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> |
ffcc3936 | 22 | #include <linux/wait.h> |
2eb6e1e3 | 23 | #include <linux/kthread.h> |
0ce65797 | 24 | #include <linux/ktime.h> |
de3ec86d | 25 | #include <linux/elevator.h> /* for rq_end_sector() */ |
bfebd1cd | 26 | #include <linux/blk-mq.h> |
55782138 LZ |
27 | |
28 | #include <trace/events/block.h> | |
1da177e4 | 29 | |
72d94861 AK |
30 | #define DM_MSG_PREFIX "core" |
31 | ||
71a16736 NK |
32 | #ifdef CONFIG_PRINTK |
33 | /* | |
34 | * ratelimit state to be used in DMXXX_LIMIT(). | |
35 | */ | |
36 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
37 | DEFAULT_RATELIMIT_INTERVAL, | |
38 | DEFAULT_RATELIMIT_BURST); | |
39 | EXPORT_SYMBOL(dm_ratelimit_state); | |
40 | #endif | |
41 | ||
60935eb2 MB |
42 | /* |
43 | * Cookies are numeric values sent with CHANGE and REMOVE | |
44 | * uevents while resuming, removing or renaming the device. | |
45 | */ | |
46 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
47 | #define DM_COOKIE_LENGTH 24 | |
48 | ||
1da177e4 LT |
49 | static const char *_name = DM_NAME; |
50 | ||
51 | static unsigned int major = 0; | |
52 | static unsigned int _major = 0; | |
53 | ||
d15b774c AK |
54 | static DEFINE_IDR(_minor_idr); |
55 | ||
f32c10b0 | 56 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
57 | |
58 | static void do_deferred_remove(struct work_struct *w); | |
59 | ||
60 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
61 | ||
acfe0ad7 MP |
62 | static struct workqueue_struct *deferred_remove_workqueue; |
63 | ||
1da177e4 | 64 | /* |
8fbf26ad | 65 | * For bio-based dm. |
1da177e4 LT |
66 | * One of these is allocated per bio. |
67 | */ | |
68 | struct dm_io { | |
69 | struct mapped_device *md; | |
70 | int error; | |
1da177e4 | 71 | atomic_t io_count; |
6ae2fa67 | 72 | struct bio *bio; |
3eaf840e | 73 | unsigned long start_time; |
f88fb981 | 74 | spinlock_t endio_lock; |
fd2ed4d2 | 75 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
76 | }; |
77 | ||
8fbf26ad KU |
78 | /* |
79 | * For request-based dm. | |
80 | * One of these is allocated per request. | |
81 | */ | |
82 | struct dm_rq_target_io { | |
83 | struct mapped_device *md; | |
84 | struct dm_target *ti; | |
1ae49ea2 | 85 | struct request *orig, *clone; |
2eb6e1e3 | 86 | struct kthread_work work; |
8fbf26ad KU |
87 | int error; |
88 | union map_info info; | |
89 | }; | |
90 | ||
91 | /* | |
94818742 KO |
92 | * For request-based dm - the bio clones we allocate are embedded in these |
93 | * structs. | |
94 | * | |
95 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
96 | * the bioset is created - this means the bio has to come at the end of the | |
97 | * struct. | |
8fbf26ad KU |
98 | */ |
99 | struct dm_rq_clone_bio_info { | |
100 | struct bio *orig; | |
cec47e3d | 101 | struct dm_rq_target_io *tio; |
94818742 | 102 | struct bio clone; |
8fbf26ad KU |
103 | }; |
104 | ||
cec47e3d KU |
105 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
106 | { | |
107 | if (rq && rq->end_io_data) | |
108 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
109 | return NULL; | |
110 | } | |
111 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
112 | ||
ba61fdd1 JM |
113 | #define MINOR_ALLOCED ((void *)-1) |
114 | ||
1da177e4 LT |
115 | /* |
116 | * Bits for the md->flags field. | |
117 | */ | |
1eb787ec | 118 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 119 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 120 | #define DMF_FROZEN 2 |
fba9f90e | 121 | #define DMF_FREEING 3 |
5c6bd75d | 122 | #define DMF_DELETING 4 |
2e93ccc1 | 123 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 124 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 125 | #define DMF_DEFERRED_REMOVE 7 |
ffcc3936 | 126 | #define DMF_SUSPENDED_INTERNALLY 8 |
1da177e4 | 127 | |
83d5e5b0 MP |
128 | /* |
129 | * A dummy definition to make RCU happy. | |
130 | * struct dm_table should never be dereferenced in this file. | |
131 | */ | |
132 | struct dm_table { | |
133 | int undefined__; | |
134 | }; | |
135 | ||
304f3f6a MB |
136 | /* |
137 | * Work processed by per-device workqueue. | |
138 | */ | |
1da177e4 | 139 | struct mapped_device { |
83d5e5b0 | 140 | struct srcu_struct io_barrier; |
e61290a4 | 141 | struct mutex suspend_lock; |
1da177e4 | 142 | atomic_t holders; |
5c6bd75d | 143 | atomic_t open_count; |
1da177e4 | 144 | |
2a7faeb1 MP |
145 | /* |
146 | * The current mapping. | |
147 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
148 | * dereference. | |
149 | */ | |
6fa99520 | 150 | struct dm_table __rcu *map; |
2a7faeb1 | 151 | |
86f1152b BM |
152 | struct list_head table_devices; |
153 | struct mutex table_devices_lock; | |
154 | ||
1da177e4 LT |
155 | unsigned long flags; |
156 | ||
165125e1 | 157 | struct request_queue *queue; |
a5664dad | 158 | unsigned type; |
4a0b4ddf | 159 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
160 | struct mutex type_lock; |
161 | ||
36a0456f AK |
162 | struct target_type *immutable_target_type; |
163 | ||
1da177e4 | 164 | struct gendisk *disk; |
7e51f257 | 165 | char name[16]; |
1da177e4 LT |
166 | |
167 | void *interface_ptr; | |
168 | ||
169 | /* | |
170 | * A list of ios that arrived while we were suspended. | |
171 | */ | |
316d315b | 172 | atomic_t pending[2]; |
1da177e4 | 173 | wait_queue_head_t wait; |
53d5914f | 174 | struct work_struct work; |
74859364 | 175 | struct bio_list deferred; |
022c2611 | 176 | spinlock_t deferred_lock; |
1da177e4 | 177 | |
af7e466a | 178 | /* |
29e4013d | 179 | * Processing queue (flush) |
304f3f6a MB |
180 | */ |
181 | struct workqueue_struct *wq; | |
182 | ||
1da177e4 LT |
183 | /* |
184 | * io objects are allocated from here. | |
185 | */ | |
186 | mempool_t *io_pool; | |
1ae49ea2 | 187 | mempool_t *rq_pool; |
1da177e4 | 188 | |
9faf400f SB |
189 | struct bio_set *bs; |
190 | ||
1da177e4 LT |
191 | /* |
192 | * Event handling. | |
193 | */ | |
194 | atomic_t event_nr; | |
195 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
196 | atomic_t uevent_seq; |
197 | struct list_head uevent_list; | |
198 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
199 | |
200 | /* | |
201 | * freeze/thaw support require holding onto a super block | |
202 | */ | |
203 | struct super_block *frozen_sb; | |
db8fef4f | 204 | struct block_device *bdev; |
3ac51e74 DW |
205 | |
206 | /* forced geometry settings */ | |
207 | struct hd_geometry geometry; | |
784aae73 | 208 | |
2995fa78 MP |
209 | /* kobject and completion */ |
210 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 211 | |
d87f4c14 TH |
212 | /* zero-length flush that will be cloned and submitted to targets */ |
213 | struct bio flush_bio; | |
fd2ed4d2 | 214 | |
96b26c8c MP |
215 | /* the number of internal suspends */ |
216 | unsigned internal_suspend_count; | |
217 | ||
fd2ed4d2 | 218 | struct dm_stats stats; |
2eb6e1e3 KB |
219 | |
220 | struct kthread_worker kworker; | |
221 | struct task_struct *kworker_task; | |
de3ec86d MS |
222 | |
223 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 224 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 225 | int last_rq_rw; |
0ce65797 MS |
226 | sector_t last_rq_pos; |
227 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
228 | |
229 | /* for blk-mq request-based DM support */ | |
230 | struct blk_mq_tag_set tag_set; | |
1da177e4 LT |
231 | }; |
232 | ||
e6ee8c0b KU |
233 | /* |
234 | * For mempools pre-allocation at the table loading time. | |
235 | */ | |
236 | struct dm_md_mempools { | |
237 | mempool_t *io_pool; | |
1ae49ea2 | 238 | mempool_t *rq_pool; |
e6ee8c0b KU |
239 | struct bio_set *bs; |
240 | }; | |
241 | ||
86f1152b BM |
242 | struct table_device { |
243 | struct list_head list; | |
244 | atomic_t count; | |
245 | struct dm_dev dm_dev; | |
246 | }; | |
247 | ||
6cfa5857 MS |
248 | #define RESERVED_BIO_BASED_IOS 16 |
249 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 250 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 251 | static struct kmem_cache *_io_cache; |
8fbf26ad | 252 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 253 | static struct kmem_cache *_rq_cache; |
94818742 | 254 | |
e8603136 MS |
255 | /* |
256 | * Bio-based DM's mempools' reserved IOs set by the user. | |
257 | */ | |
258 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
259 | ||
f4790826 MS |
260 | /* |
261 | * Request-based DM's mempools' reserved IOs set by the user. | |
262 | */ | |
263 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
264 | ||
09c2d531 | 265 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
266 | unsigned def, unsigned max) |
267 | { | |
09c2d531 MS |
268 | unsigned param = ACCESS_ONCE(*module_param); |
269 | unsigned modified_param = 0; | |
f4790826 | 270 | |
09c2d531 MS |
271 | if (!param) |
272 | modified_param = def; | |
273 | else if (param > max) | |
274 | modified_param = max; | |
f4790826 | 275 | |
09c2d531 MS |
276 | if (modified_param) { |
277 | (void)cmpxchg(module_param, param, modified_param); | |
278 | param = modified_param; | |
f4790826 MS |
279 | } |
280 | ||
09c2d531 | 281 | return param; |
f4790826 MS |
282 | } |
283 | ||
e8603136 MS |
284 | unsigned dm_get_reserved_bio_based_ios(void) |
285 | { | |
09c2d531 | 286 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
287 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
288 | } | |
289 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
290 | ||
f4790826 MS |
291 | unsigned dm_get_reserved_rq_based_ios(void) |
292 | { | |
09c2d531 | 293 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
294 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
295 | } | |
296 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
297 | ||
1da177e4 LT |
298 | static int __init local_init(void) |
299 | { | |
51157b4a | 300 | int r = -ENOMEM; |
1da177e4 | 301 | |
1da177e4 | 302 | /* allocate a slab for the dm_ios */ |
028867ac | 303 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 304 | if (!_io_cache) |
51157b4a | 305 | return r; |
1da177e4 | 306 | |
8fbf26ad KU |
307 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
308 | if (!_rq_tio_cache) | |
dba14160 | 309 | goto out_free_io_cache; |
8fbf26ad | 310 | |
1ae49ea2 MS |
311 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
312 | __alignof__(struct request), 0, NULL); | |
313 | if (!_rq_cache) | |
314 | goto out_free_rq_tio_cache; | |
315 | ||
51e5b2bd | 316 | r = dm_uevent_init(); |
51157b4a | 317 | if (r) |
1ae49ea2 | 318 | goto out_free_rq_cache; |
51e5b2bd | 319 | |
acfe0ad7 MP |
320 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
321 | if (!deferred_remove_workqueue) { | |
322 | r = -ENOMEM; | |
323 | goto out_uevent_exit; | |
324 | } | |
325 | ||
1da177e4 LT |
326 | _major = major; |
327 | r = register_blkdev(_major, _name); | |
51157b4a | 328 | if (r < 0) |
acfe0ad7 | 329 | goto out_free_workqueue; |
1da177e4 LT |
330 | |
331 | if (!_major) | |
332 | _major = r; | |
333 | ||
334 | return 0; | |
51157b4a | 335 | |
acfe0ad7 MP |
336 | out_free_workqueue: |
337 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
338 | out_uevent_exit: |
339 | dm_uevent_exit(); | |
1ae49ea2 MS |
340 | out_free_rq_cache: |
341 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
342 | out_free_rq_tio_cache: |
343 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
344 | out_free_io_cache: |
345 | kmem_cache_destroy(_io_cache); | |
346 | ||
347 | return r; | |
1da177e4 LT |
348 | } |
349 | ||
350 | static void local_exit(void) | |
351 | { | |
2c140a24 | 352 | flush_scheduled_work(); |
acfe0ad7 | 353 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 354 | |
1ae49ea2 | 355 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 356 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 357 | kmem_cache_destroy(_io_cache); |
00d59405 | 358 | unregister_blkdev(_major, _name); |
51e5b2bd | 359 | dm_uevent_exit(); |
1da177e4 LT |
360 | |
361 | _major = 0; | |
362 | ||
363 | DMINFO("cleaned up"); | |
364 | } | |
365 | ||
b9249e55 | 366 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
367 | local_init, |
368 | dm_target_init, | |
369 | dm_linear_init, | |
370 | dm_stripe_init, | |
952b3557 | 371 | dm_io_init, |
945fa4d2 | 372 | dm_kcopyd_init, |
1da177e4 | 373 | dm_interface_init, |
fd2ed4d2 | 374 | dm_statistics_init, |
1da177e4 LT |
375 | }; |
376 | ||
b9249e55 | 377 | static void (*_exits[])(void) = { |
1da177e4 LT |
378 | local_exit, |
379 | dm_target_exit, | |
380 | dm_linear_exit, | |
381 | dm_stripe_exit, | |
952b3557 | 382 | dm_io_exit, |
945fa4d2 | 383 | dm_kcopyd_exit, |
1da177e4 | 384 | dm_interface_exit, |
fd2ed4d2 | 385 | dm_statistics_exit, |
1da177e4 LT |
386 | }; |
387 | ||
388 | static int __init dm_init(void) | |
389 | { | |
390 | const int count = ARRAY_SIZE(_inits); | |
391 | ||
392 | int r, i; | |
393 | ||
394 | for (i = 0; i < count; i++) { | |
395 | r = _inits[i](); | |
396 | if (r) | |
397 | goto bad; | |
398 | } | |
399 | ||
400 | return 0; | |
401 | ||
402 | bad: | |
403 | while (i--) | |
404 | _exits[i](); | |
405 | ||
406 | return r; | |
407 | } | |
408 | ||
409 | static void __exit dm_exit(void) | |
410 | { | |
411 | int i = ARRAY_SIZE(_exits); | |
412 | ||
413 | while (i--) | |
414 | _exits[i](); | |
d15b774c AK |
415 | |
416 | /* | |
417 | * Should be empty by this point. | |
418 | */ | |
d15b774c | 419 | idr_destroy(&_minor_idr); |
1da177e4 LT |
420 | } |
421 | ||
422 | /* | |
423 | * Block device functions | |
424 | */ | |
432a212c MA |
425 | int dm_deleting_md(struct mapped_device *md) |
426 | { | |
427 | return test_bit(DMF_DELETING, &md->flags); | |
428 | } | |
429 | ||
fe5f9f2c | 430 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
431 | { |
432 | struct mapped_device *md; | |
433 | ||
fba9f90e JM |
434 | spin_lock(&_minor_lock); |
435 | ||
fe5f9f2c | 436 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
437 | if (!md) |
438 | goto out; | |
439 | ||
5c6bd75d | 440 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 441 | dm_deleting_md(md)) { |
fba9f90e JM |
442 | md = NULL; |
443 | goto out; | |
444 | } | |
445 | ||
1da177e4 | 446 | dm_get(md); |
5c6bd75d | 447 | atomic_inc(&md->open_count); |
fba9f90e JM |
448 | out: |
449 | spin_unlock(&_minor_lock); | |
450 | ||
451 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
452 | } |
453 | ||
db2a144b | 454 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 455 | { |
63a4f065 | 456 | struct mapped_device *md; |
6e9624b8 | 457 | |
4a1aeb98 MB |
458 | spin_lock(&_minor_lock); |
459 | ||
63a4f065 MS |
460 | md = disk->private_data; |
461 | if (WARN_ON(!md)) | |
462 | goto out; | |
463 | ||
2c140a24 MP |
464 | if (atomic_dec_and_test(&md->open_count) && |
465 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 466 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 467 | |
1da177e4 | 468 | dm_put(md); |
63a4f065 | 469 | out: |
4a1aeb98 | 470 | spin_unlock(&_minor_lock); |
1da177e4 LT |
471 | } |
472 | ||
5c6bd75d AK |
473 | int dm_open_count(struct mapped_device *md) |
474 | { | |
475 | return atomic_read(&md->open_count); | |
476 | } | |
477 | ||
478 | /* | |
479 | * Guarantees nothing is using the device before it's deleted. | |
480 | */ | |
2c140a24 | 481 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
482 | { |
483 | int r = 0; | |
484 | ||
485 | spin_lock(&_minor_lock); | |
486 | ||
2c140a24 | 487 | if (dm_open_count(md)) { |
5c6bd75d | 488 | r = -EBUSY; |
2c140a24 MP |
489 | if (mark_deferred) |
490 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
491 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
492 | r = -EEXIST; | |
5c6bd75d AK |
493 | else |
494 | set_bit(DMF_DELETING, &md->flags); | |
495 | ||
496 | spin_unlock(&_minor_lock); | |
497 | ||
498 | return r; | |
499 | } | |
500 | ||
2c140a24 MP |
501 | int dm_cancel_deferred_remove(struct mapped_device *md) |
502 | { | |
503 | int r = 0; | |
504 | ||
505 | spin_lock(&_minor_lock); | |
506 | ||
507 | if (test_bit(DMF_DELETING, &md->flags)) | |
508 | r = -EBUSY; | |
509 | else | |
510 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
511 | ||
512 | spin_unlock(&_minor_lock); | |
513 | ||
514 | return r; | |
515 | } | |
516 | ||
517 | static void do_deferred_remove(struct work_struct *w) | |
518 | { | |
519 | dm_deferred_remove(); | |
520 | } | |
521 | ||
fd2ed4d2 MP |
522 | sector_t dm_get_size(struct mapped_device *md) |
523 | { | |
524 | return get_capacity(md->disk); | |
525 | } | |
526 | ||
9974fa2c MS |
527 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
528 | { | |
529 | return md->queue; | |
530 | } | |
531 | ||
fd2ed4d2 MP |
532 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
533 | { | |
534 | return &md->stats; | |
535 | } | |
536 | ||
3ac51e74 DW |
537 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
538 | { | |
539 | struct mapped_device *md = bdev->bd_disk->private_data; | |
540 | ||
541 | return dm_get_geometry(md, geo); | |
542 | } | |
543 | ||
fe5f9f2c | 544 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
545 | unsigned int cmd, unsigned long arg) |
546 | { | |
fe5f9f2c | 547 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 548 | int srcu_idx; |
6c182cd8 | 549 | struct dm_table *map; |
aa129a22 MB |
550 | struct dm_target *tgt; |
551 | int r = -ENOTTY; | |
552 | ||
6c182cd8 | 553 | retry: |
83d5e5b0 MP |
554 | map = dm_get_live_table(md, &srcu_idx); |
555 | ||
aa129a22 MB |
556 | if (!map || !dm_table_get_size(map)) |
557 | goto out; | |
558 | ||
559 | /* We only support devices that have a single target */ | |
560 | if (dm_table_get_num_targets(map) != 1) | |
561 | goto out; | |
562 | ||
563 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
564 | if (!tgt->type->ioctl) |
565 | goto out; | |
aa129a22 | 566 | |
4f186f8b | 567 | if (dm_suspended_md(md)) { |
aa129a22 MB |
568 | r = -EAGAIN; |
569 | goto out; | |
570 | } | |
571 | ||
4d341d82 | 572 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
573 | |
574 | out: | |
83d5e5b0 | 575 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 576 | |
6c182cd8 HR |
577 | if (r == -ENOTCONN) { |
578 | msleep(10); | |
579 | goto retry; | |
580 | } | |
581 | ||
aa129a22 MB |
582 | return r; |
583 | } | |
584 | ||
028867ac | 585 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
586 | { |
587 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
588 | } | |
589 | ||
028867ac | 590 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
591 | { |
592 | mempool_free(io, md->io_pool); | |
593 | } | |
594 | ||
028867ac | 595 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 596 | { |
dba14160 | 597 | bio_put(&tio->clone); |
1da177e4 LT |
598 | } |
599 | ||
08885643 KU |
600 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
601 | gfp_t gfp_mask) | |
cec47e3d | 602 | { |
5f015204 | 603 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
604 | } |
605 | ||
606 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
607 | { | |
5f015204 | 608 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
609 | } |
610 | ||
1ae49ea2 MS |
611 | static struct request *alloc_clone_request(struct mapped_device *md, |
612 | gfp_t gfp_mask) | |
613 | { | |
614 | return mempool_alloc(md->rq_pool, gfp_mask); | |
615 | } | |
616 | ||
617 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
618 | { | |
619 | mempool_free(rq, md->rq_pool); | |
620 | } | |
621 | ||
90abb8c4 KU |
622 | static int md_in_flight(struct mapped_device *md) |
623 | { | |
624 | return atomic_read(&md->pending[READ]) + | |
625 | atomic_read(&md->pending[WRITE]); | |
626 | } | |
627 | ||
3eaf840e JNN |
628 | static void start_io_acct(struct dm_io *io) |
629 | { | |
630 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 631 | struct bio *bio = io->bio; |
c9959059 | 632 | int cpu; |
fd2ed4d2 | 633 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
634 | |
635 | io->start_time = jiffies; | |
636 | ||
074a7aca TH |
637 | cpu = part_stat_lock(); |
638 | part_round_stats(cpu, &dm_disk(md)->part0); | |
639 | part_stat_unlock(); | |
1e9bb880 SL |
640 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
641 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
642 | |
643 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 644 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 645 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
646 | } |
647 | ||
d221d2e7 | 648 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
649 | { |
650 | struct mapped_device *md = io->md; | |
651 | struct bio *bio = io->bio; | |
652 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 653 | int pending; |
3eaf840e JNN |
654 | int rw = bio_data_dir(bio); |
655 | ||
18c0b223 | 656 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 657 | |
fd2ed4d2 | 658 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 659 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
660 | bio_sectors(bio), true, duration, &io->stats_aux); |
661 | ||
af7e466a MP |
662 | /* |
663 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 664 | * a flush. |
af7e466a | 665 | */ |
1e9bb880 SL |
666 | pending = atomic_dec_return(&md->pending[rw]); |
667 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 668 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 669 | |
d221d2e7 MP |
670 | /* nudge anyone waiting on suspend queue */ |
671 | if (!pending) | |
672 | wake_up(&md->wait); | |
3eaf840e JNN |
673 | } |
674 | ||
1da177e4 LT |
675 | /* |
676 | * Add the bio to the list of deferred io. | |
677 | */ | |
92c63902 | 678 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 679 | { |
05447420 | 680 | unsigned long flags; |
1da177e4 | 681 | |
05447420 | 682 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 683 | bio_list_add(&md->deferred, bio); |
05447420 | 684 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 685 | queue_work(md->wq, &md->work); |
1da177e4 LT |
686 | } |
687 | ||
688 | /* | |
689 | * Everyone (including functions in this file), should use this | |
690 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 691 | * dm_put_live_table() when finished. |
1da177e4 | 692 | */ |
83d5e5b0 | 693 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 694 | { |
83d5e5b0 MP |
695 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
696 | ||
697 | return srcu_dereference(md->map, &md->io_barrier); | |
698 | } | |
1da177e4 | 699 | |
83d5e5b0 MP |
700 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
701 | { | |
702 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
703 | } | |
704 | ||
705 | void dm_sync_table(struct mapped_device *md) | |
706 | { | |
707 | synchronize_srcu(&md->io_barrier); | |
708 | synchronize_rcu_expedited(); | |
709 | } | |
710 | ||
711 | /* | |
712 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
713 | * The caller must not block between these two functions. | |
714 | */ | |
715 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
716 | { | |
717 | rcu_read_lock(); | |
718 | return rcu_dereference(md->map); | |
719 | } | |
1da177e4 | 720 | |
83d5e5b0 MP |
721 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
722 | { | |
723 | rcu_read_unlock(); | |
1da177e4 LT |
724 | } |
725 | ||
86f1152b BM |
726 | /* |
727 | * Open a table device so we can use it as a map destination. | |
728 | */ | |
729 | static int open_table_device(struct table_device *td, dev_t dev, | |
730 | struct mapped_device *md) | |
731 | { | |
732 | static char *_claim_ptr = "I belong to device-mapper"; | |
733 | struct block_device *bdev; | |
734 | ||
735 | int r; | |
736 | ||
737 | BUG_ON(td->dm_dev.bdev); | |
738 | ||
739 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
740 | if (IS_ERR(bdev)) | |
741 | return PTR_ERR(bdev); | |
742 | ||
743 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
744 | if (r) { | |
745 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
746 | return r; | |
747 | } | |
748 | ||
749 | td->dm_dev.bdev = bdev; | |
750 | return 0; | |
751 | } | |
752 | ||
753 | /* | |
754 | * Close a table device that we've been using. | |
755 | */ | |
756 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
757 | { | |
758 | if (!td->dm_dev.bdev) | |
759 | return; | |
760 | ||
761 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
762 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
763 | td->dm_dev.bdev = NULL; | |
764 | } | |
765 | ||
766 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
767 | fmode_t mode) { | |
768 | struct table_device *td; | |
769 | ||
770 | list_for_each_entry(td, l, list) | |
771 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
772 | return td; | |
773 | ||
774 | return NULL; | |
775 | } | |
776 | ||
777 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
778 | struct dm_dev **result) { | |
779 | int r; | |
780 | struct table_device *td; | |
781 | ||
782 | mutex_lock(&md->table_devices_lock); | |
783 | td = find_table_device(&md->table_devices, dev, mode); | |
784 | if (!td) { | |
785 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
786 | if (!td) { | |
787 | mutex_unlock(&md->table_devices_lock); | |
788 | return -ENOMEM; | |
789 | } | |
790 | ||
791 | td->dm_dev.mode = mode; | |
792 | td->dm_dev.bdev = NULL; | |
793 | ||
794 | if ((r = open_table_device(td, dev, md))) { | |
795 | mutex_unlock(&md->table_devices_lock); | |
796 | kfree(td); | |
797 | return r; | |
798 | } | |
799 | ||
800 | format_dev_t(td->dm_dev.name, dev); | |
801 | ||
802 | atomic_set(&td->count, 0); | |
803 | list_add(&td->list, &md->table_devices); | |
804 | } | |
805 | atomic_inc(&td->count); | |
806 | mutex_unlock(&md->table_devices_lock); | |
807 | ||
808 | *result = &td->dm_dev; | |
809 | return 0; | |
810 | } | |
811 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
812 | ||
813 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
814 | { | |
815 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
816 | ||
817 | mutex_lock(&md->table_devices_lock); | |
818 | if (atomic_dec_and_test(&td->count)) { | |
819 | close_table_device(td, md); | |
820 | list_del(&td->list); | |
821 | kfree(td); | |
822 | } | |
823 | mutex_unlock(&md->table_devices_lock); | |
824 | } | |
825 | EXPORT_SYMBOL(dm_put_table_device); | |
826 | ||
827 | static void free_table_devices(struct list_head *devices) | |
828 | { | |
829 | struct list_head *tmp, *next; | |
830 | ||
831 | list_for_each_safe(tmp, next, devices) { | |
832 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
833 | ||
834 | DMWARN("dm_destroy: %s still exists with %d references", | |
835 | td->dm_dev.name, atomic_read(&td->count)); | |
836 | kfree(td); | |
837 | } | |
838 | } | |
839 | ||
3ac51e74 DW |
840 | /* |
841 | * Get the geometry associated with a dm device | |
842 | */ | |
843 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
844 | { | |
845 | *geo = md->geometry; | |
846 | ||
847 | return 0; | |
848 | } | |
849 | ||
850 | /* | |
851 | * Set the geometry of a device. | |
852 | */ | |
853 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
854 | { | |
855 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
856 | ||
857 | if (geo->start > sz) { | |
858 | DMWARN("Start sector is beyond the geometry limits."); | |
859 | return -EINVAL; | |
860 | } | |
861 | ||
862 | md->geometry = *geo; | |
863 | ||
864 | return 0; | |
865 | } | |
866 | ||
1da177e4 LT |
867 | /*----------------------------------------------------------------- |
868 | * CRUD START: | |
869 | * A more elegant soln is in the works that uses the queue | |
870 | * merge fn, unfortunately there are a couple of changes to | |
871 | * the block layer that I want to make for this. So in the | |
872 | * interests of getting something for people to use I give | |
873 | * you this clearly demarcated crap. | |
874 | *---------------------------------------------------------------*/ | |
875 | ||
2e93ccc1 KU |
876 | static int __noflush_suspending(struct mapped_device *md) |
877 | { | |
878 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
879 | } | |
880 | ||
1da177e4 LT |
881 | /* |
882 | * Decrements the number of outstanding ios that a bio has been | |
883 | * cloned into, completing the original io if necc. | |
884 | */ | |
858119e1 | 885 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 886 | { |
2e93ccc1 | 887 | unsigned long flags; |
b35f8caa MB |
888 | int io_error; |
889 | struct bio *bio; | |
890 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
891 | |
892 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
893 | if (unlikely(error)) { |
894 | spin_lock_irqsave(&io->endio_lock, flags); | |
895 | if (!(io->error > 0 && __noflush_suspending(md))) | |
896 | io->error = error; | |
897 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
898 | } | |
1da177e4 LT |
899 | |
900 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
901 | if (io->error == DM_ENDIO_REQUEUE) { |
902 | /* | |
903 | * Target requested pushing back the I/O. | |
2e93ccc1 | 904 | */ |
022c2611 | 905 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
906 | if (__noflush_suspending(md)) |
907 | bio_list_add_head(&md->deferred, io->bio); | |
908 | else | |
2e93ccc1 KU |
909 | /* noflush suspend was interrupted. */ |
910 | io->error = -EIO; | |
022c2611 | 911 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
912 | } |
913 | ||
b35f8caa MB |
914 | io_error = io->error; |
915 | bio = io->bio; | |
6a8736d1 TH |
916 | end_io_acct(io); |
917 | free_io(md, io); | |
918 | ||
919 | if (io_error == DM_ENDIO_REQUEUE) | |
920 | return; | |
2e93ccc1 | 921 | |
4f024f37 | 922 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 923 | /* |
6a8736d1 TH |
924 | * Preflush done for flush with data, reissue |
925 | * without REQ_FLUSH. | |
af7e466a | 926 | */ |
6a8736d1 TH |
927 | bio->bi_rw &= ~REQ_FLUSH; |
928 | queue_io(md, bio); | |
af7e466a | 929 | } else { |
b372d360 | 930 | /* done with normal IO or empty flush */ |
0a82a8d1 | 931 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 932 | bio_endio(bio, io_error); |
b35f8caa | 933 | } |
1da177e4 LT |
934 | } |
935 | } | |
936 | ||
7eee4ae2 MS |
937 | static void disable_write_same(struct mapped_device *md) |
938 | { | |
939 | struct queue_limits *limits = dm_get_queue_limits(md); | |
940 | ||
941 | /* device doesn't really support WRITE SAME, disable it */ | |
942 | limits->max_write_same_sectors = 0; | |
943 | } | |
944 | ||
6712ecf8 | 945 | static void clone_endio(struct bio *bio, int error) |
1da177e4 | 946 | { |
5164bece | 947 | int r = error; |
bfc6d41c | 948 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 949 | struct dm_io *io = tio->io; |
9faf400f | 950 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
951 | dm_endio_fn endio = tio->ti->type->end_io; |
952 | ||
1da177e4 LT |
953 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
954 | error = -EIO; | |
955 | ||
956 | if (endio) { | |
7de3ee57 | 957 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
958 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
959 | /* | |
960 | * error and requeue request are handled | |
961 | * in dec_pending(). | |
962 | */ | |
1da177e4 | 963 | error = r; |
45cbcd79 KU |
964 | else if (r == DM_ENDIO_INCOMPLETE) |
965 | /* The target will handle the io */ | |
6712ecf8 | 966 | return; |
45cbcd79 KU |
967 | else if (r) { |
968 | DMWARN("unimplemented target endio return value: %d", r); | |
969 | BUG(); | |
970 | } | |
1da177e4 LT |
971 | } |
972 | ||
7eee4ae2 MS |
973 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
974 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
975 | disable_write_same(md); | |
976 | ||
9faf400f | 977 | free_tio(md, tio); |
b35f8caa | 978 | dec_pending(io, error); |
1da177e4 LT |
979 | } |
980 | ||
cec47e3d KU |
981 | /* |
982 | * Partial completion handling for request-based dm | |
983 | */ | |
984 | static void end_clone_bio(struct bio *clone, int error) | |
985 | { | |
bfc6d41c MP |
986 | struct dm_rq_clone_bio_info *info = |
987 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
988 | struct dm_rq_target_io *tio = info->tio; |
989 | struct bio *bio = info->orig; | |
4f024f37 | 990 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
991 | |
992 | bio_put(clone); | |
993 | ||
994 | if (tio->error) | |
995 | /* | |
996 | * An error has already been detected on the request. | |
997 | * Once error occurred, just let clone->end_io() handle | |
998 | * the remainder. | |
999 | */ | |
1000 | return; | |
1001 | else if (error) { | |
1002 | /* | |
1003 | * Don't notice the error to the upper layer yet. | |
1004 | * The error handling decision is made by the target driver, | |
1005 | * when the request is completed. | |
1006 | */ | |
1007 | tio->error = error; | |
1008 | return; | |
1009 | } | |
1010 | ||
1011 | /* | |
1012 | * I/O for the bio successfully completed. | |
1013 | * Notice the data completion to the upper layer. | |
1014 | */ | |
1015 | ||
1016 | /* | |
1017 | * bios are processed from the head of the list. | |
1018 | * So the completing bio should always be rq->bio. | |
1019 | * If it's not, something wrong is happening. | |
1020 | */ | |
1021 | if (tio->orig->bio != bio) | |
1022 | DMERR("bio completion is going in the middle of the request"); | |
1023 | ||
1024 | /* | |
1025 | * Update the original request. | |
1026 | * Do not use blk_end_request() here, because it may complete | |
1027 | * the original request before the clone, and break the ordering. | |
1028 | */ | |
1029 | blk_update_request(tio->orig, 0, nr_bytes); | |
1030 | } | |
1031 | ||
bfebd1cd MS |
1032 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1033 | { | |
1034 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1035 | } | |
1036 | ||
cec47e3d KU |
1037 | /* |
1038 | * Don't touch any member of the md after calling this function because | |
1039 | * the md may be freed in dm_put() at the end of this function. | |
1040 | * Or do dm_get() before calling this function and dm_put() later. | |
1041 | */ | |
466d89a6 | 1042 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1043 | { |
9a0e609e MS |
1044 | int nr_requests_pending; |
1045 | ||
b4324fee | 1046 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1047 | |
1048 | /* nudge anyone waiting on suspend queue */ | |
9a0e609e MS |
1049 | nr_requests_pending = md_in_flight(md); |
1050 | if (!nr_requests_pending) | |
cec47e3d KU |
1051 | wake_up(&md->wait); |
1052 | ||
a8c32a5c JA |
1053 | /* |
1054 | * Run this off this callpath, as drivers could invoke end_io while | |
1055 | * inside their request_fn (and holding the queue lock). Calling | |
1056 | * back into ->request_fn() could deadlock attempting to grab the | |
1057 | * queue lock again. | |
1058 | */ | |
9a0e609e | 1059 | if (run_queue) { |
bfebd1cd MS |
1060 | if (md->queue->mq_ops) |
1061 | blk_mq_run_hw_queues(md->queue, true); | |
1062 | else if (!nr_requests_pending || | |
1063 | (nr_requests_pending >= md->queue->nr_congestion_on)) | |
9a0e609e MS |
1064 | blk_run_queue_async(md->queue); |
1065 | } | |
cec47e3d KU |
1066 | |
1067 | /* | |
1068 | * dm_put() must be at the end of this function. See the comment above | |
1069 | */ | |
1070 | dm_put(md); | |
1071 | } | |
1072 | ||
a77e28c7 KU |
1073 | static void free_rq_clone(struct request *clone) |
1074 | { | |
1075 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1076 | struct mapped_device *md = tio->md; |
a77e28c7 KU |
1077 | |
1078 | blk_rq_unprep_clone(clone); | |
bfebd1cd | 1079 | |
e5863d9a MS |
1080 | if (clone->q && clone->q->mq_ops) |
1081 | tio->ti->type->release_clone_rq(clone); | |
1082 | else | |
bfebd1cd MS |
1083 | free_clone_request(md, clone); |
1084 | ||
1085 | if (!md->queue->mq_ops) | |
1086 | free_rq_tio(tio); | |
a77e28c7 KU |
1087 | } |
1088 | ||
980691e5 KU |
1089 | /* |
1090 | * Complete the clone and the original request. | |
466d89a6 KB |
1091 | * Must be called without clone's queue lock held, |
1092 | * see end_clone_request() for more details. | |
980691e5 KU |
1093 | */ |
1094 | static void dm_end_request(struct request *clone, int error) | |
1095 | { | |
1096 | int rw = rq_data_dir(clone); | |
1097 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1098 | struct mapped_device *md = tio->md; | |
1099 | struct request *rq = tio->orig; | |
1100 | ||
29e4013d | 1101 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1102 | rq->errors = clone->errors; |
1103 | rq->resid_len = clone->resid_len; | |
1104 | ||
1105 | if (rq->sense) | |
1106 | /* | |
1107 | * We are using the sense buffer of the original | |
1108 | * request. | |
1109 | * So setting the length of the sense data is enough. | |
1110 | */ | |
1111 | rq->sense_len = clone->sense_len; | |
1112 | } | |
1113 | ||
1114 | free_rq_clone(clone); | |
bfebd1cd MS |
1115 | if (!rq->q->mq_ops) |
1116 | blk_end_request_all(rq, error); | |
1117 | else | |
1118 | blk_mq_end_request(rq, error); | |
29e4013d | 1119 | rq_completed(md, rw, true); |
980691e5 KU |
1120 | } |
1121 | ||
cec47e3d KU |
1122 | static void dm_unprep_request(struct request *rq) |
1123 | { | |
bfebd1cd | 1124 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1125 | struct request *clone = tio->clone; |
cec47e3d | 1126 | |
bfebd1cd MS |
1127 | if (!rq->q->mq_ops) { |
1128 | rq->special = NULL; | |
1129 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1130 | } | |
cec47e3d | 1131 | |
e5863d9a MS |
1132 | if (clone) |
1133 | free_rq_clone(clone); | |
cec47e3d KU |
1134 | } |
1135 | ||
1136 | /* | |
1137 | * Requeue the original request of a clone. | |
1138 | */ | |
bfebd1cd | 1139 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1140 | { |
cec47e3d KU |
1141 | struct request_queue *q = rq->q; |
1142 | unsigned long flags; | |
1143 | ||
cec47e3d | 1144 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d KU |
1145 | blk_requeue_request(q, rq); |
1146 | spin_unlock_irqrestore(q->queue_lock, flags); | |
bfebd1cd MS |
1147 | } |
1148 | ||
1149 | static void dm_requeue_unmapped_original_request(struct mapped_device *md, | |
1150 | struct request *rq) | |
1151 | { | |
1152 | int rw = rq_data_dir(rq); | |
1153 | ||
1154 | dm_unprep_request(rq); | |
1155 | ||
1156 | if (!rq->q->mq_ops) | |
1157 | old_requeue_request(rq); | |
1158 | else { | |
1159 | blk_mq_requeue_request(rq); | |
1160 | blk_mq_kick_requeue_list(rq->q); | |
1161 | } | |
cec47e3d | 1162 | |
466d89a6 KB |
1163 | rq_completed(md, rw, false); |
1164 | } | |
1165 | ||
1166 | static void dm_requeue_unmapped_request(struct request *clone) | |
1167 | { | |
1168 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1169 | ||
1170 | dm_requeue_unmapped_original_request(tio->md, tio->orig); | |
cec47e3d | 1171 | } |
cec47e3d | 1172 | |
bfebd1cd | 1173 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1174 | { |
1175 | unsigned long flags; | |
1176 | ||
bfebd1cd MS |
1177 | if (blk_queue_stopped(q)) |
1178 | return; | |
1179 | ||
cec47e3d | 1180 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1181 | blk_stop_queue(q); |
cec47e3d KU |
1182 | spin_unlock_irqrestore(q->queue_lock, flags); |
1183 | } | |
1184 | ||
bfebd1cd | 1185 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1186 | { |
bfebd1cd MS |
1187 | if (!q->mq_ops) |
1188 | old_stop_queue(q); | |
1189 | else | |
1190 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1191 | } |
1192 | ||
bfebd1cd | 1193 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1194 | { |
1195 | unsigned long flags; | |
1196 | ||
1197 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1198 | if (blk_queue_stopped(q)) |
1199 | blk_start_queue(q); | |
cec47e3d KU |
1200 | spin_unlock_irqrestore(q->queue_lock, flags); |
1201 | } | |
1202 | ||
bfebd1cd MS |
1203 | static void start_queue(struct request_queue *q) |
1204 | { | |
1205 | if (!q->mq_ops) | |
1206 | old_start_queue(q); | |
1207 | else | |
1208 | blk_mq_start_stopped_hw_queues(q, true); | |
1209 | } | |
1210 | ||
11a68244 | 1211 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1212 | { |
11a68244 | 1213 | int r = error; |
cec47e3d | 1214 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1215 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1216 | |
ba1cbad9 MS |
1217 | if (tio->ti) { |
1218 | rq_end_io = tio->ti->type->rq_end_io; | |
1219 | ||
1220 | if (mapped && rq_end_io) | |
1221 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1222 | } | |
cec47e3d | 1223 | |
7eee4ae2 MS |
1224 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1225 | !clone->q->limits.max_write_same_sectors)) | |
1226 | disable_write_same(tio->md); | |
1227 | ||
11a68244 | 1228 | if (r <= 0) |
cec47e3d | 1229 | /* The target wants to complete the I/O */ |
11a68244 KU |
1230 | dm_end_request(clone, r); |
1231 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1232 | /* The target will handle the I/O */ |
1233 | return; | |
11a68244 | 1234 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1235 | /* The target wants to requeue the I/O */ |
1236 | dm_requeue_unmapped_request(clone); | |
1237 | else { | |
11a68244 | 1238 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1239 | BUG(); |
1240 | } | |
1241 | } | |
1242 | ||
11a68244 KU |
1243 | /* |
1244 | * Request completion handler for request-based dm | |
1245 | */ | |
1246 | static void dm_softirq_done(struct request *rq) | |
1247 | { | |
1248 | bool mapped = true; | |
bfebd1cd | 1249 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1250 | struct request *clone = tio->clone; |
bfebd1cd | 1251 | int rw; |
11a68244 | 1252 | |
e5863d9a | 1253 | if (!clone) { |
bfebd1cd MS |
1254 | rw = rq_data_dir(rq); |
1255 | if (!rq->q->mq_ops) { | |
1256 | blk_end_request_all(rq, tio->error); | |
1257 | rq_completed(tio->md, rw, false); | |
1258 | free_rq_tio(tio); | |
1259 | } else { | |
1260 | blk_mq_end_request(rq, tio->error); | |
1261 | rq_completed(tio->md, rw, false); | |
1262 | } | |
e5863d9a MS |
1263 | return; |
1264 | } | |
11a68244 KU |
1265 | |
1266 | if (rq->cmd_flags & REQ_FAILED) | |
1267 | mapped = false; | |
1268 | ||
1269 | dm_done(clone, tio->error, mapped); | |
1270 | } | |
1271 | ||
cec47e3d KU |
1272 | /* |
1273 | * Complete the clone and the original request with the error status | |
1274 | * through softirq context. | |
1275 | */ | |
466d89a6 | 1276 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1277 | { |
bfebd1cd | 1278 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1279 | |
1280 | tio->error = error; | |
cec47e3d KU |
1281 | blk_complete_request(rq); |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * Complete the not-mapped clone and the original request with the error status | |
1286 | * through softirq context. | |
1287 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1288 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1289 | */ |
466d89a6 | 1290 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1291 | { |
cec47e3d | 1292 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1293 | dm_complete_request(rq, error); |
cec47e3d | 1294 | } |
cec47e3d KU |
1295 | |
1296 | /* | |
bfebd1cd | 1297 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1298 | */ |
1299 | static void end_clone_request(struct request *clone, int error) | |
1300 | { | |
466d89a6 KB |
1301 | struct dm_rq_target_io *tio = clone->end_io_data; |
1302 | ||
e5863d9a MS |
1303 | if (!clone->q->mq_ops) { |
1304 | /* | |
1305 | * For just cleaning up the information of the queue in which | |
1306 | * the clone was dispatched. | |
1307 | * The clone is *NOT* freed actually here because it is alloced | |
1308 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1309 | */ | |
1310 | __blk_put_request(clone->q, clone); | |
1311 | } | |
cec47e3d KU |
1312 | |
1313 | /* | |
1314 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1315 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1316 | * - another request may be submitted by the upper level driver |
1317 | * of the stacking during the completion | |
1318 | * - the submission which requires queue lock may be done | |
466d89a6 | 1319 | * against this clone's queue |
cec47e3d | 1320 | */ |
466d89a6 | 1321 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1322 | } |
1323 | ||
56a67df7 MS |
1324 | /* |
1325 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1326 | * target boundary. | |
1327 | */ | |
1328 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1329 | { | |
1330 | sector_t target_offset = dm_target_offset(ti, sector); | |
1331 | ||
1332 | return ti->len - target_offset; | |
1333 | } | |
1334 | ||
1335 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1336 | { |
56a67df7 | 1337 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1338 | sector_t offset, max_len; |
1da177e4 LT |
1339 | |
1340 | /* | |
542f9038 | 1341 | * Does the target need to split even further? |
1da177e4 | 1342 | */ |
542f9038 MS |
1343 | if (ti->max_io_len) { |
1344 | offset = dm_target_offset(ti, sector); | |
1345 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1346 | max_len = sector_div(offset, ti->max_io_len); | |
1347 | else | |
1348 | max_len = offset & (ti->max_io_len - 1); | |
1349 | max_len = ti->max_io_len - max_len; | |
1350 | ||
1351 | if (len > max_len) | |
1352 | len = max_len; | |
1da177e4 LT |
1353 | } |
1354 | ||
1355 | return len; | |
1356 | } | |
1357 | ||
542f9038 MS |
1358 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1359 | { | |
1360 | if (len > UINT_MAX) { | |
1361 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1362 | (unsigned long long)len, UINT_MAX); | |
1363 | ti->error = "Maximum size of target IO is too large"; | |
1364 | return -EINVAL; | |
1365 | } | |
1366 | ||
1367 | ti->max_io_len = (uint32_t) len; | |
1368 | ||
1369 | return 0; | |
1370 | } | |
1371 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1372 | ||
1dd40c3e MP |
1373 | /* |
1374 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1375 | * allowed for all bio types except REQ_FLUSH. | |
1376 | * | |
1377 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1378 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1379 | * sent in a next bio. | |
1380 | * | |
1381 | * A diagram that explains the arithmetics: | |
1382 | * +--------------------+---------------+-------+ | |
1383 | * | 1 | 2 | 3 | | |
1384 | * +--------------------+---------------+-------+ | |
1385 | * | |
1386 | * <-------------- *tio->len_ptr ---------------> | |
1387 | * <------- bi_size -------> | |
1388 | * <-- n_sectors --> | |
1389 | * | |
1390 | * Region 1 was already iterated over with bio_advance or similar function. | |
1391 | * (it may be empty if the target doesn't use bio_advance) | |
1392 | * Region 2 is the remaining bio size that the target wants to process. | |
1393 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1394 | * to make it empty) | |
1395 | * The target requires that region 3 is to be sent in the next bio. | |
1396 | * | |
1397 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1398 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1399 | * copies of the bio. | |
1400 | */ | |
1401 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1402 | { | |
1403 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1404 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1405 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1406 | BUG_ON(bi_size > *tio->len_ptr); | |
1407 | BUG_ON(n_sectors > bi_size); | |
1408 | *tio->len_ptr -= bi_size - n_sectors; | |
1409 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1410 | } | |
1411 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1412 | ||
bd2a49b8 | 1413 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1414 | { |
1415 | int r; | |
2056a782 | 1416 | sector_t sector; |
9faf400f | 1417 | struct mapped_device *md; |
dba14160 | 1418 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1419 | struct dm_target *ti = tio->ti; |
1da177e4 | 1420 | |
1da177e4 | 1421 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1422 | |
1423 | /* | |
1424 | * Map the clone. If r == 0 we don't need to do | |
1425 | * anything, the target has assumed ownership of | |
1426 | * this io. | |
1427 | */ | |
1428 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1429 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1430 | r = ti->type->map(ti, clone); |
45cbcd79 | 1431 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1432 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1433 | |
d07335e5 MS |
1434 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1435 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1436 | |
1da177e4 | 1437 | generic_make_request(clone); |
2e93ccc1 KU |
1438 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1439 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1440 | md = tio->io->md; |
1441 | dec_pending(tio->io, r); | |
9faf400f | 1442 | free_tio(md, tio); |
45cbcd79 KU |
1443 | } else if (r) { |
1444 | DMWARN("unimplemented target map return value: %d", r); | |
1445 | BUG(); | |
1da177e4 LT |
1446 | } |
1447 | } | |
1448 | ||
1449 | struct clone_info { | |
1450 | struct mapped_device *md; | |
1451 | struct dm_table *map; | |
1452 | struct bio *bio; | |
1453 | struct dm_io *io; | |
1454 | sector_t sector; | |
e0d6609a | 1455 | unsigned sector_count; |
1da177e4 LT |
1456 | }; |
1457 | ||
e0d6609a | 1458 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1459 | { |
4f024f37 KO |
1460 | bio->bi_iter.bi_sector = sector; |
1461 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1462 | } |
1463 | ||
1464 | /* | |
1465 | * Creates a bio that consists of range of complete bvecs. | |
1466 | */ | |
dba14160 | 1467 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1468 | sector_t sector, unsigned len) |
1da177e4 | 1469 | { |
dba14160 | 1470 | struct bio *clone = &tio->clone; |
1da177e4 | 1471 | |
1c3b13e6 KO |
1472 | __bio_clone_fast(clone, bio); |
1473 | ||
1474 | if (bio_integrity(bio)) | |
1475 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1476 | |
1c3b13e6 KO |
1477 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1478 | clone->bi_iter.bi_size = to_bytes(len); | |
1479 | ||
1480 | if (bio_integrity(bio)) | |
1481 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1482 | } |
1483 | ||
9015df24 | 1484 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1485 | struct dm_target *ti, |
55a62eef | 1486 | unsigned target_bio_nr) |
f9ab94ce | 1487 | { |
dba14160 MP |
1488 | struct dm_target_io *tio; |
1489 | struct bio *clone; | |
1490 | ||
99778273 | 1491 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1492 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1493 | |
1494 | tio->io = ci->io; | |
1495 | tio->ti = ti; | |
55a62eef | 1496 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1497 | |
1498 | return tio; | |
1499 | } | |
1500 | ||
14fe594d AK |
1501 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1502 | struct dm_target *ti, | |
1dd40c3e | 1503 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1504 | { |
99778273 | 1505 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1506 | struct bio *clone = &tio->clone; |
9015df24 | 1507 | |
1dd40c3e MP |
1508 | tio->len_ptr = len; |
1509 | ||
99778273 | 1510 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1511 | if (len) |
1dd40c3e | 1512 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1513 | |
bd2a49b8 | 1514 | __map_bio(tio); |
f9ab94ce MP |
1515 | } |
1516 | ||
14fe594d | 1517 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1518 | unsigned num_bios, unsigned *len) |
06a426ce | 1519 | { |
55a62eef | 1520 | unsigned target_bio_nr; |
06a426ce | 1521 | |
55a62eef | 1522 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1523 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1524 | } |
1525 | ||
14fe594d | 1526 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1527 | { |
06a426ce | 1528 | unsigned target_nr = 0; |
f9ab94ce MP |
1529 | struct dm_target *ti; |
1530 | ||
b372d360 | 1531 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1532 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1533 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1534 | |
f9ab94ce MP |
1535 | return 0; |
1536 | } | |
1537 | ||
e4c93811 | 1538 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1539 | sector_t sector, unsigned *len) |
5ae89a87 | 1540 | { |
dba14160 | 1541 | struct bio *bio = ci->bio; |
5ae89a87 | 1542 | struct dm_target_io *tio; |
b0d8ed4d AK |
1543 | unsigned target_bio_nr; |
1544 | unsigned num_target_bios = 1; | |
5ae89a87 | 1545 | |
b0d8ed4d AK |
1546 | /* |
1547 | * Does the target want to receive duplicate copies of the bio? | |
1548 | */ | |
1549 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1550 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1551 | |
b0d8ed4d | 1552 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1553 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1554 | tio->len_ptr = len; |
1555 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1556 | __map_bio(tio); |
1557 | } | |
5ae89a87 MS |
1558 | } |
1559 | ||
55a62eef | 1560 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1561 | |
55a62eef | 1562 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1563 | { |
55a62eef | 1564 | return ti->num_discard_bios; |
23508a96 MS |
1565 | } |
1566 | ||
55a62eef | 1567 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1568 | { |
55a62eef | 1569 | return ti->num_write_same_bios; |
23508a96 MS |
1570 | } |
1571 | ||
1572 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1573 | ||
1574 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1575 | { | |
55a62eef | 1576 | return ti->split_discard_bios; |
23508a96 MS |
1577 | } |
1578 | ||
14fe594d AK |
1579 | static int __send_changing_extent_only(struct clone_info *ci, |
1580 | get_num_bios_fn get_num_bios, | |
1581 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1582 | { |
1583 | struct dm_target *ti; | |
e0d6609a | 1584 | unsigned len; |
55a62eef | 1585 | unsigned num_bios; |
5ae89a87 | 1586 | |
a79245b3 MS |
1587 | do { |
1588 | ti = dm_table_find_target(ci->map, ci->sector); | |
1589 | if (!dm_target_is_valid(ti)) | |
1590 | return -EIO; | |
5ae89a87 | 1591 | |
5ae89a87 | 1592 | /* |
23508a96 MS |
1593 | * Even though the device advertised support for this type of |
1594 | * request, that does not mean every target supports it, and | |
936688d7 | 1595 | * reconfiguration might also have changed that since the |
a79245b3 | 1596 | * check was performed. |
5ae89a87 | 1597 | */ |
55a62eef AK |
1598 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1599 | if (!num_bios) | |
a79245b3 | 1600 | return -EOPNOTSUPP; |
5ae89a87 | 1601 | |
23508a96 | 1602 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1603 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1604 | else |
e0d6609a | 1605 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1606 | |
1dd40c3e | 1607 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1608 | |
1609 | ci->sector += len; | |
1610 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1611 | |
1612 | return 0; | |
1613 | } | |
1614 | ||
14fe594d | 1615 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1616 | { |
14fe594d AK |
1617 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1618 | is_split_required_for_discard); | |
23508a96 MS |
1619 | } |
1620 | ||
14fe594d | 1621 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1622 | { |
14fe594d | 1623 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1624 | } |
1625 | ||
e4c93811 AK |
1626 | /* |
1627 | * Select the correct strategy for processing a non-flush bio. | |
1628 | */ | |
14fe594d | 1629 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1630 | { |
dba14160 | 1631 | struct bio *bio = ci->bio; |
512875bd | 1632 | struct dm_target *ti; |
1c3b13e6 | 1633 | unsigned len; |
1da177e4 | 1634 | |
5ae89a87 | 1635 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1636 | return __send_discard(ci); |
23508a96 | 1637 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1638 | return __send_write_same(ci); |
5ae89a87 | 1639 | |
512875bd JN |
1640 | ti = dm_table_find_target(ci->map, ci->sector); |
1641 | if (!dm_target_is_valid(ti)) | |
1642 | return -EIO; | |
1643 | ||
1c3b13e6 | 1644 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1645 | |
1dd40c3e | 1646 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1647 | |
1c3b13e6 KO |
1648 | ci->sector += len; |
1649 | ci->sector_count -= len; | |
1da177e4 | 1650 | |
1c3b13e6 | 1651 | return 0; |
1da177e4 LT |
1652 | } |
1653 | ||
1654 | /* | |
14fe594d | 1655 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1656 | */ |
83d5e5b0 MP |
1657 | static void __split_and_process_bio(struct mapped_device *md, |
1658 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1659 | { |
1660 | struct clone_info ci; | |
512875bd | 1661 | int error = 0; |
1da177e4 | 1662 | |
83d5e5b0 | 1663 | if (unlikely(!map)) { |
6a8736d1 | 1664 | bio_io_error(bio); |
f0b9a450 MP |
1665 | return; |
1666 | } | |
692d0eb9 | 1667 | |
83d5e5b0 | 1668 | ci.map = map; |
1da177e4 | 1669 | ci.md = md; |
1da177e4 LT |
1670 | ci.io = alloc_io(md); |
1671 | ci.io->error = 0; | |
1672 | atomic_set(&ci.io->io_count, 1); | |
1673 | ci.io->bio = bio; | |
1674 | ci.io->md = md; | |
f88fb981 | 1675 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1676 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1677 | |
3eaf840e | 1678 | start_io_acct(ci.io); |
bd2a49b8 | 1679 | |
b372d360 MS |
1680 | if (bio->bi_rw & REQ_FLUSH) { |
1681 | ci.bio = &ci.md->flush_bio; | |
1682 | ci.sector_count = 0; | |
14fe594d | 1683 | error = __send_empty_flush(&ci); |
b372d360 MS |
1684 | /* dec_pending submits any data associated with flush */ |
1685 | } else { | |
6a8736d1 | 1686 | ci.bio = bio; |
d87f4c14 | 1687 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1688 | while (ci.sector_count && !error) |
14fe594d | 1689 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1690 | } |
1da177e4 LT |
1691 | |
1692 | /* drop the extra reference count */ | |
512875bd | 1693 | dec_pending(ci.io, error); |
1da177e4 LT |
1694 | } |
1695 | /*----------------------------------------------------------------- | |
1696 | * CRUD END | |
1697 | *---------------------------------------------------------------*/ | |
1698 | ||
f6fccb12 MB |
1699 | static int dm_merge_bvec(struct request_queue *q, |
1700 | struct bvec_merge_data *bvm, | |
1701 | struct bio_vec *biovec) | |
1702 | { | |
1703 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1704 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1705 | struct dm_target *ti; |
1706 | sector_t max_sectors; | |
5037108a | 1707 | int max_size = 0; |
f6fccb12 MB |
1708 | |
1709 | if (unlikely(!map)) | |
5037108a | 1710 | goto out; |
f6fccb12 MB |
1711 | |
1712 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1713 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1714 | goto out; |
f6fccb12 MB |
1715 | |
1716 | /* | |
1717 | * Find maximum amount of I/O that won't need splitting | |
1718 | */ | |
56a67df7 | 1719 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1720 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1721 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1722 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1723 | max_size = 0; |
1724 | ||
1725 | /* | |
1726 | * merge_bvec_fn() returns number of bytes | |
1727 | * it can accept at this offset | |
1728 | * max is precomputed maximal io size | |
1729 | */ | |
1730 | if (max_size && ti->type->merge) | |
1731 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1732 | /* |
1733 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1734 | * provided their merge_bvec method (we know this by looking for the |
1735 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1736 | * allow bios with multiple vector entries. So always set max_size | |
1737 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1738 | */ |
1739 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1740 | max_size = 0; |
f6fccb12 | 1741 | |
5037108a | 1742 | out: |
83d5e5b0 | 1743 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1744 | /* |
1745 | * Always allow an entire first page | |
1746 | */ | |
1747 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1748 | max_size = biovec->bv_len; | |
1749 | ||
f6fccb12 MB |
1750 | return max_size; |
1751 | } | |
1752 | ||
1da177e4 LT |
1753 | /* |
1754 | * The request function that just remaps the bio built up by | |
1755 | * dm_merge_bvec. | |
1756 | */ | |
ff36ab34 | 1757 | static void dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1758 | { |
12f03a49 | 1759 | int rw = bio_data_dir(bio); |
1da177e4 | 1760 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1761 | int srcu_idx; |
1762 | struct dm_table *map; | |
1da177e4 | 1763 | |
83d5e5b0 | 1764 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1765 | |
18c0b223 | 1766 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1767 | |
6a8736d1 TH |
1768 | /* if we're suspended, we have to queue this io for later */ |
1769 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1770 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1771 | |
6a8736d1 TH |
1772 | if (bio_rw(bio) != READA) |
1773 | queue_io(md, bio); | |
1774 | else | |
54d9a1b4 | 1775 | bio_io_error(bio); |
5a7bbad2 | 1776 | return; |
1da177e4 LT |
1777 | } |
1778 | ||
83d5e5b0 MP |
1779 | __split_and_process_bio(md, map, bio); |
1780 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1781 | return; |
cec47e3d KU |
1782 | } |
1783 | ||
fd2ed4d2 | 1784 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1785 | { |
1786 | return blk_queue_stackable(md->queue); | |
1787 | } | |
1788 | ||
466d89a6 | 1789 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1790 | { |
1791 | int r; | |
1792 | ||
466d89a6 KB |
1793 | if (blk_queue_io_stat(clone->q)) |
1794 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1795 | |
466d89a6 KB |
1796 | clone->start_time = jiffies; |
1797 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1798 | if (r) |
466d89a6 | 1799 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1800 | dm_complete_request(rq, r); |
1801 | } | |
cec47e3d | 1802 | |
cec47e3d KU |
1803 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1804 | void *data) | |
1805 | { | |
1806 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1807 | struct dm_rq_clone_bio_info *info = |
1808 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1809 | |
1810 | info->orig = bio_orig; | |
1811 | info->tio = tio; | |
1812 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1813 | |
1814 | return 0; | |
1815 | } | |
1816 | ||
1817 | static int setup_clone(struct request *clone, struct request *rq, | |
1ae49ea2 | 1818 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
cec47e3d | 1819 | { |
d0bcb878 | 1820 | int r; |
cec47e3d | 1821 | |
1ae49ea2 | 1822 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, |
29e4013d TH |
1823 | dm_rq_bio_constructor, tio); |
1824 | if (r) | |
1825 | return r; | |
cec47e3d | 1826 | |
29e4013d TH |
1827 | clone->cmd = rq->cmd; |
1828 | clone->cmd_len = rq->cmd_len; | |
1829 | clone->sense = rq->sense; | |
cec47e3d KU |
1830 | clone->end_io = end_clone_request; |
1831 | clone->end_io_data = tio; | |
1832 | ||
1ae49ea2 MS |
1833 | tio->clone = clone; |
1834 | ||
cec47e3d KU |
1835 | return 0; |
1836 | } | |
1837 | ||
6facdaff | 1838 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1839 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 MS |
1840 | { |
1841 | struct request *clone = alloc_clone_request(md, gfp_mask); | |
1842 | ||
1843 | if (!clone) | |
1844 | return NULL; | |
1845 | ||
1846 | blk_rq_init(NULL, clone); | |
1847 | if (setup_clone(clone, rq, tio, gfp_mask)) { | |
1848 | /* -ENOMEM */ | |
1849 | free_clone_request(md, clone); | |
1850 | return NULL; | |
1851 | } | |
1852 | ||
1853 | return clone; | |
1854 | } | |
1855 | ||
2eb6e1e3 KB |
1856 | static void map_tio_request(struct kthread_work *work); |
1857 | ||
bfebd1cd MS |
1858 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1859 | struct mapped_device *md) | |
1860 | { | |
1861 | tio->md = md; | |
1862 | tio->ti = NULL; | |
1863 | tio->clone = NULL; | |
1864 | tio->orig = rq; | |
1865 | tio->error = 0; | |
1866 | memset(&tio->info, 0, sizeof(tio->info)); | |
1867 | init_kthread_work(&tio->work, map_tio_request); | |
1868 | } | |
1869 | ||
466d89a6 KB |
1870 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1871 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1872 | { |
6facdaff | 1873 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1874 | int srcu_idx; |
1875 | struct dm_table *table; | |
6facdaff KU |
1876 | |
1877 | tio = alloc_rq_tio(md, gfp_mask); | |
1878 | if (!tio) | |
1879 | return NULL; | |
1880 | ||
bfebd1cd | 1881 | init_tio(tio, rq, md); |
6facdaff | 1882 | |
e5863d9a MS |
1883 | table = dm_get_live_table(md, &srcu_idx); |
1884 | if (!dm_table_mq_request_based(table)) { | |
1885 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1886 | dm_put_live_table(md, srcu_idx); | |
1887 | free_rq_tio(tio); | |
1888 | return NULL; | |
1889 | } | |
6facdaff | 1890 | } |
e5863d9a | 1891 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1892 | |
466d89a6 | 1893 | return tio; |
6facdaff KU |
1894 | } |
1895 | ||
cec47e3d KU |
1896 | /* |
1897 | * Called with the queue lock held. | |
1898 | */ | |
1899 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1900 | { | |
1901 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1902 | struct dm_rq_target_io *tio; |
cec47e3d | 1903 | |
cec47e3d KU |
1904 | if (unlikely(rq->special)) { |
1905 | DMWARN("Already has something in rq->special."); | |
1906 | return BLKPREP_KILL; | |
1907 | } | |
1908 | ||
466d89a6 KB |
1909 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1910 | if (!tio) | |
cec47e3d | 1911 | return BLKPREP_DEFER; |
cec47e3d | 1912 | |
466d89a6 | 1913 | rq->special = tio; |
cec47e3d KU |
1914 | rq->cmd_flags |= REQ_DONTPREP; |
1915 | ||
1916 | return BLKPREP_OK; | |
1917 | } | |
1918 | ||
9eef87da KU |
1919 | /* |
1920 | * Returns: | |
e5863d9a MS |
1921 | * 0 : the request has been processed |
1922 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1923 | * < 0 : the request was completed due to failure | |
9eef87da | 1924 | */ |
bfebd1cd | 1925 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1926 | struct mapped_device *md) |
cec47e3d | 1927 | { |
e5863d9a | 1928 | int r; |
bfebd1cd | 1929 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1930 | struct request *clone = NULL; |
1931 | ||
1932 | if (tio->clone) { | |
1933 | clone = tio->clone; | |
1934 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1935 | } else { | |
1936 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1937 | if (r < 0) { | |
1938 | /* The target wants to complete the I/O */ | |
1939 | dm_kill_unmapped_request(rq, r); | |
1940 | return r; | |
1941 | } | |
1942 | if (IS_ERR(clone)) | |
1943 | return DM_MAPIO_REQUEUE; | |
bfebd1cd | 1944 | if (setup_clone(clone, rq, tio, GFP_NOIO)) { |
e5863d9a MS |
1945 | /* -ENOMEM */ |
1946 | ti->type->release_clone_rq(clone); | |
1947 | return DM_MAPIO_REQUEUE; | |
1948 | } | |
1949 | } | |
cec47e3d | 1950 | |
cec47e3d KU |
1951 | switch (r) { |
1952 | case DM_MAPIO_SUBMITTED: | |
1953 | /* The target has taken the I/O to submit by itself later */ | |
1954 | break; | |
1955 | case DM_MAPIO_REMAPPED: | |
1956 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1957 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1958 | blk_rq_pos(rq)); |
1959 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1960 | break; |
1961 | case DM_MAPIO_REQUEUE: | |
1962 | /* The target wants to requeue the I/O */ | |
1963 | dm_requeue_unmapped_request(clone); | |
1964 | break; | |
1965 | default: | |
1966 | if (r > 0) { | |
1967 | DMWARN("unimplemented target map return value: %d", r); | |
1968 | BUG(); | |
1969 | } | |
1970 | ||
1971 | /* The target wants to complete the I/O */ | |
466d89a6 | 1972 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1973 | return r; |
cec47e3d | 1974 | } |
9eef87da | 1975 | |
e5863d9a | 1976 | return 0; |
cec47e3d KU |
1977 | } |
1978 | ||
2eb6e1e3 | 1979 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 1980 | { |
2eb6e1e3 | 1981 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
1982 | struct request *rq = tio->orig; |
1983 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 1984 | |
bfebd1cd | 1985 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
e5863d9a | 1986 | dm_requeue_unmapped_original_request(md, rq); |
2eb6e1e3 KB |
1987 | } |
1988 | ||
466d89a6 | 1989 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 1990 | { |
bfebd1cd MS |
1991 | if (!orig->q->mq_ops) |
1992 | blk_start_request(orig); | |
1993 | else | |
1994 | blk_mq_start_request(orig); | |
466d89a6 | 1995 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 1996 | |
0ce65797 MS |
1997 | if (md->seq_rq_merge_deadline_usecs) { |
1998 | md->last_rq_pos = rq_end_sector(orig); | |
1999 | md->last_rq_rw = rq_data_dir(orig); | |
2000 | md->last_rq_start_time = ktime_get(); | |
2001 | } | |
de3ec86d | 2002 | |
ba1cbad9 MS |
2003 | /* |
2004 | * Hold the md reference here for the in-flight I/O. | |
2005 | * We can't rely on the reference count by device opener, | |
2006 | * because the device may be closed during the request completion | |
2007 | * when all bios are completed. | |
2008 | * See the comment in rq_completed() too. | |
2009 | */ | |
2010 | dm_get(md); | |
ba1cbad9 MS |
2011 | } |
2012 | ||
0ce65797 MS |
2013 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2014 | ||
2015 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2016 | { | |
2017 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2018 | } | |
2019 | ||
2020 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2021 | const char *buf, size_t count) | |
2022 | { | |
2023 | unsigned deadline; | |
2024 | ||
2025 | if (!dm_request_based(md)) | |
2026 | return count; | |
2027 | ||
2028 | if (kstrtouint(buf, 10, &deadline)) | |
2029 | return -EINVAL; | |
2030 | ||
2031 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2032 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2033 | ||
2034 | md->seq_rq_merge_deadline_usecs = deadline; | |
2035 | ||
2036 | return count; | |
2037 | } | |
2038 | ||
2039 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2040 | { | |
2041 | ktime_t kt_deadline; | |
2042 | ||
2043 | if (!md->seq_rq_merge_deadline_usecs) | |
2044 | return false; | |
2045 | ||
2046 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2047 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2048 | ||
2049 | return !ktime_after(ktime_get(), kt_deadline); | |
2050 | } | |
2051 | ||
cec47e3d KU |
2052 | /* |
2053 | * q->request_fn for request-based dm. | |
2054 | * Called with the queue lock held. | |
2055 | */ | |
2056 | static void dm_request_fn(struct request_queue *q) | |
2057 | { | |
2058 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2059 | int srcu_idx; |
2060 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2061 | struct dm_target *ti; |
466d89a6 | 2062 | struct request *rq; |
2eb6e1e3 | 2063 | struct dm_rq_target_io *tio; |
29e4013d | 2064 | sector_t pos; |
cec47e3d KU |
2065 | |
2066 | /* | |
b4324fee KU |
2067 | * For suspend, check blk_queue_stopped() and increment |
2068 | * ->pending within a single queue_lock not to increment the | |
2069 | * number of in-flight I/Os after the queue is stopped in | |
2070 | * dm_suspend(). | |
cec47e3d | 2071 | */ |
7eaceacc | 2072 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2073 | rq = blk_peek_request(q); |
2074 | if (!rq) | |
9d1deb83 | 2075 | goto out; |
cec47e3d | 2076 | |
29e4013d TH |
2077 | /* always use block 0 to find the target for flushes for now */ |
2078 | pos = 0; | |
2079 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2080 | pos = blk_rq_pos(rq); | |
2081 | ||
2082 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2083 | if (!dm_target_is_valid(ti)) { |
2084 | /* | |
466d89a6 | 2085 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2086 | * before calling dm_kill_unmapped_request |
2087 | */ | |
2088 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2089 | dm_start_request(md, rq); |
2090 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2091 | continue; |
2092 | } | |
d0bcb878 | 2093 | |
0ce65797 MS |
2094 | if (dm_request_peeked_before_merge_deadline(md) && |
2095 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2096 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2097 | goto delay_and_out; | |
2098 | ||
cec47e3d | 2099 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2100 | goto delay_and_out; |
cec47e3d | 2101 | |
466d89a6 | 2102 | dm_start_request(md, rq); |
9eef87da | 2103 | |
bfebd1cd | 2104 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2105 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2106 | tio->ti = ti; | |
2107 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2108 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2109 | } |
2110 | ||
2111 | goto out; | |
2112 | ||
7eaceacc | 2113 | delay_and_out: |
d548b34b | 2114 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2115 | out: |
83d5e5b0 | 2116 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2117 | } |
2118 | ||
1da177e4 LT |
2119 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2120 | { | |
8a57dfc6 CS |
2121 | int r = bdi_bits; |
2122 | struct mapped_device *md = congested_data; | |
2123 | struct dm_table *map; | |
1da177e4 | 2124 | |
1eb787ec | 2125 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2126 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2127 | if (map) { |
cec47e3d KU |
2128 | /* |
2129 | * Request-based dm cares about only own queue for | |
2130 | * the query about congestion status of request_queue | |
2131 | */ | |
2132 | if (dm_request_based(md)) | |
2133 | r = md->queue->backing_dev_info.state & | |
2134 | bdi_bits; | |
2135 | else | |
2136 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2137 | } |
83d5e5b0 | 2138 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2139 | } |
2140 | ||
1da177e4 LT |
2141 | return r; |
2142 | } | |
2143 | ||
2144 | /*----------------------------------------------------------------- | |
2145 | * An IDR is used to keep track of allocated minor numbers. | |
2146 | *---------------------------------------------------------------*/ | |
2b06cfff | 2147 | static void free_minor(int minor) |
1da177e4 | 2148 | { |
f32c10b0 | 2149 | spin_lock(&_minor_lock); |
1da177e4 | 2150 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2151 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2152 | } |
2153 | ||
2154 | /* | |
2155 | * See if the device with a specific minor # is free. | |
2156 | */ | |
cf13ab8e | 2157 | static int specific_minor(int minor) |
1da177e4 | 2158 | { |
c9d76be6 | 2159 | int r; |
1da177e4 LT |
2160 | |
2161 | if (minor >= (1 << MINORBITS)) | |
2162 | return -EINVAL; | |
2163 | ||
c9d76be6 | 2164 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2165 | spin_lock(&_minor_lock); |
1da177e4 | 2166 | |
c9d76be6 | 2167 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2168 | |
f32c10b0 | 2169 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2170 | idr_preload_end(); |
2171 | if (r < 0) | |
2172 | return r == -ENOSPC ? -EBUSY : r; | |
2173 | return 0; | |
1da177e4 LT |
2174 | } |
2175 | ||
cf13ab8e | 2176 | static int next_free_minor(int *minor) |
1da177e4 | 2177 | { |
c9d76be6 | 2178 | int r; |
62f75c2f | 2179 | |
c9d76be6 | 2180 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2181 | spin_lock(&_minor_lock); |
1da177e4 | 2182 | |
c9d76be6 | 2183 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2184 | |
f32c10b0 | 2185 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2186 | idr_preload_end(); |
2187 | if (r < 0) | |
2188 | return r; | |
2189 | *minor = r; | |
2190 | return 0; | |
1da177e4 LT |
2191 | } |
2192 | ||
83d5cde4 | 2193 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2194 | |
53d5914f MP |
2195 | static void dm_wq_work(struct work_struct *work); |
2196 | ||
4a0b4ddf MS |
2197 | static void dm_init_md_queue(struct mapped_device *md) |
2198 | { | |
2199 | /* | |
2200 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2201 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2202 | * The type is decided at the first table loading time. |
2203 | * To prevent problematic device stacking, clear the queue flag | |
2204 | * for request stacking support until then. | |
2205 | * | |
2206 | * This queue is new, so no concurrency on the queue_flags. | |
2207 | */ | |
2208 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
bfebd1cd | 2209 | } |
4a0b4ddf | 2210 | |
bfebd1cd MS |
2211 | static void dm_init_old_md_queue(struct mapped_device *md) |
2212 | { | |
2213 | dm_init_md_queue(md); | |
2214 | ||
2215 | /* | |
2216 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2217 | */ | |
4a0b4ddf MS |
2218 | md->queue->queuedata = md; |
2219 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2220 | md->queue->backing_dev_info.congested_data = md; | |
ff36ab34 | 2221 | |
4a0b4ddf | 2222 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2223 | } |
2224 | ||
1da177e4 LT |
2225 | /* |
2226 | * Allocate and initialise a blank device with a given minor. | |
2227 | */ | |
2b06cfff | 2228 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2229 | { |
2230 | int r; | |
cf13ab8e | 2231 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2232 | void *old_md; |
1da177e4 LT |
2233 | |
2234 | if (!md) { | |
2235 | DMWARN("unable to allocate device, out of memory."); | |
2236 | return NULL; | |
2237 | } | |
2238 | ||
10da4f79 | 2239 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2240 | goto bad_module_get; |
10da4f79 | 2241 | |
1da177e4 | 2242 | /* get a minor number for the dev */ |
2b06cfff | 2243 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2244 | r = next_free_minor(&minor); |
2b06cfff | 2245 | else |
cf13ab8e | 2246 | r = specific_minor(minor); |
1da177e4 | 2247 | if (r < 0) |
6ed7ade8 | 2248 | goto bad_minor; |
1da177e4 | 2249 | |
83d5e5b0 MP |
2250 | r = init_srcu_struct(&md->io_barrier); |
2251 | if (r < 0) | |
2252 | goto bad_io_barrier; | |
2253 | ||
a5664dad | 2254 | md->type = DM_TYPE_NONE; |
e61290a4 | 2255 | mutex_init(&md->suspend_lock); |
a5664dad | 2256 | mutex_init(&md->type_lock); |
86f1152b | 2257 | mutex_init(&md->table_devices_lock); |
022c2611 | 2258 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2259 | atomic_set(&md->holders, 1); |
5c6bd75d | 2260 | atomic_set(&md->open_count, 0); |
1da177e4 | 2261 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2262 | atomic_set(&md->uevent_seq, 0); |
2263 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2264 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2265 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2266 | |
4a0b4ddf | 2267 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2268 | if (!md->queue) |
6ed7ade8 | 2269 | goto bad_queue; |
1da177e4 | 2270 | |
4a0b4ddf | 2271 | dm_init_md_queue(md); |
9faf400f | 2272 | |
1da177e4 LT |
2273 | md->disk = alloc_disk(1); |
2274 | if (!md->disk) | |
6ed7ade8 | 2275 | goto bad_disk; |
1da177e4 | 2276 | |
316d315b NK |
2277 | atomic_set(&md->pending[0], 0); |
2278 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2279 | init_waitqueue_head(&md->wait); |
53d5914f | 2280 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2281 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2282 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2283 | md->kworker_task = NULL; |
f0b04115 | 2284 | |
1da177e4 LT |
2285 | md->disk->major = _major; |
2286 | md->disk->first_minor = minor; | |
2287 | md->disk->fops = &dm_blk_dops; | |
2288 | md->disk->queue = md->queue; | |
2289 | md->disk->private_data = md; | |
2290 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2291 | add_disk(md->disk); | |
7e51f257 | 2292 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2293 | |
670368a8 | 2294 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2295 | if (!md->wq) |
2296 | goto bad_thread; | |
2297 | ||
32a926da MP |
2298 | md->bdev = bdget_disk(md->disk, 0); |
2299 | if (!md->bdev) | |
2300 | goto bad_bdev; | |
2301 | ||
6a8736d1 TH |
2302 | bio_init(&md->flush_bio); |
2303 | md->flush_bio.bi_bdev = md->bdev; | |
2304 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2305 | ||
fd2ed4d2 MP |
2306 | dm_stats_init(&md->stats); |
2307 | ||
ba61fdd1 | 2308 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2309 | spin_lock(&_minor_lock); |
ba61fdd1 | 2310 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2311 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2312 | |
2313 | BUG_ON(old_md != MINOR_ALLOCED); | |
2314 | ||
1da177e4 LT |
2315 | return md; |
2316 | ||
32a926da MP |
2317 | bad_bdev: |
2318 | destroy_workqueue(md->wq); | |
304f3f6a | 2319 | bad_thread: |
03022c54 | 2320 | del_gendisk(md->disk); |
304f3f6a | 2321 | put_disk(md->disk); |
6ed7ade8 | 2322 | bad_disk: |
1312f40e | 2323 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2324 | bad_queue: |
83d5e5b0 MP |
2325 | cleanup_srcu_struct(&md->io_barrier); |
2326 | bad_io_barrier: | |
1da177e4 | 2327 | free_minor(minor); |
6ed7ade8 | 2328 | bad_minor: |
10da4f79 | 2329 | module_put(THIS_MODULE); |
6ed7ade8 | 2330 | bad_module_get: |
1da177e4 LT |
2331 | kfree(md); |
2332 | return NULL; | |
2333 | } | |
2334 | ||
ae9da83f JN |
2335 | static void unlock_fs(struct mapped_device *md); |
2336 | ||
1da177e4 LT |
2337 | static void free_dev(struct mapped_device *md) |
2338 | { | |
f331c029 | 2339 | int minor = MINOR(disk_devt(md->disk)); |
bfebd1cd | 2340 | bool using_blk_mq = !!md->queue->mq_ops; |
63d94e48 | 2341 | |
32a926da | 2342 | unlock_fs(md); |
304f3f6a | 2343 | destroy_workqueue(md->wq); |
2eb6e1e3 KB |
2344 | |
2345 | if (md->kworker_task) | |
2346 | kthread_stop(md->kworker_task); | |
e6ee8c0b KU |
2347 | if (md->io_pool) |
2348 | mempool_destroy(md->io_pool); | |
1ae49ea2 MS |
2349 | if (md->rq_pool) |
2350 | mempool_destroy(md->rq_pool); | |
e6ee8c0b KU |
2351 | if (md->bs) |
2352 | bioset_free(md->bs); | |
63a4f065 | 2353 | |
83d5e5b0 | 2354 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2355 | free_table_devices(&md->table_devices); |
63a4f065 | 2356 | dm_stats_cleanup(&md->stats); |
fba9f90e JM |
2357 | |
2358 | spin_lock(&_minor_lock); | |
2359 | md->disk->private_data = NULL; | |
2360 | spin_unlock(&_minor_lock); | |
63a4f065 MS |
2361 | if (blk_get_integrity(md->disk)) |
2362 | blk_integrity_unregister(md->disk); | |
2363 | del_gendisk(md->disk); | |
1da177e4 | 2364 | put_disk(md->disk); |
1312f40e | 2365 | blk_cleanup_queue(md->queue); |
bfebd1cd MS |
2366 | if (using_blk_mq) |
2367 | blk_mq_free_tag_set(&md->tag_set); | |
63a4f065 MS |
2368 | bdput(md->bdev); |
2369 | free_minor(minor); | |
2370 | ||
10da4f79 | 2371 | module_put(THIS_MODULE); |
1da177e4 LT |
2372 | kfree(md); |
2373 | } | |
2374 | ||
e6ee8c0b KU |
2375 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2376 | { | |
c0820cf5 | 2377 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2378 | |
5f015204 | 2379 | if (md->io_pool && md->bs) { |
16245bdc JN |
2380 | /* The md already has necessary mempools. */ |
2381 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2382 | /* | |
2383 | * Reload bioset because front_pad may have changed | |
2384 | * because a different table was loaded. | |
2385 | */ | |
2386 | bioset_free(md->bs); | |
2387 | md->bs = p->bs; | |
2388 | p->bs = NULL; | |
16245bdc | 2389 | } |
466d89a6 KB |
2390 | /* |
2391 | * There's no need to reload with request-based dm | |
2392 | * because the size of front_pad doesn't change. | |
2393 | * Note for future: If you are to reload bioset, | |
2394 | * prep-ed requests in the queue may refer | |
2395 | * to bio from the old bioset, so you must walk | |
2396 | * through the queue to unprep. | |
2397 | */ | |
e6ee8c0b | 2398 | goto out; |
c0820cf5 | 2399 | } |
e6ee8c0b | 2400 | |
1ae49ea2 | 2401 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
e6ee8c0b KU |
2402 | |
2403 | md->io_pool = p->io_pool; | |
2404 | p->io_pool = NULL; | |
1ae49ea2 MS |
2405 | md->rq_pool = p->rq_pool; |
2406 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2407 | md->bs = p->bs; |
2408 | p->bs = NULL; | |
2409 | ||
2410 | out: | |
2411 | /* mempool bind completed, now no need any mempools in the table */ | |
2412 | dm_table_free_md_mempools(t); | |
2413 | } | |
2414 | ||
1da177e4 LT |
2415 | /* |
2416 | * Bind a table to the device. | |
2417 | */ | |
2418 | static void event_callback(void *context) | |
2419 | { | |
7a8c3d3b MA |
2420 | unsigned long flags; |
2421 | LIST_HEAD(uevents); | |
1da177e4 LT |
2422 | struct mapped_device *md = (struct mapped_device *) context; |
2423 | ||
7a8c3d3b MA |
2424 | spin_lock_irqsave(&md->uevent_lock, flags); |
2425 | list_splice_init(&md->uevent_list, &uevents); | |
2426 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2427 | ||
ed9e1982 | 2428 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2429 | |
1da177e4 LT |
2430 | atomic_inc(&md->event_nr); |
2431 | wake_up(&md->eventq); | |
2432 | } | |
2433 | ||
c217649b MS |
2434 | /* |
2435 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2436 | */ | |
4e90188b | 2437 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2438 | { |
4e90188b | 2439 | set_capacity(md->disk, size); |
1da177e4 | 2440 | |
db8fef4f | 2441 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2442 | } |
2443 | ||
d5b9dd04 MP |
2444 | /* |
2445 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2446 | * | |
2447 | * If this function returns 0, then the device is either a non-dm | |
2448 | * device without a merge_bvec_fn, or it is a dm device that is | |
2449 | * able to split any bios it receives that are too big. | |
2450 | */ | |
2451 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2452 | { | |
2453 | struct mapped_device *dev_md; | |
2454 | ||
2455 | if (!q->merge_bvec_fn) | |
2456 | return 0; | |
2457 | ||
ff36ab34 | 2458 | if (q->make_request_fn == dm_make_request) { |
d5b9dd04 MP |
2459 | dev_md = q->queuedata; |
2460 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2461 | return 0; | |
2462 | } | |
2463 | ||
2464 | return 1; | |
2465 | } | |
2466 | ||
2467 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2468 | struct dm_dev *dev, sector_t start, | |
2469 | sector_t len, void *data) | |
2470 | { | |
2471 | struct block_device *bdev = dev->bdev; | |
2472 | struct request_queue *q = bdev_get_queue(bdev); | |
2473 | ||
2474 | return dm_queue_merge_is_compulsory(q); | |
2475 | } | |
2476 | ||
2477 | /* | |
2478 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2479 | * on the properties of the underlying devices. | |
2480 | */ | |
2481 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2482 | { | |
2483 | unsigned i = 0; | |
2484 | struct dm_target *ti; | |
2485 | ||
2486 | while (i < dm_table_get_num_targets(table)) { | |
2487 | ti = dm_table_get_target(table, i++); | |
2488 | ||
2489 | if (ti->type->iterate_devices && | |
2490 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2491 | return 0; | |
2492 | } | |
2493 | ||
2494 | return 1; | |
2495 | } | |
2496 | ||
042d2a9b AK |
2497 | /* |
2498 | * Returns old map, which caller must destroy. | |
2499 | */ | |
2500 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2501 | struct queue_limits *limits) | |
1da177e4 | 2502 | { |
042d2a9b | 2503 | struct dm_table *old_map; |
165125e1 | 2504 | struct request_queue *q = md->queue; |
1da177e4 | 2505 | sector_t size; |
d5b9dd04 | 2506 | int merge_is_optional; |
1da177e4 LT |
2507 | |
2508 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2509 | |
2510 | /* | |
2511 | * Wipe any geometry if the size of the table changed. | |
2512 | */ | |
fd2ed4d2 | 2513 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2514 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2515 | ||
32a926da | 2516 | __set_size(md, size); |
d5816876 | 2517 | |
2ca3310e AK |
2518 | dm_table_event_callback(t, event_callback, md); |
2519 | ||
e6ee8c0b KU |
2520 | /* |
2521 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2522 | * for request-based during suspension. So stop it to prevent | |
2523 | * I/O mapping before resume. | |
2524 | * This must be done before setting the queue restrictions, | |
2525 | * because request-based dm may be run just after the setting. | |
2526 | */ | |
bfebd1cd | 2527 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2528 | stop_queue(q); |
2529 | ||
2530 | __bind_mempools(md, t); | |
2531 | ||
d5b9dd04 MP |
2532 | merge_is_optional = dm_table_merge_is_optional(t); |
2533 | ||
a12f5d48 | 2534 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2535 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2536 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2537 | ||
754c5fc7 | 2538 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2539 | if (merge_is_optional) |
2540 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2541 | else | |
2542 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2543 | if (old_map) |
2544 | dm_sync_table(md); | |
1da177e4 | 2545 | |
042d2a9b | 2546 | return old_map; |
1da177e4 LT |
2547 | } |
2548 | ||
a7940155 AK |
2549 | /* |
2550 | * Returns unbound table for the caller to free. | |
2551 | */ | |
2552 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2553 | { |
a12f5d48 | 2554 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2555 | |
2556 | if (!map) | |
a7940155 | 2557 | return NULL; |
1da177e4 LT |
2558 | |
2559 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2560 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2561 | dm_sync_table(md); |
a7940155 AK |
2562 | |
2563 | return map; | |
1da177e4 LT |
2564 | } |
2565 | ||
2566 | /* | |
2567 | * Constructor for a new device. | |
2568 | */ | |
2b06cfff | 2569 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2570 | { |
2571 | struct mapped_device *md; | |
2572 | ||
2b06cfff | 2573 | md = alloc_dev(minor); |
1da177e4 LT |
2574 | if (!md) |
2575 | return -ENXIO; | |
2576 | ||
784aae73 MB |
2577 | dm_sysfs_init(md); |
2578 | ||
1da177e4 LT |
2579 | *result = md; |
2580 | return 0; | |
2581 | } | |
2582 | ||
a5664dad MS |
2583 | /* |
2584 | * Functions to manage md->type. | |
2585 | * All are required to hold md->type_lock. | |
2586 | */ | |
2587 | void dm_lock_md_type(struct mapped_device *md) | |
2588 | { | |
2589 | mutex_lock(&md->type_lock); | |
2590 | } | |
2591 | ||
2592 | void dm_unlock_md_type(struct mapped_device *md) | |
2593 | { | |
2594 | mutex_unlock(&md->type_lock); | |
2595 | } | |
2596 | ||
2597 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2598 | { | |
00c4fc3b | 2599 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2600 | md->type = type; |
2601 | } | |
2602 | ||
2603 | unsigned dm_get_md_type(struct mapped_device *md) | |
2604 | { | |
00c4fc3b | 2605 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2606 | return md->type; |
2607 | } | |
2608 | ||
36a0456f AK |
2609 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2610 | { | |
2611 | return md->immutable_target_type; | |
2612 | } | |
2613 | ||
f84cb8a4 MS |
2614 | /* |
2615 | * The queue_limits are only valid as long as you have a reference | |
2616 | * count on 'md'. | |
2617 | */ | |
2618 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2619 | { | |
2620 | BUG_ON(!atomic_read(&md->holders)); | |
2621 | return &md->queue->limits; | |
2622 | } | |
2623 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2624 | ||
bfebd1cd MS |
2625 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2626 | { | |
2627 | /* Initialize the request-based DM worker thread */ | |
2628 | init_kthread_worker(&md->kworker); | |
2629 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2630 | "kdmwork-%s", dm_device_name(md)); | |
2631 | } | |
2632 | ||
4a0b4ddf MS |
2633 | /* |
2634 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2635 | */ | |
2636 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2637 | { | |
2638 | struct request_queue *q = NULL; | |
2639 | ||
2640 | if (md->queue->elevator) | |
bfebd1cd | 2641 | return 0; |
4a0b4ddf MS |
2642 | |
2643 | /* Fully initialize the queue */ | |
2644 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2645 | if (!q) | |
bfebd1cd | 2646 | return -EINVAL; |
4a0b4ddf | 2647 | |
0ce65797 MS |
2648 | /* disable dm_request_fn's merge heuristic by default */ |
2649 | md->seq_rq_merge_deadline_usecs = 0; | |
2650 | ||
4a0b4ddf | 2651 | md->queue = q; |
bfebd1cd | 2652 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2653 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2654 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2655 | |
bfebd1cd | 2656 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2657 | |
4a0b4ddf MS |
2658 | elv_register_queue(md->queue); |
2659 | ||
bfebd1cd MS |
2660 | return 0; |
2661 | } | |
2662 | ||
2663 | static int dm_mq_init_request(void *data, struct request *rq, | |
2664 | unsigned int hctx_idx, unsigned int request_idx, | |
2665 | unsigned int numa_node) | |
2666 | { | |
2667 | struct mapped_device *md = data; | |
2668 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2669 | ||
2670 | /* | |
2671 | * Must initialize md member of tio, otherwise it won't | |
2672 | * be available in dm_mq_queue_rq. | |
2673 | */ | |
2674 | tio->md = md; | |
2675 | ||
2676 | return 0; | |
2677 | } | |
2678 | ||
2679 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2680 | const struct blk_mq_queue_data *bd) | |
2681 | { | |
2682 | struct request *rq = bd->rq; | |
2683 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2684 | struct mapped_device *md = tio->md; | |
2685 | int srcu_idx; | |
2686 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2687 | struct dm_target *ti; | |
2688 | sector_t pos; | |
2689 | ||
2690 | /* always use block 0 to find the target for flushes for now */ | |
2691 | pos = 0; | |
2692 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2693 | pos = blk_rq_pos(rq); | |
2694 | ||
2695 | ti = dm_table_find_target(map, pos); | |
2696 | if (!dm_target_is_valid(ti)) { | |
2697 | dm_put_live_table(md, srcu_idx); | |
2698 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2699 | /* | |
2700 | * Must perform setup, that rq_completed() requires, | |
2701 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2702 | */ | |
2703 | dm_start_request(md, rq); | |
2704 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2705 | } | |
2706 | dm_put_live_table(md, srcu_idx); | |
2707 | ||
2708 | if (ti->type->busy && ti->type->busy(ti)) | |
2709 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2710 | ||
2711 | dm_start_request(md, rq); | |
2712 | ||
2713 | /* Init tio using md established in .init_request */ | |
2714 | init_tio(tio, rq, md); | |
2715 | ||
2716 | /* Establish tio->ti before queuing work (map_tio_request) */ | |
2717 | tio->ti = ti; | |
2718 | queue_kthread_work(&md->kworker, &tio->work); | |
2719 | ||
2720 | return BLK_MQ_RQ_QUEUE_OK; | |
2721 | } | |
2722 | ||
2723 | static struct blk_mq_ops dm_mq_ops = { | |
2724 | .queue_rq = dm_mq_queue_rq, | |
2725 | .map_queue = blk_mq_map_queue, | |
2726 | .complete = dm_softirq_done, | |
2727 | .init_request = dm_mq_init_request, | |
2728 | }; | |
2729 | ||
2730 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2731 | { | |
2732 | struct request_queue *q; | |
2733 | int err; | |
2734 | ||
2735 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2736 | md->tag_set.ops = &dm_mq_ops; | |
2737 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2738 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2739 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2740 | md->tag_set.nr_hw_queues = 1; | |
2741 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
2742 | md->tag_set.driver_data = md; | |
2743 | ||
2744 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2745 | if (err) | |
2746 | return err; | |
2747 | ||
2748 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2749 | if (IS_ERR(q)) { | |
2750 | err = PTR_ERR(q); | |
2751 | goto out_tag_set; | |
2752 | } | |
2753 | md->queue = q; | |
2754 | dm_init_md_queue(md); | |
2755 | ||
2756 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2757 | blk_mq_register_disk(md->disk); | |
2758 | ||
2759 | init_rq_based_worker_thread(md); | |
2760 | ||
2761 | return 0; | |
2762 | ||
2763 | out_tag_set: | |
2764 | blk_mq_free_tag_set(&md->tag_set); | |
2765 | return err; | |
4a0b4ddf MS |
2766 | } |
2767 | ||
2768 | /* | |
2769 | * Setup the DM device's queue based on md's type | |
2770 | */ | |
2771 | int dm_setup_md_queue(struct mapped_device *md) | |
2772 | { | |
bfebd1cd MS |
2773 | int r; |
2774 | unsigned md_type = dm_get_md_type(md); | |
2775 | ||
2776 | switch (md_type) { | |
2777 | case DM_TYPE_REQUEST_BASED: | |
2778 | r = dm_init_request_based_queue(md); | |
2779 | if (r) { | |
ff36ab34 | 2780 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2781 | return r; |
ff36ab34 | 2782 | } |
bfebd1cd MS |
2783 | break; |
2784 | case DM_TYPE_MQ_REQUEST_BASED: | |
2785 | r = dm_init_request_based_blk_mq_queue(md); | |
2786 | if (r) { | |
2787 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2788 | return r; | |
2789 | } | |
2790 | break; | |
2791 | case DM_TYPE_BIO_BASED: | |
2792 | dm_init_old_md_queue(md); | |
ff36ab34 MS |
2793 | blk_queue_make_request(md->queue, dm_make_request); |
2794 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); | |
bfebd1cd | 2795 | break; |
4a0b4ddf MS |
2796 | } |
2797 | ||
2798 | return 0; | |
2799 | } | |
2800 | ||
2bec1f4a | 2801 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2802 | { |
2803 | struct mapped_device *md; | |
1da177e4 LT |
2804 | unsigned minor = MINOR(dev); |
2805 | ||
2806 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2807 | return NULL; | |
2808 | ||
f32c10b0 | 2809 | spin_lock(&_minor_lock); |
1da177e4 LT |
2810 | |
2811 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2812 | if (md) { |
2813 | if ((md == MINOR_ALLOCED || | |
2814 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2815 | dm_deleting_md(md) || | |
2816 | test_bit(DMF_FREEING, &md->flags))) { | |
2817 | md = NULL; | |
2818 | goto out; | |
2819 | } | |
2820 | dm_get(md); | |
fba9f90e | 2821 | } |
1da177e4 | 2822 | |
fba9f90e | 2823 | out: |
f32c10b0 | 2824 | spin_unlock(&_minor_lock); |
1da177e4 | 2825 | |
637842cf DT |
2826 | return md; |
2827 | } | |
3cf2e4ba | 2828 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2829 | |
9ade92a9 | 2830 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2831 | { |
9ade92a9 | 2832 | return md->interface_ptr; |
1da177e4 LT |
2833 | } |
2834 | ||
2835 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2836 | { | |
2837 | md->interface_ptr = ptr; | |
2838 | } | |
2839 | ||
2840 | void dm_get(struct mapped_device *md) | |
2841 | { | |
2842 | atomic_inc(&md->holders); | |
3f77316d | 2843 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2844 | } |
2845 | ||
09ee96b2 MP |
2846 | int dm_hold(struct mapped_device *md) |
2847 | { | |
2848 | spin_lock(&_minor_lock); | |
2849 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2850 | spin_unlock(&_minor_lock); | |
2851 | return -EBUSY; | |
2852 | } | |
2853 | dm_get(md); | |
2854 | spin_unlock(&_minor_lock); | |
2855 | return 0; | |
2856 | } | |
2857 | EXPORT_SYMBOL_GPL(dm_hold); | |
2858 | ||
72d94861 AK |
2859 | const char *dm_device_name(struct mapped_device *md) |
2860 | { | |
2861 | return md->name; | |
2862 | } | |
2863 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2864 | ||
3f77316d | 2865 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2866 | { |
1134e5ae | 2867 | struct dm_table *map; |
83d5e5b0 | 2868 | int srcu_idx; |
1da177e4 | 2869 | |
3f77316d | 2870 | might_sleep(); |
fba9f90e | 2871 | |
83d5e5b0 | 2872 | map = dm_get_live_table(md, &srcu_idx); |
63a4f065 MS |
2873 | |
2874 | spin_lock(&_minor_lock); | |
3f77316d KU |
2875 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2876 | set_bit(DMF_FREEING, &md->flags); | |
2877 | spin_unlock(&_minor_lock); | |
2878 | ||
2eb6e1e3 KB |
2879 | if (dm_request_based(md)) |
2880 | flush_kthread_worker(&md->kworker); | |
2881 | ||
ab7c7bb6 MP |
2882 | /* |
2883 | * Take suspend_lock so that presuspend and postsuspend methods | |
2884 | * do not race with internal suspend. | |
2885 | */ | |
2886 | mutex_lock(&md->suspend_lock); | |
3f77316d KU |
2887 | if (!dm_suspended_md(md)) { |
2888 | dm_table_presuspend_targets(map); | |
2889 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2890 | } |
ab7c7bb6 | 2891 | mutex_unlock(&md->suspend_lock); |
3f77316d | 2892 | |
83d5e5b0 MP |
2893 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2894 | dm_put_live_table(md, srcu_idx); | |
2895 | ||
3f77316d KU |
2896 | /* |
2897 | * Rare, but there may be I/O requests still going to complete, | |
2898 | * for example. Wait for all references to disappear. | |
2899 | * No one should increment the reference count of the mapped_device, | |
2900 | * after the mapped_device state becomes DMF_FREEING. | |
2901 | */ | |
2902 | if (wait) | |
2903 | while (atomic_read(&md->holders)) | |
2904 | msleep(1); | |
2905 | else if (atomic_read(&md->holders)) | |
2906 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2907 | dm_device_name(md), atomic_read(&md->holders)); | |
2908 | ||
2909 | dm_sysfs_exit(md); | |
3f77316d KU |
2910 | dm_table_destroy(__unbind(md)); |
2911 | free_dev(md); | |
2912 | } | |
2913 | ||
2914 | void dm_destroy(struct mapped_device *md) | |
2915 | { | |
2916 | __dm_destroy(md, true); | |
2917 | } | |
2918 | ||
2919 | void dm_destroy_immediate(struct mapped_device *md) | |
2920 | { | |
2921 | __dm_destroy(md, false); | |
2922 | } | |
2923 | ||
2924 | void dm_put(struct mapped_device *md) | |
2925 | { | |
2926 | atomic_dec(&md->holders); | |
1da177e4 | 2927 | } |
79eb885c | 2928 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2929 | |
401600df | 2930 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2931 | { |
2932 | int r = 0; | |
b44ebeb0 MP |
2933 | DECLARE_WAITQUEUE(wait, current); |
2934 | ||
b44ebeb0 | 2935 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2936 | |
2937 | while (1) { | |
401600df | 2938 | set_current_state(interruptible); |
46125c1c | 2939 | |
b4324fee | 2940 | if (!md_in_flight(md)) |
46125c1c MB |
2941 | break; |
2942 | ||
401600df MP |
2943 | if (interruptible == TASK_INTERRUPTIBLE && |
2944 | signal_pending(current)) { | |
46125c1c MB |
2945 | r = -EINTR; |
2946 | break; | |
2947 | } | |
2948 | ||
2949 | io_schedule(); | |
2950 | } | |
2951 | set_current_state(TASK_RUNNING); | |
2952 | ||
b44ebeb0 MP |
2953 | remove_wait_queue(&md->wait, &wait); |
2954 | ||
46125c1c MB |
2955 | return r; |
2956 | } | |
2957 | ||
1da177e4 LT |
2958 | /* |
2959 | * Process the deferred bios | |
2960 | */ | |
ef208587 | 2961 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2962 | { |
ef208587 MP |
2963 | struct mapped_device *md = container_of(work, struct mapped_device, |
2964 | work); | |
6d6f10df | 2965 | struct bio *c; |
83d5e5b0 MP |
2966 | int srcu_idx; |
2967 | struct dm_table *map; | |
1da177e4 | 2968 | |
83d5e5b0 | 2969 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2970 | |
3b00b203 | 2971 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2972 | spin_lock_irq(&md->deferred_lock); |
2973 | c = bio_list_pop(&md->deferred); | |
2974 | spin_unlock_irq(&md->deferred_lock); | |
2975 | ||
6a8736d1 | 2976 | if (!c) |
df12ee99 | 2977 | break; |
022c2611 | 2978 | |
e6ee8c0b KU |
2979 | if (dm_request_based(md)) |
2980 | generic_make_request(c); | |
6a8736d1 | 2981 | else |
83d5e5b0 | 2982 | __split_and_process_bio(md, map, c); |
022c2611 | 2983 | } |
73d410c0 | 2984 | |
83d5e5b0 | 2985 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2986 | } |
2987 | ||
9a1fb464 | 2988 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2989 | { |
3b00b203 | 2990 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2991 | smp_mb__after_atomic(); |
53d5914f | 2992 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2993 | } |
2994 | ||
1da177e4 | 2995 | /* |
042d2a9b | 2996 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2997 | */ |
042d2a9b | 2998 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2999 | { |
87eb5b21 | 3000 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3001 | struct queue_limits limits; |
042d2a9b | 3002 | int r; |
1da177e4 | 3003 | |
e61290a4 | 3004 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3005 | |
3006 | /* device must be suspended */ | |
4f186f8b | 3007 | if (!dm_suspended_md(md)) |
93c534ae | 3008 | goto out; |
1da177e4 | 3009 | |
3ae70656 MS |
3010 | /* |
3011 | * If the new table has no data devices, retain the existing limits. | |
3012 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3013 | * then new I/O is queued based on these limits, and then some paths | |
3014 | * reappear. | |
3015 | */ | |
3016 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3017 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3018 | if (live_map) |
3019 | limits = md->queue->limits; | |
83d5e5b0 | 3020 | dm_put_live_table_fast(md); |
3ae70656 MS |
3021 | } |
3022 | ||
87eb5b21 MC |
3023 | if (!live_map) { |
3024 | r = dm_calculate_queue_limits(table, &limits); | |
3025 | if (r) { | |
3026 | map = ERR_PTR(r); | |
3027 | goto out; | |
3028 | } | |
042d2a9b | 3029 | } |
754c5fc7 | 3030 | |
042d2a9b | 3031 | map = __bind(md, table, &limits); |
1da177e4 | 3032 | |
93c534ae | 3033 | out: |
e61290a4 | 3034 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3035 | return map; |
1da177e4 LT |
3036 | } |
3037 | ||
3038 | /* | |
3039 | * Functions to lock and unlock any filesystem running on the | |
3040 | * device. | |
3041 | */ | |
2ca3310e | 3042 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3043 | { |
e39e2e95 | 3044 | int r; |
1da177e4 LT |
3045 | |
3046 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3047 | |
db8fef4f | 3048 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3049 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3050 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3051 | md->frozen_sb = NULL; |
3052 | return r; | |
dfbe03f6 AK |
3053 | } |
3054 | ||
aa8d7c2f AK |
3055 | set_bit(DMF_FROZEN, &md->flags); |
3056 | ||
1da177e4 LT |
3057 | return 0; |
3058 | } | |
3059 | ||
2ca3310e | 3060 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3061 | { |
aa8d7c2f AK |
3062 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3063 | return; | |
3064 | ||
db8fef4f | 3065 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3066 | md->frozen_sb = NULL; |
aa8d7c2f | 3067 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3068 | } |
3069 | ||
3070 | /* | |
ffcc3936 MS |
3071 | * If __dm_suspend returns 0, the device is completely quiescent |
3072 | * now. There is no request-processing activity. All new requests | |
3073 | * are being added to md->deferred list. | |
cec47e3d | 3074 | * |
ffcc3936 | 3075 | * Caller must hold md->suspend_lock |
cec47e3d | 3076 | */ |
ffcc3936 MS |
3077 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3078 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3079 | { |
ffcc3936 MS |
3080 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3081 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3082 | int r; | |
1da177e4 | 3083 | |
2e93ccc1 KU |
3084 | /* |
3085 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3086 | * This flag is cleared before dm_suspend returns. | |
3087 | */ | |
3088 | if (noflush) | |
3089 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3090 | ||
d67ee213 MS |
3091 | /* |
3092 | * This gets reverted if there's an error later and the targets | |
3093 | * provide the .presuspend_undo hook. | |
3094 | */ | |
cf222b37 AK |
3095 | dm_table_presuspend_targets(map); |
3096 | ||
32a926da | 3097 | /* |
9f518b27 KU |
3098 | * Flush I/O to the device. |
3099 | * Any I/O submitted after lock_fs() may not be flushed. | |
3100 | * noflush takes precedence over do_lockfs. | |
3101 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3102 | */ |
3103 | if (!noflush && do_lockfs) { | |
3104 | r = lock_fs(md); | |
d67ee213 MS |
3105 | if (r) { |
3106 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3107 | return r; |
d67ee213 | 3108 | } |
aa8d7c2f | 3109 | } |
1da177e4 LT |
3110 | |
3111 | /* | |
3b00b203 MP |
3112 | * Here we must make sure that no processes are submitting requests |
3113 | * to target drivers i.e. no one may be executing | |
3114 | * __split_and_process_bio. This is called from dm_request and | |
3115 | * dm_wq_work. | |
3116 | * | |
3117 | * To get all processes out of __split_and_process_bio in dm_request, | |
3118 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3119 | * __split_and_process_bio from dm_request and quiesce the thread |
3120 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3121 | * flush_workqueue(md->wq). | |
1da177e4 | 3122 | */ |
1eb787ec | 3123 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3124 | if (map) |
3125 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3126 | |
d0bcb878 | 3127 | /* |
29e4013d TH |
3128 | * Stop md->queue before flushing md->wq in case request-based |
3129 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3130 | */ |
2eb6e1e3 | 3131 | if (dm_request_based(md)) { |
9f518b27 | 3132 | stop_queue(md->queue); |
2eb6e1e3 KB |
3133 | flush_kthread_worker(&md->kworker); |
3134 | } | |
cec47e3d | 3135 | |
d0bcb878 KU |
3136 | flush_workqueue(md->wq); |
3137 | ||
1da177e4 | 3138 | /* |
3b00b203 MP |
3139 | * At this point no more requests are entering target request routines. |
3140 | * We call dm_wait_for_completion to wait for all existing requests | |
3141 | * to finish. | |
1da177e4 | 3142 | */ |
ffcc3936 | 3143 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3144 | |
6d6f10df | 3145 | if (noflush) |
022c2611 | 3146 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3147 | if (map) |
3148 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3149 | |
1da177e4 | 3150 | /* were we interrupted ? */ |
46125c1c | 3151 | if (r < 0) { |
9a1fb464 | 3152 | dm_queue_flush(md); |
73d410c0 | 3153 | |
cec47e3d | 3154 | if (dm_request_based(md)) |
9f518b27 | 3155 | start_queue(md->queue); |
cec47e3d | 3156 | |
2ca3310e | 3157 | unlock_fs(md); |
d67ee213 | 3158 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3159 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3160 | } |
1da177e4 | 3161 | |
ffcc3936 MS |
3162 | return r; |
3163 | } | |
3164 | ||
3165 | /* | |
3166 | * We need to be able to change a mapping table under a mounted | |
3167 | * filesystem. For example we might want to move some data in | |
3168 | * the background. Before the table can be swapped with | |
3169 | * dm_bind_table, dm_suspend must be called to flush any in | |
3170 | * flight bios and ensure that any further io gets deferred. | |
3171 | */ | |
3172 | /* | |
3173 | * Suspend mechanism in request-based dm. | |
3174 | * | |
3175 | * 1. Flush all I/Os by lock_fs() if needed. | |
3176 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3177 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3178 | * | |
3179 | * To abort suspend, start the request_queue. | |
3180 | */ | |
3181 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3182 | { | |
3183 | struct dm_table *map = NULL; | |
3184 | int r = 0; | |
3185 | ||
3186 | retry: | |
3187 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3188 | ||
3189 | if (dm_suspended_md(md)) { | |
3190 | r = -EINVAL; | |
3191 | goto out_unlock; | |
3192 | } | |
3193 | ||
3194 | if (dm_suspended_internally_md(md)) { | |
3195 | /* already internally suspended, wait for internal resume */ | |
3196 | mutex_unlock(&md->suspend_lock); | |
3197 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3198 | if (r) | |
3199 | return r; | |
3200 | goto retry; | |
3201 | } | |
3202 | ||
a12f5d48 | 3203 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3204 | |
3205 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3206 | if (r) | |
3207 | goto out_unlock; | |
3b00b203 | 3208 | |
2ca3310e | 3209 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3210 | |
4d4471cb KU |
3211 | dm_table_postsuspend_targets(map); |
3212 | ||
d287483d | 3213 | out_unlock: |
e61290a4 | 3214 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3215 | return r; |
1da177e4 LT |
3216 | } |
3217 | ||
ffcc3936 MS |
3218 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3219 | { | |
3220 | if (map) { | |
3221 | int r = dm_table_resume_targets(map); | |
3222 | if (r) | |
3223 | return r; | |
3224 | } | |
3225 | ||
3226 | dm_queue_flush(md); | |
3227 | ||
3228 | /* | |
3229 | * Flushing deferred I/Os must be done after targets are resumed | |
3230 | * so that mapping of targets can work correctly. | |
3231 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3232 | */ | |
3233 | if (dm_request_based(md)) | |
3234 | start_queue(md->queue); | |
3235 | ||
3236 | unlock_fs(md); | |
3237 | ||
3238 | return 0; | |
3239 | } | |
3240 | ||
1da177e4 LT |
3241 | int dm_resume(struct mapped_device *md) |
3242 | { | |
cf222b37 | 3243 | int r = -EINVAL; |
cf222b37 | 3244 | struct dm_table *map = NULL; |
1da177e4 | 3245 | |
ffcc3936 MS |
3246 | retry: |
3247 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3248 | ||
4f186f8b | 3249 | if (!dm_suspended_md(md)) |
cf222b37 | 3250 | goto out; |
cf222b37 | 3251 | |
ffcc3936 MS |
3252 | if (dm_suspended_internally_md(md)) { |
3253 | /* already internally suspended, wait for internal resume */ | |
3254 | mutex_unlock(&md->suspend_lock); | |
3255 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3256 | if (r) | |
3257 | return r; | |
3258 | goto retry; | |
3259 | } | |
3260 | ||
a12f5d48 | 3261 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3262 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3263 | goto out; |
1da177e4 | 3264 | |
ffcc3936 | 3265 | r = __dm_resume(md, map); |
8757b776 MB |
3266 | if (r) |
3267 | goto out; | |
2ca3310e | 3268 | |
2ca3310e AK |
3269 | clear_bit(DMF_SUSPENDED, &md->flags); |
3270 | ||
cf222b37 AK |
3271 | r = 0; |
3272 | out: | |
e61290a4 | 3273 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3274 | |
cf222b37 | 3275 | return r; |
1da177e4 LT |
3276 | } |
3277 | ||
fd2ed4d2 MP |
3278 | /* |
3279 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3280 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3281 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3282 | */ |
3283 | ||
ffcc3936 | 3284 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3285 | { |
ffcc3936 MS |
3286 | struct dm_table *map = NULL; |
3287 | ||
96b26c8c | 3288 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3289 | return; /* nested internal suspend */ |
3290 | ||
3291 | if (dm_suspended_md(md)) { | |
3292 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3293 | return; /* nest suspend */ | |
3294 | } | |
3295 | ||
a12f5d48 | 3296 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3297 | |
3298 | /* | |
3299 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3300 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3301 | * would require changing .presuspend to return an error -- avoid this | |
3302 | * until there is a need for more elaborate variants of internal suspend. | |
3303 | */ | |
3304 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3305 | ||
3306 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3307 | ||
3308 | dm_table_postsuspend_targets(map); | |
3309 | } | |
3310 | ||
3311 | static void __dm_internal_resume(struct mapped_device *md) | |
3312 | { | |
96b26c8c MP |
3313 | BUG_ON(!md->internal_suspend_count); |
3314 | ||
3315 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3316 | return; /* resume from nested internal suspend */ |
3317 | ||
fd2ed4d2 | 3318 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3319 | goto done; /* resume from nested suspend */ |
3320 | ||
3321 | /* | |
3322 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3323 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3324 | */ | |
3325 | (void) __dm_resume(md, NULL); | |
3326 | ||
3327 | done: | |
3328 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3329 | smp_mb__after_atomic(); | |
3330 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3331 | } | |
3332 | ||
3333 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3334 | { | |
3335 | mutex_lock(&md->suspend_lock); | |
3336 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3337 | mutex_unlock(&md->suspend_lock); | |
3338 | } | |
3339 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3340 | ||
3341 | void dm_internal_resume(struct mapped_device *md) | |
3342 | { | |
3343 | mutex_lock(&md->suspend_lock); | |
3344 | __dm_internal_resume(md); | |
3345 | mutex_unlock(&md->suspend_lock); | |
3346 | } | |
3347 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3348 | ||
3349 | /* | |
3350 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3351 | * which prevents interaction with userspace-driven suspend. | |
3352 | */ | |
3353 | ||
3354 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3355 | { | |
3356 | mutex_lock(&md->suspend_lock); | |
3357 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3358 | return; |
3359 | ||
3360 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3361 | synchronize_srcu(&md->io_barrier); | |
3362 | flush_workqueue(md->wq); | |
3363 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3364 | } | |
b735fede | 3365 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3366 | |
ffcc3936 | 3367 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3368 | { |
ffcc3936 | 3369 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3370 | goto done; |
3371 | ||
3372 | dm_queue_flush(md); | |
3373 | ||
3374 | done: | |
3375 | mutex_unlock(&md->suspend_lock); | |
3376 | } | |
b735fede | 3377 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3378 | |
1da177e4 LT |
3379 | /*----------------------------------------------------------------- |
3380 | * Event notification. | |
3381 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3382 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3383 | unsigned cookie) |
69267a30 | 3384 | { |
60935eb2 MB |
3385 | char udev_cookie[DM_COOKIE_LENGTH]; |
3386 | char *envp[] = { udev_cookie, NULL }; | |
3387 | ||
3388 | if (!cookie) | |
3abf85b5 | 3389 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3390 | else { |
3391 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3392 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3393 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3394 | action, envp); | |
60935eb2 | 3395 | } |
69267a30 AK |
3396 | } |
3397 | ||
7a8c3d3b MA |
3398 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3399 | { | |
3400 | return atomic_add_return(1, &md->uevent_seq); | |
3401 | } | |
3402 | ||
1da177e4 LT |
3403 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3404 | { | |
3405 | return atomic_read(&md->event_nr); | |
3406 | } | |
3407 | ||
3408 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3409 | { | |
3410 | return wait_event_interruptible(md->eventq, | |
3411 | (event_nr != atomic_read(&md->event_nr))); | |
3412 | } | |
3413 | ||
7a8c3d3b MA |
3414 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3415 | { | |
3416 | unsigned long flags; | |
3417 | ||
3418 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3419 | list_add(elist, &md->uevent_list); | |
3420 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3421 | } | |
3422 | ||
1da177e4 LT |
3423 | /* |
3424 | * The gendisk is only valid as long as you have a reference | |
3425 | * count on 'md'. | |
3426 | */ | |
3427 | struct gendisk *dm_disk(struct mapped_device *md) | |
3428 | { | |
3429 | return md->disk; | |
3430 | } | |
3431 | ||
784aae73 MB |
3432 | struct kobject *dm_kobject(struct mapped_device *md) |
3433 | { | |
2995fa78 | 3434 | return &md->kobj_holder.kobj; |
784aae73 MB |
3435 | } |
3436 | ||
784aae73 MB |
3437 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3438 | { | |
3439 | struct mapped_device *md; | |
3440 | ||
2995fa78 | 3441 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3442 | |
4d89b7b4 | 3443 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3444 | dm_deleting_md(md)) |
4d89b7b4 MB |
3445 | return NULL; |
3446 | ||
784aae73 MB |
3447 | dm_get(md); |
3448 | return md; | |
3449 | } | |
3450 | ||
4f186f8b | 3451 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3452 | { |
3453 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3454 | } | |
3455 | ||
ffcc3936 MS |
3456 | int dm_suspended_internally_md(struct mapped_device *md) |
3457 | { | |
3458 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3459 | } | |
3460 | ||
2c140a24 MP |
3461 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3462 | { | |
3463 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3464 | } | |
3465 | ||
64dbce58 KU |
3466 | int dm_suspended(struct dm_target *ti) |
3467 | { | |
ecdb2e25 | 3468 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3469 | } |
3470 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3471 | ||
2e93ccc1 KU |
3472 | int dm_noflush_suspending(struct dm_target *ti) |
3473 | { | |
ecdb2e25 | 3474 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3475 | } |
3476 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3477 | ||
c0820cf5 | 3478 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 3479 | { |
5f015204 JN |
3480 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3481 | struct kmem_cache *cachep; | |
e5863d9a | 3482 | unsigned int pool_size = 0; |
5f015204 | 3483 | unsigned int front_pad; |
e6ee8c0b KU |
3484 | |
3485 | if (!pools) | |
3486 | return NULL; | |
3487 | ||
e5863d9a MS |
3488 | switch (type) { |
3489 | case DM_TYPE_BIO_BASED: | |
5f015204 | 3490 | cachep = _io_cache; |
e8603136 | 3491 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 | 3492 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
e5863d9a MS |
3493 | break; |
3494 | case DM_TYPE_REQUEST_BASED: | |
f4790826 | 3495 | pool_size = dm_get_reserved_rq_based_ios(); |
1ae49ea2 MS |
3496 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); |
3497 | if (!pools->rq_pool) | |
3498 | goto out; | |
e5863d9a MS |
3499 | /* fall through to setup remaining rq-based pools */ |
3500 | case DM_TYPE_MQ_REQUEST_BASED: | |
3501 | cachep = _rq_tio_cache; | |
3502 | if (!pool_size) | |
3503 | pool_size = dm_get_reserved_rq_based_ios(); | |
5f015204 JN |
3504 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3505 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3506 | WARN_ON(per_bio_data_size != 0); | |
e5863d9a MS |
3507 | break; |
3508 | default: | |
5f015204 | 3509 | goto out; |
e5863d9a | 3510 | } |
e6ee8c0b | 3511 | |
6cfa5857 | 3512 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
3513 | if (!pools->io_pool) |
3514 | goto out; | |
e6ee8c0b | 3515 | |
3d8aab2d | 3516 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3517 | if (!pools->bs) |
5f015204 | 3518 | goto out; |
e6ee8c0b | 3519 | |
a91a2785 | 3520 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3521 | goto out; |
a91a2785 | 3522 | |
e6ee8c0b KU |
3523 | return pools; |
3524 | ||
5f015204 JN |
3525 | out: |
3526 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3527 | |
3528 | return NULL; | |
3529 | } | |
3530 | ||
3531 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3532 | { | |
3533 | if (!pools) | |
3534 | return; | |
3535 | ||
3536 | if (pools->io_pool) | |
3537 | mempool_destroy(pools->io_pool); | |
3538 | ||
1ae49ea2 MS |
3539 | if (pools->rq_pool) |
3540 | mempool_destroy(pools->rq_pool); | |
3541 | ||
e6ee8c0b KU |
3542 | if (pools->bs) |
3543 | bioset_free(pools->bs); | |
3544 | ||
3545 | kfree(pools); | |
3546 | } | |
3547 | ||
83d5cde4 | 3548 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3549 | .open = dm_blk_open, |
3550 | .release = dm_blk_close, | |
aa129a22 | 3551 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3552 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3553 | .owner = THIS_MODULE |
3554 | }; | |
3555 | ||
1da177e4 LT |
3556 | /* |
3557 | * module hooks | |
3558 | */ | |
3559 | module_init(dm_init); | |
3560 | module_exit(dm_exit); | |
3561 | ||
3562 | module_param(major, uint, 0); | |
3563 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3564 | |
e8603136 MS |
3565 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3566 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3567 | ||
f4790826 MS |
3568 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3569 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3570 | ||
1da177e4 LT |
3571 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3572 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3573 | MODULE_LICENSE("GPL"); |