]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
fd2ed4d2 | 63 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
64 | }; |
65 | ||
8fbf26ad KU |
66 | /* |
67 | * For request-based dm. | |
68 | * One of these is allocated per request. | |
69 | */ | |
70 | struct dm_rq_target_io { | |
71 | struct mapped_device *md; | |
72 | struct dm_target *ti; | |
73 | struct request *orig, clone; | |
74 | int error; | |
75 | union map_info info; | |
76 | }; | |
77 | ||
78 | /* | |
94818742 KO |
79 | * For request-based dm - the bio clones we allocate are embedded in these |
80 | * structs. | |
81 | * | |
82 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
83 | * the bioset is created - this means the bio has to come at the end of the | |
84 | * struct. | |
8fbf26ad KU |
85 | */ |
86 | struct dm_rq_clone_bio_info { | |
87 | struct bio *orig; | |
cec47e3d | 88 | struct dm_rq_target_io *tio; |
94818742 | 89 | struct bio clone; |
8fbf26ad KU |
90 | }; |
91 | ||
1da177e4 LT |
92 | union map_info *dm_get_mapinfo(struct bio *bio) |
93 | { | |
17b2f66f | 94 | if (bio && bio->bi_private) |
028867ac | 95 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 96 | return NULL; |
1da177e4 LT |
97 | } |
98 | ||
cec47e3d KU |
99 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
100 | { | |
101 | if (rq && rq->end_io_data) | |
102 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
103 | return NULL; | |
104 | } | |
105 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
106 | ||
ba61fdd1 JM |
107 | #define MINOR_ALLOCED ((void *)-1) |
108 | ||
1da177e4 LT |
109 | /* |
110 | * Bits for the md->flags field. | |
111 | */ | |
1eb787ec | 112 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 113 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 114 | #define DMF_FROZEN 2 |
fba9f90e | 115 | #define DMF_FREEING 3 |
5c6bd75d | 116 | #define DMF_DELETING 4 |
2e93ccc1 | 117 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 118 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 119 | |
83d5e5b0 MP |
120 | /* |
121 | * A dummy definition to make RCU happy. | |
122 | * struct dm_table should never be dereferenced in this file. | |
123 | */ | |
124 | struct dm_table { | |
125 | int undefined__; | |
126 | }; | |
127 | ||
304f3f6a MB |
128 | /* |
129 | * Work processed by per-device workqueue. | |
130 | */ | |
1da177e4 | 131 | struct mapped_device { |
83d5e5b0 | 132 | struct srcu_struct io_barrier; |
e61290a4 | 133 | struct mutex suspend_lock; |
1da177e4 | 134 | atomic_t holders; |
5c6bd75d | 135 | atomic_t open_count; |
1da177e4 | 136 | |
2a7faeb1 MP |
137 | /* |
138 | * The current mapping. | |
139 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
140 | * dereference. | |
141 | */ | |
142 | struct dm_table *map; | |
143 | ||
1da177e4 LT |
144 | unsigned long flags; |
145 | ||
165125e1 | 146 | struct request_queue *queue; |
a5664dad | 147 | unsigned type; |
4a0b4ddf | 148 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
149 | struct mutex type_lock; |
150 | ||
36a0456f AK |
151 | struct target_type *immutable_target_type; |
152 | ||
1da177e4 | 153 | struct gendisk *disk; |
7e51f257 | 154 | char name[16]; |
1da177e4 LT |
155 | |
156 | void *interface_ptr; | |
157 | ||
158 | /* | |
159 | * A list of ios that arrived while we were suspended. | |
160 | */ | |
316d315b | 161 | atomic_t pending[2]; |
1da177e4 | 162 | wait_queue_head_t wait; |
53d5914f | 163 | struct work_struct work; |
74859364 | 164 | struct bio_list deferred; |
022c2611 | 165 | spinlock_t deferred_lock; |
1da177e4 | 166 | |
af7e466a | 167 | /* |
29e4013d | 168 | * Processing queue (flush) |
304f3f6a MB |
169 | */ |
170 | struct workqueue_struct *wq; | |
171 | ||
1da177e4 LT |
172 | /* |
173 | * io objects are allocated from here. | |
174 | */ | |
175 | mempool_t *io_pool; | |
1da177e4 | 176 | |
9faf400f SB |
177 | struct bio_set *bs; |
178 | ||
1da177e4 LT |
179 | /* |
180 | * Event handling. | |
181 | */ | |
182 | atomic_t event_nr; | |
183 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
184 | atomic_t uevent_seq; |
185 | struct list_head uevent_list; | |
186 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
187 | |
188 | /* | |
189 | * freeze/thaw support require holding onto a super block | |
190 | */ | |
191 | struct super_block *frozen_sb; | |
db8fef4f | 192 | struct block_device *bdev; |
3ac51e74 DW |
193 | |
194 | /* forced geometry settings */ | |
195 | struct hd_geometry geometry; | |
784aae73 MB |
196 | |
197 | /* sysfs handle */ | |
198 | struct kobject kobj; | |
52b1fd5a | 199 | |
d87f4c14 TH |
200 | /* zero-length flush that will be cloned and submitted to targets */ |
201 | struct bio flush_bio; | |
fd2ed4d2 MP |
202 | |
203 | struct dm_stats stats; | |
1da177e4 LT |
204 | }; |
205 | ||
e6ee8c0b KU |
206 | /* |
207 | * For mempools pre-allocation at the table loading time. | |
208 | */ | |
209 | struct dm_md_mempools { | |
210 | mempool_t *io_pool; | |
e6ee8c0b KU |
211 | struct bio_set *bs; |
212 | }; | |
213 | ||
6cfa5857 MS |
214 | #define RESERVED_BIO_BASED_IOS 16 |
215 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 216 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 217 | static struct kmem_cache *_io_cache; |
8fbf26ad | 218 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 219 | |
e8603136 MS |
220 | /* |
221 | * Bio-based DM's mempools' reserved IOs set by the user. | |
222 | */ | |
223 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
224 | ||
f4790826 MS |
225 | /* |
226 | * Request-based DM's mempools' reserved IOs set by the user. | |
227 | */ | |
228 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
229 | ||
230 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
231 | unsigned def, unsigned max) | |
232 | { | |
233 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
234 | unsigned modified_ios = 0; | |
235 | ||
236 | if (!ios) | |
237 | modified_ios = def; | |
238 | else if (ios > max) | |
239 | modified_ios = max; | |
240 | ||
241 | if (modified_ios) { | |
242 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
243 | ios = modified_ios; | |
244 | } | |
245 | ||
246 | return ios; | |
247 | } | |
248 | ||
e8603136 MS |
249 | unsigned dm_get_reserved_bio_based_ios(void) |
250 | { | |
251 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
252 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
253 | } | |
254 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
255 | ||
f4790826 MS |
256 | unsigned dm_get_reserved_rq_based_ios(void) |
257 | { | |
258 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
259 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
260 | } | |
261 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
262 | ||
1da177e4 LT |
263 | static int __init local_init(void) |
264 | { | |
51157b4a | 265 | int r = -ENOMEM; |
1da177e4 | 266 | |
1da177e4 | 267 | /* allocate a slab for the dm_ios */ |
028867ac | 268 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 269 | if (!_io_cache) |
51157b4a | 270 | return r; |
1da177e4 | 271 | |
8fbf26ad KU |
272 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
273 | if (!_rq_tio_cache) | |
dba14160 | 274 | goto out_free_io_cache; |
8fbf26ad | 275 | |
51e5b2bd | 276 | r = dm_uevent_init(); |
51157b4a | 277 | if (r) |
23e5083b | 278 | goto out_free_rq_tio_cache; |
51e5b2bd | 279 | |
1da177e4 LT |
280 | _major = major; |
281 | r = register_blkdev(_major, _name); | |
51157b4a KU |
282 | if (r < 0) |
283 | goto out_uevent_exit; | |
1da177e4 LT |
284 | |
285 | if (!_major) | |
286 | _major = r; | |
287 | ||
288 | return 0; | |
51157b4a KU |
289 | |
290 | out_uevent_exit: | |
291 | dm_uevent_exit(); | |
8fbf26ad KU |
292 | out_free_rq_tio_cache: |
293 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
294 | out_free_io_cache: |
295 | kmem_cache_destroy(_io_cache); | |
296 | ||
297 | return r; | |
1da177e4 LT |
298 | } |
299 | ||
300 | static void local_exit(void) | |
301 | { | |
8fbf26ad | 302 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 303 | kmem_cache_destroy(_io_cache); |
00d59405 | 304 | unregister_blkdev(_major, _name); |
51e5b2bd | 305 | dm_uevent_exit(); |
1da177e4 LT |
306 | |
307 | _major = 0; | |
308 | ||
309 | DMINFO("cleaned up"); | |
310 | } | |
311 | ||
b9249e55 | 312 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
313 | local_init, |
314 | dm_target_init, | |
315 | dm_linear_init, | |
316 | dm_stripe_init, | |
952b3557 | 317 | dm_io_init, |
945fa4d2 | 318 | dm_kcopyd_init, |
1da177e4 | 319 | dm_interface_init, |
fd2ed4d2 | 320 | dm_statistics_init, |
1da177e4 LT |
321 | }; |
322 | ||
b9249e55 | 323 | static void (*_exits[])(void) = { |
1da177e4 LT |
324 | local_exit, |
325 | dm_target_exit, | |
326 | dm_linear_exit, | |
327 | dm_stripe_exit, | |
952b3557 | 328 | dm_io_exit, |
945fa4d2 | 329 | dm_kcopyd_exit, |
1da177e4 | 330 | dm_interface_exit, |
fd2ed4d2 | 331 | dm_statistics_exit, |
1da177e4 LT |
332 | }; |
333 | ||
334 | static int __init dm_init(void) | |
335 | { | |
336 | const int count = ARRAY_SIZE(_inits); | |
337 | ||
338 | int r, i; | |
339 | ||
340 | for (i = 0; i < count; i++) { | |
341 | r = _inits[i](); | |
342 | if (r) | |
343 | goto bad; | |
344 | } | |
345 | ||
346 | return 0; | |
347 | ||
348 | bad: | |
349 | while (i--) | |
350 | _exits[i](); | |
351 | ||
352 | return r; | |
353 | } | |
354 | ||
355 | static void __exit dm_exit(void) | |
356 | { | |
357 | int i = ARRAY_SIZE(_exits); | |
358 | ||
359 | while (i--) | |
360 | _exits[i](); | |
d15b774c AK |
361 | |
362 | /* | |
363 | * Should be empty by this point. | |
364 | */ | |
d15b774c | 365 | idr_destroy(&_minor_idr); |
1da177e4 LT |
366 | } |
367 | ||
368 | /* | |
369 | * Block device functions | |
370 | */ | |
432a212c MA |
371 | int dm_deleting_md(struct mapped_device *md) |
372 | { | |
373 | return test_bit(DMF_DELETING, &md->flags); | |
374 | } | |
375 | ||
fe5f9f2c | 376 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
377 | { |
378 | struct mapped_device *md; | |
379 | ||
fba9f90e JM |
380 | spin_lock(&_minor_lock); |
381 | ||
fe5f9f2c | 382 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
383 | if (!md) |
384 | goto out; | |
385 | ||
5c6bd75d | 386 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 387 | dm_deleting_md(md)) { |
fba9f90e JM |
388 | md = NULL; |
389 | goto out; | |
390 | } | |
391 | ||
1da177e4 | 392 | dm_get(md); |
5c6bd75d | 393 | atomic_inc(&md->open_count); |
fba9f90e JM |
394 | |
395 | out: | |
396 | spin_unlock(&_minor_lock); | |
397 | ||
398 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
399 | } |
400 | ||
db2a144b | 401 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 402 | { |
fe5f9f2c | 403 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 404 | |
4a1aeb98 MB |
405 | spin_lock(&_minor_lock); |
406 | ||
5c6bd75d | 407 | atomic_dec(&md->open_count); |
1da177e4 | 408 | dm_put(md); |
4a1aeb98 MB |
409 | |
410 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
411 | } |
412 | ||
5c6bd75d AK |
413 | int dm_open_count(struct mapped_device *md) |
414 | { | |
415 | return atomic_read(&md->open_count); | |
416 | } | |
417 | ||
418 | /* | |
419 | * Guarantees nothing is using the device before it's deleted. | |
420 | */ | |
421 | int dm_lock_for_deletion(struct mapped_device *md) | |
422 | { | |
423 | int r = 0; | |
424 | ||
425 | spin_lock(&_minor_lock); | |
426 | ||
427 | if (dm_open_count(md)) | |
428 | r = -EBUSY; | |
429 | else | |
430 | set_bit(DMF_DELETING, &md->flags); | |
431 | ||
432 | spin_unlock(&_minor_lock); | |
433 | ||
434 | return r; | |
435 | } | |
436 | ||
fd2ed4d2 MP |
437 | sector_t dm_get_size(struct mapped_device *md) |
438 | { | |
439 | return get_capacity(md->disk); | |
440 | } | |
441 | ||
442 | struct dm_stats *dm_get_stats(struct mapped_device *md) | |
443 | { | |
444 | return &md->stats; | |
445 | } | |
446 | ||
3ac51e74 DW |
447 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
448 | { | |
449 | struct mapped_device *md = bdev->bd_disk->private_data; | |
450 | ||
451 | return dm_get_geometry(md, geo); | |
452 | } | |
453 | ||
fe5f9f2c | 454 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
455 | unsigned int cmd, unsigned long arg) |
456 | { | |
fe5f9f2c | 457 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 458 | int srcu_idx; |
6c182cd8 | 459 | struct dm_table *map; |
aa129a22 MB |
460 | struct dm_target *tgt; |
461 | int r = -ENOTTY; | |
462 | ||
6c182cd8 | 463 | retry: |
83d5e5b0 MP |
464 | map = dm_get_live_table(md, &srcu_idx); |
465 | ||
aa129a22 MB |
466 | if (!map || !dm_table_get_size(map)) |
467 | goto out; | |
468 | ||
469 | /* We only support devices that have a single target */ | |
470 | if (dm_table_get_num_targets(map) != 1) | |
471 | goto out; | |
472 | ||
473 | tgt = dm_table_get_target(map, 0); | |
474 | ||
4f186f8b | 475 | if (dm_suspended_md(md)) { |
aa129a22 MB |
476 | r = -EAGAIN; |
477 | goto out; | |
478 | } | |
479 | ||
480 | if (tgt->type->ioctl) | |
647b3d00 | 481 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
482 | |
483 | out: | |
83d5e5b0 | 484 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 485 | |
6c182cd8 HR |
486 | if (r == -ENOTCONN) { |
487 | msleep(10); | |
488 | goto retry; | |
489 | } | |
490 | ||
aa129a22 MB |
491 | return r; |
492 | } | |
493 | ||
028867ac | 494 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
495 | { |
496 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
497 | } | |
498 | ||
028867ac | 499 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
500 | { |
501 | mempool_free(io, md->io_pool); | |
502 | } | |
503 | ||
028867ac | 504 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 505 | { |
dba14160 | 506 | bio_put(&tio->clone); |
1da177e4 LT |
507 | } |
508 | ||
08885643 KU |
509 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
510 | gfp_t gfp_mask) | |
cec47e3d | 511 | { |
5f015204 | 512 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
513 | } |
514 | ||
515 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
516 | { | |
5f015204 | 517 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
518 | } |
519 | ||
90abb8c4 KU |
520 | static int md_in_flight(struct mapped_device *md) |
521 | { | |
522 | return atomic_read(&md->pending[READ]) + | |
523 | atomic_read(&md->pending[WRITE]); | |
524 | } | |
525 | ||
3eaf840e JNN |
526 | static void start_io_acct(struct dm_io *io) |
527 | { | |
528 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 529 | struct bio *bio = io->bio; |
c9959059 | 530 | int cpu; |
fd2ed4d2 | 531 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
532 | |
533 | io->start_time = jiffies; | |
534 | ||
074a7aca TH |
535 | cpu = part_stat_lock(); |
536 | part_round_stats(cpu, &dm_disk(md)->part0); | |
537 | part_stat_unlock(); | |
1e9bb880 SL |
538 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
539 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
540 | |
541 | if (unlikely(dm_stats_used(&md->stats))) | |
542 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector, | |
543 | bio_sectors(bio), false, 0, &io->stats_aux); | |
3eaf840e JNN |
544 | } |
545 | ||
d221d2e7 | 546 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
547 | { |
548 | struct mapped_device *md = io->md; | |
549 | struct bio *bio = io->bio; | |
550 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 551 | int pending, cpu; |
3eaf840e JNN |
552 | int rw = bio_data_dir(bio); |
553 | ||
074a7aca TH |
554 | cpu = part_stat_lock(); |
555 | part_round_stats(cpu, &dm_disk(md)->part0); | |
556 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
557 | part_stat_unlock(); | |
3eaf840e | 558 | |
fd2ed4d2 MP |
559 | if (unlikely(dm_stats_used(&md->stats))) |
560 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector, | |
561 | bio_sectors(bio), true, duration, &io->stats_aux); | |
562 | ||
af7e466a MP |
563 | /* |
564 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 565 | * a flush. |
af7e466a | 566 | */ |
1e9bb880 SL |
567 | pending = atomic_dec_return(&md->pending[rw]); |
568 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 569 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 570 | |
d221d2e7 MP |
571 | /* nudge anyone waiting on suspend queue */ |
572 | if (!pending) | |
573 | wake_up(&md->wait); | |
3eaf840e JNN |
574 | } |
575 | ||
1da177e4 LT |
576 | /* |
577 | * Add the bio to the list of deferred io. | |
578 | */ | |
92c63902 | 579 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 580 | { |
05447420 | 581 | unsigned long flags; |
1da177e4 | 582 | |
05447420 | 583 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 584 | bio_list_add(&md->deferred, bio); |
05447420 | 585 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 586 | queue_work(md->wq, &md->work); |
1da177e4 LT |
587 | } |
588 | ||
589 | /* | |
590 | * Everyone (including functions in this file), should use this | |
591 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 592 | * dm_put_live_table() when finished. |
1da177e4 | 593 | */ |
83d5e5b0 | 594 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 595 | { |
83d5e5b0 MP |
596 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
597 | ||
598 | return srcu_dereference(md->map, &md->io_barrier); | |
599 | } | |
1da177e4 | 600 | |
83d5e5b0 MP |
601 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
602 | { | |
603 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
604 | } | |
605 | ||
606 | void dm_sync_table(struct mapped_device *md) | |
607 | { | |
608 | synchronize_srcu(&md->io_barrier); | |
609 | synchronize_rcu_expedited(); | |
610 | } | |
611 | ||
612 | /* | |
613 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
614 | * The caller must not block between these two functions. | |
615 | */ | |
616 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
617 | { | |
618 | rcu_read_lock(); | |
619 | return rcu_dereference(md->map); | |
620 | } | |
1da177e4 | 621 | |
83d5e5b0 MP |
622 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
623 | { | |
624 | rcu_read_unlock(); | |
1da177e4 LT |
625 | } |
626 | ||
3ac51e74 DW |
627 | /* |
628 | * Get the geometry associated with a dm device | |
629 | */ | |
630 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
631 | { | |
632 | *geo = md->geometry; | |
633 | ||
634 | return 0; | |
635 | } | |
636 | ||
637 | /* | |
638 | * Set the geometry of a device. | |
639 | */ | |
640 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
641 | { | |
642 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
643 | ||
644 | if (geo->start > sz) { | |
645 | DMWARN("Start sector is beyond the geometry limits."); | |
646 | return -EINVAL; | |
647 | } | |
648 | ||
649 | md->geometry = *geo; | |
650 | ||
651 | return 0; | |
652 | } | |
653 | ||
1da177e4 LT |
654 | /*----------------------------------------------------------------- |
655 | * CRUD START: | |
656 | * A more elegant soln is in the works that uses the queue | |
657 | * merge fn, unfortunately there are a couple of changes to | |
658 | * the block layer that I want to make for this. So in the | |
659 | * interests of getting something for people to use I give | |
660 | * you this clearly demarcated crap. | |
661 | *---------------------------------------------------------------*/ | |
662 | ||
2e93ccc1 KU |
663 | static int __noflush_suspending(struct mapped_device *md) |
664 | { | |
665 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
666 | } | |
667 | ||
1da177e4 LT |
668 | /* |
669 | * Decrements the number of outstanding ios that a bio has been | |
670 | * cloned into, completing the original io if necc. | |
671 | */ | |
858119e1 | 672 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 673 | { |
2e93ccc1 | 674 | unsigned long flags; |
b35f8caa MB |
675 | int io_error; |
676 | struct bio *bio; | |
677 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
678 | |
679 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
680 | if (unlikely(error)) { |
681 | spin_lock_irqsave(&io->endio_lock, flags); | |
682 | if (!(io->error > 0 && __noflush_suspending(md))) | |
683 | io->error = error; | |
684 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
685 | } | |
1da177e4 LT |
686 | |
687 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
688 | if (io->error == DM_ENDIO_REQUEUE) { |
689 | /* | |
690 | * Target requested pushing back the I/O. | |
2e93ccc1 | 691 | */ |
022c2611 | 692 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
693 | if (__noflush_suspending(md)) |
694 | bio_list_add_head(&md->deferred, io->bio); | |
695 | else | |
2e93ccc1 KU |
696 | /* noflush suspend was interrupted. */ |
697 | io->error = -EIO; | |
022c2611 | 698 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
699 | } |
700 | ||
b35f8caa MB |
701 | io_error = io->error; |
702 | bio = io->bio; | |
6a8736d1 TH |
703 | end_io_acct(io); |
704 | free_io(md, io); | |
705 | ||
706 | if (io_error == DM_ENDIO_REQUEUE) | |
707 | return; | |
2e93ccc1 | 708 | |
b372d360 | 709 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 710 | /* |
6a8736d1 TH |
711 | * Preflush done for flush with data, reissue |
712 | * without REQ_FLUSH. | |
af7e466a | 713 | */ |
6a8736d1 TH |
714 | bio->bi_rw &= ~REQ_FLUSH; |
715 | queue_io(md, bio); | |
af7e466a | 716 | } else { |
b372d360 | 717 | /* done with normal IO or empty flush */ |
0a82a8d1 | 718 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 719 | bio_endio(bio, io_error); |
b35f8caa | 720 | } |
1da177e4 LT |
721 | } |
722 | } | |
723 | ||
6712ecf8 | 724 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
725 | { |
726 | int r = 0; | |
028867ac | 727 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 728 | struct dm_io *io = tio->io; |
9faf400f | 729 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
730 | dm_endio_fn endio = tio->ti->type->end_io; |
731 | ||
1da177e4 LT |
732 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
733 | error = -EIO; | |
734 | ||
735 | if (endio) { | |
7de3ee57 | 736 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
737 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
738 | /* | |
739 | * error and requeue request are handled | |
740 | * in dec_pending(). | |
741 | */ | |
1da177e4 | 742 | error = r; |
45cbcd79 KU |
743 | else if (r == DM_ENDIO_INCOMPLETE) |
744 | /* The target will handle the io */ | |
6712ecf8 | 745 | return; |
45cbcd79 KU |
746 | else if (r) { |
747 | DMWARN("unimplemented target endio return value: %d", r); | |
748 | BUG(); | |
749 | } | |
1da177e4 LT |
750 | } |
751 | ||
9faf400f | 752 | free_tio(md, tio); |
b35f8caa | 753 | dec_pending(io, error); |
1da177e4 LT |
754 | } |
755 | ||
cec47e3d KU |
756 | /* |
757 | * Partial completion handling for request-based dm | |
758 | */ | |
759 | static void end_clone_bio(struct bio *clone, int error) | |
760 | { | |
761 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
762 | struct dm_rq_target_io *tio = info->tio; | |
763 | struct bio *bio = info->orig; | |
764 | unsigned int nr_bytes = info->orig->bi_size; | |
765 | ||
766 | bio_put(clone); | |
767 | ||
768 | if (tio->error) | |
769 | /* | |
770 | * An error has already been detected on the request. | |
771 | * Once error occurred, just let clone->end_io() handle | |
772 | * the remainder. | |
773 | */ | |
774 | return; | |
775 | else if (error) { | |
776 | /* | |
777 | * Don't notice the error to the upper layer yet. | |
778 | * The error handling decision is made by the target driver, | |
779 | * when the request is completed. | |
780 | */ | |
781 | tio->error = error; | |
782 | return; | |
783 | } | |
784 | ||
785 | /* | |
786 | * I/O for the bio successfully completed. | |
787 | * Notice the data completion to the upper layer. | |
788 | */ | |
789 | ||
790 | /* | |
791 | * bios are processed from the head of the list. | |
792 | * So the completing bio should always be rq->bio. | |
793 | * If it's not, something wrong is happening. | |
794 | */ | |
795 | if (tio->orig->bio != bio) | |
796 | DMERR("bio completion is going in the middle of the request"); | |
797 | ||
798 | /* | |
799 | * Update the original request. | |
800 | * Do not use blk_end_request() here, because it may complete | |
801 | * the original request before the clone, and break the ordering. | |
802 | */ | |
803 | blk_update_request(tio->orig, 0, nr_bytes); | |
804 | } | |
805 | ||
806 | /* | |
807 | * Don't touch any member of the md after calling this function because | |
808 | * the md may be freed in dm_put() at the end of this function. | |
809 | * Or do dm_get() before calling this function and dm_put() later. | |
810 | */ | |
b4324fee | 811 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 812 | { |
b4324fee | 813 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
814 | |
815 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 816 | if (!md_in_flight(md)) |
cec47e3d KU |
817 | wake_up(&md->wait); |
818 | ||
a8c32a5c JA |
819 | /* |
820 | * Run this off this callpath, as drivers could invoke end_io while | |
821 | * inside their request_fn (and holding the queue lock). Calling | |
822 | * back into ->request_fn() could deadlock attempting to grab the | |
823 | * queue lock again. | |
824 | */ | |
cec47e3d | 825 | if (run_queue) |
a8c32a5c | 826 | blk_run_queue_async(md->queue); |
cec47e3d KU |
827 | |
828 | /* | |
829 | * dm_put() must be at the end of this function. See the comment above | |
830 | */ | |
831 | dm_put(md); | |
832 | } | |
833 | ||
a77e28c7 KU |
834 | static void free_rq_clone(struct request *clone) |
835 | { | |
836 | struct dm_rq_target_io *tio = clone->end_io_data; | |
837 | ||
838 | blk_rq_unprep_clone(clone); | |
839 | free_rq_tio(tio); | |
840 | } | |
841 | ||
980691e5 KU |
842 | /* |
843 | * Complete the clone and the original request. | |
844 | * Must be called without queue lock. | |
845 | */ | |
846 | static void dm_end_request(struct request *clone, int error) | |
847 | { | |
848 | int rw = rq_data_dir(clone); | |
849 | struct dm_rq_target_io *tio = clone->end_io_data; | |
850 | struct mapped_device *md = tio->md; | |
851 | struct request *rq = tio->orig; | |
852 | ||
29e4013d | 853 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
854 | rq->errors = clone->errors; |
855 | rq->resid_len = clone->resid_len; | |
856 | ||
857 | if (rq->sense) | |
858 | /* | |
859 | * We are using the sense buffer of the original | |
860 | * request. | |
861 | * So setting the length of the sense data is enough. | |
862 | */ | |
863 | rq->sense_len = clone->sense_len; | |
864 | } | |
865 | ||
866 | free_rq_clone(clone); | |
29e4013d TH |
867 | blk_end_request_all(rq, error); |
868 | rq_completed(md, rw, true); | |
980691e5 KU |
869 | } |
870 | ||
cec47e3d KU |
871 | static void dm_unprep_request(struct request *rq) |
872 | { | |
873 | struct request *clone = rq->special; | |
cec47e3d KU |
874 | |
875 | rq->special = NULL; | |
876 | rq->cmd_flags &= ~REQ_DONTPREP; | |
877 | ||
a77e28c7 | 878 | free_rq_clone(clone); |
cec47e3d KU |
879 | } |
880 | ||
881 | /* | |
882 | * Requeue the original request of a clone. | |
883 | */ | |
884 | void dm_requeue_unmapped_request(struct request *clone) | |
885 | { | |
b4324fee | 886 | int rw = rq_data_dir(clone); |
cec47e3d KU |
887 | struct dm_rq_target_io *tio = clone->end_io_data; |
888 | struct mapped_device *md = tio->md; | |
889 | struct request *rq = tio->orig; | |
890 | struct request_queue *q = rq->q; | |
891 | unsigned long flags; | |
892 | ||
893 | dm_unprep_request(rq); | |
894 | ||
895 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
896 | blk_requeue_request(q, rq); |
897 | spin_unlock_irqrestore(q->queue_lock, flags); | |
898 | ||
b4324fee | 899 | rq_completed(md, rw, 0); |
cec47e3d KU |
900 | } |
901 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
902 | ||
903 | static void __stop_queue(struct request_queue *q) | |
904 | { | |
905 | blk_stop_queue(q); | |
906 | } | |
907 | ||
908 | static void stop_queue(struct request_queue *q) | |
909 | { | |
910 | unsigned long flags; | |
911 | ||
912 | spin_lock_irqsave(q->queue_lock, flags); | |
913 | __stop_queue(q); | |
914 | spin_unlock_irqrestore(q->queue_lock, flags); | |
915 | } | |
916 | ||
917 | static void __start_queue(struct request_queue *q) | |
918 | { | |
919 | if (blk_queue_stopped(q)) | |
920 | blk_start_queue(q); | |
921 | } | |
922 | ||
923 | static void start_queue(struct request_queue *q) | |
924 | { | |
925 | unsigned long flags; | |
926 | ||
927 | spin_lock_irqsave(q->queue_lock, flags); | |
928 | __start_queue(q); | |
929 | spin_unlock_irqrestore(q->queue_lock, flags); | |
930 | } | |
931 | ||
11a68244 | 932 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 933 | { |
11a68244 | 934 | int r = error; |
cec47e3d | 935 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 936 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 937 | |
ba1cbad9 MS |
938 | if (tio->ti) { |
939 | rq_end_io = tio->ti->type->rq_end_io; | |
940 | ||
941 | if (mapped && rq_end_io) | |
942 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
943 | } | |
cec47e3d | 944 | |
11a68244 | 945 | if (r <= 0) |
cec47e3d | 946 | /* The target wants to complete the I/O */ |
11a68244 KU |
947 | dm_end_request(clone, r); |
948 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
949 | /* The target will handle the I/O */ |
950 | return; | |
11a68244 | 951 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
952 | /* The target wants to requeue the I/O */ |
953 | dm_requeue_unmapped_request(clone); | |
954 | else { | |
11a68244 | 955 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
956 | BUG(); |
957 | } | |
958 | } | |
959 | ||
11a68244 KU |
960 | /* |
961 | * Request completion handler for request-based dm | |
962 | */ | |
963 | static void dm_softirq_done(struct request *rq) | |
964 | { | |
965 | bool mapped = true; | |
966 | struct request *clone = rq->completion_data; | |
967 | struct dm_rq_target_io *tio = clone->end_io_data; | |
968 | ||
969 | if (rq->cmd_flags & REQ_FAILED) | |
970 | mapped = false; | |
971 | ||
972 | dm_done(clone, tio->error, mapped); | |
973 | } | |
974 | ||
cec47e3d KU |
975 | /* |
976 | * Complete the clone and the original request with the error status | |
977 | * through softirq context. | |
978 | */ | |
979 | static void dm_complete_request(struct request *clone, int error) | |
980 | { | |
981 | struct dm_rq_target_io *tio = clone->end_io_data; | |
982 | struct request *rq = tio->orig; | |
983 | ||
984 | tio->error = error; | |
985 | rq->completion_data = clone; | |
986 | blk_complete_request(rq); | |
987 | } | |
988 | ||
989 | /* | |
990 | * Complete the not-mapped clone and the original request with the error status | |
991 | * through softirq context. | |
992 | * Target's rq_end_io() function isn't called. | |
993 | * This may be used when the target's map_rq() function fails. | |
994 | */ | |
995 | void dm_kill_unmapped_request(struct request *clone, int error) | |
996 | { | |
997 | struct dm_rq_target_io *tio = clone->end_io_data; | |
998 | struct request *rq = tio->orig; | |
999 | ||
1000 | rq->cmd_flags |= REQ_FAILED; | |
1001 | dm_complete_request(clone, error); | |
1002 | } | |
1003 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1004 | ||
1005 | /* | |
1006 | * Called with the queue lock held | |
1007 | */ | |
1008 | static void end_clone_request(struct request *clone, int error) | |
1009 | { | |
1010 | /* | |
1011 | * For just cleaning up the information of the queue in which | |
1012 | * the clone was dispatched. | |
1013 | * The clone is *NOT* freed actually here because it is alloced from | |
1014 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1015 | */ | |
1016 | __blk_put_request(clone->q, clone); | |
1017 | ||
1018 | /* | |
1019 | * Actual request completion is done in a softirq context which doesn't | |
1020 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1021 | * - another request may be submitted by the upper level driver | |
1022 | * of the stacking during the completion | |
1023 | * - the submission which requires queue lock may be done | |
1024 | * against this queue | |
1025 | */ | |
1026 | dm_complete_request(clone, error); | |
1027 | } | |
1028 | ||
56a67df7 MS |
1029 | /* |
1030 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1031 | * target boundary. | |
1032 | */ | |
1033 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1034 | { | |
1035 | sector_t target_offset = dm_target_offset(ti, sector); | |
1036 | ||
1037 | return ti->len - target_offset; | |
1038 | } | |
1039 | ||
1040 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1041 | { |
56a67df7 | 1042 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1043 | sector_t offset, max_len; |
1da177e4 LT |
1044 | |
1045 | /* | |
542f9038 | 1046 | * Does the target need to split even further? |
1da177e4 | 1047 | */ |
542f9038 MS |
1048 | if (ti->max_io_len) { |
1049 | offset = dm_target_offset(ti, sector); | |
1050 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1051 | max_len = sector_div(offset, ti->max_io_len); | |
1052 | else | |
1053 | max_len = offset & (ti->max_io_len - 1); | |
1054 | max_len = ti->max_io_len - max_len; | |
1055 | ||
1056 | if (len > max_len) | |
1057 | len = max_len; | |
1da177e4 LT |
1058 | } |
1059 | ||
1060 | return len; | |
1061 | } | |
1062 | ||
542f9038 MS |
1063 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1064 | { | |
1065 | if (len > UINT_MAX) { | |
1066 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1067 | (unsigned long long)len, UINT_MAX); | |
1068 | ti->error = "Maximum size of target IO is too large"; | |
1069 | return -EINVAL; | |
1070 | } | |
1071 | ||
1072 | ti->max_io_len = (uint32_t) len; | |
1073 | ||
1074 | return 0; | |
1075 | } | |
1076 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1077 | ||
bd2a49b8 | 1078 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1079 | { |
1080 | int r; | |
2056a782 | 1081 | sector_t sector; |
9faf400f | 1082 | struct mapped_device *md; |
dba14160 | 1083 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1084 | struct dm_target *ti = tio->ti; |
1da177e4 | 1085 | |
1da177e4 LT |
1086 | clone->bi_end_io = clone_endio; |
1087 | clone->bi_private = tio; | |
1088 | ||
1089 | /* | |
1090 | * Map the clone. If r == 0 we don't need to do | |
1091 | * anything, the target has assumed ownership of | |
1092 | * this io. | |
1093 | */ | |
1094 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1095 | sector = clone->bi_sector; |
7de3ee57 | 1096 | r = ti->type->map(ti, clone); |
45cbcd79 | 1097 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1098 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1099 | |
d07335e5 MS |
1100 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1101 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1102 | |
1da177e4 | 1103 | generic_make_request(clone); |
2e93ccc1 KU |
1104 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1105 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1106 | md = tio->io->md; |
1107 | dec_pending(tio->io, r); | |
9faf400f | 1108 | free_tio(md, tio); |
45cbcd79 KU |
1109 | } else if (r) { |
1110 | DMWARN("unimplemented target map return value: %d", r); | |
1111 | BUG(); | |
1da177e4 LT |
1112 | } |
1113 | } | |
1114 | ||
1115 | struct clone_info { | |
1116 | struct mapped_device *md; | |
1117 | struct dm_table *map; | |
1118 | struct bio *bio; | |
1119 | struct dm_io *io; | |
1120 | sector_t sector; | |
1121 | sector_t sector_count; | |
1122 | unsigned short idx; | |
1123 | }; | |
1124 | ||
bd2a49b8 AK |
1125 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1126 | { | |
1127 | bio->bi_sector = sector; | |
1128 | bio->bi_size = to_bytes(len); | |
1129 | } | |
1130 | ||
1131 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1132 | { | |
1133 | bio->bi_idx = idx; | |
1134 | bio->bi_vcnt = idx + bv_count; | |
1135 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1136 | } | |
1137 | ||
1138 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1139 | unsigned short idx, unsigned len, unsigned offset, | |
1140 | unsigned trim) | |
1141 | { | |
1142 | if (!bio_integrity(bio)) | |
1143 | return; | |
1144 | ||
1145 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1146 | ||
1147 | if (trim) | |
1148 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1149 | } | |
1150 | ||
1da177e4 | 1151 | /* |
d87f4c14 | 1152 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1153 | */ |
14fe594d AK |
1154 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1155 | sector_t sector, unsigned short idx, | |
1156 | unsigned offset, unsigned len) | |
1da177e4 | 1157 | { |
dba14160 | 1158 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1159 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1160 | ||
1da177e4 LT |
1161 | *clone->bi_io_vec = *bv; |
1162 | ||
bd2a49b8 AK |
1163 | bio_setup_sector(clone, sector, len); |
1164 | ||
1da177e4 | 1165 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1166 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1167 | clone->bi_vcnt = 1; |
1da177e4 LT |
1168 | clone->bi_io_vec->bv_offset = offset; |
1169 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1170 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1171 | |
bd2a49b8 | 1172 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1173 | } |
1174 | ||
1175 | /* | |
1176 | * Creates a bio that consists of range of complete bvecs. | |
1177 | */ | |
dba14160 MP |
1178 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1179 | sector_t sector, unsigned short idx, | |
e4c93811 | 1180 | unsigned short bv_count, unsigned len) |
1da177e4 | 1181 | { |
dba14160 | 1182 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1183 | unsigned trim = 0; |
1da177e4 | 1184 | |
9faf400f | 1185 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1186 | bio_setup_sector(clone, sector, len); |
1187 | bio_setup_bv(clone, idx, bv_count); | |
1188 | ||
1189 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1190 | trim = 1; | |
1191 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1192 | } |
1193 | ||
9015df24 | 1194 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1195 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1196 | unsigned target_bio_nr) |
f9ab94ce | 1197 | { |
dba14160 MP |
1198 | struct dm_target_io *tio; |
1199 | struct bio *clone; | |
1200 | ||
1201 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1202 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1203 | |
1204 | tio->io = ci->io; | |
1205 | tio->ti = ti; | |
f9ab94ce | 1206 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1207 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1208 | |
1209 | return tio; | |
1210 | } | |
1211 | ||
14fe594d AK |
1212 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1213 | struct dm_target *ti, | |
1214 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1215 | { |
55a62eef | 1216 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1217 | struct bio *clone = &tio->clone; |
9015df24 | 1218 | |
06a426ce MS |
1219 | /* |
1220 | * Discard requests require the bio's inline iovecs be initialized. | |
1221 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1222 | * and discard, so no need for concern about wasted bvec allocations. | |
1223 | */ | |
dba14160 | 1224 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1225 | if (len) |
1226 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1227 | |
bd2a49b8 | 1228 | __map_bio(tio); |
f9ab94ce MP |
1229 | } |
1230 | ||
14fe594d AK |
1231 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1232 | unsigned num_bios, sector_t len) | |
06a426ce | 1233 | { |
55a62eef | 1234 | unsigned target_bio_nr; |
06a426ce | 1235 | |
55a62eef | 1236 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1237 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1238 | } |
1239 | ||
14fe594d | 1240 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1241 | { |
06a426ce | 1242 | unsigned target_nr = 0; |
f9ab94ce MP |
1243 | struct dm_target *ti; |
1244 | ||
b372d360 | 1245 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1246 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1247 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1248 | |
f9ab94ce MP |
1249 | return 0; |
1250 | } | |
1251 | ||
e4c93811 AK |
1252 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1253 | sector_t sector, int nr_iovecs, | |
1254 | unsigned short idx, unsigned short bv_count, | |
1255 | unsigned offset, unsigned len, | |
1256 | unsigned split_bvec) | |
5ae89a87 | 1257 | { |
dba14160 | 1258 | struct bio *bio = ci->bio; |
5ae89a87 | 1259 | struct dm_target_io *tio; |
b0d8ed4d AK |
1260 | unsigned target_bio_nr; |
1261 | unsigned num_target_bios = 1; | |
5ae89a87 | 1262 | |
b0d8ed4d AK |
1263 | /* |
1264 | * Does the target want to receive duplicate copies of the bio? | |
1265 | */ | |
1266 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1267 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1268 | |
b0d8ed4d AK |
1269 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
1270 | tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr); | |
1271 | if (split_bvec) | |
1272 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1273 | else | |
1274 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1275 | __map_bio(tio); | |
1276 | } | |
5ae89a87 MS |
1277 | } |
1278 | ||
55a62eef | 1279 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1280 | |
55a62eef | 1281 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1282 | { |
55a62eef | 1283 | return ti->num_discard_bios; |
23508a96 MS |
1284 | } |
1285 | ||
55a62eef | 1286 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1287 | { |
55a62eef | 1288 | return ti->num_write_same_bios; |
23508a96 MS |
1289 | } |
1290 | ||
1291 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1292 | ||
1293 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1294 | { | |
55a62eef | 1295 | return ti->split_discard_bios; |
23508a96 MS |
1296 | } |
1297 | ||
14fe594d AK |
1298 | static int __send_changing_extent_only(struct clone_info *ci, |
1299 | get_num_bios_fn get_num_bios, | |
1300 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1301 | { |
1302 | struct dm_target *ti; | |
a79245b3 | 1303 | sector_t len; |
55a62eef | 1304 | unsigned num_bios; |
5ae89a87 | 1305 | |
a79245b3 MS |
1306 | do { |
1307 | ti = dm_table_find_target(ci->map, ci->sector); | |
1308 | if (!dm_target_is_valid(ti)) | |
1309 | return -EIO; | |
5ae89a87 | 1310 | |
5ae89a87 | 1311 | /* |
23508a96 MS |
1312 | * Even though the device advertised support for this type of |
1313 | * request, that does not mean every target supports it, and | |
936688d7 | 1314 | * reconfiguration might also have changed that since the |
a79245b3 | 1315 | * check was performed. |
5ae89a87 | 1316 | */ |
55a62eef AK |
1317 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1318 | if (!num_bios) | |
a79245b3 | 1319 | return -EOPNOTSUPP; |
5ae89a87 | 1320 | |
23508a96 | 1321 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1322 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1323 | else | |
1324 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1325 | |
14fe594d | 1326 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1327 | |
1328 | ci->sector += len; | |
1329 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1330 | |
1331 | return 0; | |
1332 | } | |
1333 | ||
14fe594d | 1334 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1335 | { |
14fe594d AK |
1336 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1337 | is_split_required_for_discard); | |
23508a96 MS |
1338 | } |
1339 | ||
14fe594d | 1340 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1341 | { |
14fe594d | 1342 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1343 | } |
1344 | ||
e4c93811 AK |
1345 | /* |
1346 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1347 | */ | |
1348 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1349 | { | |
1350 | struct bio *bio = ci->bio; | |
1351 | sector_t bv_len, total_len = 0; | |
1352 | ||
1353 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1354 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1355 | ||
1356 | if (bv_len > max) | |
1357 | break; | |
1358 | ||
1359 | max -= bv_len; | |
1360 | total_len += bv_len; | |
1361 | } | |
1362 | ||
1363 | return total_len; | |
1364 | } | |
1365 | ||
1366 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1367 | struct dm_target *ti, sector_t max) | |
1368 | { | |
1369 | struct bio *bio = ci->bio; | |
1370 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1371 | sector_t remaining = to_sector(bv->bv_len); | |
1372 | unsigned offset = 0; | |
1373 | sector_t len; | |
1374 | ||
1375 | do { | |
1376 | if (offset) { | |
1377 | ti = dm_table_find_target(ci->map, ci->sector); | |
1378 | if (!dm_target_is_valid(ti)) | |
1379 | return -EIO; | |
1380 | ||
1381 | max = max_io_len(ci->sector, ti); | |
1382 | } | |
1383 | ||
1384 | len = min(remaining, max); | |
1385 | ||
1386 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1387 | bv->bv_offset + offset, len, 1); | |
1388 | ||
1389 | ci->sector += len; | |
1390 | ci->sector_count -= len; | |
1391 | offset += to_bytes(len); | |
1392 | } while (remaining -= len); | |
1393 | ||
1394 | ci->idx++; | |
1395 | ||
1396 | return 0; | |
1397 | } | |
1398 | ||
1399 | /* | |
1400 | * Select the correct strategy for processing a non-flush bio. | |
1401 | */ | |
14fe594d | 1402 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1403 | { |
dba14160 | 1404 | struct bio *bio = ci->bio; |
512875bd | 1405 | struct dm_target *ti; |
e4c93811 AK |
1406 | sector_t len, max; |
1407 | int idx; | |
1da177e4 | 1408 | |
5ae89a87 | 1409 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1410 | return __send_discard(ci); |
23508a96 | 1411 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1412 | return __send_write_same(ci); |
5ae89a87 | 1413 | |
512875bd JN |
1414 | ti = dm_table_find_target(ci->map, ci->sector); |
1415 | if (!dm_target_is_valid(ti)) | |
1416 | return -EIO; | |
1417 | ||
56a67df7 | 1418 | max = max_io_len(ci->sector, ti); |
512875bd | 1419 | |
e4c93811 AK |
1420 | /* |
1421 | * Optimise for the simple case where we can do all of | |
1422 | * the remaining io with a single clone. | |
1423 | */ | |
1da177e4 | 1424 | if (ci->sector_count <= max) { |
e4c93811 AK |
1425 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1426 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1427 | ci->sector_count, 0); | |
1428 | ci->sector_count = 0; | |
1429 | return 0; | |
1430 | } | |
1da177e4 | 1431 | |
e4c93811 AK |
1432 | /* |
1433 | * There are some bvecs that don't span targets. | |
1434 | * Do as many of these as possible. | |
1435 | */ | |
1436 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1437 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1438 | |
e4c93811 AK |
1439 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1440 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1441 | |
1442 | ci->sector += len; | |
1443 | ci->sector_count -= len; | |
e4c93811 | 1444 | ci->idx = idx; |
d2044a94 | 1445 | |
e4c93811 | 1446 | return 0; |
1da177e4 | 1447 | } |
512875bd | 1448 | |
e4c93811 AK |
1449 | /* |
1450 | * Handle a bvec that must be split between two or more targets. | |
1451 | */ | |
1452 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1453 | } |
1454 | ||
1455 | /* | |
14fe594d | 1456 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1457 | */ |
83d5e5b0 MP |
1458 | static void __split_and_process_bio(struct mapped_device *md, |
1459 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1460 | { |
1461 | struct clone_info ci; | |
512875bd | 1462 | int error = 0; |
1da177e4 | 1463 | |
83d5e5b0 | 1464 | if (unlikely(!map)) { |
6a8736d1 | 1465 | bio_io_error(bio); |
f0b9a450 MP |
1466 | return; |
1467 | } | |
692d0eb9 | 1468 | |
83d5e5b0 | 1469 | ci.map = map; |
1da177e4 | 1470 | ci.md = md; |
1da177e4 LT |
1471 | ci.io = alloc_io(md); |
1472 | ci.io->error = 0; | |
1473 | atomic_set(&ci.io->io_count, 1); | |
1474 | ci.io->bio = bio; | |
1475 | ci.io->md = md; | |
f88fb981 | 1476 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1477 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1478 | ci.idx = bio->bi_idx; |
1479 | ||
3eaf840e | 1480 | start_io_acct(ci.io); |
bd2a49b8 | 1481 | |
b372d360 MS |
1482 | if (bio->bi_rw & REQ_FLUSH) { |
1483 | ci.bio = &ci.md->flush_bio; | |
1484 | ci.sector_count = 0; | |
14fe594d | 1485 | error = __send_empty_flush(&ci); |
b372d360 MS |
1486 | /* dec_pending submits any data associated with flush */ |
1487 | } else { | |
6a8736d1 | 1488 | ci.bio = bio; |
d87f4c14 | 1489 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1490 | while (ci.sector_count && !error) |
14fe594d | 1491 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1492 | } |
1da177e4 LT |
1493 | |
1494 | /* drop the extra reference count */ | |
512875bd | 1495 | dec_pending(ci.io, error); |
1da177e4 LT |
1496 | } |
1497 | /*----------------------------------------------------------------- | |
1498 | * CRUD END | |
1499 | *---------------------------------------------------------------*/ | |
1500 | ||
f6fccb12 MB |
1501 | static int dm_merge_bvec(struct request_queue *q, |
1502 | struct bvec_merge_data *bvm, | |
1503 | struct bio_vec *biovec) | |
1504 | { | |
1505 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1506 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1507 | struct dm_target *ti; |
1508 | sector_t max_sectors; | |
5037108a | 1509 | int max_size = 0; |
f6fccb12 MB |
1510 | |
1511 | if (unlikely(!map)) | |
5037108a | 1512 | goto out; |
f6fccb12 MB |
1513 | |
1514 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1515 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1516 | goto out; |
f6fccb12 MB |
1517 | |
1518 | /* | |
1519 | * Find maximum amount of I/O that won't need splitting | |
1520 | */ | |
56a67df7 | 1521 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1522 | (sector_t) BIO_MAX_SECTORS); |
1523 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1524 | if (max_size < 0) | |
1525 | max_size = 0; | |
1526 | ||
1527 | /* | |
1528 | * merge_bvec_fn() returns number of bytes | |
1529 | * it can accept at this offset | |
1530 | * max is precomputed maximal io size | |
1531 | */ | |
1532 | if (max_size && ti->type->merge) | |
1533 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1534 | /* |
1535 | * If the target doesn't support merge method and some of the devices | |
1536 | * provided their merge_bvec method (we know this by looking at | |
1537 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1538 | * entries. So always set max_size to 0, and the code below allows | |
1539 | * just one page. | |
1540 | */ | |
1541 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1542 | ||
1543 | max_size = 0; | |
f6fccb12 | 1544 | |
5037108a | 1545 | out: |
83d5e5b0 | 1546 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1547 | /* |
1548 | * Always allow an entire first page | |
1549 | */ | |
1550 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1551 | max_size = biovec->bv_len; | |
1552 | ||
f6fccb12 MB |
1553 | return max_size; |
1554 | } | |
1555 | ||
1da177e4 LT |
1556 | /* |
1557 | * The request function that just remaps the bio built up by | |
1558 | * dm_merge_bvec. | |
1559 | */ | |
5a7bbad2 | 1560 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1561 | { |
12f03a49 | 1562 | int rw = bio_data_dir(bio); |
1da177e4 | 1563 | struct mapped_device *md = q->queuedata; |
c9959059 | 1564 | int cpu; |
83d5e5b0 MP |
1565 | int srcu_idx; |
1566 | struct dm_table *map; | |
1da177e4 | 1567 | |
83d5e5b0 | 1568 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1569 | |
074a7aca TH |
1570 | cpu = part_stat_lock(); |
1571 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1572 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1573 | part_stat_unlock(); | |
12f03a49 | 1574 | |
6a8736d1 TH |
1575 | /* if we're suspended, we have to queue this io for later */ |
1576 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1577 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1578 | |
6a8736d1 TH |
1579 | if (bio_rw(bio) != READA) |
1580 | queue_io(md, bio); | |
1581 | else | |
54d9a1b4 | 1582 | bio_io_error(bio); |
5a7bbad2 | 1583 | return; |
1da177e4 LT |
1584 | } |
1585 | ||
83d5e5b0 MP |
1586 | __split_and_process_bio(md, map, bio); |
1587 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1588 | return; |
cec47e3d KU |
1589 | } |
1590 | ||
fd2ed4d2 | 1591 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1592 | { |
1593 | return blk_queue_stackable(md->queue); | |
1594 | } | |
1595 | ||
5a7bbad2 | 1596 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1597 | { |
1598 | struct mapped_device *md = q->queuedata; | |
1599 | ||
1600 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1601 | blk_queue_bio(q, bio); |
1602 | else | |
1603 | _dm_request(q, bio); | |
cec47e3d KU |
1604 | } |
1605 | ||
1606 | void dm_dispatch_request(struct request *rq) | |
1607 | { | |
1608 | int r; | |
1609 | ||
1610 | if (blk_queue_io_stat(rq->q)) | |
1611 | rq->cmd_flags |= REQ_IO_STAT; | |
1612 | ||
1613 | rq->start_time = jiffies; | |
1614 | r = blk_insert_cloned_request(rq->q, rq); | |
1615 | if (r) | |
1616 | dm_complete_request(rq, r); | |
1617 | } | |
1618 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1619 | ||
cec47e3d KU |
1620 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1621 | void *data) | |
1622 | { | |
1623 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1624 | struct dm_rq_clone_bio_info *info = |
1625 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1626 | |
1627 | info->orig = bio_orig; | |
1628 | info->tio = tio; | |
1629 | bio->bi_end_io = end_clone_bio; | |
1630 | bio->bi_private = info; | |
cec47e3d KU |
1631 | |
1632 | return 0; | |
1633 | } | |
1634 | ||
1635 | static int setup_clone(struct request *clone, struct request *rq, | |
1636 | struct dm_rq_target_io *tio) | |
1637 | { | |
d0bcb878 | 1638 | int r; |
cec47e3d | 1639 | |
29e4013d TH |
1640 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1641 | dm_rq_bio_constructor, tio); | |
1642 | if (r) | |
1643 | return r; | |
cec47e3d | 1644 | |
29e4013d TH |
1645 | clone->cmd = rq->cmd; |
1646 | clone->cmd_len = rq->cmd_len; | |
1647 | clone->sense = rq->sense; | |
1648 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1649 | clone->end_io = end_clone_request; |
1650 | clone->end_io_data = tio; | |
1651 | ||
1652 | return 0; | |
1653 | } | |
1654 | ||
6facdaff KU |
1655 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1656 | gfp_t gfp_mask) | |
1657 | { | |
1658 | struct request *clone; | |
1659 | struct dm_rq_target_io *tio; | |
1660 | ||
1661 | tio = alloc_rq_tio(md, gfp_mask); | |
1662 | if (!tio) | |
1663 | return NULL; | |
1664 | ||
1665 | tio->md = md; | |
1666 | tio->ti = NULL; | |
1667 | tio->orig = rq; | |
1668 | tio->error = 0; | |
1669 | memset(&tio->info, 0, sizeof(tio->info)); | |
1670 | ||
1671 | clone = &tio->clone; | |
1672 | if (setup_clone(clone, rq, tio)) { | |
1673 | /* -ENOMEM */ | |
1674 | free_rq_tio(tio); | |
1675 | return NULL; | |
1676 | } | |
1677 | ||
1678 | return clone; | |
1679 | } | |
1680 | ||
cec47e3d KU |
1681 | /* |
1682 | * Called with the queue lock held. | |
1683 | */ | |
1684 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1685 | { | |
1686 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1687 | struct request *clone; |
1688 | ||
cec47e3d KU |
1689 | if (unlikely(rq->special)) { |
1690 | DMWARN("Already has something in rq->special."); | |
1691 | return BLKPREP_KILL; | |
1692 | } | |
1693 | ||
6facdaff KU |
1694 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1695 | if (!clone) | |
cec47e3d | 1696 | return BLKPREP_DEFER; |
cec47e3d KU |
1697 | |
1698 | rq->special = clone; | |
1699 | rq->cmd_flags |= REQ_DONTPREP; | |
1700 | ||
1701 | return BLKPREP_OK; | |
1702 | } | |
1703 | ||
9eef87da KU |
1704 | /* |
1705 | * Returns: | |
1706 | * 0 : the request has been processed (not requeued) | |
1707 | * !0 : the request has been requeued | |
1708 | */ | |
1709 | static int map_request(struct dm_target *ti, struct request *clone, | |
1710 | struct mapped_device *md) | |
cec47e3d | 1711 | { |
9eef87da | 1712 | int r, requeued = 0; |
cec47e3d KU |
1713 | struct dm_rq_target_io *tio = clone->end_io_data; |
1714 | ||
cec47e3d KU |
1715 | tio->ti = ti; |
1716 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1717 | switch (r) { | |
1718 | case DM_MAPIO_SUBMITTED: | |
1719 | /* The target has taken the I/O to submit by itself later */ | |
1720 | break; | |
1721 | case DM_MAPIO_REMAPPED: | |
1722 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1723 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1724 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1725 | dm_dispatch_request(clone); |
1726 | break; | |
1727 | case DM_MAPIO_REQUEUE: | |
1728 | /* The target wants to requeue the I/O */ | |
1729 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1730 | requeued = 1; |
cec47e3d KU |
1731 | break; |
1732 | default: | |
1733 | if (r > 0) { | |
1734 | DMWARN("unimplemented target map return value: %d", r); | |
1735 | BUG(); | |
1736 | } | |
1737 | ||
1738 | /* The target wants to complete the I/O */ | |
1739 | dm_kill_unmapped_request(clone, r); | |
1740 | break; | |
1741 | } | |
9eef87da KU |
1742 | |
1743 | return requeued; | |
cec47e3d KU |
1744 | } |
1745 | ||
ba1cbad9 MS |
1746 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1747 | { | |
1748 | struct request *clone; | |
1749 | ||
1750 | blk_start_request(orig); | |
1751 | clone = orig->special; | |
1752 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1753 | ||
1754 | /* | |
1755 | * Hold the md reference here for the in-flight I/O. | |
1756 | * We can't rely on the reference count by device opener, | |
1757 | * because the device may be closed during the request completion | |
1758 | * when all bios are completed. | |
1759 | * See the comment in rq_completed() too. | |
1760 | */ | |
1761 | dm_get(md); | |
1762 | ||
1763 | return clone; | |
1764 | } | |
1765 | ||
cec47e3d KU |
1766 | /* |
1767 | * q->request_fn for request-based dm. | |
1768 | * Called with the queue lock held. | |
1769 | */ | |
1770 | static void dm_request_fn(struct request_queue *q) | |
1771 | { | |
1772 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1773 | int srcu_idx; |
1774 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1775 | struct dm_target *ti; |
b4324fee | 1776 | struct request *rq, *clone; |
29e4013d | 1777 | sector_t pos; |
cec47e3d KU |
1778 | |
1779 | /* | |
b4324fee KU |
1780 | * For suspend, check blk_queue_stopped() and increment |
1781 | * ->pending within a single queue_lock not to increment the | |
1782 | * number of in-flight I/Os after the queue is stopped in | |
1783 | * dm_suspend(). | |
cec47e3d | 1784 | */ |
7eaceacc | 1785 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1786 | rq = blk_peek_request(q); |
1787 | if (!rq) | |
7eaceacc | 1788 | goto delay_and_out; |
cec47e3d | 1789 | |
29e4013d TH |
1790 | /* always use block 0 to find the target for flushes for now */ |
1791 | pos = 0; | |
1792 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1793 | pos = blk_rq_pos(rq); | |
1794 | ||
1795 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1796 | if (!dm_target_is_valid(ti)) { |
1797 | /* | |
1798 | * Must perform setup, that dm_done() requires, | |
1799 | * before calling dm_kill_unmapped_request | |
1800 | */ | |
1801 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1802 | clone = dm_start_request(md, rq); | |
1803 | dm_kill_unmapped_request(clone, -EIO); | |
1804 | continue; | |
1805 | } | |
d0bcb878 | 1806 | |
cec47e3d | 1807 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1808 | goto delay_and_out; |
cec47e3d | 1809 | |
ba1cbad9 | 1810 | clone = dm_start_request(md, rq); |
b4324fee | 1811 | |
cec47e3d | 1812 | spin_unlock(q->queue_lock); |
9eef87da KU |
1813 | if (map_request(ti, clone, md)) |
1814 | goto requeued; | |
1815 | ||
052189a2 KU |
1816 | BUG_ON(!irqs_disabled()); |
1817 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1818 | } |
1819 | ||
1820 | goto out; | |
1821 | ||
9eef87da | 1822 | requeued: |
052189a2 KU |
1823 | BUG_ON(!irqs_disabled()); |
1824 | spin_lock(q->queue_lock); | |
9eef87da | 1825 | |
7eaceacc JA |
1826 | delay_and_out: |
1827 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1828 | out: |
83d5e5b0 | 1829 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1830 | } |
1831 | ||
1832 | int dm_underlying_device_busy(struct request_queue *q) | |
1833 | { | |
1834 | return blk_lld_busy(q); | |
1835 | } | |
1836 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1837 | ||
1838 | static int dm_lld_busy(struct request_queue *q) | |
1839 | { | |
1840 | int r; | |
1841 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1842 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1843 | |
1844 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1845 | r = 1; | |
1846 | else | |
1847 | r = dm_table_any_busy_target(map); | |
1848 | ||
83d5e5b0 | 1849 | dm_put_live_table_fast(md); |
cec47e3d KU |
1850 | |
1851 | return r; | |
1852 | } | |
1853 | ||
1da177e4 LT |
1854 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1855 | { | |
8a57dfc6 CS |
1856 | int r = bdi_bits; |
1857 | struct mapped_device *md = congested_data; | |
1858 | struct dm_table *map; | |
1da177e4 | 1859 | |
1eb787ec | 1860 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1861 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1862 | if (map) { |
cec47e3d KU |
1863 | /* |
1864 | * Request-based dm cares about only own queue for | |
1865 | * the query about congestion status of request_queue | |
1866 | */ | |
1867 | if (dm_request_based(md)) | |
1868 | r = md->queue->backing_dev_info.state & | |
1869 | bdi_bits; | |
1870 | else | |
1871 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1872 | } |
83d5e5b0 | 1873 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1874 | } |
1875 | ||
1da177e4 LT |
1876 | return r; |
1877 | } | |
1878 | ||
1879 | /*----------------------------------------------------------------- | |
1880 | * An IDR is used to keep track of allocated minor numbers. | |
1881 | *---------------------------------------------------------------*/ | |
2b06cfff | 1882 | static void free_minor(int minor) |
1da177e4 | 1883 | { |
f32c10b0 | 1884 | spin_lock(&_minor_lock); |
1da177e4 | 1885 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1886 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1887 | } |
1888 | ||
1889 | /* | |
1890 | * See if the device with a specific minor # is free. | |
1891 | */ | |
cf13ab8e | 1892 | static int specific_minor(int minor) |
1da177e4 | 1893 | { |
c9d76be6 | 1894 | int r; |
1da177e4 LT |
1895 | |
1896 | if (minor >= (1 << MINORBITS)) | |
1897 | return -EINVAL; | |
1898 | ||
c9d76be6 | 1899 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1900 | spin_lock(&_minor_lock); |
1da177e4 | 1901 | |
c9d76be6 | 1902 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1903 | |
f32c10b0 | 1904 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1905 | idr_preload_end(); |
1906 | if (r < 0) | |
1907 | return r == -ENOSPC ? -EBUSY : r; | |
1908 | return 0; | |
1da177e4 LT |
1909 | } |
1910 | ||
cf13ab8e | 1911 | static int next_free_minor(int *minor) |
1da177e4 | 1912 | { |
c9d76be6 | 1913 | int r; |
62f75c2f | 1914 | |
c9d76be6 | 1915 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1916 | spin_lock(&_minor_lock); |
1da177e4 | 1917 | |
c9d76be6 | 1918 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1919 | |
f32c10b0 | 1920 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1921 | idr_preload_end(); |
1922 | if (r < 0) | |
1923 | return r; | |
1924 | *minor = r; | |
1925 | return 0; | |
1da177e4 LT |
1926 | } |
1927 | ||
83d5cde4 | 1928 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1929 | |
53d5914f MP |
1930 | static void dm_wq_work(struct work_struct *work); |
1931 | ||
4a0b4ddf MS |
1932 | static void dm_init_md_queue(struct mapped_device *md) |
1933 | { | |
1934 | /* | |
1935 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1936 | * devices. The type of this dm device has not been decided yet. | |
1937 | * The type is decided at the first table loading time. | |
1938 | * To prevent problematic device stacking, clear the queue flag | |
1939 | * for request stacking support until then. | |
1940 | * | |
1941 | * This queue is new, so no concurrency on the queue_flags. | |
1942 | */ | |
1943 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1944 | ||
1945 | md->queue->queuedata = md; | |
1946 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1947 | md->queue->backing_dev_info.congested_data = md; | |
1948 | blk_queue_make_request(md->queue, dm_request); | |
1949 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1950 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1951 | } | |
1952 | ||
1da177e4 LT |
1953 | /* |
1954 | * Allocate and initialise a blank device with a given minor. | |
1955 | */ | |
2b06cfff | 1956 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1957 | { |
1958 | int r; | |
cf13ab8e | 1959 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1960 | void *old_md; |
1da177e4 LT |
1961 | |
1962 | if (!md) { | |
1963 | DMWARN("unable to allocate device, out of memory."); | |
1964 | return NULL; | |
1965 | } | |
1966 | ||
10da4f79 | 1967 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1968 | goto bad_module_get; |
10da4f79 | 1969 | |
1da177e4 | 1970 | /* get a minor number for the dev */ |
2b06cfff | 1971 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1972 | r = next_free_minor(&minor); |
2b06cfff | 1973 | else |
cf13ab8e | 1974 | r = specific_minor(minor); |
1da177e4 | 1975 | if (r < 0) |
6ed7ade8 | 1976 | goto bad_minor; |
1da177e4 | 1977 | |
83d5e5b0 MP |
1978 | r = init_srcu_struct(&md->io_barrier); |
1979 | if (r < 0) | |
1980 | goto bad_io_barrier; | |
1981 | ||
a5664dad | 1982 | md->type = DM_TYPE_NONE; |
e61290a4 | 1983 | mutex_init(&md->suspend_lock); |
a5664dad | 1984 | mutex_init(&md->type_lock); |
022c2611 | 1985 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1986 | atomic_set(&md->holders, 1); |
5c6bd75d | 1987 | atomic_set(&md->open_count, 0); |
1da177e4 | 1988 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1989 | atomic_set(&md->uevent_seq, 0); |
1990 | INIT_LIST_HEAD(&md->uevent_list); | |
1991 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1992 | |
4a0b4ddf | 1993 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1994 | if (!md->queue) |
6ed7ade8 | 1995 | goto bad_queue; |
1da177e4 | 1996 | |
4a0b4ddf | 1997 | dm_init_md_queue(md); |
9faf400f | 1998 | |
1da177e4 LT |
1999 | md->disk = alloc_disk(1); |
2000 | if (!md->disk) | |
6ed7ade8 | 2001 | goto bad_disk; |
1da177e4 | 2002 | |
316d315b NK |
2003 | atomic_set(&md->pending[0], 0); |
2004 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2005 | init_waitqueue_head(&md->wait); |
53d5914f | 2006 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
2007 | init_waitqueue_head(&md->eventq); |
2008 | ||
1da177e4 LT |
2009 | md->disk->major = _major; |
2010 | md->disk->first_minor = minor; | |
2011 | md->disk->fops = &dm_blk_dops; | |
2012 | md->disk->queue = md->queue; | |
2013 | md->disk->private_data = md; | |
2014 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2015 | add_disk(md->disk); | |
7e51f257 | 2016 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2017 | |
670368a8 | 2018 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2019 | if (!md->wq) |
2020 | goto bad_thread; | |
2021 | ||
32a926da MP |
2022 | md->bdev = bdget_disk(md->disk, 0); |
2023 | if (!md->bdev) | |
2024 | goto bad_bdev; | |
2025 | ||
6a8736d1 TH |
2026 | bio_init(&md->flush_bio); |
2027 | md->flush_bio.bi_bdev = md->bdev; | |
2028 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2029 | ||
fd2ed4d2 MP |
2030 | dm_stats_init(&md->stats); |
2031 | ||
ba61fdd1 | 2032 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2033 | spin_lock(&_minor_lock); |
ba61fdd1 | 2034 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2035 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2036 | |
2037 | BUG_ON(old_md != MINOR_ALLOCED); | |
2038 | ||
1da177e4 LT |
2039 | return md; |
2040 | ||
32a926da MP |
2041 | bad_bdev: |
2042 | destroy_workqueue(md->wq); | |
304f3f6a | 2043 | bad_thread: |
03022c54 | 2044 | del_gendisk(md->disk); |
304f3f6a | 2045 | put_disk(md->disk); |
6ed7ade8 | 2046 | bad_disk: |
1312f40e | 2047 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2048 | bad_queue: |
83d5e5b0 MP |
2049 | cleanup_srcu_struct(&md->io_barrier); |
2050 | bad_io_barrier: | |
1da177e4 | 2051 | free_minor(minor); |
6ed7ade8 | 2052 | bad_minor: |
10da4f79 | 2053 | module_put(THIS_MODULE); |
6ed7ade8 | 2054 | bad_module_get: |
1da177e4 LT |
2055 | kfree(md); |
2056 | return NULL; | |
2057 | } | |
2058 | ||
ae9da83f JN |
2059 | static void unlock_fs(struct mapped_device *md); |
2060 | ||
1da177e4 LT |
2061 | static void free_dev(struct mapped_device *md) |
2062 | { | |
f331c029 | 2063 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2064 | |
32a926da MP |
2065 | unlock_fs(md); |
2066 | bdput(md->bdev); | |
304f3f6a | 2067 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
2068 | if (md->io_pool) |
2069 | mempool_destroy(md->io_pool); | |
2070 | if (md->bs) | |
2071 | bioset_free(md->bs); | |
9c47008d | 2072 | blk_integrity_unregister(md->disk); |
1da177e4 | 2073 | del_gendisk(md->disk); |
83d5e5b0 | 2074 | cleanup_srcu_struct(&md->io_barrier); |
63d94e48 | 2075 | free_minor(minor); |
fba9f90e JM |
2076 | |
2077 | spin_lock(&_minor_lock); | |
2078 | md->disk->private_data = NULL; | |
2079 | spin_unlock(&_minor_lock); | |
2080 | ||
1da177e4 | 2081 | put_disk(md->disk); |
1312f40e | 2082 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2083 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2084 | module_put(THIS_MODULE); |
1da177e4 LT |
2085 | kfree(md); |
2086 | } | |
2087 | ||
e6ee8c0b KU |
2088 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2089 | { | |
c0820cf5 | 2090 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2091 | |
5f015204 | 2092 | if (md->io_pool && md->bs) { |
16245bdc JN |
2093 | /* The md already has necessary mempools. */ |
2094 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2095 | /* | |
2096 | * Reload bioset because front_pad may have changed | |
2097 | * because a different table was loaded. | |
2098 | */ | |
2099 | bioset_free(md->bs); | |
2100 | md->bs = p->bs; | |
2101 | p->bs = NULL; | |
2102 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2103 | /* |
2104 | * There's no need to reload with request-based dm | |
2105 | * because the size of front_pad doesn't change. | |
2106 | * Note for future: If you are to reload bioset, | |
2107 | * prep-ed requests in the queue may refer | |
2108 | * to bio from the old bioset, so you must walk | |
2109 | * through the queue to unprep. | |
2110 | */ | |
2111 | } | |
e6ee8c0b | 2112 | goto out; |
c0820cf5 | 2113 | } |
e6ee8c0b | 2114 | |
5f015204 | 2115 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2116 | |
2117 | md->io_pool = p->io_pool; | |
2118 | p->io_pool = NULL; | |
e6ee8c0b KU |
2119 | md->bs = p->bs; |
2120 | p->bs = NULL; | |
2121 | ||
2122 | out: | |
2123 | /* mempool bind completed, now no need any mempools in the table */ | |
2124 | dm_table_free_md_mempools(t); | |
2125 | } | |
2126 | ||
1da177e4 LT |
2127 | /* |
2128 | * Bind a table to the device. | |
2129 | */ | |
2130 | static void event_callback(void *context) | |
2131 | { | |
7a8c3d3b MA |
2132 | unsigned long flags; |
2133 | LIST_HEAD(uevents); | |
1da177e4 LT |
2134 | struct mapped_device *md = (struct mapped_device *) context; |
2135 | ||
7a8c3d3b MA |
2136 | spin_lock_irqsave(&md->uevent_lock, flags); |
2137 | list_splice_init(&md->uevent_list, &uevents); | |
2138 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2139 | ||
ed9e1982 | 2140 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2141 | |
1da177e4 LT |
2142 | atomic_inc(&md->event_nr); |
2143 | wake_up(&md->eventq); | |
2144 | } | |
2145 | ||
c217649b MS |
2146 | /* |
2147 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2148 | */ | |
4e90188b | 2149 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2150 | { |
4e90188b | 2151 | set_capacity(md->disk, size); |
1da177e4 | 2152 | |
db8fef4f | 2153 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2154 | } |
2155 | ||
d5b9dd04 MP |
2156 | /* |
2157 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2158 | * | |
2159 | * If this function returns 0, then the device is either a non-dm | |
2160 | * device without a merge_bvec_fn, or it is a dm device that is | |
2161 | * able to split any bios it receives that are too big. | |
2162 | */ | |
2163 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2164 | { | |
2165 | struct mapped_device *dev_md; | |
2166 | ||
2167 | if (!q->merge_bvec_fn) | |
2168 | return 0; | |
2169 | ||
2170 | if (q->make_request_fn == dm_request) { | |
2171 | dev_md = q->queuedata; | |
2172 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2173 | return 0; | |
2174 | } | |
2175 | ||
2176 | return 1; | |
2177 | } | |
2178 | ||
2179 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2180 | struct dm_dev *dev, sector_t start, | |
2181 | sector_t len, void *data) | |
2182 | { | |
2183 | struct block_device *bdev = dev->bdev; | |
2184 | struct request_queue *q = bdev_get_queue(bdev); | |
2185 | ||
2186 | return dm_queue_merge_is_compulsory(q); | |
2187 | } | |
2188 | ||
2189 | /* | |
2190 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2191 | * on the properties of the underlying devices. | |
2192 | */ | |
2193 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2194 | { | |
2195 | unsigned i = 0; | |
2196 | struct dm_target *ti; | |
2197 | ||
2198 | while (i < dm_table_get_num_targets(table)) { | |
2199 | ti = dm_table_get_target(table, i++); | |
2200 | ||
2201 | if (ti->type->iterate_devices && | |
2202 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2203 | return 0; | |
2204 | } | |
2205 | ||
2206 | return 1; | |
2207 | } | |
2208 | ||
042d2a9b AK |
2209 | /* |
2210 | * Returns old map, which caller must destroy. | |
2211 | */ | |
2212 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2213 | struct queue_limits *limits) | |
1da177e4 | 2214 | { |
042d2a9b | 2215 | struct dm_table *old_map; |
165125e1 | 2216 | struct request_queue *q = md->queue; |
1da177e4 | 2217 | sector_t size; |
d5b9dd04 | 2218 | int merge_is_optional; |
1da177e4 LT |
2219 | |
2220 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2221 | |
2222 | /* | |
2223 | * Wipe any geometry if the size of the table changed. | |
2224 | */ | |
fd2ed4d2 | 2225 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2226 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2227 | ||
32a926da | 2228 | __set_size(md, size); |
d5816876 | 2229 | |
2ca3310e AK |
2230 | dm_table_event_callback(t, event_callback, md); |
2231 | ||
e6ee8c0b KU |
2232 | /* |
2233 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2234 | * for request-based during suspension. So stop it to prevent | |
2235 | * I/O mapping before resume. | |
2236 | * This must be done before setting the queue restrictions, | |
2237 | * because request-based dm may be run just after the setting. | |
2238 | */ | |
2239 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2240 | stop_queue(q); | |
2241 | ||
2242 | __bind_mempools(md, t); | |
2243 | ||
d5b9dd04 MP |
2244 | merge_is_optional = dm_table_merge_is_optional(t); |
2245 | ||
042d2a9b | 2246 | old_map = md->map; |
83d5e5b0 | 2247 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2248 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2249 | ||
754c5fc7 | 2250 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2251 | if (merge_is_optional) |
2252 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2253 | else | |
2254 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
83d5e5b0 | 2255 | dm_sync_table(md); |
1da177e4 | 2256 | |
042d2a9b | 2257 | return old_map; |
1da177e4 LT |
2258 | } |
2259 | ||
a7940155 AK |
2260 | /* |
2261 | * Returns unbound table for the caller to free. | |
2262 | */ | |
2263 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2264 | { |
2265 | struct dm_table *map = md->map; | |
2266 | ||
2267 | if (!map) | |
a7940155 | 2268 | return NULL; |
1da177e4 LT |
2269 | |
2270 | dm_table_event_callback(map, NULL, NULL); | |
83d5e5b0 MP |
2271 | rcu_assign_pointer(md->map, NULL); |
2272 | dm_sync_table(md); | |
a7940155 AK |
2273 | |
2274 | return map; | |
1da177e4 LT |
2275 | } |
2276 | ||
2277 | /* | |
2278 | * Constructor for a new device. | |
2279 | */ | |
2b06cfff | 2280 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2281 | { |
2282 | struct mapped_device *md; | |
2283 | ||
2b06cfff | 2284 | md = alloc_dev(minor); |
1da177e4 LT |
2285 | if (!md) |
2286 | return -ENXIO; | |
2287 | ||
784aae73 MB |
2288 | dm_sysfs_init(md); |
2289 | ||
1da177e4 LT |
2290 | *result = md; |
2291 | return 0; | |
2292 | } | |
2293 | ||
a5664dad MS |
2294 | /* |
2295 | * Functions to manage md->type. | |
2296 | * All are required to hold md->type_lock. | |
2297 | */ | |
2298 | void dm_lock_md_type(struct mapped_device *md) | |
2299 | { | |
2300 | mutex_lock(&md->type_lock); | |
2301 | } | |
2302 | ||
2303 | void dm_unlock_md_type(struct mapped_device *md) | |
2304 | { | |
2305 | mutex_unlock(&md->type_lock); | |
2306 | } | |
2307 | ||
2308 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2309 | { | |
00c4fc3b | 2310 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2311 | md->type = type; |
2312 | } | |
2313 | ||
2314 | unsigned dm_get_md_type(struct mapped_device *md) | |
2315 | { | |
00c4fc3b | 2316 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2317 | return md->type; |
2318 | } | |
2319 | ||
36a0456f AK |
2320 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2321 | { | |
2322 | return md->immutable_target_type; | |
2323 | } | |
2324 | ||
f84cb8a4 MS |
2325 | /* |
2326 | * The queue_limits are only valid as long as you have a reference | |
2327 | * count on 'md'. | |
2328 | */ | |
2329 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2330 | { | |
2331 | BUG_ON(!atomic_read(&md->holders)); | |
2332 | return &md->queue->limits; | |
2333 | } | |
2334 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2335 | ||
4a0b4ddf MS |
2336 | /* |
2337 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2338 | */ | |
2339 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2340 | { | |
2341 | struct request_queue *q = NULL; | |
2342 | ||
2343 | if (md->queue->elevator) | |
2344 | return 1; | |
2345 | ||
2346 | /* Fully initialize the queue */ | |
2347 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2348 | if (!q) | |
2349 | return 0; | |
2350 | ||
2351 | md->queue = q; | |
4a0b4ddf MS |
2352 | dm_init_md_queue(md); |
2353 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2354 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2355 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2356 | |
2357 | elv_register_queue(md->queue); | |
2358 | ||
2359 | return 1; | |
2360 | } | |
2361 | ||
2362 | /* | |
2363 | * Setup the DM device's queue based on md's type | |
2364 | */ | |
2365 | int dm_setup_md_queue(struct mapped_device *md) | |
2366 | { | |
2367 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2368 | !dm_init_request_based_queue(md)) { | |
2369 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2370 | return -EINVAL; | |
2371 | } | |
2372 | ||
2373 | return 0; | |
2374 | } | |
2375 | ||
637842cf | 2376 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2377 | { |
2378 | struct mapped_device *md; | |
1da177e4 LT |
2379 | unsigned minor = MINOR(dev); |
2380 | ||
2381 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2382 | return NULL; | |
2383 | ||
f32c10b0 | 2384 | spin_lock(&_minor_lock); |
1da177e4 LT |
2385 | |
2386 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2387 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2388 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2389 | dm_deleting_md(md) || |
17b2f66f | 2390 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2391 | md = NULL; |
fba9f90e JM |
2392 | goto out; |
2393 | } | |
1da177e4 | 2394 | |
fba9f90e | 2395 | out: |
f32c10b0 | 2396 | spin_unlock(&_minor_lock); |
1da177e4 | 2397 | |
637842cf DT |
2398 | return md; |
2399 | } | |
2400 | ||
d229a958 DT |
2401 | struct mapped_device *dm_get_md(dev_t dev) |
2402 | { | |
2403 | struct mapped_device *md = dm_find_md(dev); | |
2404 | ||
2405 | if (md) | |
2406 | dm_get(md); | |
2407 | ||
2408 | return md; | |
2409 | } | |
3cf2e4ba | 2410 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2411 | |
9ade92a9 | 2412 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2413 | { |
9ade92a9 | 2414 | return md->interface_ptr; |
1da177e4 LT |
2415 | } |
2416 | ||
2417 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2418 | { | |
2419 | md->interface_ptr = ptr; | |
2420 | } | |
2421 | ||
2422 | void dm_get(struct mapped_device *md) | |
2423 | { | |
2424 | atomic_inc(&md->holders); | |
3f77316d | 2425 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2426 | } |
2427 | ||
72d94861 AK |
2428 | const char *dm_device_name(struct mapped_device *md) |
2429 | { | |
2430 | return md->name; | |
2431 | } | |
2432 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2433 | ||
3f77316d | 2434 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2435 | { |
1134e5ae | 2436 | struct dm_table *map; |
83d5e5b0 | 2437 | int srcu_idx; |
1da177e4 | 2438 | |
3f77316d | 2439 | might_sleep(); |
fba9f90e | 2440 | |
3f77316d | 2441 | spin_lock(&_minor_lock); |
83d5e5b0 | 2442 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2443 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2444 | set_bit(DMF_FREEING, &md->flags); | |
2445 | spin_unlock(&_minor_lock); | |
2446 | ||
2447 | if (!dm_suspended_md(md)) { | |
2448 | dm_table_presuspend_targets(map); | |
2449 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2450 | } |
3f77316d | 2451 | |
83d5e5b0 MP |
2452 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2453 | dm_put_live_table(md, srcu_idx); | |
2454 | ||
3f77316d KU |
2455 | /* |
2456 | * Rare, but there may be I/O requests still going to complete, | |
2457 | * for example. Wait for all references to disappear. | |
2458 | * No one should increment the reference count of the mapped_device, | |
2459 | * after the mapped_device state becomes DMF_FREEING. | |
2460 | */ | |
2461 | if (wait) | |
2462 | while (atomic_read(&md->holders)) | |
2463 | msleep(1); | |
2464 | else if (atomic_read(&md->holders)) | |
2465 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2466 | dm_device_name(md), atomic_read(&md->holders)); | |
2467 | ||
2468 | dm_sysfs_exit(md); | |
3f77316d KU |
2469 | dm_table_destroy(__unbind(md)); |
2470 | free_dev(md); | |
2471 | } | |
2472 | ||
2473 | void dm_destroy(struct mapped_device *md) | |
2474 | { | |
2475 | __dm_destroy(md, true); | |
2476 | } | |
2477 | ||
2478 | void dm_destroy_immediate(struct mapped_device *md) | |
2479 | { | |
2480 | __dm_destroy(md, false); | |
2481 | } | |
2482 | ||
2483 | void dm_put(struct mapped_device *md) | |
2484 | { | |
2485 | atomic_dec(&md->holders); | |
1da177e4 | 2486 | } |
79eb885c | 2487 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2488 | |
401600df | 2489 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2490 | { |
2491 | int r = 0; | |
b44ebeb0 MP |
2492 | DECLARE_WAITQUEUE(wait, current); |
2493 | ||
b44ebeb0 | 2494 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2495 | |
2496 | while (1) { | |
401600df | 2497 | set_current_state(interruptible); |
46125c1c | 2498 | |
b4324fee | 2499 | if (!md_in_flight(md)) |
46125c1c MB |
2500 | break; |
2501 | ||
401600df MP |
2502 | if (interruptible == TASK_INTERRUPTIBLE && |
2503 | signal_pending(current)) { | |
46125c1c MB |
2504 | r = -EINTR; |
2505 | break; | |
2506 | } | |
2507 | ||
2508 | io_schedule(); | |
2509 | } | |
2510 | set_current_state(TASK_RUNNING); | |
2511 | ||
b44ebeb0 MP |
2512 | remove_wait_queue(&md->wait, &wait); |
2513 | ||
46125c1c MB |
2514 | return r; |
2515 | } | |
2516 | ||
1da177e4 LT |
2517 | /* |
2518 | * Process the deferred bios | |
2519 | */ | |
ef208587 | 2520 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2521 | { |
ef208587 MP |
2522 | struct mapped_device *md = container_of(work, struct mapped_device, |
2523 | work); | |
6d6f10df | 2524 | struct bio *c; |
83d5e5b0 MP |
2525 | int srcu_idx; |
2526 | struct dm_table *map; | |
1da177e4 | 2527 | |
83d5e5b0 | 2528 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2529 | |
3b00b203 | 2530 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2531 | spin_lock_irq(&md->deferred_lock); |
2532 | c = bio_list_pop(&md->deferred); | |
2533 | spin_unlock_irq(&md->deferred_lock); | |
2534 | ||
6a8736d1 | 2535 | if (!c) |
df12ee99 | 2536 | break; |
022c2611 | 2537 | |
e6ee8c0b KU |
2538 | if (dm_request_based(md)) |
2539 | generic_make_request(c); | |
6a8736d1 | 2540 | else |
83d5e5b0 | 2541 | __split_and_process_bio(md, map, c); |
022c2611 | 2542 | } |
73d410c0 | 2543 | |
83d5e5b0 | 2544 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2545 | } |
2546 | ||
9a1fb464 | 2547 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2548 | { |
3b00b203 MP |
2549 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2550 | smp_mb__after_clear_bit(); | |
53d5914f | 2551 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2552 | } |
2553 | ||
1da177e4 | 2554 | /* |
042d2a9b | 2555 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2556 | */ |
042d2a9b | 2557 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2558 | { |
87eb5b21 | 2559 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2560 | struct queue_limits limits; |
042d2a9b | 2561 | int r; |
1da177e4 | 2562 | |
e61290a4 | 2563 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2564 | |
2565 | /* device must be suspended */ | |
4f186f8b | 2566 | if (!dm_suspended_md(md)) |
93c534ae | 2567 | goto out; |
1da177e4 | 2568 | |
3ae70656 MS |
2569 | /* |
2570 | * If the new table has no data devices, retain the existing limits. | |
2571 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2572 | * then new I/O is queued based on these limits, and then some paths | |
2573 | * reappear. | |
2574 | */ | |
2575 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2576 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2577 | if (live_map) |
2578 | limits = md->queue->limits; | |
83d5e5b0 | 2579 | dm_put_live_table_fast(md); |
3ae70656 MS |
2580 | } |
2581 | ||
87eb5b21 MC |
2582 | if (!live_map) { |
2583 | r = dm_calculate_queue_limits(table, &limits); | |
2584 | if (r) { | |
2585 | map = ERR_PTR(r); | |
2586 | goto out; | |
2587 | } | |
042d2a9b | 2588 | } |
754c5fc7 | 2589 | |
042d2a9b | 2590 | map = __bind(md, table, &limits); |
1da177e4 | 2591 | |
93c534ae | 2592 | out: |
e61290a4 | 2593 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2594 | return map; |
1da177e4 LT |
2595 | } |
2596 | ||
2597 | /* | |
2598 | * Functions to lock and unlock any filesystem running on the | |
2599 | * device. | |
2600 | */ | |
2ca3310e | 2601 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2602 | { |
e39e2e95 | 2603 | int r; |
1da177e4 LT |
2604 | |
2605 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2606 | |
db8fef4f | 2607 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2608 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2609 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2610 | md->frozen_sb = NULL; |
2611 | return r; | |
dfbe03f6 AK |
2612 | } |
2613 | ||
aa8d7c2f AK |
2614 | set_bit(DMF_FROZEN, &md->flags); |
2615 | ||
1da177e4 LT |
2616 | return 0; |
2617 | } | |
2618 | ||
2ca3310e | 2619 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2620 | { |
aa8d7c2f AK |
2621 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2622 | return; | |
2623 | ||
db8fef4f | 2624 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2625 | md->frozen_sb = NULL; |
aa8d7c2f | 2626 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2627 | } |
2628 | ||
2629 | /* | |
2630 | * We need to be able to change a mapping table under a mounted | |
2631 | * filesystem. For example we might want to move some data in | |
2632 | * the background. Before the table can be swapped with | |
2633 | * dm_bind_table, dm_suspend must be called to flush any in | |
2634 | * flight bios and ensure that any further io gets deferred. | |
2635 | */ | |
cec47e3d KU |
2636 | /* |
2637 | * Suspend mechanism in request-based dm. | |
2638 | * | |
9f518b27 KU |
2639 | * 1. Flush all I/Os by lock_fs() if needed. |
2640 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2641 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2642 | * |
9f518b27 | 2643 | * To abort suspend, start the request_queue. |
cec47e3d | 2644 | */ |
a3d77d35 | 2645 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2646 | { |
2ca3310e | 2647 | struct dm_table *map = NULL; |
46125c1c | 2648 | int r = 0; |
a3d77d35 | 2649 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2650 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2651 | |
e61290a4 | 2652 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2653 | |
4f186f8b | 2654 | if (dm_suspended_md(md)) { |
73d410c0 | 2655 | r = -EINVAL; |
d287483d | 2656 | goto out_unlock; |
73d410c0 | 2657 | } |
1da177e4 | 2658 | |
83d5e5b0 | 2659 | map = md->map; |
1da177e4 | 2660 | |
2e93ccc1 KU |
2661 | /* |
2662 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2663 | * This flag is cleared before dm_suspend returns. | |
2664 | */ | |
2665 | if (noflush) | |
2666 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2667 | ||
cf222b37 AK |
2668 | /* This does not get reverted if there's an error later. */ |
2669 | dm_table_presuspend_targets(map); | |
2670 | ||
32a926da | 2671 | /* |
9f518b27 KU |
2672 | * Flush I/O to the device. |
2673 | * Any I/O submitted after lock_fs() may not be flushed. | |
2674 | * noflush takes precedence over do_lockfs. | |
2675 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2676 | */ |
2677 | if (!noflush && do_lockfs) { | |
2678 | r = lock_fs(md); | |
2679 | if (r) | |
83d5e5b0 | 2680 | goto out_unlock; |
aa8d7c2f | 2681 | } |
1da177e4 LT |
2682 | |
2683 | /* | |
3b00b203 MP |
2684 | * Here we must make sure that no processes are submitting requests |
2685 | * to target drivers i.e. no one may be executing | |
2686 | * __split_and_process_bio. This is called from dm_request and | |
2687 | * dm_wq_work. | |
2688 | * | |
2689 | * To get all processes out of __split_and_process_bio in dm_request, | |
2690 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2691 | * __split_and_process_bio from dm_request and quiesce the thread |
2692 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2693 | * flush_workqueue(md->wq). | |
1da177e4 | 2694 | */ |
1eb787ec | 2695 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
83d5e5b0 | 2696 | synchronize_srcu(&md->io_barrier); |
1da177e4 | 2697 | |
d0bcb878 | 2698 | /* |
29e4013d TH |
2699 | * Stop md->queue before flushing md->wq in case request-based |
2700 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2701 | */ |
cec47e3d | 2702 | if (dm_request_based(md)) |
9f518b27 | 2703 | stop_queue(md->queue); |
cec47e3d | 2704 | |
d0bcb878 KU |
2705 | flush_workqueue(md->wq); |
2706 | ||
1da177e4 | 2707 | /* |
3b00b203 MP |
2708 | * At this point no more requests are entering target request routines. |
2709 | * We call dm_wait_for_completion to wait for all existing requests | |
2710 | * to finish. | |
1da177e4 | 2711 | */ |
401600df | 2712 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2713 | |
6d6f10df | 2714 | if (noflush) |
022c2611 | 2715 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
83d5e5b0 | 2716 | synchronize_srcu(&md->io_barrier); |
2e93ccc1 | 2717 | |
1da177e4 | 2718 | /* were we interrupted ? */ |
46125c1c | 2719 | if (r < 0) { |
9a1fb464 | 2720 | dm_queue_flush(md); |
73d410c0 | 2721 | |
cec47e3d | 2722 | if (dm_request_based(md)) |
9f518b27 | 2723 | start_queue(md->queue); |
cec47e3d | 2724 | |
2ca3310e | 2725 | unlock_fs(md); |
83d5e5b0 | 2726 | goto out_unlock; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2727 | } |
1da177e4 | 2728 | |
3b00b203 MP |
2729 | /* |
2730 | * If dm_wait_for_completion returned 0, the device is completely | |
2731 | * quiescent now. There is no request-processing activity. All new | |
2732 | * requests are being added to md->deferred list. | |
2733 | */ | |
2734 | ||
2ca3310e | 2735 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2736 | |
4d4471cb KU |
2737 | dm_table_postsuspend_targets(map); |
2738 | ||
d287483d | 2739 | out_unlock: |
e61290a4 | 2740 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2741 | return r; |
1da177e4 LT |
2742 | } |
2743 | ||
2744 | int dm_resume(struct mapped_device *md) | |
2745 | { | |
cf222b37 | 2746 | int r = -EINVAL; |
cf222b37 | 2747 | struct dm_table *map = NULL; |
1da177e4 | 2748 | |
e61290a4 | 2749 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2750 | if (!dm_suspended_md(md)) |
cf222b37 | 2751 | goto out; |
cf222b37 | 2752 | |
83d5e5b0 | 2753 | map = md->map; |
2ca3310e | 2754 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2755 | goto out; |
1da177e4 | 2756 | |
8757b776 MB |
2757 | r = dm_table_resume_targets(map); |
2758 | if (r) | |
2759 | goto out; | |
2ca3310e | 2760 | |
9a1fb464 | 2761 | dm_queue_flush(md); |
2ca3310e | 2762 | |
cec47e3d KU |
2763 | /* |
2764 | * Flushing deferred I/Os must be done after targets are resumed | |
2765 | * so that mapping of targets can work correctly. | |
2766 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2767 | */ | |
2768 | if (dm_request_based(md)) | |
2769 | start_queue(md->queue); | |
2770 | ||
2ca3310e AK |
2771 | unlock_fs(md); |
2772 | ||
2773 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2774 | ||
cf222b37 AK |
2775 | r = 0; |
2776 | out: | |
e61290a4 | 2777 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2778 | |
cf222b37 | 2779 | return r; |
1da177e4 LT |
2780 | } |
2781 | ||
fd2ed4d2 MP |
2782 | /* |
2783 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2784 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2785 | * It may be used only from the kernel. | |
2786 | * | |
2787 | * Internal suspend holds md->suspend_lock, which prevents interaction with | |
2788 | * userspace-driven suspend. | |
2789 | */ | |
2790 | ||
2791 | void dm_internal_suspend(struct mapped_device *md) | |
2792 | { | |
2793 | mutex_lock(&md->suspend_lock); | |
2794 | if (dm_suspended_md(md)) | |
2795 | return; | |
2796 | ||
2797 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2798 | synchronize_srcu(&md->io_barrier); | |
2799 | flush_workqueue(md->wq); | |
2800 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2801 | } | |
2802 | ||
2803 | void dm_internal_resume(struct mapped_device *md) | |
2804 | { | |
2805 | if (dm_suspended_md(md)) | |
2806 | goto done; | |
2807 | ||
2808 | dm_queue_flush(md); | |
2809 | ||
2810 | done: | |
2811 | mutex_unlock(&md->suspend_lock); | |
2812 | } | |
2813 | ||
1da177e4 LT |
2814 | /*----------------------------------------------------------------- |
2815 | * Event notification. | |
2816 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2817 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2818 | unsigned cookie) |
69267a30 | 2819 | { |
60935eb2 MB |
2820 | char udev_cookie[DM_COOKIE_LENGTH]; |
2821 | char *envp[] = { udev_cookie, NULL }; | |
2822 | ||
2823 | if (!cookie) | |
3abf85b5 | 2824 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2825 | else { |
2826 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2827 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2828 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2829 | action, envp); | |
60935eb2 | 2830 | } |
69267a30 AK |
2831 | } |
2832 | ||
7a8c3d3b MA |
2833 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2834 | { | |
2835 | return atomic_add_return(1, &md->uevent_seq); | |
2836 | } | |
2837 | ||
1da177e4 LT |
2838 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2839 | { | |
2840 | return atomic_read(&md->event_nr); | |
2841 | } | |
2842 | ||
2843 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2844 | { | |
2845 | return wait_event_interruptible(md->eventq, | |
2846 | (event_nr != atomic_read(&md->event_nr))); | |
2847 | } | |
2848 | ||
7a8c3d3b MA |
2849 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2850 | { | |
2851 | unsigned long flags; | |
2852 | ||
2853 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2854 | list_add(elist, &md->uevent_list); | |
2855 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2856 | } | |
2857 | ||
1da177e4 LT |
2858 | /* |
2859 | * The gendisk is only valid as long as you have a reference | |
2860 | * count on 'md'. | |
2861 | */ | |
2862 | struct gendisk *dm_disk(struct mapped_device *md) | |
2863 | { | |
2864 | return md->disk; | |
2865 | } | |
2866 | ||
784aae73 MB |
2867 | struct kobject *dm_kobject(struct mapped_device *md) |
2868 | { | |
2869 | return &md->kobj; | |
2870 | } | |
2871 | ||
2872 | /* | |
2873 | * struct mapped_device should not be exported outside of dm.c | |
2874 | * so use this check to verify that kobj is part of md structure | |
2875 | */ | |
2876 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2877 | { | |
2878 | struct mapped_device *md; | |
2879 | ||
2880 | md = container_of(kobj, struct mapped_device, kobj); | |
2881 | if (&md->kobj != kobj) | |
2882 | return NULL; | |
2883 | ||
4d89b7b4 | 2884 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2885 | dm_deleting_md(md)) |
4d89b7b4 MB |
2886 | return NULL; |
2887 | ||
784aae73 MB |
2888 | dm_get(md); |
2889 | return md; | |
2890 | } | |
2891 | ||
4f186f8b | 2892 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2893 | { |
2894 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2895 | } | |
2896 | ||
64dbce58 KU |
2897 | int dm_suspended(struct dm_target *ti) |
2898 | { | |
ecdb2e25 | 2899 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2900 | } |
2901 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2902 | ||
2e93ccc1 KU |
2903 | int dm_noflush_suspending(struct dm_target *ti) |
2904 | { | |
ecdb2e25 | 2905 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2906 | } |
2907 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2908 | ||
c0820cf5 | 2909 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 2910 | { |
5f015204 JN |
2911 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
2912 | struct kmem_cache *cachep; | |
2913 | unsigned int pool_size; | |
2914 | unsigned int front_pad; | |
e6ee8c0b KU |
2915 | |
2916 | if (!pools) | |
2917 | return NULL; | |
2918 | ||
23e5083b | 2919 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 | 2920 | cachep = _io_cache; |
e8603136 | 2921 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 JN |
2922 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
2923 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
2924 | cachep = _rq_tio_cache; | |
f4790826 | 2925 | pool_size = dm_get_reserved_rq_based_ios(); |
5f015204 JN |
2926 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
2927 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
2928 | WARN_ON(per_bio_data_size != 0); | |
2929 | } else | |
2930 | goto out; | |
e6ee8c0b | 2931 | |
6cfa5857 | 2932 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
2933 | if (!pools->io_pool) |
2934 | goto out; | |
e6ee8c0b | 2935 | |
5f015204 | 2936 | pools->bs = bioset_create(pool_size, front_pad); |
e6ee8c0b | 2937 | if (!pools->bs) |
5f015204 | 2938 | goto out; |
e6ee8c0b | 2939 | |
a91a2785 | 2940 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2941 | goto out; |
a91a2785 | 2942 | |
e6ee8c0b KU |
2943 | return pools; |
2944 | ||
5f015204 JN |
2945 | out: |
2946 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
2947 | |
2948 | return NULL; | |
2949 | } | |
2950 | ||
2951 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2952 | { | |
2953 | if (!pools) | |
2954 | return; | |
2955 | ||
2956 | if (pools->io_pool) | |
2957 | mempool_destroy(pools->io_pool); | |
2958 | ||
e6ee8c0b KU |
2959 | if (pools->bs) |
2960 | bioset_free(pools->bs); | |
2961 | ||
2962 | kfree(pools); | |
2963 | } | |
2964 | ||
83d5cde4 | 2965 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2966 | .open = dm_blk_open, |
2967 | .release = dm_blk_close, | |
aa129a22 | 2968 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2969 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2970 | .owner = THIS_MODULE |
2971 | }; | |
2972 | ||
2973 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2974 | ||
2975 | /* | |
2976 | * module hooks | |
2977 | */ | |
2978 | module_init(dm_init); | |
2979 | module_exit(dm_exit); | |
2980 | ||
2981 | module_param(major, uint, 0); | |
2982 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2983 | |
e8603136 MS |
2984 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2985 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2986 | ||
f4790826 MS |
2987 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
2988 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
2989 | ||
1da177e4 LT |
2990 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2991 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2992 | MODULE_LICENSE("GPL"); |