->error) > bbio->max_errors) { |
4246a0b6 | 6056 | bio->bi_error = -EIO; |
5dbc8fca | 6057 | } else { |
1259ab75 CM |
6058 | /* |
6059 | * this bio is actually up to date, we didn't | |
6060 | * go over the max number of errors | |
6061 | */ | |
4246a0b6 | 6062 | bio->bi_error = 0; |
1259ab75 | 6063 | } |
c55f1396 | 6064 | |
4246a0b6 | 6065 | btrfs_end_bbio(bbio, bio); |
7d2b4daa | 6066 | } else if (!is_orig_bio) { |
8790d502 CM |
6067 | bio_put(bio); |
6068 | } | |
8790d502 CM |
6069 | } |
6070 | ||
8b712842 CM |
6071 | /* |
6072 | * see run_scheduled_bios for a description of why bios are collected for | |
6073 | * async submit. | |
6074 | * | |
6075 | * This will add one bio to the pending list for a device and make sure | |
6076 | * the work struct is scheduled. | |
6077 | */ | |
48a3b636 ES |
6078 | static noinline void btrfs_schedule_bio(struct btrfs_root *root, |
6079 | struct btrfs_device *device, | |
4e49ea4a | 6080 | struct bio *bio) |
8b712842 | 6081 | { |
0b246afa | 6082 | struct btrfs_fs_info *fs_info = device->fs_info; |
8b712842 | 6083 | int should_queue = 1; |
ffbd517d | 6084 | struct btrfs_pending_bios *pending_bios; |
8b712842 | 6085 | |
53b381b3 | 6086 | if (device->missing || !device->bdev) { |
4246a0b6 | 6087 | bio_io_error(bio); |
53b381b3 DW |
6088 | return; |
6089 | } | |
6090 | ||
8b712842 | 6091 | /* don't bother with additional async steps for reads, right now */ |
37226b21 | 6092 | if (bio_op(bio) == REQ_OP_READ) { |
492bb6de | 6093 | bio_get(bio); |
4e49ea4a | 6094 | btrfsic_submit_bio(bio); |
492bb6de | 6095 | bio_put(bio); |
143bede5 | 6096 | return; |
8b712842 CM |
6097 | } |
6098 | ||
6099 | /* | |
0986fe9e | 6100 | * nr_async_bios allows us to reliably return congestion to the |
8b712842 CM |
6101 | * higher layers. Otherwise, the async bio makes it appear we have |
6102 | * made progress against dirty pages when we've really just put it | |
6103 | * on a queue for later | |
6104 | */ | |
0b246afa | 6105 | atomic_inc(&fs_info->nr_async_bios); |
492bb6de | 6106 | WARN_ON(bio->bi_next); |
8b712842 | 6107 | bio->bi_next = NULL; |
8b712842 CM |
6108 | |
6109 | spin_lock(&device->io_lock); | |
1eff9d32 | 6110 | if (bio->bi_opf & REQ_SYNC) |
ffbd517d CM |
6111 | pending_bios = &device->pending_sync_bios; |
6112 | else | |
6113 | pending_bios = &device->pending_bios; | |
8b712842 | 6114 | |
ffbd517d CM |
6115 | if (pending_bios->tail) |
6116 | pending_bios->tail->bi_next = bio; | |
8b712842 | 6117 | |
ffbd517d CM |
6118 | pending_bios->tail = bio; |
6119 | if (!pending_bios->head) | |
6120 | pending_bios->head = bio; | |
8b712842 CM |
6121 | if (device->running_pending) |
6122 | should_queue = 0; | |
6123 | ||
6124 | spin_unlock(&device->io_lock); | |
6125 | ||
6126 | if (should_queue) | |
0b246afa | 6127 | btrfs_queue_work(fs_info->submit_workers, &device->work); |
8b712842 CM |
6128 | } |
6129 | ||
de1ee92a JB |
6130 | static void submit_stripe_bio(struct btrfs_root *root, struct btrfs_bio *bbio, |
6131 | struct bio *bio, u64 physical, int dev_nr, | |
81a75f67 | 6132 | int async) |
de1ee92a JB |
6133 | { |
6134 | struct btrfs_device *dev = bbio->stripes[dev_nr].dev; | |
6135 | ||
6136 | bio->bi_private = bbio; | |
9be3395b | 6137 | btrfs_io_bio(bio)->stripe_index = dev_nr; |
de1ee92a | 6138 | bio->bi_end_io = btrfs_end_bio; |
4f024f37 | 6139 | bio->bi_iter.bi_sector = physical >> 9; |
de1ee92a JB |
6140 | #ifdef DEBUG |
6141 | { | |
6142 | struct rcu_string *name; | |
6143 | ||
6144 | rcu_read_lock(); | |
6145 | name = rcu_dereference(dev->name); | |
ab8d0fc4 JM |
6146 | btrfs_debug(fs_info, |
6147 | "btrfs_map_bio: rw %d 0x%x, sector=%llu, dev=%lu (%s id %llu), size=%u", | |
6148 | bio_op(bio), bio->bi_opf, | |
6149 | (u64)bio->bi_iter.bi_sector, | |
6150 | (u_long)dev->bdev->bd_dev, name->str, dev->devid, | |
6151 | bio->bi_iter.bi_size); | |
de1ee92a JB |
6152 | rcu_read_unlock(); |
6153 | } | |
6154 | #endif | |
6155 | bio->bi_bdev = dev->bdev; | |
c404e0dc MX |
6156 | |
6157 | btrfs_bio_counter_inc_noblocked(root->fs_info); | |
6158 | ||
de1ee92a | 6159 | if (async) |
4e49ea4a | 6160 | btrfs_schedule_bio(root, dev, bio); |
de1ee92a | 6161 | else |
4e49ea4a | 6162 | btrfsic_submit_bio(bio); |
de1ee92a JB |
6163 | } |
6164 | ||
de1ee92a JB |
6165 | static void bbio_error(struct btrfs_bio *bbio, struct bio *bio, u64 logical) |
6166 | { | |
6167 | atomic_inc(&bbio->error); | |
6168 | if (atomic_dec_and_test(&bbio->stripes_pending)) { | |
01327610 | 6169 | /* Should be the original bio. */ |
8408c716 MX |
6170 | WARN_ON(bio != bbio->orig_bio); |
6171 | ||
9be3395b | 6172 | btrfs_io_bio(bio)->mirror_num = bbio->mirror_num; |
4f024f37 | 6173 | bio->bi_iter.bi_sector = logical >> 9; |
4246a0b6 CH |
6174 | bio->bi_error = -EIO; |
6175 | btrfs_end_bbio(bbio, bio); | |
de1ee92a JB |
6176 | } |
6177 | } | |
6178 | ||
81a75f67 | 6179 | int btrfs_map_bio(struct btrfs_root *root, struct bio *bio, |
8b712842 | 6180 | int mirror_num, int async_submit) |
0b86a832 | 6181 | { |
0b246afa | 6182 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b86a832 | 6183 | struct btrfs_device *dev; |
8790d502 | 6184 | struct bio *first_bio = bio; |
4f024f37 | 6185 | u64 logical = (u64)bio->bi_iter.bi_sector << 9; |
0b86a832 CM |
6186 | u64 length = 0; |
6187 | u64 map_length; | |
0b86a832 | 6188 | int ret; |
08da757d ZL |
6189 | int dev_nr; |
6190 | int total_devs; | |
a1d3c478 | 6191 | struct btrfs_bio *bbio = NULL; |
0b86a832 | 6192 | |
4f024f37 | 6193 | length = bio->bi_iter.bi_size; |
0b86a832 | 6194 | map_length = length; |
cea9e445 | 6195 | |
0b246afa JM |
6196 | btrfs_bio_counter_inc_blocked(fs_info); |
6197 | ret = __btrfs_map_block(fs_info, bio_op(bio), logical, | |
37226b21 | 6198 | &map_length, &bbio, mirror_num, 1); |
c404e0dc | 6199 | if (ret) { |
0b246afa | 6200 | btrfs_bio_counter_dec(fs_info); |
79787eaa | 6201 | return ret; |
c404e0dc | 6202 | } |
cea9e445 | 6203 | |
a1d3c478 | 6204 | total_devs = bbio->num_stripes; |
53b381b3 DW |
6205 | bbio->orig_bio = first_bio; |
6206 | bbio->private = first_bio->bi_private; | |
6207 | bbio->end_io = first_bio->bi_end_io; | |
0b246afa | 6208 | bbio->fs_info = fs_info; |
53b381b3 DW |
6209 | atomic_set(&bbio->stripes_pending, bbio->num_stripes); |
6210 | ||
ad1ba2a0 | 6211 | if ((bbio->map_type & BTRFS_BLOCK_GROUP_RAID56_MASK) && |
37226b21 | 6212 | ((bio_op(bio) == REQ_OP_WRITE) || (mirror_num > 1))) { |
53b381b3 DW |
6213 | /* In this case, map_length has been set to the length of |
6214 | a single stripe; not the whole write */ | |
37226b21 | 6215 | if (bio_op(bio) == REQ_OP_WRITE) { |
8e5cfb55 | 6216 | ret = raid56_parity_write(root, bio, bbio, map_length); |
53b381b3 | 6217 | } else { |
8e5cfb55 | 6218 | ret = raid56_parity_recover(root, bio, bbio, map_length, |
4245215d | 6219 | mirror_num, 1); |
53b381b3 | 6220 | } |
4245215d | 6221 | |
0b246afa | 6222 | btrfs_bio_counter_dec(fs_info); |
c404e0dc | 6223 | return ret; |
53b381b3 DW |
6224 | } |
6225 | ||
cea9e445 | 6226 | if (map_length < length) { |
0b246afa | 6227 | btrfs_crit(fs_info, |
5d163e0e JM |
6228 | "mapping failed logical %llu bio len %llu len %llu", |
6229 | logical, length, map_length); | |
cea9e445 CM |
6230 | BUG(); |
6231 | } | |
a1d3c478 | 6232 | |
08da757d | 6233 | for (dev_nr = 0; dev_nr < total_devs; dev_nr++) { |
de1ee92a | 6234 | dev = bbio->stripes[dev_nr].dev; |
37226b21 MC |
6235 | if (!dev || !dev->bdev || |
6236 | (bio_op(bio) == REQ_OP_WRITE && !dev->writeable)) { | |
de1ee92a | 6237 | bbio_error(bbio, first_bio, logical); |
de1ee92a JB |
6238 | continue; |
6239 | } | |
6240 | ||
a1d3c478 | 6241 | if (dev_nr < total_devs - 1) { |
9be3395b | 6242 | bio = btrfs_bio_clone(first_bio, GFP_NOFS); |
79787eaa | 6243 | BUG_ON(!bio); /* -ENOMEM */ |
326e1dbb | 6244 | } else |
a1d3c478 | 6245 | bio = first_bio; |
de1ee92a JB |
6246 | |
6247 | submit_stripe_bio(root, bbio, bio, | |
81a75f67 | 6248 | bbio->stripes[dev_nr].physical, dev_nr, |
de1ee92a | 6249 | async_submit); |
8790d502 | 6250 | } |
0b246afa | 6251 | btrfs_bio_counter_dec(fs_info); |
0b86a832 CM |
6252 | return 0; |
6253 | } | |
6254 | ||
aa1b8cd4 | 6255 | struct btrfs_device *btrfs_find_device(struct btrfs_fs_info *fs_info, u64 devid, |
2b82032c | 6256 | u8 *uuid, u8 *fsid) |
0b86a832 | 6257 | { |
2b82032c YZ |
6258 | struct btrfs_device *device; |
6259 | struct btrfs_fs_devices *cur_devices; | |
6260 | ||
aa1b8cd4 | 6261 | cur_devices = fs_info->fs_devices; |
2b82032c YZ |
6262 | while (cur_devices) { |
6263 | if (!fsid || | |
6264 | !memcmp(cur_devices->fsid, fsid, BTRFS_UUID_SIZE)) { | |
6265 | device = __find_device(&cur_devices->devices, | |
6266 | devid, uuid); | |
6267 | if (device) | |
6268 | return device; | |
6269 | } | |
6270 | cur_devices = cur_devices->seed; | |
6271 | } | |
6272 | return NULL; | |
0b86a832 CM |
6273 | } |
6274 | ||
dfe25020 | 6275 | static struct btrfs_device *add_missing_dev(struct btrfs_root *root, |
5f375835 | 6276 | struct btrfs_fs_devices *fs_devices, |
dfe25020 CM |
6277 | u64 devid, u8 *dev_uuid) |
6278 | { | |
6279 | struct btrfs_device *device; | |
dfe25020 | 6280 | |
12bd2fc0 ID |
6281 | device = btrfs_alloc_device(NULL, &devid, dev_uuid); |
6282 | if (IS_ERR(device)) | |
7cbd8a83 | 6283 | return NULL; |
12bd2fc0 ID |
6284 | |
6285 | list_add(&device->dev_list, &fs_devices->devices); | |
e4404d6e | 6286 | device->fs_devices = fs_devices; |
dfe25020 | 6287 | fs_devices->num_devices++; |
12bd2fc0 ID |
6288 | |
6289 | device->missing = 1; | |
cd02dca5 | 6290 | fs_devices->missing_devices++; |
12bd2fc0 | 6291 | |
dfe25020 CM |
6292 | return device; |
6293 | } | |
6294 | ||
12bd2fc0 ID |
6295 | /** |
6296 | * btrfs_alloc_device - allocate struct btrfs_device | |
6297 | * @fs_info: used only for generating a new devid, can be NULL if | |
6298 | * devid is provided (i.e. @devid != NULL). | |
6299 | * @devid: a pointer to devid for this device. If NULL a new devid | |
6300 | * is generated. | |
6301 | * @uuid: a pointer to UUID for this device. If NULL a new UUID | |
6302 | * is generated. | |
6303 | * | |
6304 | * Return: a pointer to a new &struct btrfs_device on success; ERR_PTR() | |
6305 | * on error. Returned struct is not linked onto any lists and can be | |
6306 | * destroyed with kfree() right away. | |
6307 | */ | |
6308 | struct btrfs_device *btrfs_alloc_device(struct btrfs_fs_info *fs_info, | |
6309 | const u64 *devid, | |
6310 | const u8 *uuid) | |
6311 | { | |
6312 | struct btrfs_device *dev; | |
6313 | u64 tmp; | |
6314 | ||
fae7f21c | 6315 | if (WARN_ON(!devid && !fs_info)) |
12bd2fc0 | 6316 | return ERR_PTR(-EINVAL); |
12bd2fc0 ID |
6317 | |
6318 | dev = __alloc_device(); | |
6319 | if (IS_ERR(dev)) | |
6320 | return dev; | |
6321 | ||
6322 | if (devid) | |
6323 | tmp = *devid; | |
6324 | else { | |
6325 | int ret; | |
6326 | ||
6327 | ret = find_next_devid(fs_info, &tmp); | |
6328 | if (ret) { | |
6329 | kfree(dev); | |
6330 | return ERR_PTR(ret); | |
6331 | } | |
6332 | } | |
6333 | dev->devid = tmp; | |
6334 | ||
6335 | if (uuid) | |
6336 | memcpy(dev->uuid, uuid, BTRFS_UUID_SIZE); | |
6337 | else | |
6338 | generate_random_uuid(dev->uuid); | |
6339 | ||
9e0af237 LB |
6340 | btrfs_init_work(&dev->work, btrfs_submit_helper, |
6341 | pending_bios_fn, NULL, NULL); | |
12bd2fc0 ID |
6342 | |
6343 | return dev; | |
6344 | } | |
6345 | ||
e06cd3dd LB |
6346 | /* Return -EIO if any error, otherwise return 0. */ |
6347 | static int btrfs_check_chunk_valid(struct btrfs_root *root, | |
6348 | struct extent_buffer *leaf, | |
6349 | struct btrfs_chunk *chunk, u64 logical) | |
0b86a832 | 6350 | { |
0b246afa | 6351 | struct btrfs_fs_info *fs_info = root->fs_info; |
0b86a832 | 6352 | u64 length; |
f04b772b | 6353 | u64 stripe_len; |
e06cd3dd LB |
6354 | u16 num_stripes; |
6355 | u16 sub_stripes; | |
6356 | u64 type; | |
0b86a832 | 6357 | |
e17cade2 | 6358 | length = btrfs_chunk_length(leaf, chunk); |
f04b772b QW |
6359 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); |
6360 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
e06cd3dd LB |
6361 | sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
6362 | type = btrfs_chunk_type(leaf, chunk); | |
6363 | ||
f04b772b | 6364 | if (!num_stripes) { |
0b246afa | 6365 | btrfs_err(fs_info, "invalid chunk num_stripes: %u", |
f04b772b QW |
6366 | num_stripes); |
6367 | return -EIO; | |
6368 | } | |
0b246afa JM |
6369 | if (!IS_ALIGNED(logical, fs_info->sectorsize)) { |
6370 | btrfs_err(fs_info, "invalid chunk logical %llu", logical); | |
f04b772b QW |
6371 | return -EIO; |
6372 | } | |
0b246afa JM |
6373 | if (btrfs_chunk_sector_size(leaf, chunk) != fs_info->sectorsize) { |
6374 | btrfs_err(fs_info, "invalid chunk sectorsize %u", | |
e06cd3dd LB |
6375 | btrfs_chunk_sector_size(leaf, chunk)); |
6376 | return -EIO; | |
6377 | } | |
0b246afa JM |
6378 | if (!length || !IS_ALIGNED(length, fs_info->sectorsize)) { |
6379 | btrfs_err(fs_info, "invalid chunk length %llu", length); | |
f04b772b QW |
6380 | return -EIO; |
6381 | } | |
3d8da678 | 6382 | if (!is_power_of_2(stripe_len) || stripe_len != BTRFS_STRIPE_LEN) { |
0b246afa | 6383 | btrfs_err(fs_info, "invalid chunk stripe length: %llu", |
f04b772b QW |
6384 | stripe_len); |
6385 | return -EIO; | |
6386 | } | |
6387 | if (~(BTRFS_BLOCK_GROUP_TYPE_MASK | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
e06cd3dd | 6388 | type) { |
0b246afa | 6389 | btrfs_err(fs_info, "unrecognized chunk type: %llu", |
f04b772b QW |
6390 | ~(BTRFS_BLOCK_GROUP_TYPE_MASK | |
6391 | BTRFS_BLOCK_GROUP_PROFILE_MASK) & | |
6392 | btrfs_chunk_type(leaf, chunk)); | |
6393 | return -EIO; | |
6394 | } | |
e06cd3dd LB |
6395 | if ((type & BTRFS_BLOCK_GROUP_RAID10 && sub_stripes != 2) || |
6396 | (type & BTRFS_BLOCK_GROUP_RAID1 && num_stripes < 1) || | |
6397 | (type & BTRFS_BLOCK_GROUP_RAID5 && num_stripes < 2) || | |
6398 | (type & BTRFS_BLOCK_GROUP_RAID6 && num_stripes < 3) || | |
6399 | (type & BTRFS_BLOCK_GROUP_DUP && num_stripes > 2) || | |
6400 | ((type & BTRFS_BLOCK_GROUP_PROFILE_MASK) == 0 && | |
6401 | num_stripes != 1)) { | |
0b246afa | 6402 | btrfs_err(fs_info, |
e06cd3dd LB |
6403 | "invalid num_stripes:sub_stripes %u:%u for profile %llu", |
6404 | num_stripes, sub_stripes, | |
6405 | type & BTRFS_BLOCK_GROUP_PROFILE_MASK); | |
6406 | return -EIO; | |
6407 | } | |
6408 | ||
6409 | return 0; | |
6410 | } | |
6411 | ||
6412 | static int read_one_chunk(struct btrfs_root *root, struct btrfs_key *key, | |
6413 | struct extent_buffer *leaf, | |
6414 | struct btrfs_chunk *chunk) | |
6415 | { | |
0b246afa JM |
6416 | struct btrfs_fs_info *fs_info = root->fs_info; |
6417 | struct btrfs_mapping_tree *map_tree = &fs_info->mapping_tree; | |
e06cd3dd LB |
6418 | struct map_lookup *map; |
6419 | struct extent_map *em; | |
6420 | u64 logical; | |
6421 | u64 length; | |
6422 | u64 stripe_len; | |
6423 | u64 devid; | |
6424 | u8 uuid[BTRFS_UUID_SIZE]; | |
6425 | int num_stripes; | |
6426 | int ret; | |
6427 | int i; | |
6428 | ||
6429 | logical = key->offset; | |
6430 | length = btrfs_chunk_length(leaf, chunk); | |
6431 | stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6432 | num_stripes = btrfs_chunk_num_stripes(leaf, chunk); | |
6433 | ||
6434 | ret = btrfs_check_chunk_valid(root, leaf, chunk, logical); | |
6435 | if (ret) | |
6436 | return ret; | |
a061fc8d | 6437 | |
890871be | 6438 | read_lock(&map_tree->map_tree.lock); |
0b86a832 | 6439 | em = lookup_extent_mapping(&map_tree->map_tree, logical, 1); |
890871be | 6440 | read_unlock(&map_tree->map_tree.lock); |
0b86a832 CM |
6441 | |
6442 | /* already mapped? */ | |
6443 | if (em && em->start <= logical && em->start + em->len > logical) { | |
6444 | free_extent_map(em); | |
0b86a832 CM |
6445 | return 0; |
6446 | } else if (em) { | |
6447 | free_extent_map(em); | |
6448 | } | |
0b86a832 | 6449 | |
172ddd60 | 6450 | em = alloc_extent_map(); |
0b86a832 CM |
6451 | if (!em) |
6452 | return -ENOMEM; | |
593060d7 | 6453 | map = kmalloc(map_lookup_size(num_stripes), GFP_NOFS); |
0b86a832 CM |
6454 | if (!map) { |
6455 | free_extent_map(em); | |
6456 | return -ENOMEM; | |
6457 | } | |
6458 | ||
298a8f9c | 6459 | set_bit(EXTENT_FLAG_FS_MAPPING, &em->flags); |
95617d69 | 6460 | em->map_lookup = map; |
0b86a832 CM |
6461 | em->start = logical; |
6462 | em->len = length; | |
70c8a91c | 6463 | em->orig_start = 0; |
0b86a832 | 6464 | em->block_start = 0; |
c8b97818 | 6465 | em->block_len = em->len; |
0b86a832 | 6466 | |
593060d7 CM |
6467 | map->num_stripes = num_stripes; |
6468 | map->io_width = btrfs_chunk_io_width(leaf, chunk); | |
6469 | map->io_align = btrfs_chunk_io_align(leaf, chunk); | |
6470 | map->sector_size = btrfs_chunk_sector_size(leaf, chunk); | |
6471 | map->stripe_len = btrfs_chunk_stripe_len(leaf, chunk); | |
6472 | map->type = btrfs_chunk_type(leaf, chunk); | |
321aecc6 | 6473 | map->sub_stripes = btrfs_chunk_sub_stripes(leaf, chunk); |
593060d7 CM |
6474 | for (i = 0; i < num_stripes; i++) { |
6475 | map->stripes[i].physical = | |
6476 | btrfs_stripe_offset_nr(leaf, chunk, i); | |
6477 | devid = btrfs_stripe_devid_nr(leaf, chunk, i); | |
a443755f CM |
6478 | read_extent_buffer(leaf, uuid, (unsigned long) |
6479 | btrfs_stripe_dev_uuid_nr(chunk, i), | |
6480 | BTRFS_UUID_SIZE); | |
0b246afa | 6481 | map->stripes[i].dev = btrfs_find_device(fs_info, devid, |
aa1b8cd4 | 6482 | uuid, NULL); |
3cdde224 | 6483 | if (!map->stripes[i].dev && |
0b246afa | 6484 | !btrfs_test_opt(fs_info, DEGRADED)) { |
593060d7 CM |
6485 | free_extent_map(em); |
6486 | return -EIO; | |
6487 | } | |
dfe25020 CM |
6488 | if (!map->stripes[i].dev) { |
6489 | map->stripes[i].dev = | |
0b246afa | 6490 | add_missing_dev(root, fs_info->fs_devices, |
5f375835 | 6491 | devid, uuid); |
dfe25020 | 6492 | if (!map->stripes[i].dev) { |
dfe25020 CM |
6493 | free_extent_map(em); |
6494 | return -EIO; | |
6495 | } | |
0b246afa | 6496 | btrfs_warn(fs_info, "devid %llu uuid %pU is missing", |
5d163e0e | 6497 | devid, uuid); |
dfe25020 CM |
6498 | } |
6499 | map->stripes[i].dev->in_fs_metadata = 1; | |
0b86a832 CM |
6500 | } |
6501 | ||
890871be | 6502 | write_lock(&map_tree->map_tree.lock); |
09a2a8f9 | 6503 | ret = add_extent_mapping(&map_tree->map_tree, em, 0); |
890871be | 6504 | write_unlock(&map_tree->map_tree.lock); |
79787eaa | 6505 | BUG_ON(ret); /* Tree corruption */ |
0b86a832 CM |
6506 | free_extent_map(em); |
6507 | ||
6508 | return 0; | |
6509 | } | |
6510 | ||
143bede5 | 6511 | static void fill_device_from_item(struct extent_buffer *leaf, |
0b86a832 CM |
6512 | struct btrfs_dev_item *dev_item, |
6513 | struct btrfs_device *device) | |
6514 | { | |
6515 | unsigned long ptr; | |
0b86a832 CM |
6516 | |
6517 | device->devid = btrfs_device_id(leaf, dev_item); | |
d6397bae CB |
6518 | device->disk_total_bytes = btrfs_device_total_bytes(leaf, dev_item); |
6519 | device->total_bytes = device->disk_total_bytes; | |
935e5cc9 | 6520 | device->commit_total_bytes = device->disk_total_bytes; |
0b86a832 | 6521 | device->bytes_used = btrfs_device_bytes_used(leaf, dev_item); |
ce7213c7 | 6522 | device->commit_bytes_used = device->bytes_used; |
0b86a832 CM |
6523 | device->type = btrfs_device_type(leaf, dev_item); |
6524 | device->io_align = btrfs_device_io_align(leaf, dev_item); | |
6525 | device->io_width = btrfs_device_io_width(leaf, dev_item); | |
6526 | device->sector_size = btrfs_device_sector_size(leaf, dev_item); | |
8dabb742 | 6527 | WARN_ON(device->devid == BTRFS_DEV_REPLACE_DEVID); |
63a212ab | 6528 | device->is_tgtdev_for_dev_replace = 0; |
0b86a832 | 6529 | |
410ba3a2 | 6530 | ptr = btrfs_device_uuid(dev_item); |
e17cade2 | 6531 | read_extent_buffer(leaf, device->uuid, ptr, BTRFS_UUID_SIZE); |
0b86a832 CM |
6532 | } |
6533 | ||
5f375835 MX |
6534 | static struct btrfs_fs_devices *open_seed_devices(struct btrfs_root *root, |
6535 | u8 *fsid) | |
2b82032c | 6536 | { |
0b246afa | 6537 | struct btrfs_fs_info *fs_info = root->fs_info; |
2b82032c YZ |
6538 | struct btrfs_fs_devices *fs_devices; |
6539 | int ret; | |
6540 | ||
b367e47f | 6541 | BUG_ON(!mutex_is_locked(&uuid_mutex)); |
2b82032c | 6542 | |
0b246afa | 6543 | fs_devices = fs_info->fs_devices->seed; |
2b82032c | 6544 | while (fs_devices) { |
5f375835 MX |
6545 | if (!memcmp(fs_devices->fsid, fsid, BTRFS_UUID_SIZE)) |
6546 | return fs_devices; | |
6547 | ||
2b82032c YZ |
6548 | fs_devices = fs_devices->seed; |
6549 | } | |
6550 | ||
6551 | fs_devices = find_fsid(fsid); | |
6552 | if (!fs_devices) { | |
0b246afa | 6553 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6554 | return ERR_PTR(-ENOENT); |
6555 | ||
6556 | fs_devices = alloc_fs_devices(fsid); | |
6557 | if (IS_ERR(fs_devices)) | |
6558 | return fs_devices; | |
6559 | ||
6560 | fs_devices->seeding = 1; | |
6561 | fs_devices->opened = 1; | |
6562 | return fs_devices; | |
2b82032c | 6563 | } |
e4404d6e YZ |
6564 | |
6565 | fs_devices = clone_fs_devices(fs_devices); | |
5f375835 MX |
6566 | if (IS_ERR(fs_devices)) |
6567 | return fs_devices; | |
2b82032c | 6568 | |
97288f2c | 6569 | ret = __btrfs_open_devices(fs_devices, FMODE_READ, |
0b246afa | 6570 | fs_info->bdev_holder); |
48d28232 JL |
6571 | if (ret) { |
6572 | free_fs_devices(fs_devices); | |
5f375835 | 6573 | fs_devices = ERR_PTR(ret); |
2b82032c | 6574 | goto out; |
48d28232 | 6575 | } |
2b82032c YZ |
6576 | |
6577 | if (!fs_devices->seeding) { | |
6578 | __btrfs_close_devices(fs_devices); | |
e4404d6e | 6579 | free_fs_devices(fs_devices); |
5f375835 | 6580 | fs_devices = ERR_PTR(-EINVAL); |
2b82032c YZ |
6581 | goto out; |
6582 | } | |
6583 | ||
0b246afa JM |
6584 | fs_devices->seed = fs_info->fs_devices->seed; |
6585 | fs_info->fs_devices->seed = fs_devices; | |
2b82032c | 6586 | out: |
5f375835 | 6587 | return fs_devices; |
2b82032c YZ |
6588 | } |
6589 | ||
0d81ba5d | 6590 | static int read_one_dev(struct btrfs_root *root, |
0b86a832 CM |
6591 | struct extent_buffer *leaf, |
6592 | struct btrfs_dev_item *dev_item) | |
6593 | { | |
0b246afa JM |
6594 | struct btrfs_fs_info *fs_info = root->fs_info; |
6595 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
0b86a832 CM |
6596 | struct btrfs_device *device; |
6597 | u64 devid; | |
6598 | int ret; | |
2b82032c | 6599 | u8 fs_uuid[BTRFS_UUID_SIZE]; |
a443755f CM |
6600 | u8 dev_uuid[BTRFS_UUID_SIZE]; |
6601 | ||
0b86a832 | 6602 | devid = btrfs_device_id(leaf, dev_item); |
410ba3a2 | 6603 | read_extent_buffer(leaf, dev_uuid, btrfs_device_uuid(dev_item), |
a443755f | 6604 | BTRFS_UUID_SIZE); |
1473b24e | 6605 | read_extent_buffer(leaf, fs_uuid, btrfs_device_fsid(dev_item), |
2b82032c YZ |
6606 | BTRFS_UUID_SIZE); |
6607 | ||
0b246afa | 6608 | if (memcmp(fs_uuid, fs_info->fsid, BTRFS_UUID_SIZE)) { |
5f375835 MX |
6609 | fs_devices = open_seed_devices(root, fs_uuid); |
6610 | if (IS_ERR(fs_devices)) | |
6611 | return PTR_ERR(fs_devices); | |
2b82032c YZ |
6612 | } |
6613 | ||
0b246afa | 6614 | device = btrfs_find_device(fs_info, devid, dev_uuid, fs_uuid); |
5f375835 | 6615 | if (!device) { |
0b246afa | 6616 | if (!btrfs_test_opt(fs_info, DEGRADED)) |
2b82032c YZ |
6617 | return -EIO; |
6618 | ||
5f375835 MX |
6619 | device = add_missing_dev(root, fs_devices, devid, dev_uuid); |
6620 | if (!device) | |
6621 | return -ENOMEM; | |
0b246afa | 6622 | btrfs_warn(fs_info, "devid %llu uuid %pU missing", |
33b97e43 | 6623 | devid, dev_uuid); |
5f375835 | 6624 | } else { |
0b246afa | 6625 | if (!device->bdev && !btrfs_test_opt(fs_info, DEGRADED)) |
5f375835 MX |
6626 | return -EIO; |
6627 | ||
6628 | if(!device->bdev && !device->missing) { | |
cd02dca5 CM |
6629 | /* |
6630 | * this happens when a device that was properly setup | |
6631 | * in the device info lists suddenly goes bad. | |
6632 | * device->bdev is NULL, and so we have to set | |
6633 | * device->missing to one here | |
6634 | */ | |
5f375835 | 6635 | device->fs_devices->missing_devices++; |
cd02dca5 | 6636 | device->missing = 1; |
2b82032c | 6637 | } |
5f375835 MX |
6638 | |
6639 | /* Move the device to its own fs_devices */ | |
6640 | if (device->fs_devices != fs_devices) { | |
6641 | ASSERT(device->missing); | |
6642 | ||
6643 | list_move(&device->dev_list, &fs_devices->devices); | |
6644 | device->fs_devices->num_devices--; | |
6645 | fs_devices->num_devices++; | |
6646 | ||
6647 | device->fs_devices->missing_devices--; | |
6648 | fs_devices->missing_devices++; | |
6649 | ||
6650 | device->fs_devices = fs_devices; | |
6651 | } | |
2b82032c YZ |
6652 | } |
6653 | ||
0b246afa | 6654 | if (device->fs_devices != fs_info->fs_devices) { |
2b82032c YZ |
6655 | BUG_ON(device->writeable); |
6656 | if (device->generation != | |
6657 | btrfs_device_generation(leaf, dev_item)) | |
6658 | return -EINVAL; | |
6324fbf3 | 6659 | } |
0b86a832 CM |
6660 | |
6661 | fill_device_from_item(leaf, dev_item, device); | |
dfe25020 | 6662 | device->in_fs_metadata = 1; |
63a212ab | 6663 | if (device->writeable && !device->is_tgtdev_for_dev_replace) { |
2b82032c | 6664 | device->fs_devices->total_rw_bytes += device->total_bytes; |
0b246afa JM |
6665 | spin_lock(&fs_info->free_chunk_lock); |
6666 | fs_info->free_chunk_space += device->total_bytes - | |
2bf64758 | 6667 | device->bytes_used; |
0b246afa | 6668 | spin_unlock(&fs_info->free_chunk_lock); |
2bf64758 | 6669 | } |
0b86a832 | 6670 | ret = 0; |
0b86a832 CM |
6671 | return ret; |
6672 | } | |
6673 | ||
6bccf3ab | 6674 | int btrfs_read_sys_array(struct btrfs_fs_info *fs_info) |
0b86a832 | 6675 | { |
6bccf3ab | 6676 | struct btrfs_root *root = fs_info->tree_root; |
ab8d0fc4 | 6677 | struct btrfs_super_block *super_copy = fs_info->super_copy; |
a061fc8d | 6678 | struct extent_buffer *sb; |
0b86a832 | 6679 | struct btrfs_disk_key *disk_key; |
0b86a832 | 6680 | struct btrfs_chunk *chunk; |
1ffb22cf DS |
6681 | u8 *array_ptr; |
6682 | unsigned long sb_array_offset; | |
84eed90f | 6683 | int ret = 0; |
0b86a832 CM |
6684 | u32 num_stripes; |
6685 | u32 array_size; | |
6686 | u32 len = 0; | |
1ffb22cf | 6687 | u32 cur_offset; |
e06cd3dd | 6688 | u64 type; |
84eed90f | 6689 | struct btrfs_key key; |
0b86a832 | 6690 | |
0b246afa | 6691 | ASSERT(BTRFS_SUPER_INFO_SIZE <= fs_info->nodesize); |
a83fffb7 DS |
6692 | /* |
6693 | * This will create extent buffer of nodesize, superblock size is | |
6694 | * fixed to BTRFS_SUPER_INFO_SIZE. If nodesize > sb size, this will | |
6695 | * overallocate but we can keep it as-is, only the first page is used. | |
6696 | */ | |
6697 | sb = btrfs_find_create_tree_block(root, BTRFS_SUPER_INFO_OFFSET); | |
c871b0f2 LB |
6698 | if (IS_ERR(sb)) |
6699 | return PTR_ERR(sb); | |
4db8c528 | 6700 | set_extent_buffer_uptodate(sb); |
85d4e461 | 6701 | btrfs_set_buffer_lockdep_class(root->root_key.objectid, sb, 0); |
8a334426 | 6702 | /* |
01327610 | 6703 | * The sb extent buffer is artificial and just used to read the system array. |
4db8c528 | 6704 | * set_extent_buffer_uptodate() call does not properly mark all it's |
8a334426 DS |
6705 | * pages up-to-date when the page is larger: extent does not cover the |
6706 | * whole page and consequently check_page_uptodate does not find all | |
6707 | * the page's extents up-to-date (the hole beyond sb), | |
6708 | * write_extent_buffer then triggers a WARN_ON. | |
6709 | * | |
6710 | * Regular short extents go through mark_extent_buffer_dirty/writeback cycle, | |
6711 | * but sb spans only this function. Add an explicit SetPageUptodate call | |
6712 | * to silence the warning eg. on PowerPC 64. | |
6713 | */ | |
09cbfeaf | 6714 | if (PAGE_SIZE > BTRFS_SUPER_INFO_SIZE) |
727011e0 | 6715 | SetPageUptodate(sb->pages[0]); |
4008c04a | 6716 | |
a061fc8d | 6717 | write_extent_buffer(sb, super_copy, 0, BTRFS_SUPER_INFO_SIZE); |
0b86a832 CM |
6718 | array_size = btrfs_super_sys_array_size(super_copy); |
6719 | ||
1ffb22cf DS |
6720 | array_ptr = super_copy->sys_chunk_array; |
6721 | sb_array_offset = offsetof(struct btrfs_super_block, sys_chunk_array); | |
6722 | cur_offset = 0; | |
0b86a832 | 6723 | |
1ffb22cf DS |
6724 | while (cur_offset < array_size) { |
6725 | disk_key = (struct btrfs_disk_key *)array_ptr; | |
e3540eab DS |
6726 | len = sizeof(*disk_key); |
6727 | if (cur_offset + len > array_size) | |
6728 | goto out_short_read; | |
6729 | ||
0b86a832 CM |
6730 | btrfs_disk_key_to_cpu(&key, disk_key); |
6731 | ||
1ffb22cf DS |
6732 | array_ptr += len; |
6733 | sb_array_offset += len; | |
6734 | cur_offset += len; | |
0b86a832 | 6735 | |
0d81ba5d | 6736 | if (key.type == BTRFS_CHUNK_ITEM_KEY) { |
1ffb22cf | 6737 | chunk = (struct btrfs_chunk *)sb_array_offset; |
e3540eab DS |
6738 | /* |
6739 | * At least one btrfs_chunk with one stripe must be | |
6740 | * present, exact stripe count check comes afterwards | |
6741 | */ | |
6742 | len = btrfs_chunk_item_size(1); | |
6743 | if (cur_offset + len > array_size) | |
6744 | goto out_short_read; | |
6745 | ||
6746 | num_stripes = btrfs_chunk_num_stripes(sb, chunk); | |
f5cdedd7 | 6747 | if (!num_stripes) { |
ab8d0fc4 JM |
6748 | btrfs_err(fs_info, |
6749 | "invalid number of stripes %u in sys_array at offset %u", | |
f5cdedd7 DS |
6750 | num_stripes, cur_offset); |
6751 | ret = -EIO; | |
6752 | break; | |
6753 | } | |
6754 | ||
e06cd3dd LB |
6755 | type = btrfs_chunk_type(sb, chunk); |
6756 | if ((type & BTRFS_BLOCK_GROUP_SYSTEM) == 0) { | |
ab8d0fc4 | 6757 | btrfs_err(fs_info, |
e06cd3dd LB |
6758 | "invalid chunk type %llu in sys_array at offset %u", |
6759 | type, cur_offset); | |
6760 | ret = -EIO; | |
6761 | break; | |
6762 | } | |
6763 | ||
e3540eab DS |
6764 | len = btrfs_chunk_item_size(num_stripes); |
6765 | if (cur_offset + len > array_size) | |
6766 | goto out_short_read; | |
6767 | ||
0d81ba5d | 6768 | ret = read_one_chunk(root, &key, sb, chunk); |
84eed90f CM |
6769 | if (ret) |
6770 | break; | |
0b86a832 | 6771 | } else { |
ab8d0fc4 JM |
6772 | btrfs_err(fs_info, |
6773 | "unexpected item type %u in sys_array at offset %u", | |
6774 | (u32)key.type, cur_offset); | |
84eed90f CM |
6775 | ret = -EIO; |
6776 | break; | |
0b86a832 | 6777 | } |
1ffb22cf DS |
6778 | array_ptr += len; |
6779 | sb_array_offset += len; | |
6780 | cur_offset += len; | |
0b86a832 | 6781 | } |
d865177a | 6782 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6783 | free_extent_buffer_stale(sb); |
84eed90f | 6784 | return ret; |
e3540eab DS |
6785 | |
6786 | out_short_read: | |
ab8d0fc4 | 6787 | btrfs_err(fs_info, "sys_array too short to read %u bytes at offset %u", |
e3540eab | 6788 | len, cur_offset); |
d865177a | 6789 | clear_extent_buffer_uptodate(sb); |
1c8b5b6e | 6790 | free_extent_buffer_stale(sb); |
e3540eab | 6791 | return -EIO; |
0b86a832 CM |
6792 | } |
6793 | ||
5b4aacef | 6794 | int btrfs_read_chunk_tree(struct btrfs_fs_info *fs_info) |
0b86a832 | 6795 | { |
5b4aacef | 6796 | struct btrfs_root *root = fs_info->chunk_root; |
0b86a832 CM |
6797 | struct btrfs_path *path; |
6798 | struct extent_buffer *leaf; | |
6799 | struct btrfs_key key; | |
6800 | struct btrfs_key found_key; | |
6801 | int ret; | |
6802 | int slot; | |
99e3ecfc | 6803 | u64 total_dev = 0; |
0b86a832 | 6804 | |
0b86a832 CM |
6805 | path = btrfs_alloc_path(); |
6806 | if (!path) | |
6807 | return -ENOMEM; | |
6808 | ||
b367e47f | 6809 | mutex_lock(&uuid_mutex); |
0b246afa | 6810 | lock_chunks(fs_info); |
b367e47f | 6811 | |
395927a9 FDBM |
6812 | /* |
6813 | * Read all device items, and then all the chunk items. All | |
6814 | * device items are found before any chunk item (their object id | |
6815 | * is smaller than the lowest possible object id for a chunk | |
6816 | * item - BTRFS_FIRST_CHUNK_TREE_OBJECTID). | |
0b86a832 CM |
6817 | */ |
6818 | key.objectid = BTRFS_DEV_ITEMS_OBJECTID; | |
6819 | key.offset = 0; | |
6820 | key.type = 0; | |
0b86a832 | 6821 | ret = btrfs_search_slot(NULL, root, &key, path, 0, 0); |
ab59381e ZL |
6822 | if (ret < 0) |
6823 | goto error; | |
d397712b | 6824 | while (1) { |
0b86a832 CM |
6825 | leaf = path->nodes[0]; |
6826 | slot = path->slots[0]; | |
6827 | if (slot >= btrfs_header_nritems(leaf)) { | |
6828 | ret = btrfs_next_leaf(root, path); | |
6829 | if (ret == 0) | |
6830 | continue; | |
6831 | if (ret < 0) | |
6832 | goto error; | |
6833 | break; | |
6834 | } | |
6835 | btrfs_item_key_to_cpu(leaf, &found_key, slot); | |
395927a9 FDBM |
6836 | if (found_key.type == BTRFS_DEV_ITEM_KEY) { |
6837 | struct btrfs_dev_item *dev_item; | |
6838 | dev_item = btrfs_item_ptr(leaf, slot, | |
0b86a832 | 6839 | struct btrfs_dev_item); |
395927a9 FDBM |
6840 | ret = read_one_dev(root, leaf, dev_item); |
6841 | if (ret) | |
6842 | goto error; | |
99e3ecfc | 6843 | total_dev++; |
0b86a832 CM |
6844 | } else if (found_key.type == BTRFS_CHUNK_ITEM_KEY) { |
6845 | struct btrfs_chunk *chunk; | |
6846 | chunk = btrfs_item_ptr(leaf, slot, struct btrfs_chunk); | |
6847 | ret = read_one_chunk(root, &found_key, leaf, chunk); | |
2b82032c YZ |
6848 | if (ret) |
6849 | goto error; | |
0b86a832 CM |
6850 | } |
6851 | path->slots[0]++; | |
6852 | } | |
99e3ecfc LB |
6853 | |
6854 | /* | |
6855 | * After loading chunk tree, we've got all device information, | |
6856 | * do another round of validation checks. | |
6857 | */ | |
0b246afa JM |
6858 | if (total_dev != fs_info->fs_devices->total_devices) { |
6859 | btrfs_err(fs_info, | |
99e3ecfc | 6860 | "super_num_devices %llu mismatch with num_devices %llu found here", |
0b246afa | 6861 | btrfs_super_num_devices(fs_info->super_copy), |
99e3ecfc LB |
6862 | total_dev); |
6863 | ret = -EINVAL; | |
6864 | goto error; | |
6865 | } | |
0b246afa JM |
6866 | if (btrfs_super_total_bytes(fs_info->super_copy) < |
6867 | fs_info->fs_devices->total_rw_bytes) { | |
6868 | btrfs_err(fs_info, | |
99e3ecfc | 6869 | "super_total_bytes %llu mismatch with fs_devices total_rw_bytes %llu", |
0b246afa JM |
6870 | btrfs_super_total_bytes(fs_info->super_copy), |
6871 | fs_info->fs_devices->total_rw_bytes); | |
99e3ecfc LB |
6872 | ret = -EINVAL; |
6873 | goto error; | |
6874 | } | |
0b86a832 CM |
6875 | ret = 0; |
6876 | error: | |
0b246afa | 6877 | unlock_chunks(fs_info); |
b367e47f LZ |
6878 | mutex_unlock(&uuid_mutex); |
6879 | ||
2b82032c | 6880 | btrfs_free_path(path); |
0b86a832 CM |
6881 | return ret; |
6882 | } | |
442a4f63 | 6883 | |
cb517eab MX |
6884 | void btrfs_init_devices_late(struct btrfs_fs_info *fs_info) |
6885 | { | |
6886 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6887 | struct btrfs_device *device; | |
6888 | ||
29cc83f6 LB |
6889 | while (fs_devices) { |
6890 | mutex_lock(&fs_devices->device_list_mutex); | |
6891 | list_for_each_entry(device, &fs_devices->devices, dev_list) | |
fb456252 | 6892 | device->fs_info = fs_info; |
29cc83f6 LB |
6893 | mutex_unlock(&fs_devices->device_list_mutex); |
6894 | ||
6895 | fs_devices = fs_devices->seed; | |
6896 | } | |
cb517eab MX |
6897 | } |
6898 | ||
733f4fbb SB |
6899 | static void __btrfs_reset_dev_stats(struct btrfs_device *dev) |
6900 | { | |
6901 | int i; | |
6902 | ||
6903 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
6904 | btrfs_dev_stat_reset(dev, i); | |
6905 | } | |
6906 | ||
6907 | int btrfs_init_dev_stats(struct btrfs_fs_info *fs_info) | |
6908 | { | |
6909 | struct btrfs_key key; | |
6910 | struct btrfs_key found_key; | |
6911 | struct btrfs_root *dev_root = fs_info->dev_root; | |
6912 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
6913 | struct extent_buffer *eb; | |
6914 | int slot; | |
6915 | int ret = 0; | |
6916 | struct btrfs_device *device; | |
6917 | struct btrfs_path *path = NULL; | |
6918 | int i; | |
6919 | ||
6920 | path = btrfs_alloc_path(); | |
6921 | if (!path) { | |
6922 | ret = -ENOMEM; | |
6923 | goto out; | |
6924 | } | |
6925 | ||
6926 | mutex_lock(&fs_devices->device_list_mutex); | |
6927 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
6928 | int item_size; | |
6929 | struct btrfs_dev_stats_item *ptr; | |
6930 | ||
242e2956 DS |
6931 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6932 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6933 | key.offset = device->devid; |
6934 | ret = btrfs_search_slot(NULL, dev_root, &key, path, 0, 0); | |
6935 | if (ret) { | |
733f4fbb SB |
6936 | __btrfs_reset_dev_stats(device); |
6937 | device->dev_stats_valid = 1; | |
6938 | btrfs_release_path(path); | |
6939 | continue; | |
6940 | } | |
6941 | slot = path->slots[0]; | |
6942 | eb = path->nodes[0]; | |
6943 | btrfs_item_key_to_cpu(eb, &found_key, slot); | |
6944 | item_size = btrfs_item_size_nr(eb, slot); | |
6945 | ||
6946 | ptr = btrfs_item_ptr(eb, slot, | |
6947 | struct btrfs_dev_stats_item); | |
6948 | ||
6949 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { | |
6950 | if (item_size >= (1 + i) * sizeof(__le64)) | |
6951 | btrfs_dev_stat_set(device, i, | |
6952 | btrfs_dev_stats_value(eb, ptr, i)); | |
6953 | else | |
6954 | btrfs_dev_stat_reset(device, i); | |
6955 | } | |
6956 | ||
6957 | device->dev_stats_valid = 1; | |
6958 | btrfs_dev_stat_print_on_load(device); | |
6959 | btrfs_release_path(path); | |
6960 | } | |
6961 | mutex_unlock(&fs_devices->device_list_mutex); | |
6962 | ||
6963 | out: | |
6964 | btrfs_free_path(path); | |
6965 | return ret < 0 ? ret : 0; | |
6966 | } | |
6967 | ||
6968 | static int update_dev_stat_item(struct btrfs_trans_handle *trans, | |
6bccf3ab | 6969 | struct btrfs_fs_info *fs_info, |
733f4fbb SB |
6970 | struct btrfs_device *device) |
6971 | { | |
6bccf3ab | 6972 | struct btrfs_root *dev_root = fs_info->dev_root; |
733f4fbb SB |
6973 | struct btrfs_path *path; |
6974 | struct btrfs_key key; | |
6975 | struct extent_buffer *eb; | |
6976 | struct btrfs_dev_stats_item *ptr; | |
6977 | int ret; | |
6978 | int i; | |
6979 | ||
242e2956 DS |
6980 | key.objectid = BTRFS_DEV_STATS_OBJECTID; |
6981 | key.type = BTRFS_PERSISTENT_ITEM_KEY; | |
733f4fbb SB |
6982 | key.offset = device->devid; |
6983 | ||
6984 | path = btrfs_alloc_path(); | |
6985 | BUG_ON(!path); | |
6986 | ret = btrfs_search_slot(trans, dev_root, &key, path, -1, 1); | |
6987 | if (ret < 0) { | |
0b246afa | 6988 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 6989 | "error %d while searching for dev_stats item for device %s", |
606686ee | 6990 | ret, rcu_str_deref(device->name)); |
733f4fbb SB |
6991 | goto out; |
6992 | } | |
6993 | ||
6994 | if (ret == 0 && | |
6995 | btrfs_item_size_nr(path->nodes[0], path->slots[0]) < sizeof(*ptr)) { | |
6996 | /* need to delete old one and insert a new one */ | |
6997 | ret = btrfs_del_item(trans, dev_root, path); | |
6998 | if (ret != 0) { | |
0b246afa | 6999 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b | 7000 | "delete too small dev_stats item for device %s failed %d", |
606686ee | 7001 | rcu_str_deref(device->name), ret); |
733f4fbb SB |
7002 | goto out; |
7003 | } | |
7004 | ret = 1; | |
7005 | } | |
7006 | ||
7007 | if (ret == 1) { | |
7008 | /* need to insert a new item */ | |
7009 | btrfs_release_path(path); | |
7010 | ret = btrfs_insert_empty_item(trans, dev_root, path, | |
7011 | &key, sizeof(*ptr)); | |
7012 | if (ret < 0) { | |
0b246afa | 7013 | btrfs_warn_in_rcu(fs_info, |
ecaeb14b DS |
7014 | "insert dev_stats item for device %s failed %d", |
7015 | rcu_str_deref(device->name), ret); | |
733f4fbb SB |
7016 | goto out; |
7017 | } | |
7018 | } | |
7019 | ||
7020 | eb = path->nodes[0]; | |
7021 | ptr = btrfs_item_ptr(eb, path->slots[0], struct btrfs_dev_stats_item); | |
7022 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7023 | btrfs_set_dev_stats_value(eb, ptr, i, | |
7024 | btrfs_dev_stat_read(device, i)); | |
7025 | btrfs_mark_buffer_dirty(eb); | |
7026 | ||
7027 | out: | |
7028 | btrfs_free_path(path); | |
7029 | return ret; | |
7030 | } | |
7031 | ||
7032 | /* | |
7033 | * called from commit_transaction. Writes all changed device stats to disk. | |
7034 | */ | |
7035 | int btrfs_run_dev_stats(struct btrfs_trans_handle *trans, | |
7036 | struct btrfs_fs_info *fs_info) | |
7037 | { | |
733f4fbb SB |
7038 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
7039 | struct btrfs_device *device; | |
addc3fa7 | 7040 | int stats_cnt; |
733f4fbb SB |
7041 | int ret = 0; |
7042 | ||
7043 | mutex_lock(&fs_devices->device_list_mutex); | |
7044 | list_for_each_entry(device, &fs_devices->devices, dev_list) { | |
addc3fa7 | 7045 | if (!device->dev_stats_valid || !btrfs_dev_stats_dirty(device)) |
733f4fbb SB |
7046 | continue; |
7047 | ||
addc3fa7 | 7048 | stats_cnt = atomic_read(&device->dev_stats_ccnt); |
6bccf3ab | 7049 | ret = update_dev_stat_item(trans, fs_info, device); |
733f4fbb | 7050 | if (!ret) |
addc3fa7 | 7051 | atomic_sub(stats_cnt, &device->dev_stats_ccnt); |
733f4fbb SB |
7052 | } |
7053 | mutex_unlock(&fs_devices->device_list_mutex); | |
7054 | ||
7055 | return ret; | |
7056 | } | |
7057 | ||
442a4f63 SB |
7058 | void btrfs_dev_stat_inc_and_print(struct btrfs_device *dev, int index) |
7059 | { | |
7060 | btrfs_dev_stat_inc(dev, index); | |
7061 | btrfs_dev_stat_print_on_error(dev); | |
7062 | } | |
7063 | ||
48a3b636 | 7064 | static void btrfs_dev_stat_print_on_error(struct btrfs_device *dev) |
442a4f63 | 7065 | { |
733f4fbb SB |
7066 | if (!dev->dev_stats_valid) |
7067 | return; | |
fb456252 | 7068 | btrfs_err_rl_in_rcu(dev->fs_info, |
b14af3b4 | 7069 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7070 | rcu_str_deref(dev->name), |
442a4f63 SB |
7071 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7072 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7073 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
efe120a0 FH |
7074 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), |
7075 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
442a4f63 | 7076 | } |
c11d2c23 | 7077 | |
733f4fbb SB |
7078 | static void btrfs_dev_stat_print_on_load(struct btrfs_device *dev) |
7079 | { | |
a98cdb85 SB |
7080 | int i; |
7081 | ||
7082 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7083 | if (btrfs_dev_stat_read(dev, i) != 0) | |
7084 | break; | |
7085 | if (i == BTRFS_DEV_STAT_VALUES_MAX) | |
7086 | return; /* all values == 0, suppress message */ | |
7087 | ||
fb456252 | 7088 | btrfs_info_in_rcu(dev->fs_info, |
ecaeb14b | 7089 | "bdev %s errs: wr %u, rd %u, flush %u, corrupt %u, gen %u", |
606686ee | 7090 | rcu_str_deref(dev->name), |
733f4fbb SB |
7091 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_WRITE_ERRS), |
7092 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_READ_ERRS), | |
7093 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_FLUSH_ERRS), | |
7094 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_CORRUPTION_ERRS), | |
7095 | btrfs_dev_stat_read(dev, BTRFS_DEV_STAT_GENERATION_ERRS)); | |
7096 | } | |
7097 | ||
c11d2c23 | 7098 | int btrfs_get_dev_stats(struct btrfs_root *root, |
b27f7c0c | 7099 | struct btrfs_ioctl_get_dev_stats *stats) |
c11d2c23 | 7100 | { |
0b246afa | 7101 | struct btrfs_fs_info *fs_info = root->fs_info; |
c11d2c23 | 7102 | struct btrfs_device *dev; |
0b246afa | 7103 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; |
c11d2c23 SB |
7104 | int i; |
7105 | ||
7106 | mutex_lock(&fs_devices->device_list_mutex); | |
0b246afa | 7107 | dev = btrfs_find_device(fs_info, stats->devid, NULL, NULL); |
c11d2c23 SB |
7108 | mutex_unlock(&fs_devices->device_list_mutex); |
7109 | ||
7110 | if (!dev) { | |
0b246afa | 7111 | btrfs_warn(fs_info, "get dev_stats failed, device not found"); |
c11d2c23 | 7112 | return -ENODEV; |
733f4fbb | 7113 | } else if (!dev->dev_stats_valid) { |
0b246afa | 7114 | btrfs_warn(fs_info, "get dev_stats failed, not yet valid"); |
733f4fbb | 7115 | return -ENODEV; |
b27f7c0c | 7116 | } else if (stats->flags & BTRFS_DEV_STATS_RESET) { |
c11d2c23 SB |
7117 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) { |
7118 | if (stats->nr_items > i) | |
7119 | stats->values[i] = | |
7120 | btrfs_dev_stat_read_and_reset(dev, i); | |
7121 | else | |
7122 | btrfs_dev_stat_reset(dev, i); | |
7123 | } | |
7124 | } else { | |
7125 | for (i = 0; i < BTRFS_DEV_STAT_VALUES_MAX; i++) | |
7126 | if (stats->nr_items > i) | |
7127 | stats->values[i] = btrfs_dev_stat_read(dev, i); | |
7128 | } | |
7129 | if (stats->nr_items > BTRFS_DEV_STAT_VALUES_MAX) | |
7130 | stats->nr_items = BTRFS_DEV_STAT_VALUES_MAX; | |
7131 | return 0; | |
7132 | } | |
a8a6dab7 | 7133 | |
12b1c263 | 7134 | void btrfs_scratch_superblocks(struct block_device *bdev, char *device_path) |
a8a6dab7 SB |
7135 | { |
7136 | struct buffer_head *bh; | |
7137 | struct btrfs_super_block *disk_super; | |
12b1c263 | 7138 | int copy_num; |
a8a6dab7 | 7139 | |
12b1c263 AJ |
7140 | if (!bdev) |
7141 | return; | |
a8a6dab7 | 7142 | |
12b1c263 AJ |
7143 | for (copy_num = 0; copy_num < BTRFS_SUPER_MIRROR_MAX; |
7144 | copy_num++) { | |
a8a6dab7 | 7145 | |
12b1c263 AJ |
7146 | if (btrfs_read_dev_one_super(bdev, copy_num, &bh)) |
7147 | continue; | |
7148 | ||
7149 | disk_super = (struct btrfs_super_block *)bh->b_data; | |
7150 | ||
7151 | memset(&disk_super->magic, 0, sizeof(disk_super->magic)); | |
7152 | set_buffer_dirty(bh); | |
7153 | sync_dirty_buffer(bh); | |
7154 | brelse(bh); | |
7155 | } | |
7156 | ||
7157 | /* Notify udev that device has changed */ | |
7158 | btrfs_kobject_uevent(bdev, KOBJ_CHANGE); | |
7159 | ||
7160 | /* Update ctime/mtime for device path for libblkid */ | |
7161 | update_dev_time(device_path); | |
a8a6dab7 | 7162 | } |
935e5cc9 MX |
7163 | |
7164 | /* | |
7165 | * Update the size of all devices, which is used for writing out the | |
7166 | * super blocks. | |
7167 | */ | |
7168 | void btrfs_update_commit_device_size(struct btrfs_fs_info *fs_info) | |
7169 | { | |
7170 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7171 | struct btrfs_device *curr, *next; | |
7172 | ||
7173 | if (list_empty(&fs_devices->resized_devices)) | |
7174 | return; | |
7175 | ||
7176 | mutex_lock(&fs_devices->device_list_mutex); | |
3796d335 | 7177 | lock_chunks(fs_info); |
935e5cc9 MX |
7178 | list_for_each_entry_safe(curr, next, &fs_devices->resized_devices, |
7179 | resized_list) { | |
7180 | list_del_init(&curr->resized_list); | |
7181 | curr->commit_total_bytes = curr->disk_total_bytes; | |
7182 | } | |
3796d335 | 7183 | unlock_chunks(fs_info); |
935e5cc9 MX |
7184 | mutex_unlock(&fs_devices->device_list_mutex); |
7185 | } | |
ce7213c7 MX |
7186 | |
7187 | /* Must be invoked during the transaction commit */ | |
7188 | void btrfs_update_commit_device_bytes_used(struct btrfs_root *root, | |
7189 | struct btrfs_transaction *transaction) | |
7190 | { | |
0b246afa | 7191 | struct btrfs_fs_info *fs_info = root->fs_info; |
ce7213c7 MX |
7192 | struct extent_map *em; |
7193 | struct map_lookup *map; | |
7194 | struct btrfs_device *dev; | |
7195 | int i; | |
7196 | ||
7197 | if (list_empty(&transaction->pending_chunks)) | |
7198 | return; | |
7199 | ||
7200 | /* In order to kick the device replace finish process */ | |
0b246afa | 7201 | lock_chunks(fs_info); |
ce7213c7 | 7202 | list_for_each_entry(em, &transaction->pending_chunks, list) { |
95617d69 | 7203 | map = em->map_lookup; |
ce7213c7 MX |
7204 | |
7205 | for (i = 0; i < map->num_stripes; i++) { | |
7206 | dev = map->stripes[i].dev; | |
7207 | dev->commit_bytes_used = dev->bytes_used; | |
7208 | } | |
7209 | } | |
0b246afa | 7210 | unlock_chunks(fs_info); |
ce7213c7 | 7211 | } |
5a13f430 AJ |
7212 | |
7213 | void btrfs_set_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7214 | { | |
7215 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7216 | while (fs_devices) { | |
7217 | fs_devices->fs_info = fs_info; | |
7218 | fs_devices = fs_devices->seed; | |
7219 | } | |
7220 | } | |
7221 | ||
7222 | void btrfs_reset_fs_info_ptr(struct btrfs_fs_info *fs_info) | |
7223 | { | |
7224 | struct btrfs_fs_devices *fs_devices = fs_info->fs_devices; | |
7225 | while (fs_devices) { | |
7226 | fs_devices->fs_info = NULL; | |
7227 | fs_devices = fs_devices->seed; | |
7228 | } | |
7229 | } |