]>
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; | |
e262f347 MP |
89 | struct dm_stats_aux stats_aux; |
90 | unsigned long duration_jiffies; | |
91 | unsigned n_sectors; | |
8fbf26ad KU |
92 | }; |
93 | ||
94 | /* | |
94818742 KO |
95 | * For request-based dm - the bio clones we allocate are embedded in these |
96 | * structs. | |
97 | * | |
98 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
99 | * the bioset is created - this means the bio has to come at the end of the | |
100 | * struct. | |
8fbf26ad KU |
101 | */ |
102 | struct dm_rq_clone_bio_info { | |
103 | struct bio *orig; | |
cec47e3d | 104 | struct dm_rq_target_io *tio; |
94818742 | 105 | struct bio clone; |
8fbf26ad KU |
106 | }; |
107 | ||
cec47e3d KU |
108 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
109 | { | |
110 | if (rq && rq->end_io_data) | |
111 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
112 | return NULL; | |
113 | } | |
114 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
115 | ||
ba61fdd1 JM |
116 | #define MINOR_ALLOCED ((void *)-1) |
117 | ||
1da177e4 LT |
118 | /* |
119 | * Bits for the md->flags field. | |
120 | */ | |
1eb787ec | 121 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 122 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 123 | #define DMF_FROZEN 2 |
fba9f90e | 124 | #define DMF_FREEING 3 |
5c6bd75d | 125 | #define DMF_DELETING 4 |
2e93ccc1 | 126 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
127 | #define DMF_DEFERRED_REMOVE 6 |
128 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 129 | |
83d5e5b0 MP |
130 | /* |
131 | * A dummy definition to make RCU happy. | |
132 | * struct dm_table should never be dereferenced in this file. | |
133 | */ | |
134 | struct dm_table { | |
135 | int undefined__; | |
136 | }; | |
137 | ||
304f3f6a MB |
138 | /* |
139 | * Work processed by per-device workqueue. | |
140 | */ | |
1da177e4 | 141 | struct mapped_device { |
83d5e5b0 | 142 | struct srcu_struct io_barrier; |
e61290a4 | 143 | struct mutex suspend_lock; |
1da177e4 | 144 | atomic_t holders; |
5c6bd75d | 145 | atomic_t open_count; |
1da177e4 | 146 | |
2a7faeb1 MP |
147 | /* |
148 | * The current mapping. | |
149 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
150 | * dereference. | |
151 | */ | |
6fa99520 | 152 | struct dm_table __rcu *map; |
2a7faeb1 | 153 | |
86f1152b BM |
154 | struct list_head table_devices; |
155 | struct mutex table_devices_lock; | |
156 | ||
1da177e4 LT |
157 | unsigned long flags; |
158 | ||
165125e1 | 159 | struct request_queue *queue; |
a5664dad | 160 | unsigned type; |
4a0b4ddf | 161 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
162 | struct mutex type_lock; |
163 | ||
36a0456f AK |
164 | struct target_type *immutable_target_type; |
165 | ||
1da177e4 | 166 | struct gendisk *disk; |
7e51f257 | 167 | char name[16]; |
1da177e4 LT |
168 | |
169 | void *interface_ptr; | |
170 | ||
171 | /* | |
172 | * A list of ios that arrived while we were suspended. | |
173 | */ | |
316d315b | 174 | atomic_t pending[2]; |
1da177e4 | 175 | wait_queue_head_t wait; |
53d5914f | 176 | struct work_struct work; |
74859364 | 177 | struct bio_list deferred; |
022c2611 | 178 | spinlock_t deferred_lock; |
1da177e4 | 179 | |
af7e466a | 180 | /* |
29e4013d | 181 | * Processing queue (flush) |
304f3f6a MB |
182 | */ |
183 | struct workqueue_struct *wq; | |
184 | ||
1da177e4 LT |
185 | /* |
186 | * io objects are allocated from here. | |
187 | */ | |
188 | mempool_t *io_pool; | |
1ae49ea2 | 189 | mempool_t *rq_pool; |
1da177e4 | 190 | |
9faf400f SB |
191 | struct bio_set *bs; |
192 | ||
1da177e4 LT |
193 | /* |
194 | * Event handling. | |
195 | */ | |
196 | atomic_t event_nr; | |
197 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
198 | atomic_t uevent_seq; |
199 | struct list_head uevent_list; | |
200 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
201 | |
202 | /* | |
203 | * freeze/thaw support require holding onto a super block | |
204 | */ | |
205 | struct super_block *frozen_sb; | |
db8fef4f | 206 | struct block_device *bdev; |
3ac51e74 DW |
207 | |
208 | /* forced geometry settings */ | |
209 | struct hd_geometry geometry; | |
784aae73 | 210 | |
2995fa78 MP |
211 | /* kobject and completion */ |
212 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 213 | |
d87f4c14 TH |
214 | /* zero-length flush that will be cloned and submitted to targets */ |
215 | struct bio flush_bio; | |
fd2ed4d2 | 216 | |
96b26c8c MP |
217 | /* the number of internal suspends */ |
218 | unsigned internal_suspend_count; | |
219 | ||
fd2ed4d2 | 220 | struct dm_stats stats; |
2eb6e1e3 KB |
221 | |
222 | struct kthread_worker kworker; | |
223 | struct task_struct *kworker_task; | |
de3ec86d MS |
224 | |
225 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 226 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 227 | int last_rq_rw; |
0ce65797 MS |
228 | sector_t last_rq_pos; |
229 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
230 | |
231 | /* for blk-mq request-based DM support */ | |
232 | struct blk_mq_tag_set tag_set; | |
17e149b8 | 233 | bool use_blk_mq; |
1da177e4 LT |
234 | }; |
235 | ||
17e149b8 MS |
236 | #ifdef CONFIG_DM_MQ_DEFAULT |
237 | static bool use_blk_mq = true; | |
238 | #else | |
239 | static bool use_blk_mq = false; | |
240 | #endif | |
241 | ||
242 | bool dm_use_blk_mq(struct mapped_device *md) | |
243 | { | |
244 | return md->use_blk_mq; | |
245 | } | |
246 | ||
e6ee8c0b KU |
247 | /* |
248 | * For mempools pre-allocation at the table loading time. | |
249 | */ | |
250 | struct dm_md_mempools { | |
251 | mempool_t *io_pool; | |
1ae49ea2 | 252 | mempool_t *rq_pool; |
e6ee8c0b KU |
253 | struct bio_set *bs; |
254 | }; | |
255 | ||
86f1152b BM |
256 | struct table_device { |
257 | struct list_head list; | |
258 | atomic_t count; | |
259 | struct dm_dev dm_dev; | |
260 | }; | |
261 | ||
6cfa5857 MS |
262 | #define RESERVED_BIO_BASED_IOS 16 |
263 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 264 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 265 | static struct kmem_cache *_io_cache; |
8fbf26ad | 266 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 267 | static struct kmem_cache *_rq_cache; |
94818742 | 268 | |
e8603136 MS |
269 | /* |
270 | * Bio-based DM's mempools' reserved IOs set by the user. | |
271 | */ | |
272 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
273 | ||
f4790826 MS |
274 | /* |
275 | * Request-based DM's mempools' reserved IOs set by the user. | |
276 | */ | |
277 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
278 | ||
09c2d531 | 279 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
280 | unsigned def, unsigned max) |
281 | { | |
09c2d531 MS |
282 | unsigned param = ACCESS_ONCE(*module_param); |
283 | unsigned modified_param = 0; | |
f4790826 | 284 | |
09c2d531 MS |
285 | if (!param) |
286 | modified_param = def; | |
287 | else if (param > max) | |
288 | modified_param = max; | |
f4790826 | 289 | |
09c2d531 MS |
290 | if (modified_param) { |
291 | (void)cmpxchg(module_param, param, modified_param); | |
292 | param = modified_param; | |
f4790826 MS |
293 | } |
294 | ||
09c2d531 | 295 | return param; |
f4790826 MS |
296 | } |
297 | ||
e8603136 MS |
298 | unsigned dm_get_reserved_bio_based_ios(void) |
299 | { | |
09c2d531 | 300 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
301 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
302 | } | |
303 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
304 | ||
f4790826 MS |
305 | unsigned dm_get_reserved_rq_based_ios(void) |
306 | { | |
09c2d531 | 307 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
308 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
309 | } | |
310 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
311 | ||
1da177e4 LT |
312 | static int __init local_init(void) |
313 | { | |
51157b4a | 314 | int r = -ENOMEM; |
1da177e4 | 315 | |
1da177e4 | 316 | /* allocate a slab for the dm_ios */ |
028867ac | 317 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 318 | if (!_io_cache) |
51157b4a | 319 | return r; |
1da177e4 | 320 | |
8fbf26ad KU |
321 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
322 | if (!_rq_tio_cache) | |
dba14160 | 323 | goto out_free_io_cache; |
8fbf26ad | 324 | |
1ae49ea2 MS |
325 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
326 | __alignof__(struct request), 0, NULL); | |
327 | if (!_rq_cache) | |
328 | goto out_free_rq_tio_cache; | |
329 | ||
51e5b2bd | 330 | r = dm_uevent_init(); |
51157b4a | 331 | if (r) |
1ae49ea2 | 332 | goto out_free_rq_cache; |
51e5b2bd | 333 | |
acfe0ad7 MP |
334 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
335 | if (!deferred_remove_workqueue) { | |
336 | r = -ENOMEM; | |
337 | goto out_uevent_exit; | |
338 | } | |
339 | ||
1da177e4 LT |
340 | _major = major; |
341 | r = register_blkdev(_major, _name); | |
51157b4a | 342 | if (r < 0) |
acfe0ad7 | 343 | goto out_free_workqueue; |
1da177e4 LT |
344 | |
345 | if (!_major) | |
346 | _major = r; | |
347 | ||
348 | return 0; | |
51157b4a | 349 | |
acfe0ad7 MP |
350 | out_free_workqueue: |
351 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
352 | out_uevent_exit: |
353 | dm_uevent_exit(); | |
1ae49ea2 MS |
354 | out_free_rq_cache: |
355 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
356 | out_free_rq_tio_cache: |
357 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
358 | out_free_io_cache: |
359 | kmem_cache_destroy(_io_cache); | |
360 | ||
361 | return r; | |
1da177e4 LT |
362 | } |
363 | ||
364 | static void local_exit(void) | |
365 | { | |
2c140a24 | 366 | flush_scheduled_work(); |
acfe0ad7 | 367 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 368 | |
1ae49ea2 | 369 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 370 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 371 | kmem_cache_destroy(_io_cache); |
00d59405 | 372 | unregister_blkdev(_major, _name); |
51e5b2bd | 373 | dm_uevent_exit(); |
1da177e4 LT |
374 | |
375 | _major = 0; | |
376 | ||
377 | DMINFO("cleaned up"); | |
378 | } | |
379 | ||
b9249e55 | 380 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
381 | local_init, |
382 | dm_target_init, | |
383 | dm_linear_init, | |
384 | dm_stripe_init, | |
952b3557 | 385 | dm_io_init, |
945fa4d2 | 386 | dm_kcopyd_init, |
1da177e4 | 387 | dm_interface_init, |
fd2ed4d2 | 388 | dm_statistics_init, |
1da177e4 LT |
389 | }; |
390 | ||
b9249e55 | 391 | static void (*_exits[])(void) = { |
1da177e4 LT |
392 | local_exit, |
393 | dm_target_exit, | |
394 | dm_linear_exit, | |
395 | dm_stripe_exit, | |
952b3557 | 396 | dm_io_exit, |
945fa4d2 | 397 | dm_kcopyd_exit, |
1da177e4 | 398 | dm_interface_exit, |
fd2ed4d2 | 399 | dm_statistics_exit, |
1da177e4 LT |
400 | }; |
401 | ||
402 | static int __init dm_init(void) | |
403 | { | |
404 | const int count = ARRAY_SIZE(_inits); | |
405 | ||
406 | int r, i; | |
407 | ||
408 | for (i = 0; i < count; i++) { | |
409 | r = _inits[i](); | |
410 | if (r) | |
411 | goto bad; | |
412 | } | |
413 | ||
414 | return 0; | |
415 | ||
416 | bad: | |
417 | while (i--) | |
418 | _exits[i](); | |
419 | ||
420 | return r; | |
421 | } | |
422 | ||
423 | static void __exit dm_exit(void) | |
424 | { | |
425 | int i = ARRAY_SIZE(_exits); | |
426 | ||
427 | while (i--) | |
428 | _exits[i](); | |
d15b774c AK |
429 | |
430 | /* | |
431 | * Should be empty by this point. | |
432 | */ | |
d15b774c | 433 | idr_destroy(&_minor_idr); |
1da177e4 LT |
434 | } |
435 | ||
436 | /* | |
437 | * Block device functions | |
438 | */ | |
432a212c MA |
439 | int dm_deleting_md(struct mapped_device *md) |
440 | { | |
441 | return test_bit(DMF_DELETING, &md->flags); | |
442 | } | |
443 | ||
fe5f9f2c | 444 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
445 | { |
446 | struct mapped_device *md; | |
447 | ||
fba9f90e JM |
448 | spin_lock(&_minor_lock); |
449 | ||
fe5f9f2c | 450 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
451 | if (!md) |
452 | goto out; | |
453 | ||
5c6bd75d | 454 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 455 | dm_deleting_md(md)) { |
fba9f90e JM |
456 | md = NULL; |
457 | goto out; | |
458 | } | |
459 | ||
1da177e4 | 460 | dm_get(md); |
5c6bd75d | 461 | atomic_inc(&md->open_count); |
fba9f90e JM |
462 | out: |
463 | spin_unlock(&_minor_lock); | |
464 | ||
465 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
466 | } |
467 | ||
db2a144b | 468 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 469 | { |
63a4f065 | 470 | struct mapped_device *md; |
6e9624b8 | 471 | |
4a1aeb98 MB |
472 | spin_lock(&_minor_lock); |
473 | ||
63a4f065 MS |
474 | md = disk->private_data; |
475 | if (WARN_ON(!md)) | |
476 | goto out; | |
477 | ||
2c140a24 MP |
478 | if (atomic_dec_and_test(&md->open_count) && |
479 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 480 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 481 | |
1da177e4 | 482 | dm_put(md); |
63a4f065 | 483 | out: |
4a1aeb98 | 484 | spin_unlock(&_minor_lock); |
1da177e4 LT |
485 | } |
486 | ||
5c6bd75d AK |
487 | int dm_open_count(struct mapped_device *md) |
488 | { | |
489 | return atomic_read(&md->open_count); | |
490 | } | |
491 | ||
492 | /* | |
493 | * Guarantees nothing is using the device before it's deleted. | |
494 | */ | |
2c140a24 | 495 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
496 | { |
497 | int r = 0; | |
498 | ||
499 | spin_lock(&_minor_lock); | |
500 | ||
2c140a24 | 501 | if (dm_open_count(md)) { |
5c6bd75d | 502 | r = -EBUSY; |
2c140a24 MP |
503 | if (mark_deferred) |
504 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
505 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
506 | r = -EEXIST; | |
5c6bd75d AK |
507 | else |
508 | set_bit(DMF_DELETING, &md->flags); | |
509 | ||
510 | spin_unlock(&_minor_lock); | |
511 | ||
512 | return r; | |
513 | } | |
514 | ||
2c140a24 MP |
515 | int dm_cancel_deferred_remove(struct mapped_device *md) |
516 | { | |
517 | int r = 0; | |
518 | ||
519 | spin_lock(&_minor_lock); | |
520 | ||
521 | if (test_bit(DMF_DELETING, &md->flags)) | |
522 | r = -EBUSY; | |
523 | else | |
524 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
525 | ||
526 | spin_unlock(&_minor_lock); | |
527 | ||
528 | return r; | |
529 | } | |
530 | ||
531 | static void do_deferred_remove(struct work_struct *w) | |
532 | { | |
533 | dm_deferred_remove(); | |
534 | } | |
535 | ||
fd2ed4d2 MP |
536 | sector_t dm_get_size(struct mapped_device *md) |
537 | { | |
538 | return get_capacity(md->disk); | |
539 | } | |
540 | ||
9974fa2c MS |
541 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
542 | { | |
543 | return md->queue; | |
544 | } | |
545 | ||
fd2ed4d2 MP |
546 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
547 | { | |
548 | return &md->stats; | |
549 | } | |
550 | ||
3ac51e74 DW |
551 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
552 | { | |
553 | struct mapped_device *md = bdev->bd_disk->private_data; | |
554 | ||
555 | return dm_get_geometry(md, geo); | |
556 | } | |
557 | ||
fe5f9f2c | 558 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
559 | unsigned int cmd, unsigned long arg) |
560 | { | |
fe5f9f2c | 561 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 562 | int srcu_idx; |
6c182cd8 | 563 | struct dm_table *map; |
aa129a22 MB |
564 | struct dm_target *tgt; |
565 | int r = -ENOTTY; | |
566 | ||
6c182cd8 | 567 | retry: |
83d5e5b0 MP |
568 | map = dm_get_live_table(md, &srcu_idx); |
569 | ||
aa129a22 MB |
570 | if (!map || !dm_table_get_size(map)) |
571 | goto out; | |
572 | ||
573 | /* We only support devices that have a single target */ | |
574 | if (dm_table_get_num_targets(map) != 1) | |
575 | goto out; | |
576 | ||
577 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
578 | if (!tgt->type->ioctl) |
579 | goto out; | |
aa129a22 | 580 | |
4f186f8b | 581 | if (dm_suspended_md(md)) { |
aa129a22 MB |
582 | r = -EAGAIN; |
583 | goto out; | |
584 | } | |
585 | ||
4d341d82 | 586 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
587 | |
588 | out: | |
83d5e5b0 | 589 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 590 | |
6c182cd8 HR |
591 | if (r == -ENOTCONN) { |
592 | msleep(10); | |
593 | goto retry; | |
594 | } | |
595 | ||
aa129a22 MB |
596 | return r; |
597 | } | |
598 | ||
028867ac | 599 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
600 | { |
601 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
602 | } | |
603 | ||
028867ac | 604 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
605 | { |
606 | mempool_free(io, md->io_pool); | |
607 | } | |
608 | ||
028867ac | 609 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 610 | { |
dba14160 | 611 | bio_put(&tio->clone); |
1da177e4 LT |
612 | } |
613 | ||
08885643 KU |
614 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
615 | gfp_t gfp_mask) | |
cec47e3d | 616 | { |
5f015204 | 617 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
618 | } |
619 | ||
620 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
621 | { | |
5f015204 | 622 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
623 | } |
624 | ||
1ae49ea2 MS |
625 | static struct request *alloc_clone_request(struct mapped_device *md, |
626 | gfp_t gfp_mask) | |
627 | { | |
628 | return mempool_alloc(md->rq_pool, gfp_mask); | |
629 | } | |
630 | ||
631 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
632 | { | |
633 | mempool_free(rq, md->rq_pool); | |
634 | } | |
635 | ||
90abb8c4 KU |
636 | static int md_in_flight(struct mapped_device *md) |
637 | { | |
638 | return atomic_read(&md->pending[READ]) + | |
639 | atomic_read(&md->pending[WRITE]); | |
640 | } | |
641 | ||
3eaf840e JNN |
642 | static void start_io_acct(struct dm_io *io) |
643 | { | |
644 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 645 | struct bio *bio = io->bio; |
c9959059 | 646 | int cpu; |
fd2ed4d2 | 647 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
648 | |
649 | io->start_time = jiffies; | |
650 | ||
074a7aca TH |
651 | cpu = part_stat_lock(); |
652 | part_round_stats(cpu, &dm_disk(md)->part0); | |
653 | part_stat_unlock(); | |
1e9bb880 SL |
654 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
655 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
656 | |
657 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 658 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 659 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
660 | } |
661 | ||
d221d2e7 | 662 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
663 | { |
664 | struct mapped_device *md = io->md; | |
665 | struct bio *bio = io->bio; | |
666 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 667 | int pending; |
3eaf840e JNN |
668 | int rw = bio_data_dir(bio); |
669 | ||
18c0b223 | 670 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 671 | |
fd2ed4d2 | 672 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 673 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
674 | bio_sectors(bio), true, duration, &io->stats_aux); |
675 | ||
af7e466a MP |
676 | /* |
677 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 678 | * a flush. |
af7e466a | 679 | */ |
1e9bb880 SL |
680 | pending = atomic_dec_return(&md->pending[rw]); |
681 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 682 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 683 | |
d221d2e7 MP |
684 | /* nudge anyone waiting on suspend queue */ |
685 | if (!pending) | |
686 | wake_up(&md->wait); | |
3eaf840e JNN |
687 | } |
688 | ||
1da177e4 LT |
689 | /* |
690 | * Add the bio to the list of deferred io. | |
691 | */ | |
92c63902 | 692 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 693 | { |
05447420 | 694 | unsigned long flags; |
1da177e4 | 695 | |
05447420 | 696 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 697 | bio_list_add(&md->deferred, bio); |
05447420 | 698 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 699 | queue_work(md->wq, &md->work); |
1da177e4 LT |
700 | } |
701 | ||
702 | /* | |
703 | * Everyone (including functions in this file), should use this | |
704 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 705 | * dm_put_live_table() when finished. |
1da177e4 | 706 | */ |
83d5e5b0 | 707 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 708 | { |
83d5e5b0 MP |
709 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
710 | ||
711 | return srcu_dereference(md->map, &md->io_barrier); | |
712 | } | |
1da177e4 | 713 | |
83d5e5b0 MP |
714 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
715 | { | |
716 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
717 | } | |
718 | ||
719 | void dm_sync_table(struct mapped_device *md) | |
720 | { | |
721 | synchronize_srcu(&md->io_barrier); | |
722 | synchronize_rcu_expedited(); | |
723 | } | |
724 | ||
725 | /* | |
726 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
727 | * The caller must not block between these two functions. | |
728 | */ | |
729 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
730 | { | |
731 | rcu_read_lock(); | |
732 | return rcu_dereference(md->map); | |
733 | } | |
1da177e4 | 734 | |
83d5e5b0 MP |
735 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
736 | { | |
737 | rcu_read_unlock(); | |
1da177e4 LT |
738 | } |
739 | ||
86f1152b BM |
740 | /* |
741 | * Open a table device so we can use it as a map destination. | |
742 | */ | |
743 | static int open_table_device(struct table_device *td, dev_t dev, | |
744 | struct mapped_device *md) | |
745 | { | |
746 | static char *_claim_ptr = "I belong to device-mapper"; | |
747 | struct block_device *bdev; | |
748 | ||
749 | int r; | |
750 | ||
751 | BUG_ON(td->dm_dev.bdev); | |
752 | ||
753 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
754 | if (IS_ERR(bdev)) | |
755 | return PTR_ERR(bdev); | |
756 | ||
757 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
758 | if (r) { | |
759 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
760 | return r; | |
761 | } | |
762 | ||
763 | td->dm_dev.bdev = bdev; | |
764 | return 0; | |
765 | } | |
766 | ||
767 | /* | |
768 | * Close a table device that we've been using. | |
769 | */ | |
770 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
771 | { | |
772 | if (!td->dm_dev.bdev) | |
773 | return; | |
774 | ||
775 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
776 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
777 | td->dm_dev.bdev = NULL; | |
778 | } | |
779 | ||
780 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
781 | fmode_t mode) { | |
782 | struct table_device *td; | |
783 | ||
784 | list_for_each_entry(td, l, list) | |
785 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
786 | return td; | |
787 | ||
788 | return NULL; | |
789 | } | |
790 | ||
791 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
792 | struct dm_dev **result) { | |
793 | int r; | |
794 | struct table_device *td; | |
795 | ||
796 | mutex_lock(&md->table_devices_lock); | |
797 | td = find_table_device(&md->table_devices, dev, mode); | |
798 | if (!td) { | |
799 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
800 | if (!td) { | |
801 | mutex_unlock(&md->table_devices_lock); | |
802 | return -ENOMEM; | |
803 | } | |
804 | ||
805 | td->dm_dev.mode = mode; | |
806 | td->dm_dev.bdev = NULL; | |
807 | ||
808 | if ((r = open_table_device(td, dev, md))) { | |
809 | mutex_unlock(&md->table_devices_lock); | |
810 | kfree(td); | |
811 | return r; | |
812 | } | |
813 | ||
814 | format_dev_t(td->dm_dev.name, dev); | |
815 | ||
816 | atomic_set(&td->count, 0); | |
817 | list_add(&td->list, &md->table_devices); | |
818 | } | |
819 | atomic_inc(&td->count); | |
820 | mutex_unlock(&md->table_devices_lock); | |
821 | ||
822 | *result = &td->dm_dev; | |
823 | return 0; | |
824 | } | |
825 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
826 | ||
827 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
828 | { | |
829 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
830 | ||
831 | mutex_lock(&md->table_devices_lock); | |
832 | if (atomic_dec_and_test(&td->count)) { | |
833 | close_table_device(td, md); | |
834 | list_del(&td->list); | |
835 | kfree(td); | |
836 | } | |
837 | mutex_unlock(&md->table_devices_lock); | |
838 | } | |
839 | EXPORT_SYMBOL(dm_put_table_device); | |
840 | ||
841 | static void free_table_devices(struct list_head *devices) | |
842 | { | |
843 | struct list_head *tmp, *next; | |
844 | ||
845 | list_for_each_safe(tmp, next, devices) { | |
846 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
847 | ||
848 | DMWARN("dm_destroy: %s still exists with %d references", | |
849 | td->dm_dev.name, atomic_read(&td->count)); | |
850 | kfree(td); | |
851 | } | |
852 | } | |
853 | ||
3ac51e74 DW |
854 | /* |
855 | * Get the geometry associated with a dm device | |
856 | */ | |
857 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
858 | { | |
859 | *geo = md->geometry; | |
860 | ||
861 | return 0; | |
862 | } | |
863 | ||
864 | /* | |
865 | * Set the geometry of a device. | |
866 | */ | |
867 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
868 | { | |
869 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
870 | ||
871 | if (geo->start > sz) { | |
872 | DMWARN("Start sector is beyond the geometry limits."); | |
873 | return -EINVAL; | |
874 | } | |
875 | ||
876 | md->geometry = *geo; | |
877 | ||
878 | return 0; | |
879 | } | |
880 | ||
1da177e4 LT |
881 | /*----------------------------------------------------------------- |
882 | * CRUD START: | |
883 | * A more elegant soln is in the works that uses the queue | |
884 | * merge fn, unfortunately there are a couple of changes to | |
885 | * the block layer that I want to make for this. So in the | |
886 | * interests of getting something for people to use I give | |
887 | * you this clearly demarcated crap. | |
888 | *---------------------------------------------------------------*/ | |
889 | ||
2e93ccc1 KU |
890 | static int __noflush_suspending(struct mapped_device *md) |
891 | { | |
892 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
893 | } | |
894 | ||
1da177e4 LT |
895 | /* |
896 | * Decrements the number of outstanding ios that a bio has been | |
897 | * cloned into, completing the original io if necc. | |
898 | */ | |
858119e1 | 899 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 900 | { |
2e93ccc1 | 901 | unsigned long flags; |
b35f8caa MB |
902 | int io_error; |
903 | struct bio *bio; | |
904 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
905 | |
906 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
907 | if (unlikely(error)) { |
908 | spin_lock_irqsave(&io->endio_lock, flags); | |
909 | if (!(io->error > 0 && __noflush_suspending(md))) | |
910 | io->error = error; | |
911 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
912 | } | |
1da177e4 LT |
913 | |
914 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
915 | if (io->error == DM_ENDIO_REQUEUE) { |
916 | /* | |
917 | * Target requested pushing back the I/O. | |
2e93ccc1 | 918 | */ |
022c2611 | 919 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
920 | if (__noflush_suspending(md)) |
921 | bio_list_add_head(&md->deferred, io->bio); | |
922 | else | |
2e93ccc1 KU |
923 | /* noflush suspend was interrupted. */ |
924 | io->error = -EIO; | |
022c2611 | 925 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
926 | } |
927 | ||
b35f8caa MB |
928 | io_error = io->error; |
929 | bio = io->bio; | |
6a8736d1 TH |
930 | end_io_acct(io); |
931 | free_io(md, io); | |
932 | ||
933 | if (io_error == DM_ENDIO_REQUEUE) | |
934 | return; | |
2e93ccc1 | 935 | |
4f024f37 | 936 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 937 | /* |
6a8736d1 TH |
938 | * Preflush done for flush with data, reissue |
939 | * without REQ_FLUSH. | |
af7e466a | 940 | */ |
6a8736d1 TH |
941 | bio->bi_rw &= ~REQ_FLUSH; |
942 | queue_io(md, bio); | |
af7e466a | 943 | } else { |
b372d360 | 944 | /* done with normal IO or empty flush */ |
0a82a8d1 | 945 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
946 | bio->bi_error = io_error; |
947 | bio_endio(bio); | |
b35f8caa | 948 | } |
1da177e4 LT |
949 | } |
950 | } | |
951 | ||
7eee4ae2 MS |
952 | static void disable_write_same(struct mapped_device *md) |
953 | { | |
954 | struct queue_limits *limits = dm_get_queue_limits(md); | |
955 | ||
956 | /* device doesn't really support WRITE SAME, disable it */ | |
957 | limits->max_write_same_sectors = 0; | |
958 | } | |
959 | ||
4246a0b6 | 960 | static void clone_endio(struct bio *bio) |
1da177e4 | 961 | { |
4246a0b6 | 962 | int error = bio->bi_error; |
5164bece | 963 | int r = error; |
bfc6d41c | 964 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 965 | struct dm_io *io = tio->io; |
9faf400f | 966 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
967 | dm_endio_fn endio = tio->ti->type->end_io; |
968 | ||
1da177e4 | 969 | if (endio) { |
7de3ee57 | 970 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
971 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
972 | /* | |
973 | * error and requeue request are handled | |
974 | * in dec_pending(). | |
975 | */ | |
1da177e4 | 976 | error = r; |
45cbcd79 KU |
977 | else if (r == DM_ENDIO_INCOMPLETE) |
978 | /* The target will handle the io */ | |
6712ecf8 | 979 | return; |
45cbcd79 KU |
980 | else if (r) { |
981 | DMWARN("unimplemented target endio return value: %d", r); | |
982 | BUG(); | |
983 | } | |
1da177e4 LT |
984 | } |
985 | ||
7eee4ae2 MS |
986 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
987 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
988 | disable_write_same(md); | |
989 | ||
9faf400f | 990 | free_tio(md, tio); |
b35f8caa | 991 | dec_pending(io, error); |
1da177e4 LT |
992 | } |
993 | ||
78d8e58a MS |
994 | /* |
995 | * Partial completion handling for request-based dm | |
996 | */ | |
4246a0b6 | 997 | static void end_clone_bio(struct bio *clone) |
78d8e58a MS |
998 | { |
999 | struct dm_rq_clone_bio_info *info = | |
1000 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
1001 | struct dm_rq_target_io *tio = info->tio; | |
1002 | struct bio *bio = info->orig; | |
1003 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; | |
1004 | ||
1005 | bio_put(clone); | |
1006 | ||
1007 | if (tio->error) | |
1008 | /* | |
1009 | * An error has already been detected on the request. | |
1010 | * Once error occurred, just let clone->end_io() handle | |
1011 | * the remainder. | |
1012 | */ | |
1013 | return; | |
4246a0b6 | 1014 | else if (bio->bi_error) { |
78d8e58a MS |
1015 | /* |
1016 | * Don't notice the error to the upper layer yet. | |
1017 | * The error handling decision is made by the target driver, | |
1018 | * when the request is completed. | |
1019 | */ | |
4246a0b6 | 1020 | tio->error = bio->bi_error; |
78d8e58a MS |
1021 | return; |
1022 | } | |
1023 | ||
1024 | /* | |
1025 | * I/O for the bio successfully completed. | |
1026 | * Notice the data completion to the upper layer. | |
1027 | */ | |
1028 | ||
1029 | /* | |
1030 | * bios are processed from the head of the list. | |
1031 | * So the completing bio should always be rq->bio. | |
1032 | * If it's not, something wrong is happening. | |
1033 | */ | |
1034 | if (tio->orig->bio != bio) | |
1035 | DMERR("bio completion is going in the middle of the request"); | |
1036 | ||
1037 | /* | |
1038 | * Update the original request. | |
1039 | * Do not use blk_end_request() here, because it may complete | |
1040 | * the original request before the clone, and break the ordering. | |
1041 | */ | |
1042 | blk_update_request(tio->orig, 0, nr_bytes); | |
1043 | } | |
1044 | ||
bfebd1cd MS |
1045 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1046 | { | |
1047 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1048 | } | |
1049 | ||
e262f347 MP |
1050 | static void rq_end_stats(struct mapped_device *md, struct request *orig) |
1051 | { | |
1052 | if (unlikely(dm_stats_used(&md->stats))) { | |
1053 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1054 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
1055 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1056 | tio->n_sectors, true, tio->duration_jiffies, | |
1057 | &tio->stats_aux); | |
1058 | } | |
1059 | } | |
1060 | ||
cec47e3d KU |
1061 | /* |
1062 | * Don't touch any member of the md after calling this function because | |
1063 | * the md may be freed in dm_put() at the end of this function. | |
1064 | * Or do dm_get() before calling this function and dm_put() later. | |
1065 | */ | |
466d89a6 | 1066 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1067 | { |
b4324fee | 1068 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1069 | |
1070 | /* nudge anyone waiting on suspend queue */ | |
621739b0 | 1071 | if (!md_in_flight(md)) |
cec47e3d KU |
1072 | wake_up(&md->wait); |
1073 | ||
a8c32a5c JA |
1074 | /* |
1075 | * Run this off this callpath, as drivers could invoke end_io while | |
1076 | * inside their request_fn (and holding the queue lock). Calling | |
1077 | * back into ->request_fn() could deadlock attempting to grab the | |
1078 | * queue lock again. | |
1079 | */ | |
9a0e609e | 1080 | if (run_queue) { |
bfebd1cd MS |
1081 | if (md->queue->mq_ops) |
1082 | blk_mq_run_hw_queues(md->queue, true); | |
621739b0 | 1083 | else |
9a0e609e MS |
1084 | blk_run_queue_async(md->queue); |
1085 | } | |
cec47e3d KU |
1086 | |
1087 | /* | |
1088 | * dm_put() must be at the end of this function. See the comment above | |
1089 | */ | |
1090 | dm_put(md); | |
1091 | } | |
1092 | ||
e5d8de32 | 1093 | static void free_rq_clone(struct request *clone) |
a77e28c7 KU |
1094 | { |
1095 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1096 | struct mapped_device *md = tio->md; |
a77e28c7 | 1097 | |
78d8e58a MS |
1098 | blk_rq_unprep_clone(clone); |
1099 | ||
aa6df8dd MS |
1100 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1101 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1102 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1103 | else if (!md->queue->mq_ops) |
1104 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd | 1105 | free_clone_request(md, clone); |
aa6df8dd MS |
1106 | /* |
1107 | * NOTE: for the blk-mq queue stacked on request_fn queue(s) case: | |
1108 | * no need to call free_clone_request() because we leverage blk-mq by | |
1109 | * allocating the clone at the end of the blk-mq pdu (see: clone_rq) | |
1110 | */ | |
bfebd1cd MS |
1111 | |
1112 | if (!md->queue->mq_ops) | |
1113 | free_rq_tio(tio); | |
a77e28c7 KU |
1114 | } |
1115 | ||
980691e5 KU |
1116 | /* |
1117 | * Complete the clone and the original request. | |
466d89a6 KB |
1118 | * Must be called without clone's queue lock held, |
1119 | * see end_clone_request() for more details. | |
980691e5 KU |
1120 | */ |
1121 | static void dm_end_request(struct request *clone, int error) | |
1122 | { | |
1123 | int rw = rq_data_dir(clone); | |
1124 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1125 | struct mapped_device *md = tio->md; | |
1126 | struct request *rq = tio->orig; | |
1127 | ||
29e4013d | 1128 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1129 | rq->errors = clone->errors; |
1130 | rq->resid_len = clone->resid_len; | |
1131 | ||
1132 | if (rq->sense) | |
1133 | /* | |
1134 | * We are using the sense buffer of the original | |
1135 | * request. | |
1136 | * So setting the length of the sense data is enough. | |
1137 | */ | |
1138 | rq->sense_len = clone->sense_len; | |
1139 | } | |
1140 | ||
e5d8de32 | 1141 | free_rq_clone(clone); |
e262f347 | 1142 | rq_end_stats(md, rq); |
bfebd1cd MS |
1143 | if (!rq->q->mq_ops) |
1144 | blk_end_request_all(rq, error); | |
1145 | else | |
1146 | blk_mq_end_request(rq, error); | |
29e4013d | 1147 | rq_completed(md, rw, true); |
980691e5 KU |
1148 | } |
1149 | ||
cec47e3d KU |
1150 | static void dm_unprep_request(struct request *rq) |
1151 | { | |
bfebd1cd | 1152 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1153 | struct request *clone = tio->clone; |
cec47e3d | 1154 | |
bfebd1cd MS |
1155 | if (!rq->q->mq_ops) { |
1156 | rq->special = NULL; | |
1157 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1158 | } | |
cec47e3d | 1159 | |
e5863d9a | 1160 | if (clone) |
e5d8de32 | 1161 | free_rq_clone(clone); |
cec47e3d KU |
1162 | } |
1163 | ||
1164 | /* | |
1165 | * Requeue the original request of a clone. | |
1166 | */ | |
bfebd1cd | 1167 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1168 | { |
cec47e3d KU |
1169 | struct request_queue *q = rq->q; |
1170 | unsigned long flags; | |
1171 | ||
cec47e3d | 1172 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1173 | blk_requeue_request(q, rq); |
4ae9944d | 1174 | blk_run_queue_async(q); |
cec47e3d | 1175 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1176 | } |
1177 | ||
2d76fff1 MS |
1178 | static void dm_requeue_original_request(struct mapped_device *md, |
1179 | struct request *rq) | |
bfebd1cd MS |
1180 | { |
1181 | int rw = rq_data_dir(rq); | |
1182 | ||
1183 | dm_unprep_request(rq); | |
1184 | ||
e262f347 | 1185 | rq_end_stats(md, rq); |
bfebd1cd MS |
1186 | if (!rq->q->mq_ops) |
1187 | old_requeue_request(rq); | |
1188 | else { | |
1189 | blk_mq_requeue_request(rq); | |
1190 | blk_mq_kick_requeue_list(rq->q); | |
1191 | } | |
cec47e3d | 1192 | |
466d89a6 KB |
1193 | rq_completed(md, rw, false); |
1194 | } | |
1195 | ||
bfebd1cd | 1196 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1197 | { |
1198 | unsigned long flags; | |
1199 | ||
bfebd1cd MS |
1200 | if (blk_queue_stopped(q)) |
1201 | return; | |
1202 | ||
cec47e3d | 1203 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1204 | blk_stop_queue(q); |
cec47e3d KU |
1205 | spin_unlock_irqrestore(q->queue_lock, flags); |
1206 | } | |
1207 | ||
bfebd1cd | 1208 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1209 | { |
bfebd1cd MS |
1210 | if (!q->mq_ops) |
1211 | old_stop_queue(q); | |
1212 | else | |
1213 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1214 | } |
1215 | ||
bfebd1cd | 1216 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1217 | { |
1218 | unsigned long flags; | |
1219 | ||
1220 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1221 | if (blk_queue_stopped(q)) |
1222 | blk_start_queue(q); | |
cec47e3d KU |
1223 | spin_unlock_irqrestore(q->queue_lock, flags); |
1224 | } | |
1225 | ||
bfebd1cd MS |
1226 | static void start_queue(struct request_queue *q) |
1227 | { | |
1228 | if (!q->mq_ops) | |
1229 | old_start_queue(q); | |
1230 | else | |
1231 | blk_mq_start_stopped_hw_queues(q, true); | |
1232 | } | |
1233 | ||
11a68244 | 1234 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1235 | { |
11a68244 | 1236 | int r = error; |
cec47e3d | 1237 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1238 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1239 | |
ba1cbad9 MS |
1240 | if (tio->ti) { |
1241 | rq_end_io = tio->ti->type->rq_end_io; | |
1242 | ||
1243 | if (mapped && rq_end_io) | |
1244 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1245 | } | |
cec47e3d | 1246 | |
7eee4ae2 MS |
1247 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1248 | !clone->q->limits.max_write_same_sectors)) | |
1249 | disable_write_same(tio->md); | |
1250 | ||
11a68244 | 1251 | if (r <= 0) |
cec47e3d | 1252 | /* The target wants to complete the I/O */ |
11a68244 KU |
1253 | dm_end_request(clone, r); |
1254 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1255 | /* The target will handle the I/O */ |
1256 | return; | |
11a68244 | 1257 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1258 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1259 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1260 | else { |
11a68244 | 1261 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1262 | BUG(); |
1263 | } | |
1264 | } | |
1265 | ||
11a68244 KU |
1266 | /* |
1267 | * Request completion handler for request-based dm | |
1268 | */ | |
1269 | static void dm_softirq_done(struct request *rq) | |
1270 | { | |
1271 | bool mapped = true; | |
bfebd1cd | 1272 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1273 | struct request *clone = tio->clone; |
bfebd1cd | 1274 | int rw; |
11a68244 | 1275 | |
e5863d9a | 1276 | if (!clone) { |
e262f347 | 1277 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1278 | rw = rq_data_dir(rq); |
1279 | if (!rq->q->mq_ops) { | |
1280 | blk_end_request_all(rq, tio->error); | |
1281 | rq_completed(tio->md, rw, false); | |
1282 | free_rq_tio(tio); | |
1283 | } else { | |
1284 | blk_mq_end_request(rq, tio->error); | |
1285 | rq_completed(tio->md, rw, false); | |
1286 | } | |
e5863d9a MS |
1287 | return; |
1288 | } | |
11a68244 KU |
1289 | |
1290 | if (rq->cmd_flags & REQ_FAILED) | |
1291 | mapped = false; | |
1292 | ||
1293 | dm_done(clone, tio->error, mapped); | |
1294 | } | |
1295 | ||
cec47e3d KU |
1296 | /* |
1297 | * Complete the clone and the original request with the error status | |
1298 | * through softirq context. | |
1299 | */ | |
466d89a6 | 1300 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1301 | { |
bfebd1cd | 1302 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1303 | |
1304 | tio->error = error; | |
cec47e3d KU |
1305 | blk_complete_request(rq); |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * Complete the not-mapped clone and the original request with the error status | |
1310 | * through softirq context. | |
1311 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1312 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1313 | */ |
466d89a6 | 1314 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1315 | { |
cec47e3d | 1316 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1317 | dm_complete_request(rq, error); |
cec47e3d | 1318 | } |
cec47e3d KU |
1319 | |
1320 | /* | |
bfebd1cd | 1321 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1322 | */ |
1323 | static void end_clone_request(struct request *clone, int error) | |
1324 | { | |
466d89a6 KB |
1325 | struct dm_rq_target_io *tio = clone->end_io_data; |
1326 | ||
e5863d9a MS |
1327 | if (!clone->q->mq_ops) { |
1328 | /* | |
1329 | * For just cleaning up the information of the queue in which | |
1330 | * the clone was dispatched. | |
1331 | * The clone is *NOT* freed actually here because it is alloced | |
1332 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1333 | */ | |
1334 | __blk_put_request(clone->q, clone); | |
1335 | } | |
cec47e3d KU |
1336 | |
1337 | /* | |
1338 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1339 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1340 | * - another request may be submitted by the upper level driver |
1341 | * of the stacking during the completion | |
1342 | * - the submission which requires queue lock may be done | |
466d89a6 | 1343 | * against this clone's queue |
cec47e3d | 1344 | */ |
466d89a6 | 1345 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1346 | } |
1347 | ||
56a67df7 MS |
1348 | /* |
1349 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1350 | * target boundary. | |
1351 | */ | |
1352 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1353 | { | |
1354 | sector_t target_offset = dm_target_offset(ti, sector); | |
1355 | ||
1356 | return ti->len - target_offset; | |
1357 | } | |
1358 | ||
1359 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1360 | { |
56a67df7 | 1361 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1362 | sector_t offset, max_len; |
1da177e4 LT |
1363 | |
1364 | /* | |
542f9038 | 1365 | * Does the target need to split even further? |
1da177e4 | 1366 | */ |
542f9038 MS |
1367 | if (ti->max_io_len) { |
1368 | offset = dm_target_offset(ti, sector); | |
1369 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1370 | max_len = sector_div(offset, ti->max_io_len); | |
1371 | else | |
1372 | max_len = offset & (ti->max_io_len - 1); | |
1373 | max_len = ti->max_io_len - max_len; | |
1374 | ||
1375 | if (len > max_len) | |
1376 | len = max_len; | |
1da177e4 LT |
1377 | } |
1378 | ||
1379 | return len; | |
1380 | } | |
1381 | ||
542f9038 MS |
1382 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1383 | { | |
1384 | if (len > UINT_MAX) { | |
1385 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1386 | (unsigned long long)len, UINT_MAX); | |
1387 | ti->error = "Maximum size of target IO is too large"; | |
1388 | return -EINVAL; | |
1389 | } | |
1390 | ||
1391 | ti->max_io_len = (uint32_t) len; | |
1392 | ||
1393 | return 0; | |
1394 | } | |
1395 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1396 | ||
1dd40c3e MP |
1397 | /* |
1398 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1399 | * allowed for all bio types except REQ_FLUSH. | |
1400 | * | |
1401 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1402 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1403 | * sent in a next bio. | |
1404 | * | |
1405 | * A diagram that explains the arithmetics: | |
1406 | * +--------------------+---------------+-------+ | |
1407 | * | 1 | 2 | 3 | | |
1408 | * +--------------------+---------------+-------+ | |
1409 | * | |
1410 | * <-------------- *tio->len_ptr ---------------> | |
1411 | * <------- bi_size -------> | |
1412 | * <-- n_sectors --> | |
1413 | * | |
1414 | * Region 1 was already iterated over with bio_advance or similar function. | |
1415 | * (it may be empty if the target doesn't use bio_advance) | |
1416 | * Region 2 is the remaining bio size that the target wants to process. | |
1417 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1418 | * to make it empty) | |
1419 | * The target requires that region 3 is to be sent in the next bio. | |
1420 | * | |
1421 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1422 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1423 | * copies of the bio. | |
1424 | */ | |
1425 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1426 | { | |
1427 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1428 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1429 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1430 | BUG_ON(bi_size > *tio->len_ptr); | |
1431 | BUG_ON(n_sectors > bi_size); | |
1432 | *tio->len_ptr -= bi_size - n_sectors; | |
1433 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1434 | } | |
1435 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1436 | ||
bd2a49b8 | 1437 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1438 | { |
1439 | int r; | |
2056a782 | 1440 | sector_t sector; |
9faf400f | 1441 | struct mapped_device *md; |
dba14160 | 1442 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1443 | struct dm_target *ti = tio->ti; |
1da177e4 | 1444 | |
1da177e4 | 1445 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1446 | |
1447 | /* | |
1448 | * Map the clone. If r == 0 we don't need to do | |
1449 | * anything, the target has assumed ownership of | |
1450 | * this io. | |
1451 | */ | |
1452 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1453 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1454 | r = ti->type->map(ti, clone); |
45cbcd79 | 1455 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1456 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1457 | |
d07335e5 MS |
1458 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1459 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1460 | |
1da177e4 | 1461 | generic_make_request(clone); |
2e93ccc1 KU |
1462 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1463 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1464 | md = tio->io->md; |
1465 | dec_pending(tio->io, r); | |
9faf400f | 1466 | free_tio(md, tio); |
ab37844d | 1467 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1468 | DMWARN("unimplemented target map return value: %d", r); |
1469 | BUG(); | |
1da177e4 LT |
1470 | } |
1471 | } | |
1472 | ||
1473 | struct clone_info { | |
1474 | struct mapped_device *md; | |
1475 | struct dm_table *map; | |
1476 | struct bio *bio; | |
1477 | struct dm_io *io; | |
1478 | sector_t sector; | |
e0d6609a | 1479 | unsigned sector_count; |
1da177e4 LT |
1480 | }; |
1481 | ||
e0d6609a | 1482 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1483 | { |
4f024f37 KO |
1484 | bio->bi_iter.bi_sector = sector; |
1485 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1486 | } |
1487 | ||
1488 | /* | |
1489 | * Creates a bio that consists of range of complete bvecs. | |
1490 | */ | |
dba14160 | 1491 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1492 | sector_t sector, unsigned len) |
1da177e4 | 1493 | { |
dba14160 | 1494 | struct bio *clone = &tio->clone; |
1da177e4 | 1495 | |
1c3b13e6 KO |
1496 | __bio_clone_fast(clone, bio); |
1497 | ||
1498 | if (bio_integrity(bio)) | |
1499 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1500 | |
1c3b13e6 KO |
1501 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1502 | clone->bi_iter.bi_size = to_bytes(len); | |
1503 | ||
1504 | if (bio_integrity(bio)) | |
1505 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1506 | } |
1507 | ||
9015df24 | 1508 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1509 | struct dm_target *ti, |
55a62eef | 1510 | unsigned target_bio_nr) |
f9ab94ce | 1511 | { |
dba14160 MP |
1512 | struct dm_target_io *tio; |
1513 | struct bio *clone; | |
1514 | ||
99778273 | 1515 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1516 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1517 | |
1518 | tio->io = ci->io; | |
1519 | tio->ti = ti; | |
55a62eef | 1520 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1521 | |
1522 | return tio; | |
1523 | } | |
1524 | ||
14fe594d AK |
1525 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1526 | struct dm_target *ti, | |
1dd40c3e | 1527 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1528 | { |
99778273 | 1529 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1530 | struct bio *clone = &tio->clone; |
9015df24 | 1531 | |
1dd40c3e MP |
1532 | tio->len_ptr = len; |
1533 | ||
99778273 | 1534 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1535 | if (len) |
1dd40c3e | 1536 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1537 | |
bd2a49b8 | 1538 | __map_bio(tio); |
f9ab94ce MP |
1539 | } |
1540 | ||
14fe594d | 1541 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1542 | unsigned num_bios, unsigned *len) |
06a426ce | 1543 | { |
55a62eef | 1544 | unsigned target_bio_nr; |
06a426ce | 1545 | |
55a62eef | 1546 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1547 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1548 | } |
1549 | ||
14fe594d | 1550 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1551 | { |
06a426ce | 1552 | unsigned target_nr = 0; |
f9ab94ce MP |
1553 | struct dm_target *ti; |
1554 | ||
b372d360 | 1555 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1556 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1557 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1558 | |
f9ab94ce MP |
1559 | return 0; |
1560 | } | |
1561 | ||
e4c93811 | 1562 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1563 | sector_t sector, unsigned *len) |
5ae89a87 | 1564 | { |
dba14160 | 1565 | struct bio *bio = ci->bio; |
5ae89a87 | 1566 | struct dm_target_io *tio; |
b0d8ed4d AK |
1567 | unsigned target_bio_nr; |
1568 | unsigned num_target_bios = 1; | |
5ae89a87 | 1569 | |
b0d8ed4d AK |
1570 | /* |
1571 | * Does the target want to receive duplicate copies of the bio? | |
1572 | */ | |
1573 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1574 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1575 | |
b0d8ed4d | 1576 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1577 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1578 | tio->len_ptr = len; |
1579 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1580 | __map_bio(tio); |
1581 | } | |
5ae89a87 MS |
1582 | } |
1583 | ||
55a62eef | 1584 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1585 | |
55a62eef | 1586 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1587 | { |
55a62eef | 1588 | return ti->num_discard_bios; |
23508a96 MS |
1589 | } |
1590 | ||
55a62eef | 1591 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1592 | { |
55a62eef | 1593 | return ti->num_write_same_bios; |
23508a96 MS |
1594 | } |
1595 | ||
1596 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1597 | ||
1598 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1599 | { | |
55a62eef | 1600 | return ti->split_discard_bios; |
23508a96 MS |
1601 | } |
1602 | ||
14fe594d AK |
1603 | static int __send_changing_extent_only(struct clone_info *ci, |
1604 | get_num_bios_fn get_num_bios, | |
1605 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1606 | { |
1607 | struct dm_target *ti; | |
e0d6609a | 1608 | unsigned len; |
55a62eef | 1609 | unsigned num_bios; |
5ae89a87 | 1610 | |
a79245b3 MS |
1611 | do { |
1612 | ti = dm_table_find_target(ci->map, ci->sector); | |
1613 | if (!dm_target_is_valid(ti)) | |
1614 | return -EIO; | |
5ae89a87 | 1615 | |
5ae89a87 | 1616 | /* |
23508a96 MS |
1617 | * Even though the device advertised support for this type of |
1618 | * request, that does not mean every target supports it, and | |
936688d7 | 1619 | * reconfiguration might also have changed that since the |
a79245b3 | 1620 | * check was performed. |
5ae89a87 | 1621 | */ |
55a62eef AK |
1622 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1623 | if (!num_bios) | |
a79245b3 | 1624 | return -EOPNOTSUPP; |
5ae89a87 | 1625 | |
23508a96 | 1626 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1627 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1628 | else |
e0d6609a | 1629 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1630 | |
1dd40c3e | 1631 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1632 | |
1633 | ci->sector += len; | |
1634 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1635 | |
1636 | return 0; | |
1637 | } | |
1638 | ||
14fe594d | 1639 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1640 | { |
14fe594d AK |
1641 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1642 | is_split_required_for_discard); | |
23508a96 MS |
1643 | } |
1644 | ||
14fe594d | 1645 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1646 | { |
14fe594d | 1647 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1648 | } |
1649 | ||
e4c93811 AK |
1650 | /* |
1651 | * Select the correct strategy for processing a non-flush bio. | |
1652 | */ | |
14fe594d | 1653 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1654 | { |
dba14160 | 1655 | struct bio *bio = ci->bio; |
512875bd | 1656 | struct dm_target *ti; |
1c3b13e6 | 1657 | unsigned len; |
1da177e4 | 1658 | |
5ae89a87 | 1659 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1660 | return __send_discard(ci); |
23508a96 | 1661 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1662 | return __send_write_same(ci); |
5ae89a87 | 1663 | |
512875bd JN |
1664 | ti = dm_table_find_target(ci->map, ci->sector); |
1665 | if (!dm_target_is_valid(ti)) | |
1666 | return -EIO; | |
1667 | ||
1c3b13e6 | 1668 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1669 | |
1dd40c3e | 1670 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1671 | |
1c3b13e6 KO |
1672 | ci->sector += len; |
1673 | ci->sector_count -= len; | |
1da177e4 | 1674 | |
1c3b13e6 | 1675 | return 0; |
1da177e4 LT |
1676 | } |
1677 | ||
1678 | /* | |
14fe594d | 1679 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1680 | */ |
83d5e5b0 MP |
1681 | static void __split_and_process_bio(struct mapped_device *md, |
1682 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1683 | { |
1684 | struct clone_info ci; | |
512875bd | 1685 | int error = 0; |
1da177e4 | 1686 | |
83d5e5b0 | 1687 | if (unlikely(!map)) { |
6a8736d1 | 1688 | bio_io_error(bio); |
f0b9a450 MP |
1689 | return; |
1690 | } | |
692d0eb9 | 1691 | |
83d5e5b0 | 1692 | ci.map = map; |
1da177e4 | 1693 | ci.md = md; |
1da177e4 LT |
1694 | ci.io = alloc_io(md); |
1695 | ci.io->error = 0; | |
1696 | atomic_set(&ci.io->io_count, 1); | |
1697 | ci.io->bio = bio; | |
1698 | ci.io->md = md; | |
f88fb981 | 1699 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1700 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1701 | |
3eaf840e | 1702 | start_io_acct(ci.io); |
bd2a49b8 | 1703 | |
b372d360 MS |
1704 | if (bio->bi_rw & REQ_FLUSH) { |
1705 | ci.bio = &ci.md->flush_bio; | |
1706 | ci.sector_count = 0; | |
14fe594d | 1707 | error = __send_empty_flush(&ci); |
b372d360 MS |
1708 | /* dec_pending submits any data associated with flush */ |
1709 | } else { | |
6a8736d1 | 1710 | ci.bio = bio; |
d87f4c14 | 1711 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1712 | while (ci.sector_count && !error) |
14fe594d | 1713 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1714 | } |
1da177e4 LT |
1715 | |
1716 | /* drop the extra reference count */ | |
512875bd | 1717 | dec_pending(ci.io, error); |
1da177e4 LT |
1718 | } |
1719 | /*----------------------------------------------------------------- | |
1720 | * CRUD END | |
1721 | *---------------------------------------------------------------*/ | |
1722 | ||
1723 | /* | |
1724 | * The request function that just remaps the bio built up by | |
1725 | * dm_merge_bvec. | |
1726 | */ | |
ff36ab34 | 1727 | static void dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1728 | { |
12f03a49 | 1729 | int rw = bio_data_dir(bio); |
1da177e4 | 1730 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1731 | int srcu_idx; |
1732 | struct dm_table *map; | |
1da177e4 | 1733 | |
83d5e5b0 | 1734 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1735 | |
54efd50b KO |
1736 | blk_queue_split(q, &bio, q->bio_split); |
1737 | ||
18c0b223 | 1738 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1739 | |
6a8736d1 TH |
1740 | /* if we're suspended, we have to queue this io for later */ |
1741 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1742 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1743 | |
6a8736d1 TH |
1744 | if (bio_rw(bio) != READA) |
1745 | queue_io(md, bio); | |
1746 | else | |
54d9a1b4 | 1747 | bio_io_error(bio); |
5a7bbad2 | 1748 | return; |
1da177e4 LT |
1749 | } |
1750 | ||
83d5e5b0 MP |
1751 | __split_and_process_bio(md, map, bio); |
1752 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1753 | return; |
cec47e3d KU |
1754 | } |
1755 | ||
fd2ed4d2 | 1756 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1757 | { |
1758 | return blk_queue_stackable(md->queue); | |
1759 | } | |
1760 | ||
466d89a6 | 1761 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1762 | { |
1763 | int r; | |
1764 | ||
466d89a6 KB |
1765 | if (blk_queue_io_stat(clone->q)) |
1766 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1767 | |
466d89a6 KB |
1768 | clone->start_time = jiffies; |
1769 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1770 | if (r) |
466d89a6 | 1771 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1772 | dm_complete_request(rq, r); |
1773 | } | |
cec47e3d | 1774 | |
78d8e58a MS |
1775 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1776 | void *data) | |
cec47e3d | 1777 | { |
78d8e58a MS |
1778 | struct dm_rq_target_io *tio = data; |
1779 | struct dm_rq_clone_bio_info *info = | |
1780 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
1781 | ||
1782 | info->orig = bio_orig; | |
1783 | info->tio = tio; | |
1784 | bio->bi_end_io = end_clone_bio; | |
1785 | ||
1786 | return 0; | |
1787 | } | |
1788 | ||
1789 | static int setup_clone(struct request *clone, struct request *rq, | |
1790 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
cec47e3d | 1791 | { |
78d8e58a MS |
1792 | int r; |
1793 | ||
1794 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
1795 | dm_rq_bio_constructor, tio); | |
1796 | if (r) | |
1797 | return r; | |
1798 | ||
1799 | clone->cmd = rq->cmd; | |
1800 | clone->cmd_len = rq->cmd_len; | |
1801 | clone->sense = rq->sense; | |
cec47e3d KU |
1802 | clone->end_io = end_clone_request; |
1803 | clone->end_io_data = tio; | |
78d8e58a | 1804 | |
1ae49ea2 | 1805 | tio->clone = clone; |
78d8e58a MS |
1806 | |
1807 | return 0; | |
cec47e3d KU |
1808 | } |
1809 | ||
6facdaff | 1810 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1811 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1812 | { |
02233342 MS |
1813 | /* |
1814 | * Do not allocate a clone if tio->clone was already set | |
1815 | * (see: dm_mq_queue_rq). | |
1816 | */ | |
1817 | bool alloc_clone = !tio->clone; | |
1818 | struct request *clone; | |
1ae49ea2 | 1819 | |
02233342 MS |
1820 | if (alloc_clone) { |
1821 | clone = alloc_clone_request(md, gfp_mask); | |
1822 | if (!clone) | |
1823 | return NULL; | |
1824 | } else | |
1825 | clone = tio->clone; | |
1ae49ea2 MS |
1826 | |
1827 | blk_rq_init(NULL, clone); | |
78d8e58a MS |
1828 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
1829 | /* -ENOMEM */ | |
1830 | if (alloc_clone) | |
1831 | free_clone_request(md, clone); | |
1832 | return NULL; | |
1833 | } | |
1ae49ea2 MS |
1834 | |
1835 | return clone; | |
1836 | } | |
1837 | ||
2eb6e1e3 KB |
1838 | static void map_tio_request(struct kthread_work *work); |
1839 | ||
bfebd1cd MS |
1840 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1841 | struct mapped_device *md) | |
1842 | { | |
1843 | tio->md = md; | |
1844 | tio->ti = NULL; | |
1845 | tio->clone = NULL; | |
1846 | tio->orig = rq; | |
1847 | tio->error = 0; | |
1848 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1849 | if (md->kworker_task) |
1850 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1851 | } |
1852 | ||
466d89a6 KB |
1853 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1854 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1855 | { |
6facdaff | 1856 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1857 | int srcu_idx; |
1858 | struct dm_table *table; | |
6facdaff KU |
1859 | |
1860 | tio = alloc_rq_tio(md, gfp_mask); | |
1861 | if (!tio) | |
1862 | return NULL; | |
1863 | ||
bfebd1cd | 1864 | init_tio(tio, rq, md); |
6facdaff | 1865 | |
e5863d9a MS |
1866 | table = dm_get_live_table(md, &srcu_idx); |
1867 | if (!dm_table_mq_request_based(table)) { | |
1868 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1869 | dm_put_live_table(md, srcu_idx); | |
1870 | free_rq_tio(tio); | |
1871 | return NULL; | |
1872 | } | |
6facdaff | 1873 | } |
e5863d9a | 1874 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1875 | |
466d89a6 | 1876 | return tio; |
6facdaff KU |
1877 | } |
1878 | ||
cec47e3d KU |
1879 | /* |
1880 | * Called with the queue lock held. | |
1881 | */ | |
1882 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1883 | { | |
1884 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1885 | struct dm_rq_target_io *tio; |
cec47e3d | 1886 | |
cec47e3d KU |
1887 | if (unlikely(rq->special)) { |
1888 | DMWARN("Already has something in rq->special."); | |
1889 | return BLKPREP_KILL; | |
1890 | } | |
1891 | ||
466d89a6 KB |
1892 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1893 | if (!tio) | |
cec47e3d | 1894 | return BLKPREP_DEFER; |
cec47e3d | 1895 | |
466d89a6 | 1896 | rq->special = tio; |
cec47e3d KU |
1897 | rq->cmd_flags |= REQ_DONTPREP; |
1898 | ||
1899 | return BLKPREP_OK; | |
1900 | } | |
1901 | ||
9eef87da KU |
1902 | /* |
1903 | * Returns: | |
e5863d9a MS |
1904 | * 0 : the request has been processed |
1905 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1906 | * < 0 : the request was completed due to failure | |
9eef87da | 1907 | */ |
bfebd1cd | 1908 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1909 | struct mapped_device *md) |
cec47e3d | 1910 | { |
e5863d9a | 1911 | int r; |
bfebd1cd | 1912 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1913 | struct request *clone = NULL; |
1914 | ||
1915 | if (tio->clone) { | |
1916 | clone = tio->clone; | |
1917 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1918 | } else { | |
1919 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1920 | if (r < 0) { | |
1921 | /* The target wants to complete the I/O */ | |
1922 | dm_kill_unmapped_request(rq, r); | |
1923 | return r; | |
1924 | } | |
3a140755 JN |
1925 | if (r != DM_MAPIO_REMAPPED) |
1926 | return r; | |
78d8e58a MS |
1927 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
1928 | /* -ENOMEM */ | |
1929 | ti->type->release_clone_rq(clone); | |
1930 | return DM_MAPIO_REQUEUE; | |
1931 | } | |
e5863d9a | 1932 | } |
cec47e3d | 1933 | |
cec47e3d KU |
1934 | switch (r) { |
1935 | case DM_MAPIO_SUBMITTED: | |
1936 | /* The target has taken the I/O to submit by itself later */ | |
1937 | break; | |
1938 | case DM_MAPIO_REMAPPED: | |
1939 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1940 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1941 | blk_rq_pos(rq)); |
1942 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1943 | break; |
1944 | case DM_MAPIO_REQUEUE: | |
1945 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 1946 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
1947 | break; |
1948 | default: | |
1949 | if (r > 0) { | |
1950 | DMWARN("unimplemented target map return value: %d", r); | |
1951 | BUG(); | |
1952 | } | |
1953 | ||
1954 | /* The target wants to complete the I/O */ | |
466d89a6 | 1955 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1956 | return r; |
cec47e3d | 1957 | } |
9eef87da | 1958 | |
e5863d9a | 1959 | return 0; |
cec47e3d KU |
1960 | } |
1961 | ||
2eb6e1e3 | 1962 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 1963 | { |
2eb6e1e3 | 1964 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
1965 | struct request *rq = tio->orig; |
1966 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 1967 | |
bfebd1cd | 1968 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 1969 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
1970 | } |
1971 | ||
466d89a6 | 1972 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 1973 | { |
bfebd1cd MS |
1974 | if (!orig->q->mq_ops) |
1975 | blk_start_request(orig); | |
1976 | else | |
1977 | blk_mq_start_request(orig); | |
466d89a6 | 1978 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 1979 | |
0ce65797 MS |
1980 | if (md->seq_rq_merge_deadline_usecs) { |
1981 | md->last_rq_pos = rq_end_sector(orig); | |
1982 | md->last_rq_rw = rq_data_dir(orig); | |
1983 | md->last_rq_start_time = ktime_get(); | |
1984 | } | |
de3ec86d | 1985 | |
e262f347 MP |
1986 | if (unlikely(dm_stats_used(&md->stats))) { |
1987 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1988 | tio->duration_jiffies = jiffies; | |
1989 | tio->n_sectors = blk_rq_sectors(orig); | |
1990 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1991 | tio->n_sectors, false, 0, &tio->stats_aux); | |
1992 | } | |
1993 | ||
ba1cbad9 MS |
1994 | /* |
1995 | * Hold the md reference here for the in-flight I/O. | |
1996 | * We can't rely on the reference count by device opener, | |
1997 | * because the device may be closed during the request completion | |
1998 | * when all bios are completed. | |
1999 | * See the comment in rq_completed() too. | |
2000 | */ | |
2001 | dm_get(md); | |
ba1cbad9 MS |
2002 | } |
2003 | ||
0ce65797 MS |
2004 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2005 | ||
2006 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2007 | { | |
2008 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2009 | } | |
2010 | ||
2011 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2012 | const char *buf, size_t count) | |
2013 | { | |
2014 | unsigned deadline; | |
2015 | ||
17e149b8 | 2016 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2017 | return count; |
2018 | ||
2019 | if (kstrtouint(buf, 10, &deadline)) | |
2020 | return -EINVAL; | |
2021 | ||
2022 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2023 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2024 | ||
2025 | md->seq_rq_merge_deadline_usecs = deadline; | |
2026 | ||
2027 | return count; | |
2028 | } | |
2029 | ||
2030 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2031 | { | |
2032 | ktime_t kt_deadline; | |
2033 | ||
2034 | if (!md->seq_rq_merge_deadline_usecs) | |
2035 | return false; | |
2036 | ||
2037 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2038 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2039 | ||
2040 | return !ktime_after(ktime_get(), kt_deadline); | |
2041 | } | |
2042 | ||
cec47e3d KU |
2043 | /* |
2044 | * q->request_fn for request-based dm. | |
2045 | * Called with the queue lock held. | |
2046 | */ | |
2047 | static void dm_request_fn(struct request_queue *q) | |
2048 | { | |
2049 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2050 | int srcu_idx; |
2051 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2052 | struct dm_target *ti; |
466d89a6 | 2053 | struct request *rq; |
2eb6e1e3 | 2054 | struct dm_rq_target_io *tio; |
29e4013d | 2055 | sector_t pos; |
cec47e3d KU |
2056 | |
2057 | /* | |
b4324fee KU |
2058 | * For suspend, check blk_queue_stopped() and increment |
2059 | * ->pending within a single queue_lock not to increment the | |
2060 | * number of in-flight I/Os after the queue is stopped in | |
2061 | * dm_suspend(). | |
cec47e3d | 2062 | */ |
7eaceacc | 2063 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2064 | rq = blk_peek_request(q); |
2065 | if (!rq) | |
9d1deb83 | 2066 | goto out; |
cec47e3d | 2067 | |
29e4013d TH |
2068 | /* always use block 0 to find the target for flushes for now */ |
2069 | pos = 0; | |
2070 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2071 | pos = blk_rq_pos(rq); | |
2072 | ||
2073 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2074 | if (!dm_target_is_valid(ti)) { |
2075 | /* | |
466d89a6 | 2076 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2077 | * before calling dm_kill_unmapped_request |
2078 | */ | |
2079 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2080 | dm_start_request(md, rq); |
2081 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2082 | continue; |
2083 | } | |
d0bcb878 | 2084 | |
0ce65797 MS |
2085 | if (dm_request_peeked_before_merge_deadline(md) && |
2086 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2087 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2088 | goto delay_and_out; | |
2089 | ||
cec47e3d | 2090 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2091 | goto delay_and_out; |
cec47e3d | 2092 | |
466d89a6 | 2093 | dm_start_request(md, rq); |
9eef87da | 2094 | |
bfebd1cd | 2095 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2096 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2097 | tio->ti = ti; | |
2098 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2099 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2100 | } |
2101 | ||
2102 | goto out; | |
2103 | ||
7eaceacc | 2104 | delay_and_out: |
d548b34b | 2105 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2106 | out: |
83d5e5b0 | 2107 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2108 | } |
2109 | ||
1da177e4 LT |
2110 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2111 | { | |
8a57dfc6 CS |
2112 | int r = bdi_bits; |
2113 | struct mapped_device *md = congested_data; | |
2114 | struct dm_table *map; | |
1da177e4 | 2115 | |
1eb787ec | 2116 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2117 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2118 | if (map) { |
cec47e3d KU |
2119 | /* |
2120 | * Request-based dm cares about only own queue for | |
2121 | * the query about congestion status of request_queue | |
2122 | */ | |
2123 | if (dm_request_based(md)) | |
4452226e | 2124 | r = md->queue->backing_dev_info.wb.state & |
cec47e3d KU |
2125 | bdi_bits; |
2126 | else | |
2127 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2128 | } |
83d5e5b0 | 2129 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2130 | } |
2131 | ||
1da177e4 LT |
2132 | return r; |
2133 | } | |
2134 | ||
2135 | /*----------------------------------------------------------------- | |
2136 | * An IDR is used to keep track of allocated minor numbers. | |
2137 | *---------------------------------------------------------------*/ | |
2b06cfff | 2138 | static void free_minor(int minor) |
1da177e4 | 2139 | { |
f32c10b0 | 2140 | spin_lock(&_minor_lock); |
1da177e4 | 2141 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2142 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2143 | } |
2144 | ||
2145 | /* | |
2146 | * See if the device with a specific minor # is free. | |
2147 | */ | |
cf13ab8e | 2148 | static int specific_minor(int minor) |
1da177e4 | 2149 | { |
c9d76be6 | 2150 | int r; |
1da177e4 LT |
2151 | |
2152 | if (minor >= (1 << MINORBITS)) | |
2153 | return -EINVAL; | |
2154 | ||
c9d76be6 | 2155 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2156 | spin_lock(&_minor_lock); |
1da177e4 | 2157 | |
c9d76be6 | 2158 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2159 | |
f32c10b0 | 2160 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2161 | idr_preload_end(); |
2162 | if (r < 0) | |
2163 | return r == -ENOSPC ? -EBUSY : r; | |
2164 | return 0; | |
1da177e4 LT |
2165 | } |
2166 | ||
cf13ab8e | 2167 | static int next_free_minor(int *minor) |
1da177e4 | 2168 | { |
c9d76be6 | 2169 | int r; |
62f75c2f | 2170 | |
c9d76be6 | 2171 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2172 | spin_lock(&_minor_lock); |
1da177e4 | 2173 | |
c9d76be6 | 2174 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2175 | |
f32c10b0 | 2176 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2177 | idr_preload_end(); |
2178 | if (r < 0) | |
2179 | return r; | |
2180 | *minor = r; | |
2181 | return 0; | |
1da177e4 LT |
2182 | } |
2183 | ||
83d5cde4 | 2184 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2185 | |
53d5914f MP |
2186 | static void dm_wq_work(struct work_struct *work); |
2187 | ||
4a0b4ddf MS |
2188 | static void dm_init_md_queue(struct mapped_device *md) |
2189 | { | |
2190 | /* | |
2191 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2192 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2193 | * The type is decided at the first table loading time. |
2194 | * To prevent problematic device stacking, clear the queue flag | |
2195 | * for request stacking support until then. | |
2196 | * | |
2197 | * This queue is new, so no concurrency on the queue_flags. | |
2198 | */ | |
2199 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
bfebd1cd | 2200 | } |
4a0b4ddf | 2201 | |
bfebd1cd MS |
2202 | static void dm_init_old_md_queue(struct mapped_device *md) |
2203 | { | |
17e149b8 | 2204 | md->use_blk_mq = false; |
bfebd1cd MS |
2205 | dm_init_md_queue(md); |
2206 | ||
2207 | /* | |
2208 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2209 | */ | |
4a0b4ddf MS |
2210 | md->queue->queuedata = md; |
2211 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2212 | md->queue->backing_dev_info.congested_data = md; | |
ff36ab34 | 2213 | |
4a0b4ddf | 2214 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2215 | } |
2216 | ||
0f20972f MS |
2217 | static void cleanup_mapped_device(struct mapped_device *md) |
2218 | { | |
0f20972f MS |
2219 | if (md->wq) |
2220 | destroy_workqueue(md->wq); | |
2221 | if (md->kworker_task) | |
2222 | kthread_stop(md->kworker_task); | |
2223 | if (md->io_pool) | |
2224 | mempool_destroy(md->io_pool); | |
2225 | if (md->rq_pool) | |
2226 | mempool_destroy(md->rq_pool); | |
2227 | if (md->bs) | |
2228 | bioset_free(md->bs); | |
2229 | ||
b06075a9 MP |
2230 | cleanup_srcu_struct(&md->io_barrier); |
2231 | ||
0f20972f MS |
2232 | if (md->disk) { |
2233 | spin_lock(&_minor_lock); | |
2234 | md->disk->private_data = NULL; | |
2235 | spin_unlock(&_minor_lock); | |
2236 | if (blk_get_integrity(md->disk)) | |
2237 | blk_integrity_unregister(md->disk); | |
2238 | del_gendisk(md->disk); | |
2239 | put_disk(md->disk); | |
2240 | } | |
2241 | ||
2242 | if (md->queue) | |
2243 | blk_cleanup_queue(md->queue); | |
2244 | ||
2245 | if (md->bdev) { | |
2246 | bdput(md->bdev); | |
2247 | md->bdev = NULL; | |
2248 | } | |
2249 | } | |
2250 | ||
1da177e4 LT |
2251 | /* |
2252 | * Allocate and initialise a blank device with a given minor. | |
2253 | */ | |
2b06cfff | 2254 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2255 | { |
2256 | int r; | |
cf13ab8e | 2257 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2258 | void *old_md; |
1da177e4 LT |
2259 | |
2260 | if (!md) { | |
2261 | DMWARN("unable to allocate device, out of memory."); | |
2262 | return NULL; | |
2263 | } | |
2264 | ||
10da4f79 | 2265 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2266 | goto bad_module_get; |
10da4f79 | 2267 | |
1da177e4 | 2268 | /* get a minor number for the dev */ |
2b06cfff | 2269 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2270 | r = next_free_minor(&minor); |
2b06cfff | 2271 | else |
cf13ab8e | 2272 | r = specific_minor(minor); |
1da177e4 | 2273 | if (r < 0) |
6ed7ade8 | 2274 | goto bad_minor; |
1da177e4 | 2275 | |
83d5e5b0 MP |
2276 | r = init_srcu_struct(&md->io_barrier); |
2277 | if (r < 0) | |
2278 | goto bad_io_barrier; | |
2279 | ||
17e149b8 | 2280 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2281 | md->type = DM_TYPE_NONE; |
e61290a4 | 2282 | mutex_init(&md->suspend_lock); |
a5664dad | 2283 | mutex_init(&md->type_lock); |
86f1152b | 2284 | mutex_init(&md->table_devices_lock); |
022c2611 | 2285 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2286 | atomic_set(&md->holders, 1); |
5c6bd75d | 2287 | atomic_set(&md->open_count, 0); |
1da177e4 | 2288 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2289 | atomic_set(&md->uevent_seq, 0); |
2290 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2291 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2292 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2293 | |
4a0b4ddf | 2294 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2295 | if (!md->queue) |
0f20972f | 2296 | goto bad; |
1da177e4 | 2297 | |
4a0b4ddf | 2298 | dm_init_md_queue(md); |
9faf400f | 2299 | |
1da177e4 LT |
2300 | md->disk = alloc_disk(1); |
2301 | if (!md->disk) | |
0f20972f | 2302 | goto bad; |
1da177e4 | 2303 | |
316d315b NK |
2304 | atomic_set(&md->pending[0], 0); |
2305 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2306 | init_waitqueue_head(&md->wait); |
53d5914f | 2307 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2308 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2309 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2310 | md->kworker_task = NULL; |
f0b04115 | 2311 | |
1da177e4 LT |
2312 | md->disk->major = _major; |
2313 | md->disk->first_minor = minor; | |
2314 | md->disk->fops = &dm_blk_dops; | |
2315 | md->disk->queue = md->queue; | |
2316 | md->disk->private_data = md; | |
2317 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2318 | add_disk(md->disk); | |
7e51f257 | 2319 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2320 | |
670368a8 | 2321 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2322 | if (!md->wq) |
0f20972f | 2323 | goto bad; |
304f3f6a | 2324 | |
32a926da MP |
2325 | md->bdev = bdget_disk(md->disk, 0); |
2326 | if (!md->bdev) | |
0f20972f | 2327 | goto bad; |
32a926da | 2328 | |
6a8736d1 TH |
2329 | bio_init(&md->flush_bio); |
2330 | md->flush_bio.bi_bdev = md->bdev; | |
2331 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2332 | ||
fd2ed4d2 MP |
2333 | dm_stats_init(&md->stats); |
2334 | ||
ba61fdd1 | 2335 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2336 | spin_lock(&_minor_lock); |
ba61fdd1 | 2337 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2338 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2339 | |
2340 | BUG_ON(old_md != MINOR_ALLOCED); | |
2341 | ||
1da177e4 LT |
2342 | return md; |
2343 | ||
0f20972f MS |
2344 | bad: |
2345 | cleanup_mapped_device(md); | |
83d5e5b0 | 2346 | bad_io_barrier: |
1da177e4 | 2347 | free_minor(minor); |
6ed7ade8 | 2348 | bad_minor: |
10da4f79 | 2349 | module_put(THIS_MODULE); |
6ed7ade8 | 2350 | bad_module_get: |
1da177e4 LT |
2351 | kfree(md); |
2352 | return NULL; | |
2353 | } | |
2354 | ||
ae9da83f JN |
2355 | static void unlock_fs(struct mapped_device *md); |
2356 | ||
1da177e4 LT |
2357 | static void free_dev(struct mapped_device *md) |
2358 | { | |
f331c029 | 2359 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2360 | |
32a926da | 2361 | unlock_fs(md); |
2eb6e1e3 | 2362 | |
0f20972f MS |
2363 | cleanup_mapped_device(md); |
2364 | if (md->use_blk_mq) | |
2365 | blk_mq_free_tag_set(&md->tag_set); | |
63a4f065 | 2366 | |
86f1152b | 2367 | free_table_devices(&md->table_devices); |
63a4f065 | 2368 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
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 | |
4e6e36c3 MS |
2379 | if (md->bs) { |
2380 | /* The md already has necessary mempools. */ | |
2381 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
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 | } |
4e6e36c3 MS |
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 | */ | |
2398 | goto out; | |
c0820cf5 | 2399 | } |
e6ee8c0b | 2400 | |
cbc4e3c1 MS |
2401 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
2402 | ||
e6ee8c0b KU |
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; | |
4e6e36c3 | 2409 | |
e6ee8c0b | 2410 | out: |
02233342 | 2411 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
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 | ||
042d2a9b AK |
2444 | /* |
2445 | * Returns old map, which caller must destroy. | |
2446 | */ | |
2447 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2448 | struct queue_limits *limits) | |
1da177e4 | 2449 | { |
042d2a9b | 2450 | struct dm_table *old_map; |
165125e1 | 2451 | struct request_queue *q = md->queue; |
1da177e4 LT |
2452 | sector_t size; |
2453 | ||
2454 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2455 | |
2456 | /* | |
2457 | * Wipe any geometry if the size of the table changed. | |
2458 | */ | |
fd2ed4d2 | 2459 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2460 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2461 | ||
32a926da | 2462 | __set_size(md, size); |
d5816876 | 2463 | |
2ca3310e AK |
2464 | dm_table_event_callback(t, event_callback, md); |
2465 | ||
e6ee8c0b KU |
2466 | /* |
2467 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2468 | * for request-based during suspension. So stop it to prevent | |
2469 | * I/O mapping before resume. | |
2470 | * This must be done before setting the queue restrictions, | |
2471 | * because request-based dm may be run just after the setting. | |
2472 | */ | |
bfebd1cd | 2473 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2474 | stop_queue(q); |
2475 | ||
2476 | __bind_mempools(md, t); | |
2477 | ||
a12f5d48 | 2478 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2479 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2480 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2481 | ||
754c5fc7 | 2482 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2483 | if (old_map) |
2484 | dm_sync_table(md); | |
1da177e4 | 2485 | |
042d2a9b | 2486 | return old_map; |
1da177e4 LT |
2487 | } |
2488 | ||
a7940155 AK |
2489 | /* |
2490 | * Returns unbound table for the caller to free. | |
2491 | */ | |
2492 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2493 | { |
a12f5d48 | 2494 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2495 | |
2496 | if (!map) | |
a7940155 | 2497 | return NULL; |
1da177e4 LT |
2498 | |
2499 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2500 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2501 | dm_sync_table(md); |
a7940155 AK |
2502 | |
2503 | return map; | |
1da177e4 LT |
2504 | } |
2505 | ||
2506 | /* | |
2507 | * Constructor for a new device. | |
2508 | */ | |
2b06cfff | 2509 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2510 | { |
2511 | struct mapped_device *md; | |
2512 | ||
2b06cfff | 2513 | md = alloc_dev(minor); |
1da177e4 LT |
2514 | if (!md) |
2515 | return -ENXIO; | |
2516 | ||
784aae73 MB |
2517 | dm_sysfs_init(md); |
2518 | ||
1da177e4 LT |
2519 | *result = md; |
2520 | return 0; | |
2521 | } | |
2522 | ||
a5664dad MS |
2523 | /* |
2524 | * Functions to manage md->type. | |
2525 | * All are required to hold md->type_lock. | |
2526 | */ | |
2527 | void dm_lock_md_type(struct mapped_device *md) | |
2528 | { | |
2529 | mutex_lock(&md->type_lock); | |
2530 | } | |
2531 | ||
2532 | void dm_unlock_md_type(struct mapped_device *md) | |
2533 | { | |
2534 | mutex_unlock(&md->type_lock); | |
2535 | } | |
2536 | ||
2537 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2538 | { | |
00c4fc3b | 2539 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2540 | md->type = type; |
2541 | } | |
2542 | ||
2543 | unsigned dm_get_md_type(struct mapped_device *md) | |
2544 | { | |
00c4fc3b | 2545 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2546 | return md->type; |
2547 | } | |
2548 | ||
36a0456f AK |
2549 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2550 | { | |
2551 | return md->immutable_target_type; | |
2552 | } | |
2553 | ||
f84cb8a4 MS |
2554 | /* |
2555 | * The queue_limits are only valid as long as you have a reference | |
2556 | * count on 'md'. | |
2557 | */ | |
2558 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2559 | { | |
2560 | BUG_ON(!atomic_read(&md->holders)); | |
2561 | return &md->queue->limits; | |
2562 | } | |
2563 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2564 | ||
bfebd1cd MS |
2565 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2566 | { | |
2567 | /* Initialize the request-based DM worker thread */ | |
2568 | init_kthread_worker(&md->kworker); | |
2569 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2570 | "kdmwork-%s", dm_device_name(md)); | |
2571 | } | |
2572 | ||
4a0b4ddf MS |
2573 | /* |
2574 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2575 | */ | |
2576 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2577 | { | |
2578 | struct request_queue *q = NULL; | |
2579 | ||
4a0b4ddf MS |
2580 | /* Fully initialize the queue */ |
2581 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2582 | if (!q) | |
bfebd1cd | 2583 | return -EINVAL; |
4a0b4ddf | 2584 | |
0ce65797 MS |
2585 | /* disable dm_request_fn's merge heuristic by default */ |
2586 | md->seq_rq_merge_deadline_usecs = 0; | |
2587 | ||
4a0b4ddf | 2588 | md->queue = q; |
bfebd1cd | 2589 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2590 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2591 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2592 | |
bfebd1cd | 2593 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2594 | |
4a0b4ddf MS |
2595 | elv_register_queue(md->queue); |
2596 | ||
bfebd1cd MS |
2597 | return 0; |
2598 | } | |
2599 | ||
2600 | static int dm_mq_init_request(void *data, struct request *rq, | |
2601 | unsigned int hctx_idx, unsigned int request_idx, | |
2602 | unsigned int numa_node) | |
2603 | { | |
2604 | struct mapped_device *md = data; | |
2605 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2606 | ||
2607 | /* | |
2608 | * Must initialize md member of tio, otherwise it won't | |
2609 | * be available in dm_mq_queue_rq. | |
2610 | */ | |
2611 | tio->md = md; | |
2612 | ||
2613 | return 0; | |
2614 | } | |
2615 | ||
2616 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2617 | const struct blk_mq_queue_data *bd) | |
2618 | { | |
2619 | struct request *rq = bd->rq; | |
2620 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2621 | struct mapped_device *md = tio->md; | |
2622 | int srcu_idx; | |
2623 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2624 | struct dm_target *ti; | |
2625 | sector_t pos; | |
2626 | ||
2627 | /* always use block 0 to find the target for flushes for now */ | |
2628 | pos = 0; | |
2629 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2630 | pos = blk_rq_pos(rq); | |
2631 | ||
2632 | ti = dm_table_find_target(map, pos); | |
2633 | if (!dm_target_is_valid(ti)) { | |
2634 | dm_put_live_table(md, srcu_idx); | |
2635 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2636 | /* | |
2637 | * Must perform setup, that rq_completed() requires, | |
2638 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2639 | */ | |
2640 | dm_start_request(md, rq); | |
2641 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2642 | } | |
2643 | dm_put_live_table(md, srcu_idx); | |
2644 | ||
2645 | if (ti->type->busy && ti->type->busy(ti)) | |
2646 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2647 | ||
2648 | dm_start_request(md, rq); | |
2649 | ||
2650 | /* Init tio using md established in .init_request */ | |
2651 | init_tio(tio, rq, md); | |
2652 | ||
02233342 MS |
2653 | /* |
2654 | * Establish tio->ti before queuing work (map_tio_request) | |
2655 | * or making direct call to map_request(). | |
2656 | */ | |
bfebd1cd | 2657 | tio->ti = ti; |
02233342 MS |
2658 | |
2659 | /* Clone the request if underlying devices aren't blk-mq */ | |
2660 | if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) { | |
2661 | /* clone request is allocated at the end of the pdu */ | |
2662 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
45714fbe | 2663 | (void) clone_rq(rq, md, tio, GFP_ATOMIC); |
02233342 MS |
2664 | queue_kthread_work(&md->kworker, &tio->work); |
2665 | } else { | |
2666 | /* Direct call is fine since .queue_rq allows allocations */ | |
45714fbe MS |
2667 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { |
2668 | /* Undo dm_start_request() before requeuing */ | |
e262f347 | 2669 | rq_end_stats(md, rq); |
45714fbe MS |
2670 | rq_completed(md, rq_data_dir(rq), false); |
2671 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2672 | } | |
02233342 | 2673 | } |
bfebd1cd MS |
2674 | |
2675 | return BLK_MQ_RQ_QUEUE_OK; | |
2676 | } | |
2677 | ||
2678 | static struct blk_mq_ops dm_mq_ops = { | |
2679 | .queue_rq = dm_mq_queue_rq, | |
2680 | .map_queue = blk_mq_map_queue, | |
2681 | .complete = dm_softirq_done, | |
2682 | .init_request = dm_mq_init_request, | |
2683 | }; | |
2684 | ||
2685 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2686 | { | |
02233342 | 2687 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2688 | struct request_queue *q; |
2689 | int err; | |
2690 | ||
2691 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2692 | md->tag_set.ops = &dm_mq_ops; | |
2693 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2694 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2695 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2696 | md->tag_set.nr_hw_queues = 1; | |
02233342 MS |
2697 | if (md_type == DM_TYPE_REQUEST_BASED) { |
2698 | /* make the memory for non-blk-mq clone part of the pdu */ | |
2699 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request); | |
2700 | } else | |
2701 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd MS |
2702 | md->tag_set.driver_data = md; |
2703 | ||
2704 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2705 | if (err) | |
2706 | return err; | |
2707 | ||
2708 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2709 | if (IS_ERR(q)) { | |
2710 | err = PTR_ERR(q); | |
2711 | goto out_tag_set; | |
2712 | } | |
2713 | md->queue = q; | |
2714 | dm_init_md_queue(md); | |
2715 | ||
2716 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2717 | blk_mq_register_disk(md->disk); | |
2718 | ||
02233342 MS |
2719 | if (md_type == DM_TYPE_REQUEST_BASED) |
2720 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2721 | |
2722 | return 0; | |
2723 | ||
2724 | out_tag_set: | |
2725 | blk_mq_free_tag_set(&md->tag_set); | |
2726 | return err; | |
4a0b4ddf MS |
2727 | } |
2728 | ||
4e6e36c3 MS |
2729 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2730 | { | |
2731 | if (type == DM_TYPE_BIO_BASED) | |
2732 | return type; | |
2733 | ||
2734 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2735 | } | |
2736 | ||
4a0b4ddf MS |
2737 | /* |
2738 | * Setup the DM device's queue based on md's type | |
2739 | */ | |
2740 | int dm_setup_md_queue(struct mapped_device *md) | |
2741 | { | |
bfebd1cd | 2742 | int r; |
17e149b8 | 2743 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2744 | |
2745 | switch (md_type) { | |
2746 | case DM_TYPE_REQUEST_BASED: | |
2747 | r = dm_init_request_based_queue(md); | |
2748 | if (r) { | |
ff36ab34 | 2749 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2750 | return r; |
ff36ab34 | 2751 | } |
bfebd1cd MS |
2752 | break; |
2753 | case DM_TYPE_MQ_REQUEST_BASED: | |
2754 | r = dm_init_request_based_blk_mq_queue(md); | |
2755 | if (r) { | |
2756 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2757 | return r; | |
2758 | } | |
2759 | break; | |
2760 | case DM_TYPE_BIO_BASED: | |
2761 | dm_init_old_md_queue(md); | |
ff36ab34 | 2762 | blk_queue_make_request(md->queue, dm_make_request); |
bfebd1cd | 2763 | break; |
4a0b4ddf MS |
2764 | } |
2765 | ||
2766 | return 0; | |
2767 | } | |
2768 | ||
2bec1f4a | 2769 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2770 | { |
2771 | struct mapped_device *md; | |
1da177e4 LT |
2772 | unsigned minor = MINOR(dev); |
2773 | ||
2774 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2775 | return NULL; | |
2776 | ||
f32c10b0 | 2777 | spin_lock(&_minor_lock); |
1da177e4 LT |
2778 | |
2779 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2780 | if (md) { |
2781 | if ((md == MINOR_ALLOCED || | |
2782 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2783 | dm_deleting_md(md) || | |
2784 | test_bit(DMF_FREEING, &md->flags))) { | |
2785 | md = NULL; | |
2786 | goto out; | |
2787 | } | |
2788 | dm_get(md); | |
fba9f90e | 2789 | } |
1da177e4 | 2790 | |
fba9f90e | 2791 | out: |
f32c10b0 | 2792 | spin_unlock(&_minor_lock); |
1da177e4 | 2793 | |
637842cf DT |
2794 | return md; |
2795 | } | |
3cf2e4ba | 2796 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2797 | |
9ade92a9 | 2798 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2799 | { |
9ade92a9 | 2800 | return md->interface_ptr; |
1da177e4 LT |
2801 | } |
2802 | ||
2803 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2804 | { | |
2805 | md->interface_ptr = ptr; | |
2806 | } | |
2807 | ||
2808 | void dm_get(struct mapped_device *md) | |
2809 | { | |
2810 | atomic_inc(&md->holders); | |
3f77316d | 2811 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2812 | } |
2813 | ||
09ee96b2 MP |
2814 | int dm_hold(struct mapped_device *md) |
2815 | { | |
2816 | spin_lock(&_minor_lock); | |
2817 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2818 | spin_unlock(&_minor_lock); | |
2819 | return -EBUSY; | |
2820 | } | |
2821 | dm_get(md); | |
2822 | spin_unlock(&_minor_lock); | |
2823 | return 0; | |
2824 | } | |
2825 | EXPORT_SYMBOL_GPL(dm_hold); | |
2826 | ||
72d94861 AK |
2827 | const char *dm_device_name(struct mapped_device *md) |
2828 | { | |
2829 | return md->name; | |
2830 | } | |
2831 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2832 | ||
3f77316d | 2833 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2834 | { |
1134e5ae | 2835 | struct dm_table *map; |
83d5e5b0 | 2836 | int srcu_idx; |
1da177e4 | 2837 | |
3f77316d | 2838 | might_sleep(); |
fba9f90e | 2839 | |
63a4f065 | 2840 | spin_lock(&_minor_lock); |
3f77316d KU |
2841 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2842 | set_bit(DMF_FREEING, &md->flags); | |
2843 | spin_unlock(&_minor_lock); | |
2844 | ||
02233342 | 2845 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2846 | flush_kthread_worker(&md->kworker); |
2847 | ||
ab7c7bb6 MP |
2848 | /* |
2849 | * Take suspend_lock so that presuspend and postsuspend methods | |
2850 | * do not race with internal suspend. | |
2851 | */ | |
2852 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2853 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2854 | if (!dm_suspended_md(md)) { |
2855 | dm_table_presuspend_targets(map); | |
2856 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2857 | } |
83d5e5b0 MP |
2858 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2859 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2860 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2861 | |
3f77316d KU |
2862 | /* |
2863 | * Rare, but there may be I/O requests still going to complete, | |
2864 | * for example. Wait for all references to disappear. | |
2865 | * No one should increment the reference count of the mapped_device, | |
2866 | * after the mapped_device state becomes DMF_FREEING. | |
2867 | */ | |
2868 | if (wait) | |
2869 | while (atomic_read(&md->holders)) | |
2870 | msleep(1); | |
2871 | else if (atomic_read(&md->holders)) | |
2872 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2873 | dm_device_name(md), atomic_read(&md->holders)); | |
2874 | ||
2875 | dm_sysfs_exit(md); | |
3f77316d KU |
2876 | dm_table_destroy(__unbind(md)); |
2877 | free_dev(md); | |
2878 | } | |
2879 | ||
2880 | void dm_destroy(struct mapped_device *md) | |
2881 | { | |
2882 | __dm_destroy(md, true); | |
2883 | } | |
2884 | ||
2885 | void dm_destroy_immediate(struct mapped_device *md) | |
2886 | { | |
2887 | __dm_destroy(md, false); | |
2888 | } | |
2889 | ||
2890 | void dm_put(struct mapped_device *md) | |
2891 | { | |
2892 | atomic_dec(&md->holders); | |
1da177e4 | 2893 | } |
79eb885c | 2894 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2895 | |
401600df | 2896 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2897 | { |
2898 | int r = 0; | |
b44ebeb0 MP |
2899 | DECLARE_WAITQUEUE(wait, current); |
2900 | ||
b44ebeb0 | 2901 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2902 | |
2903 | while (1) { | |
401600df | 2904 | set_current_state(interruptible); |
46125c1c | 2905 | |
b4324fee | 2906 | if (!md_in_flight(md)) |
46125c1c MB |
2907 | break; |
2908 | ||
401600df MP |
2909 | if (interruptible == TASK_INTERRUPTIBLE && |
2910 | signal_pending(current)) { | |
46125c1c MB |
2911 | r = -EINTR; |
2912 | break; | |
2913 | } | |
2914 | ||
2915 | io_schedule(); | |
2916 | } | |
2917 | set_current_state(TASK_RUNNING); | |
2918 | ||
b44ebeb0 MP |
2919 | remove_wait_queue(&md->wait, &wait); |
2920 | ||
46125c1c MB |
2921 | return r; |
2922 | } | |
2923 | ||
1da177e4 LT |
2924 | /* |
2925 | * Process the deferred bios | |
2926 | */ | |
ef208587 | 2927 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2928 | { |
ef208587 MP |
2929 | struct mapped_device *md = container_of(work, struct mapped_device, |
2930 | work); | |
6d6f10df | 2931 | struct bio *c; |
83d5e5b0 MP |
2932 | int srcu_idx; |
2933 | struct dm_table *map; | |
1da177e4 | 2934 | |
83d5e5b0 | 2935 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2936 | |
3b00b203 | 2937 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2938 | spin_lock_irq(&md->deferred_lock); |
2939 | c = bio_list_pop(&md->deferred); | |
2940 | spin_unlock_irq(&md->deferred_lock); | |
2941 | ||
6a8736d1 | 2942 | if (!c) |
df12ee99 | 2943 | break; |
022c2611 | 2944 | |
e6ee8c0b KU |
2945 | if (dm_request_based(md)) |
2946 | generic_make_request(c); | |
6a8736d1 | 2947 | else |
83d5e5b0 | 2948 | __split_and_process_bio(md, map, c); |
022c2611 | 2949 | } |
73d410c0 | 2950 | |
83d5e5b0 | 2951 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2952 | } |
2953 | ||
9a1fb464 | 2954 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2955 | { |
3b00b203 | 2956 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2957 | smp_mb__after_atomic(); |
53d5914f | 2958 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2959 | } |
2960 | ||
1da177e4 | 2961 | /* |
042d2a9b | 2962 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2963 | */ |
042d2a9b | 2964 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2965 | { |
87eb5b21 | 2966 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2967 | struct queue_limits limits; |
042d2a9b | 2968 | int r; |
1da177e4 | 2969 | |
e61290a4 | 2970 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2971 | |
2972 | /* device must be suspended */ | |
4f186f8b | 2973 | if (!dm_suspended_md(md)) |
93c534ae | 2974 | goto out; |
1da177e4 | 2975 | |
3ae70656 MS |
2976 | /* |
2977 | * If the new table has no data devices, retain the existing limits. | |
2978 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2979 | * then new I/O is queued based on these limits, and then some paths | |
2980 | * reappear. | |
2981 | */ | |
2982 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2983 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2984 | if (live_map) |
2985 | limits = md->queue->limits; | |
83d5e5b0 | 2986 | dm_put_live_table_fast(md); |
3ae70656 MS |
2987 | } |
2988 | ||
87eb5b21 MC |
2989 | if (!live_map) { |
2990 | r = dm_calculate_queue_limits(table, &limits); | |
2991 | if (r) { | |
2992 | map = ERR_PTR(r); | |
2993 | goto out; | |
2994 | } | |
042d2a9b | 2995 | } |
754c5fc7 | 2996 | |
042d2a9b | 2997 | map = __bind(md, table, &limits); |
1da177e4 | 2998 | |
93c534ae | 2999 | out: |
e61290a4 | 3000 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3001 | return map; |
1da177e4 LT |
3002 | } |
3003 | ||
3004 | /* | |
3005 | * Functions to lock and unlock any filesystem running on the | |
3006 | * device. | |
3007 | */ | |
2ca3310e | 3008 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3009 | { |
e39e2e95 | 3010 | int r; |
1da177e4 LT |
3011 | |
3012 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3013 | |
db8fef4f | 3014 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3015 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3016 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3017 | md->frozen_sb = NULL; |
3018 | return r; | |
dfbe03f6 AK |
3019 | } |
3020 | ||
aa8d7c2f AK |
3021 | set_bit(DMF_FROZEN, &md->flags); |
3022 | ||
1da177e4 LT |
3023 | return 0; |
3024 | } | |
3025 | ||
2ca3310e | 3026 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3027 | { |
aa8d7c2f AK |
3028 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3029 | return; | |
3030 | ||
db8fef4f | 3031 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3032 | md->frozen_sb = NULL; |
aa8d7c2f | 3033 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3034 | } |
3035 | ||
3036 | /* | |
ffcc3936 MS |
3037 | * If __dm_suspend returns 0, the device is completely quiescent |
3038 | * now. There is no request-processing activity. All new requests | |
3039 | * are being added to md->deferred list. | |
cec47e3d | 3040 | * |
ffcc3936 | 3041 | * Caller must hold md->suspend_lock |
cec47e3d | 3042 | */ |
ffcc3936 MS |
3043 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3044 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3045 | { |
ffcc3936 MS |
3046 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3047 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3048 | int r; | |
1da177e4 | 3049 | |
2e93ccc1 KU |
3050 | /* |
3051 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3052 | * This flag is cleared before dm_suspend returns. | |
3053 | */ | |
3054 | if (noflush) | |
3055 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3056 | ||
d67ee213 MS |
3057 | /* |
3058 | * This gets reverted if there's an error later and the targets | |
3059 | * provide the .presuspend_undo hook. | |
3060 | */ | |
cf222b37 AK |
3061 | dm_table_presuspend_targets(map); |
3062 | ||
32a926da | 3063 | /* |
9f518b27 KU |
3064 | * Flush I/O to the device. |
3065 | * Any I/O submitted after lock_fs() may not be flushed. | |
3066 | * noflush takes precedence over do_lockfs. | |
3067 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3068 | */ |
3069 | if (!noflush && do_lockfs) { | |
3070 | r = lock_fs(md); | |
d67ee213 MS |
3071 | if (r) { |
3072 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3073 | return r; |
d67ee213 | 3074 | } |
aa8d7c2f | 3075 | } |
1da177e4 LT |
3076 | |
3077 | /* | |
3b00b203 MP |
3078 | * Here we must make sure that no processes are submitting requests |
3079 | * to target drivers i.e. no one may be executing | |
3080 | * __split_and_process_bio. This is called from dm_request and | |
3081 | * dm_wq_work. | |
3082 | * | |
3083 | * To get all processes out of __split_and_process_bio in dm_request, | |
3084 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3085 | * __split_and_process_bio from dm_request and quiesce the thread |
3086 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3087 | * flush_workqueue(md->wq). | |
1da177e4 | 3088 | */ |
1eb787ec | 3089 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3090 | if (map) |
3091 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3092 | |
d0bcb878 | 3093 | /* |
29e4013d TH |
3094 | * Stop md->queue before flushing md->wq in case request-based |
3095 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3096 | */ |
2eb6e1e3 | 3097 | if (dm_request_based(md)) { |
9f518b27 | 3098 | stop_queue(md->queue); |
02233342 MS |
3099 | if (md->kworker_task) |
3100 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3101 | } |
cec47e3d | 3102 | |
d0bcb878 KU |
3103 | flush_workqueue(md->wq); |
3104 | ||
1da177e4 | 3105 | /* |
3b00b203 MP |
3106 | * At this point no more requests are entering target request routines. |
3107 | * We call dm_wait_for_completion to wait for all existing requests | |
3108 | * to finish. | |
1da177e4 | 3109 | */ |
ffcc3936 | 3110 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3111 | |
6d6f10df | 3112 | if (noflush) |
022c2611 | 3113 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3114 | if (map) |
3115 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3116 | |
1da177e4 | 3117 | /* were we interrupted ? */ |
46125c1c | 3118 | if (r < 0) { |
9a1fb464 | 3119 | dm_queue_flush(md); |
73d410c0 | 3120 | |
cec47e3d | 3121 | if (dm_request_based(md)) |
9f518b27 | 3122 | start_queue(md->queue); |
cec47e3d | 3123 | |
2ca3310e | 3124 | unlock_fs(md); |
d67ee213 | 3125 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3126 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3127 | } |
1da177e4 | 3128 | |
ffcc3936 MS |
3129 | return r; |
3130 | } | |
3131 | ||
3132 | /* | |
3133 | * We need to be able to change a mapping table under a mounted | |
3134 | * filesystem. For example we might want to move some data in | |
3135 | * the background. Before the table can be swapped with | |
3136 | * dm_bind_table, dm_suspend must be called to flush any in | |
3137 | * flight bios and ensure that any further io gets deferred. | |
3138 | */ | |
3139 | /* | |
3140 | * Suspend mechanism in request-based dm. | |
3141 | * | |
3142 | * 1. Flush all I/Os by lock_fs() if needed. | |
3143 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3144 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3145 | * | |
3146 | * To abort suspend, start the request_queue. | |
3147 | */ | |
3148 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3149 | { | |
3150 | struct dm_table *map = NULL; | |
3151 | int r = 0; | |
3152 | ||
3153 | retry: | |
3154 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3155 | ||
3156 | if (dm_suspended_md(md)) { | |
3157 | r = -EINVAL; | |
3158 | goto out_unlock; | |
3159 | } | |
3160 | ||
3161 | if (dm_suspended_internally_md(md)) { | |
3162 | /* already internally suspended, wait for internal resume */ | |
3163 | mutex_unlock(&md->suspend_lock); | |
3164 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3165 | if (r) | |
3166 | return r; | |
3167 | goto retry; | |
3168 | } | |
3169 | ||
a12f5d48 | 3170 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3171 | |
3172 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3173 | if (r) | |
3174 | goto out_unlock; | |
3b00b203 | 3175 | |
2ca3310e | 3176 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3177 | |
4d4471cb KU |
3178 | dm_table_postsuspend_targets(map); |
3179 | ||
d287483d | 3180 | out_unlock: |
e61290a4 | 3181 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3182 | return r; |
1da177e4 LT |
3183 | } |
3184 | ||
ffcc3936 MS |
3185 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3186 | { | |
3187 | if (map) { | |
3188 | int r = dm_table_resume_targets(map); | |
3189 | if (r) | |
3190 | return r; | |
3191 | } | |
3192 | ||
3193 | dm_queue_flush(md); | |
3194 | ||
3195 | /* | |
3196 | * Flushing deferred I/Os must be done after targets are resumed | |
3197 | * so that mapping of targets can work correctly. | |
3198 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3199 | */ | |
3200 | if (dm_request_based(md)) | |
3201 | start_queue(md->queue); | |
3202 | ||
3203 | unlock_fs(md); | |
3204 | ||
3205 | return 0; | |
3206 | } | |
3207 | ||
1da177e4 LT |
3208 | int dm_resume(struct mapped_device *md) |
3209 | { | |
cf222b37 | 3210 | int r = -EINVAL; |
cf222b37 | 3211 | struct dm_table *map = NULL; |
1da177e4 | 3212 | |
ffcc3936 MS |
3213 | retry: |
3214 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3215 | ||
4f186f8b | 3216 | if (!dm_suspended_md(md)) |
cf222b37 | 3217 | goto out; |
cf222b37 | 3218 | |
ffcc3936 MS |
3219 | if (dm_suspended_internally_md(md)) { |
3220 | /* already internally suspended, wait for internal resume */ | |
3221 | mutex_unlock(&md->suspend_lock); | |
3222 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3223 | if (r) | |
3224 | return r; | |
3225 | goto retry; | |
3226 | } | |
3227 | ||
a12f5d48 | 3228 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3229 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3230 | goto out; |
1da177e4 | 3231 | |
ffcc3936 | 3232 | r = __dm_resume(md, map); |
8757b776 MB |
3233 | if (r) |
3234 | goto out; | |
2ca3310e | 3235 | |
2ca3310e AK |
3236 | clear_bit(DMF_SUSPENDED, &md->flags); |
3237 | ||
cf222b37 AK |
3238 | r = 0; |
3239 | out: | |
e61290a4 | 3240 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3241 | |
cf222b37 | 3242 | return r; |
1da177e4 LT |
3243 | } |
3244 | ||
fd2ed4d2 MP |
3245 | /* |
3246 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3247 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3248 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3249 | */ |
3250 | ||
ffcc3936 | 3251 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3252 | { |
ffcc3936 MS |
3253 | struct dm_table *map = NULL; |
3254 | ||
96b26c8c | 3255 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3256 | return; /* nested internal suspend */ |
3257 | ||
3258 | if (dm_suspended_md(md)) { | |
3259 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3260 | return; /* nest suspend */ | |
3261 | } | |
3262 | ||
a12f5d48 | 3263 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3264 | |
3265 | /* | |
3266 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3267 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3268 | * would require changing .presuspend to return an error -- avoid this | |
3269 | * until there is a need for more elaborate variants of internal suspend. | |
3270 | */ | |
3271 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3272 | ||
3273 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3274 | ||
3275 | dm_table_postsuspend_targets(map); | |
3276 | } | |
3277 | ||
3278 | static void __dm_internal_resume(struct mapped_device *md) | |
3279 | { | |
96b26c8c MP |
3280 | BUG_ON(!md->internal_suspend_count); |
3281 | ||
3282 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3283 | return; /* resume from nested internal suspend */ |
3284 | ||
fd2ed4d2 | 3285 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3286 | goto done; /* resume from nested suspend */ |
3287 | ||
3288 | /* | |
3289 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3290 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3291 | */ | |
3292 | (void) __dm_resume(md, NULL); | |
3293 | ||
3294 | done: | |
3295 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3296 | smp_mb__after_atomic(); | |
3297 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3298 | } | |
3299 | ||
3300 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3301 | { | |
3302 | mutex_lock(&md->suspend_lock); | |
3303 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3304 | mutex_unlock(&md->suspend_lock); | |
3305 | } | |
3306 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3307 | ||
3308 | void dm_internal_resume(struct mapped_device *md) | |
3309 | { | |
3310 | mutex_lock(&md->suspend_lock); | |
3311 | __dm_internal_resume(md); | |
3312 | mutex_unlock(&md->suspend_lock); | |
3313 | } | |
3314 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3315 | ||
3316 | /* | |
3317 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3318 | * which prevents interaction with userspace-driven suspend. | |
3319 | */ | |
3320 | ||
3321 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3322 | { | |
3323 | mutex_lock(&md->suspend_lock); | |
3324 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3325 | return; |
3326 | ||
3327 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3328 | synchronize_srcu(&md->io_barrier); | |
3329 | flush_workqueue(md->wq); | |
3330 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3331 | } | |
b735fede | 3332 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3333 | |
ffcc3936 | 3334 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3335 | { |
ffcc3936 | 3336 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3337 | goto done; |
3338 | ||
3339 | dm_queue_flush(md); | |
3340 | ||
3341 | done: | |
3342 | mutex_unlock(&md->suspend_lock); | |
3343 | } | |
b735fede | 3344 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3345 | |
1da177e4 LT |
3346 | /*----------------------------------------------------------------- |
3347 | * Event notification. | |
3348 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3349 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3350 | unsigned cookie) |
69267a30 | 3351 | { |
60935eb2 MB |
3352 | char udev_cookie[DM_COOKIE_LENGTH]; |
3353 | char *envp[] = { udev_cookie, NULL }; | |
3354 | ||
3355 | if (!cookie) | |
3abf85b5 | 3356 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3357 | else { |
3358 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3359 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3360 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3361 | action, envp); | |
60935eb2 | 3362 | } |
69267a30 AK |
3363 | } |
3364 | ||
7a8c3d3b MA |
3365 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3366 | { | |
3367 | return atomic_add_return(1, &md->uevent_seq); | |
3368 | } | |
3369 | ||
1da177e4 LT |
3370 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3371 | { | |
3372 | return atomic_read(&md->event_nr); | |
3373 | } | |
3374 | ||
3375 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3376 | { | |
3377 | return wait_event_interruptible(md->eventq, | |
3378 | (event_nr != atomic_read(&md->event_nr))); | |
3379 | } | |
3380 | ||
7a8c3d3b MA |
3381 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3382 | { | |
3383 | unsigned long flags; | |
3384 | ||
3385 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3386 | list_add(elist, &md->uevent_list); | |
3387 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3388 | } | |
3389 | ||
1da177e4 LT |
3390 | /* |
3391 | * The gendisk is only valid as long as you have a reference | |
3392 | * count on 'md'. | |
3393 | */ | |
3394 | struct gendisk *dm_disk(struct mapped_device *md) | |
3395 | { | |
3396 | return md->disk; | |
3397 | } | |
65ff5b7d | 3398 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3399 | |
784aae73 MB |
3400 | struct kobject *dm_kobject(struct mapped_device *md) |
3401 | { | |
2995fa78 | 3402 | return &md->kobj_holder.kobj; |
784aae73 MB |
3403 | } |
3404 | ||
784aae73 MB |
3405 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3406 | { | |
3407 | struct mapped_device *md; | |
3408 | ||
2995fa78 | 3409 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3410 | |
4d89b7b4 | 3411 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3412 | dm_deleting_md(md)) |
4d89b7b4 MB |
3413 | return NULL; |
3414 | ||
784aae73 MB |
3415 | dm_get(md); |
3416 | return md; | |
3417 | } | |
3418 | ||
4f186f8b | 3419 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3420 | { |
3421 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3422 | } | |
3423 | ||
ffcc3936 MS |
3424 | int dm_suspended_internally_md(struct mapped_device *md) |
3425 | { | |
3426 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3427 | } | |
3428 | ||
2c140a24 MP |
3429 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3430 | { | |
3431 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3432 | } | |
3433 | ||
64dbce58 KU |
3434 | int dm_suspended(struct dm_target *ti) |
3435 | { | |
ecdb2e25 | 3436 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3437 | } |
3438 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3439 | ||
2e93ccc1 KU |
3440 | int dm_noflush_suspending(struct dm_target *ti) |
3441 | { | |
ecdb2e25 | 3442 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3443 | } |
3444 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3445 | ||
78d8e58a MS |
3446 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3447 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3448 | { |
78d8e58a MS |
3449 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3450 | struct kmem_cache *cachep = NULL; | |
3451 | unsigned int pool_size = 0; | |
5f015204 | 3452 | unsigned int front_pad; |
e6ee8c0b KU |
3453 | |
3454 | if (!pools) | |
4e6e36c3 | 3455 | return NULL; |
e6ee8c0b | 3456 | |
78d8e58a | 3457 | type = filter_md_type(type, md); |
17e149b8 | 3458 | |
78d8e58a MS |
3459 | switch (type) { |
3460 | case DM_TYPE_BIO_BASED: | |
3461 | cachep = _io_cache; | |
3462 | pool_size = dm_get_reserved_bio_based_ios(); | |
3463 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
3464 | break; | |
3465 | case DM_TYPE_REQUEST_BASED: | |
3466 | cachep = _rq_tio_cache; | |
3467 | pool_size = dm_get_reserved_rq_based_ios(); | |
3468 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3469 | if (!pools->rq_pool) | |
3470 | goto out; | |
3471 | /* fall through to setup remaining rq-based pools */ | |
3472 | case DM_TYPE_MQ_REQUEST_BASED: | |
3473 | if (!pool_size) | |
3474 | pool_size = dm_get_reserved_rq_based_ios(); | |
3475 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
3476 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3477 | WARN_ON(per_bio_data_size != 0); | |
3478 | break; | |
3479 | default: | |
3480 | BUG(); | |
3481 | } | |
3482 | ||
3483 | if (cachep) { | |
3484 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3485 | if (!pools->io_pool) | |
3486 | goto out; | |
3487 | } | |
e6ee8c0b | 3488 | |
3d8aab2d | 3489 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3490 | if (!pools->bs) |
5f015204 | 3491 | goto out; |
e6ee8c0b | 3492 | |
a91a2785 | 3493 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3494 | goto out; |
a91a2785 | 3495 | |
e6ee8c0b | 3496 | return pools; |
5f1b670d | 3497 | |
5f1b670d CH |
3498 | out: |
3499 | dm_free_md_mempools(pools); | |
78d8e58a | 3500 | |
4e6e36c3 | 3501 | return NULL; |
e6ee8c0b KU |
3502 | } |
3503 | ||
3504 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3505 | { | |
3506 | if (!pools) | |
3507 | return; | |
3508 | ||
3509 | if (pools->io_pool) | |
3510 | mempool_destroy(pools->io_pool); | |
3511 | ||
1ae49ea2 MS |
3512 | if (pools->rq_pool) |
3513 | mempool_destroy(pools->rq_pool); | |
3514 | ||
e6ee8c0b KU |
3515 | if (pools->bs) |
3516 | bioset_free(pools->bs); | |
3517 | ||
3518 | kfree(pools); | |
3519 | } | |
3520 | ||
83d5cde4 | 3521 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3522 | .open = dm_blk_open, |
3523 | .release = dm_blk_close, | |
aa129a22 | 3524 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3525 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3526 | .owner = THIS_MODULE |
3527 | }; | |
3528 | ||
1da177e4 LT |
3529 | /* |
3530 | * module hooks | |
3531 | */ | |
3532 | module_init(dm_init); | |
3533 | module_exit(dm_exit); | |
3534 | ||
3535 | module_param(major, uint, 0); | |
3536 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3537 | |
e8603136 MS |
3538 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3539 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3540 | ||
f4790826 MS |
3541 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3542 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3543 | ||
17e149b8 MS |
3544 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3545 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3546 | ||
1da177e4 LT |
3547 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3548 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3549 | MODULE_LICENSE("GPL"); |