Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
ffcc3936 | 22 | #include <linux/wait.h> |
2eb6e1e3 | 23 | #include <linux/kthread.h> |
0ce65797 | 24 | #include <linux/ktime.h> |
de3ec86d | 25 | #include <linux/elevator.h> /* for rq_end_sector() */ |
bfebd1cd | 26 | #include <linux/blk-mq.h> |
55782138 LZ |
27 | |
28 | #include <trace/events/block.h> | |
1da177e4 | 29 | |
72d94861 AK |
30 | #define DM_MSG_PREFIX "core" |
31 | ||
71a16736 NK |
32 | #ifdef CONFIG_PRINTK |
33 | /* | |
34 | * ratelimit state to be used in DMXXX_LIMIT(). | |
35 | */ | |
36 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
37 | DEFAULT_RATELIMIT_INTERVAL, | |
38 | DEFAULT_RATELIMIT_BURST); | |
39 | EXPORT_SYMBOL(dm_ratelimit_state); | |
40 | #endif | |
41 | ||
60935eb2 MB |
42 | /* |
43 | * Cookies are numeric values sent with CHANGE and REMOVE | |
44 | * uevents while resuming, removing or renaming the device. | |
45 | */ | |
46 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
47 | #define DM_COOKIE_LENGTH 24 | |
48 | ||
1da177e4 LT |
49 | static const char *_name = DM_NAME; |
50 | ||
51 | static unsigned int major = 0; | |
52 | static unsigned int _major = 0; | |
53 | ||
d15b774c AK |
54 | static DEFINE_IDR(_minor_idr); |
55 | ||
f32c10b0 | 56 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
57 | |
58 | static void do_deferred_remove(struct work_struct *w); | |
59 | ||
60 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
61 | ||
acfe0ad7 MP |
62 | static struct workqueue_struct *deferred_remove_workqueue; |
63 | ||
1da177e4 | 64 | /* |
8fbf26ad | 65 | * For bio-based dm. |
1da177e4 LT |
66 | * One of these is allocated per bio. |
67 | */ | |
68 | struct dm_io { | |
69 | struct mapped_device *md; | |
70 | int error; | |
1da177e4 | 71 | atomic_t io_count; |
6ae2fa67 | 72 | struct bio *bio; |
3eaf840e | 73 | unsigned long start_time; |
f88fb981 | 74 | spinlock_t endio_lock; |
fd2ed4d2 | 75 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
76 | }; |
77 | ||
8fbf26ad KU |
78 | /* |
79 | * For request-based dm. | |
80 | * One of these is allocated per request. | |
81 | */ | |
82 | struct dm_rq_target_io { | |
83 | struct mapped_device *md; | |
84 | struct dm_target *ti; | |
1ae49ea2 | 85 | struct request *orig, *clone; |
2eb6e1e3 | 86 | struct kthread_work work; |
8fbf26ad KU |
87 | int error; |
88 | union map_info info; | |
89 | }; | |
90 | ||
91 | /* | |
94818742 KO |
92 | * For request-based dm - the bio clones we allocate are embedded in these |
93 | * structs. | |
94 | * | |
95 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
96 | * the bioset is created - this means the bio has to come at the end of the | |
97 | * struct. | |
8fbf26ad KU |
98 | */ |
99 | struct dm_rq_clone_bio_info { | |
100 | struct bio *orig; | |
cec47e3d | 101 | struct dm_rq_target_io *tio; |
94818742 | 102 | struct bio clone; |
8fbf26ad KU |
103 | }; |
104 | ||
cec47e3d KU |
105 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
106 | { | |
107 | if (rq && rq->end_io_data) | |
108 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
109 | return NULL; | |
110 | } | |
111 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
112 | ||
ba61fdd1 JM |
113 | #define MINOR_ALLOCED ((void *)-1) |
114 | ||
1da177e4 LT |
115 | /* |
116 | * Bits for the md->flags field. | |
117 | */ | |
1eb787ec | 118 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 119 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 120 | #define DMF_FROZEN 2 |
fba9f90e | 121 | #define DMF_FREEING 3 |
5c6bd75d | 122 | #define DMF_DELETING 4 |
2e93ccc1 | 123 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 124 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 125 | #define DMF_DEFERRED_REMOVE 7 |
ffcc3936 | 126 | #define DMF_SUSPENDED_INTERNALLY 8 |
1da177e4 | 127 | |
83d5e5b0 MP |
128 | /* |
129 | * A dummy definition to make RCU happy. | |
130 | * struct dm_table should never be dereferenced in this file. | |
131 | */ | |
132 | struct dm_table { | |
133 | int undefined__; | |
134 | }; | |
135 | ||
304f3f6a MB |
136 | /* |
137 | * Work processed by per-device workqueue. | |
138 | */ | |
1da177e4 | 139 | struct mapped_device { |
83d5e5b0 | 140 | struct srcu_struct io_barrier; |
e61290a4 | 141 | struct mutex suspend_lock; |
1da177e4 | 142 | atomic_t holders; |
5c6bd75d | 143 | atomic_t open_count; |
1da177e4 | 144 | |
2a7faeb1 MP |
145 | /* |
146 | * The current mapping. | |
147 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
148 | * dereference. | |
149 | */ | |
6fa99520 | 150 | struct dm_table __rcu *map; |
2a7faeb1 | 151 | |
86f1152b BM |
152 | struct list_head table_devices; |
153 | struct mutex table_devices_lock; | |
154 | ||
1da177e4 LT |
155 | unsigned long flags; |
156 | ||
165125e1 | 157 | struct request_queue *queue; |
a5664dad | 158 | unsigned type; |
4a0b4ddf | 159 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
160 | struct mutex type_lock; |
161 | ||
36a0456f AK |
162 | struct target_type *immutable_target_type; |
163 | ||
1da177e4 | 164 | struct gendisk *disk; |
7e51f257 | 165 | char name[16]; |
1da177e4 LT |
166 | |
167 | void *interface_ptr; | |
168 | ||
169 | /* | |
170 | * A list of ios that arrived while we were suspended. | |
171 | */ | |
316d315b | 172 | atomic_t pending[2]; |
1da177e4 | 173 | wait_queue_head_t wait; |
53d5914f | 174 | struct work_struct work; |
74859364 | 175 | struct bio_list deferred; |
022c2611 | 176 | spinlock_t deferred_lock; |
1da177e4 | 177 | |
af7e466a | 178 | /* |
29e4013d | 179 | * Processing queue (flush) |
304f3f6a MB |
180 | */ |
181 | struct workqueue_struct *wq; | |
182 | ||
1da177e4 LT |
183 | /* |
184 | * io objects are allocated from here. | |
185 | */ | |
186 | mempool_t *io_pool; | |
1ae49ea2 | 187 | mempool_t *rq_pool; |
1da177e4 | 188 | |
9faf400f SB |
189 | struct bio_set *bs; |
190 | ||
1da177e4 LT |
191 | /* |
192 | * Event handling. | |
193 | */ | |
194 | atomic_t event_nr; | |
195 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
196 | atomic_t uevent_seq; |
197 | struct list_head uevent_list; | |
198 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
199 | |
200 | /* | |
201 | * freeze/thaw support require holding onto a super block | |
202 | */ | |
203 | struct super_block *frozen_sb; | |
db8fef4f | 204 | struct block_device *bdev; |
3ac51e74 DW |
205 | |
206 | /* forced geometry settings */ | |
207 | struct hd_geometry geometry; | |
784aae73 | 208 | |
2995fa78 MP |
209 | /* kobject and completion */ |
210 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 211 | |
d87f4c14 TH |
212 | /* zero-length flush that will be cloned and submitted to targets */ |
213 | struct bio flush_bio; | |
fd2ed4d2 | 214 | |
96b26c8c MP |
215 | /* the number of internal suspends */ |
216 | unsigned internal_suspend_count; | |
217 | ||
fd2ed4d2 | 218 | struct dm_stats stats; |
2eb6e1e3 KB |
219 | |
220 | struct kthread_worker kworker; | |
221 | struct task_struct *kworker_task; | |
de3ec86d MS |
222 | |
223 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 224 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 225 | int last_rq_rw; |
0ce65797 MS |
226 | sector_t last_rq_pos; |
227 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
228 | |
229 | /* for blk-mq request-based DM support */ | |
230 | struct blk_mq_tag_set tag_set; | |
17e149b8 | 231 | bool use_blk_mq; |
1da177e4 LT |
232 | }; |
233 | ||
17e149b8 MS |
234 | #ifdef CONFIG_DM_MQ_DEFAULT |
235 | static bool use_blk_mq = true; | |
236 | #else | |
237 | static bool use_blk_mq = false; | |
238 | #endif | |
239 | ||
240 | bool dm_use_blk_mq(struct mapped_device *md) | |
241 | { | |
242 | return md->use_blk_mq; | |
243 | } | |
244 | ||
e6ee8c0b KU |
245 | /* |
246 | * For mempools pre-allocation at the table loading time. | |
247 | */ | |
248 | struct dm_md_mempools { | |
249 | mempool_t *io_pool; | |
1ae49ea2 | 250 | mempool_t *rq_pool; |
e6ee8c0b KU |
251 | struct bio_set *bs; |
252 | }; | |
253 | ||
86f1152b BM |
254 | struct table_device { |
255 | struct list_head list; | |
256 | atomic_t count; | |
257 | struct dm_dev dm_dev; | |
258 | }; | |
259 | ||
6cfa5857 MS |
260 | #define RESERVED_BIO_BASED_IOS 16 |
261 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 262 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 263 | static struct kmem_cache *_io_cache; |
8fbf26ad | 264 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 265 | static struct kmem_cache *_rq_cache; |
94818742 | 266 | |
e8603136 MS |
267 | /* |
268 | * Bio-based DM's mempools' reserved IOs set by the user. | |
269 | */ | |
270 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
271 | ||
f4790826 MS |
272 | /* |
273 | * Request-based DM's mempools' reserved IOs set by the user. | |
274 | */ | |
275 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
276 | ||
09c2d531 | 277 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
278 | unsigned def, unsigned max) |
279 | { | |
09c2d531 MS |
280 | unsigned param = ACCESS_ONCE(*module_param); |
281 | unsigned modified_param = 0; | |
f4790826 | 282 | |
09c2d531 MS |
283 | if (!param) |
284 | modified_param = def; | |
285 | else if (param > max) | |
286 | modified_param = max; | |
f4790826 | 287 | |
09c2d531 MS |
288 | if (modified_param) { |
289 | (void)cmpxchg(module_param, param, modified_param); | |
290 | param = modified_param; | |
f4790826 MS |
291 | } |
292 | ||
09c2d531 | 293 | return param; |
f4790826 MS |
294 | } |
295 | ||
e8603136 MS |
296 | unsigned dm_get_reserved_bio_based_ios(void) |
297 | { | |
09c2d531 | 298 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
299 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
300 | } | |
301 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
302 | ||
f4790826 MS |
303 | unsigned dm_get_reserved_rq_based_ios(void) |
304 | { | |
09c2d531 | 305 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
306 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
307 | } | |
308 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
309 | ||
1da177e4 LT |
310 | static int __init local_init(void) |
311 | { | |
51157b4a | 312 | int r = -ENOMEM; |
1da177e4 | 313 | |
1da177e4 | 314 | /* allocate a slab for the dm_ios */ |
028867ac | 315 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 316 | if (!_io_cache) |
51157b4a | 317 | return r; |
1da177e4 | 318 | |
8fbf26ad KU |
319 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
320 | if (!_rq_tio_cache) | |
dba14160 | 321 | goto out_free_io_cache; |
8fbf26ad | 322 | |
1ae49ea2 MS |
323 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
324 | __alignof__(struct request), 0, NULL); | |
325 | if (!_rq_cache) | |
326 | goto out_free_rq_tio_cache; | |
327 | ||
51e5b2bd | 328 | r = dm_uevent_init(); |
51157b4a | 329 | if (r) |
1ae49ea2 | 330 | goto out_free_rq_cache; |
51e5b2bd | 331 | |
acfe0ad7 MP |
332 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
333 | if (!deferred_remove_workqueue) { | |
334 | r = -ENOMEM; | |
335 | goto out_uevent_exit; | |
336 | } | |
337 | ||
1da177e4 LT |
338 | _major = major; |
339 | r = register_blkdev(_major, _name); | |
51157b4a | 340 | if (r < 0) |
acfe0ad7 | 341 | goto out_free_workqueue; |
1da177e4 LT |
342 | |
343 | if (!_major) | |
344 | _major = r; | |
345 | ||
346 | return 0; | |
51157b4a | 347 | |
acfe0ad7 MP |
348 | out_free_workqueue: |
349 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
350 | out_uevent_exit: |
351 | dm_uevent_exit(); | |
1ae49ea2 MS |
352 | out_free_rq_cache: |
353 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
354 | out_free_rq_tio_cache: |
355 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
356 | out_free_io_cache: |
357 | kmem_cache_destroy(_io_cache); | |
358 | ||
359 | return r; | |
1da177e4 LT |
360 | } |
361 | ||
362 | static void local_exit(void) | |
363 | { | |
2c140a24 | 364 | flush_scheduled_work(); |
acfe0ad7 | 365 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 366 | |
1ae49ea2 | 367 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 368 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 369 | kmem_cache_destroy(_io_cache); |
00d59405 | 370 | unregister_blkdev(_major, _name); |
51e5b2bd | 371 | dm_uevent_exit(); |
1da177e4 LT |
372 | |
373 | _major = 0; | |
374 | ||
375 | DMINFO("cleaned up"); | |
376 | } | |
377 | ||
b9249e55 | 378 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
379 | local_init, |
380 | dm_target_init, | |
381 | dm_linear_init, | |
382 | dm_stripe_init, | |
952b3557 | 383 | dm_io_init, |
945fa4d2 | 384 | dm_kcopyd_init, |
1da177e4 | 385 | dm_interface_init, |
fd2ed4d2 | 386 | dm_statistics_init, |
1da177e4 LT |
387 | }; |
388 | ||
b9249e55 | 389 | static void (*_exits[])(void) = { |
1da177e4 LT |
390 | local_exit, |
391 | dm_target_exit, | |
392 | dm_linear_exit, | |
393 | dm_stripe_exit, | |
952b3557 | 394 | dm_io_exit, |
945fa4d2 | 395 | dm_kcopyd_exit, |
1da177e4 | 396 | dm_interface_exit, |
fd2ed4d2 | 397 | dm_statistics_exit, |
1da177e4 LT |
398 | }; |
399 | ||
400 | static int __init dm_init(void) | |
401 | { | |
402 | const int count = ARRAY_SIZE(_inits); | |
403 | ||
404 | int r, i; | |
405 | ||
406 | for (i = 0; i < count; i++) { | |
407 | r = _inits[i](); | |
408 | if (r) | |
409 | goto bad; | |
410 | } | |
411 | ||
412 | return 0; | |
413 | ||
414 | bad: | |
415 | while (i--) | |
416 | _exits[i](); | |
417 | ||
418 | return r; | |
419 | } | |
420 | ||
421 | static void __exit dm_exit(void) | |
422 | { | |
423 | int i = ARRAY_SIZE(_exits); | |
424 | ||
425 | while (i--) | |
426 | _exits[i](); | |
d15b774c AK |
427 | |
428 | /* | |
429 | * Should be empty by this point. | |
430 | */ | |
d15b774c | 431 | idr_destroy(&_minor_idr); |
1da177e4 LT |
432 | } |
433 | ||
434 | /* | |
435 | * Block device functions | |
436 | */ | |
432a212c MA |
437 | int dm_deleting_md(struct mapped_device *md) |
438 | { | |
439 | return test_bit(DMF_DELETING, &md->flags); | |
440 | } | |
441 | ||
fe5f9f2c | 442 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
443 | { |
444 | struct mapped_device *md; | |
445 | ||
fba9f90e JM |
446 | spin_lock(&_minor_lock); |
447 | ||
fe5f9f2c | 448 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
449 | if (!md) |
450 | goto out; | |
451 | ||
5c6bd75d | 452 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 453 | dm_deleting_md(md)) { |
fba9f90e JM |
454 | md = NULL; |
455 | goto out; | |
456 | } | |
457 | ||
1da177e4 | 458 | dm_get(md); |
5c6bd75d | 459 | atomic_inc(&md->open_count); |
fba9f90e JM |
460 | out: |
461 | spin_unlock(&_minor_lock); | |
462 | ||
463 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
464 | } |
465 | ||
db2a144b | 466 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 467 | { |
63a4f065 | 468 | struct mapped_device *md; |
6e9624b8 | 469 | |
4a1aeb98 MB |
470 | spin_lock(&_minor_lock); |
471 | ||
63a4f065 MS |
472 | md = disk->private_data; |
473 | if (WARN_ON(!md)) | |
474 | goto out; | |
475 | ||
2c140a24 MP |
476 | if (atomic_dec_and_test(&md->open_count) && |
477 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 478 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 479 | |
1da177e4 | 480 | dm_put(md); |
63a4f065 | 481 | out: |
4a1aeb98 | 482 | spin_unlock(&_minor_lock); |
1da177e4 LT |
483 | } |
484 | ||
5c6bd75d AK |
485 | int dm_open_count(struct mapped_device *md) |
486 | { | |
487 | return atomic_read(&md->open_count); | |
488 | } | |
489 | ||
490 | /* | |
491 | * Guarantees nothing is using the device before it's deleted. | |
492 | */ | |
2c140a24 | 493 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
494 | { |
495 | int r = 0; | |
496 | ||
497 | spin_lock(&_minor_lock); | |
498 | ||
2c140a24 | 499 | if (dm_open_count(md)) { |
5c6bd75d | 500 | r = -EBUSY; |
2c140a24 MP |
501 | if (mark_deferred) |
502 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
503 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
504 | r = -EEXIST; | |
5c6bd75d AK |
505 | else |
506 | set_bit(DMF_DELETING, &md->flags); | |
507 | ||
508 | spin_unlock(&_minor_lock); | |
509 | ||
510 | return r; | |
511 | } | |
512 | ||
2c140a24 MP |
513 | int dm_cancel_deferred_remove(struct mapped_device *md) |
514 | { | |
515 | int r = 0; | |
516 | ||
517 | spin_lock(&_minor_lock); | |
518 | ||
519 | if (test_bit(DMF_DELETING, &md->flags)) | |
520 | r = -EBUSY; | |
521 | else | |
522 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
523 | ||
524 | spin_unlock(&_minor_lock); | |
525 | ||
526 | return r; | |
527 | } | |
528 | ||
529 | static void do_deferred_remove(struct work_struct *w) | |
530 | { | |
531 | dm_deferred_remove(); | |
532 | } | |
533 | ||
fd2ed4d2 MP |
534 | sector_t dm_get_size(struct mapped_device *md) |
535 | { | |
536 | return get_capacity(md->disk); | |
537 | } | |
538 | ||
9974fa2c MS |
539 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
540 | { | |
541 | return md->queue; | |
542 | } | |
543 | ||
fd2ed4d2 MP |
544 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
545 | { | |
546 | return &md->stats; | |
547 | } | |
548 | ||
3ac51e74 DW |
549 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
550 | { | |
551 | struct mapped_device *md = bdev->bd_disk->private_data; | |
552 | ||
553 | return dm_get_geometry(md, geo); | |
554 | } | |
555 | ||
fe5f9f2c | 556 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
557 | unsigned int cmd, unsigned long arg) |
558 | { | |
fe5f9f2c | 559 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 560 | int srcu_idx; |
6c182cd8 | 561 | struct dm_table *map; |
aa129a22 MB |
562 | struct dm_target *tgt; |
563 | int r = -ENOTTY; | |
564 | ||
6c182cd8 | 565 | retry: |
83d5e5b0 MP |
566 | map = dm_get_live_table(md, &srcu_idx); |
567 | ||
aa129a22 MB |
568 | if (!map || !dm_table_get_size(map)) |
569 | goto out; | |
570 | ||
571 | /* We only support devices that have a single target */ | |
572 | if (dm_table_get_num_targets(map) != 1) | |
573 | goto out; | |
574 | ||
575 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
576 | if (!tgt->type->ioctl) |
577 | goto out; | |
aa129a22 | 578 | |
4f186f8b | 579 | if (dm_suspended_md(md)) { |
aa129a22 MB |
580 | r = -EAGAIN; |
581 | goto out; | |
582 | } | |
583 | ||
4d341d82 | 584 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
585 | |
586 | out: | |
83d5e5b0 | 587 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 588 | |
6c182cd8 HR |
589 | if (r == -ENOTCONN) { |
590 | msleep(10); | |
591 | goto retry; | |
592 | } | |
593 | ||
aa129a22 MB |
594 | return r; |
595 | } | |
596 | ||
028867ac | 597 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
598 | { |
599 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
600 | } | |
601 | ||
028867ac | 602 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
603 | { |
604 | mempool_free(io, md->io_pool); | |
605 | } | |
606 | ||
028867ac | 607 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 608 | { |
dba14160 | 609 | bio_put(&tio->clone); |
1da177e4 LT |
610 | } |
611 | ||
08885643 KU |
612 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
613 | gfp_t gfp_mask) | |
cec47e3d | 614 | { |
5f015204 | 615 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
616 | } |
617 | ||
618 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
619 | { | |
5f015204 | 620 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
621 | } |
622 | ||
1ae49ea2 MS |
623 | static struct request *alloc_clone_request(struct mapped_device *md, |
624 | gfp_t gfp_mask) | |
625 | { | |
626 | return mempool_alloc(md->rq_pool, gfp_mask); | |
627 | } | |
628 | ||
629 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
630 | { | |
631 | mempool_free(rq, md->rq_pool); | |
632 | } | |
633 | ||
90abb8c4 KU |
634 | static int md_in_flight(struct mapped_device *md) |
635 | { | |
636 | return atomic_read(&md->pending[READ]) + | |
637 | atomic_read(&md->pending[WRITE]); | |
638 | } | |
639 | ||
3eaf840e JNN |
640 | static void start_io_acct(struct dm_io *io) |
641 | { | |
642 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 643 | struct bio *bio = io->bio; |
c9959059 | 644 | int cpu; |
fd2ed4d2 | 645 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
646 | |
647 | io->start_time = jiffies; | |
648 | ||
074a7aca TH |
649 | cpu = part_stat_lock(); |
650 | part_round_stats(cpu, &dm_disk(md)->part0); | |
651 | part_stat_unlock(); | |
1e9bb880 SL |
652 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
653 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
654 | |
655 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 656 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 657 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
658 | } |
659 | ||
d221d2e7 | 660 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
661 | { |
662 | struct mapped_device *md = io->md; | |
663 | struct bio *bio = io->bio; | |
664 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 665 | int pending; |
3eaf840e JNN |
666 | int rw = bio_data_dir(bio); |
667 | ||
18c0b223 | 668 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 669 | |
fd2ed4d2 | 670 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 671 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
672 | bio_sectors(bio), true, duration, &io->stats_aux); |
673 | ||
af7e466a MP |
674 | /* |
675 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 676 | * a flush. |
af7e466a | 677 | */ |
1e9bb880 SL |
678 | pending = atomic_dec_return(&md->pending[rw]); |
679 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 680 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 681 | |
d221d2e7 MP |
682 | /* nudge anyone waiting on suspend queue */ |
683 | if (!pending) | |
684 | wake_up(&md->wait); | |
3eaf840e JNN |
685 | } |
686 | ||
1da177e4 LT |
687 | /* |
688 | * Add the bio to the list of deferred io. | |
689 | */ | |
92c63902 | 690 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 691 | { |
05447420 | 692 | unsigned long flags; |
1da177e4 | 693 | |
05447420 | 694 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 695 | bio_list_add(&md->deferred, bio); |
05447420 | 696 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 697 | queue_work(md->wq, &md->work); |
1da177e4 LT |
698 | } |
699 | ||
700 | /* | |
701 | * Everyone (including functions in this file), should use this | |
702 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 703 | * dm_put_live_table() when finished. |
1da177e4 | 704 | */ |
83d5e5b0 | 705 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 706 | { |
83d5e5b0 MP |
707 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
708 | ||
709 | return srcu_dereference(md->map, &md->io_barrier); | |
710 | } | |
1da177e4 | 711 | |
83d5e5b0 MP |
712 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
713 | { | |
714 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
715 | } | |
716 | ||
717 | void dm_sync_table(struct mapped_device *md) | |
718 | { | |
719 | synchronize_srcu(&md->io_barrier); | |
720 | synchronize_rcu_expedited(); | |
721 | } | |
722 | ||
723 | /* | |
724 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
725 | * The caller must not block between these two functions. | |
726 | */ | |
727 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
728 | { | |
729 | rcu_read_lock(); | |
730 | return rcu_dereference(md->map); | |
731 | } | |
1da177e4 | 732 | |
83d5e5b0 MP |
733 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
734 | { | |
735 | rcu_read_unlock(); | |
1da177e4 LT |
736 | } |
737 | ||
86f1152b BM |
738 | /* |
739 | * Open a table device so we can use it as a map destination. | |
740 | */ | |
741 | static int open_table_device(struct table_device *td, dev_t dev, | |
742 | struct mapped_device *md) | |
743 | { | |
744 | static char *_claim_ptr = "I belong to device-mapper"; | |
745 | struct block_device *bdev; | |
746 | ||
747 | int r; | |
748 | ||
749 | BUG_ON(td->dm_dev.bdev); | |
750 | ||
751 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
752 | if (IS_ERR(bdev)) | |
753 | return PTR_ERR(bdev); | |
754 | ||
755 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
756 | if (r) { | |
757 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
758 | return r; | |
759 | } | |
760 | ||
761 | td->dm_dev.bdev = bdev; | |
762 | return 0; | |
763 | } | |
764 | ||
765 | /* | |
766 | * Close a table device that we've been using. | |
767 | */ | |
768 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
769 | { | |
770 | if (!td->dm_dev.bdev) | |
771 | return; | |
772 | ||
773 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
774 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
775 | td->dm_dev.bdev = NULL; | |
776 | } | |
777 | ||
778 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
779 | fmode_t mode) { | |
780 | struct table_device *td; | |
781 | ||
782 | list_for_each_entry(td, l, list) | |
783 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
784 | return td; | |
785 | ||
786 | return NULL; | |
787 | } | |
788 | ||
789 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
790 | struct dm_dev **result) { | |
791 | int r; | |
792 | struct table_device *td; | |
793 | ||
794 | mutex_lock(&md->table_devices_lock); | |
795 | td = find_table_device(&md->table_devices, dev, mode); | |
796 | if (!td) { | |
797 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
798 | if (!td) { | |
799 | mutex_unlock(&md->table_devices_lock); | |
800 | return -ENOMEM; | |
801 | } | |
802 | ||
803 | td->dm_dev.mode = mode; | |
804 | td->dm_dev.bdev = NULL; | |
805 | ||
806 | if ((r = open_table_device(td, dev, md))) { | |
807 | mutex_unlock(&md->table_devices_lock); | |
808 | kfree(td); | |
809 | return r; | |
810 | } | |
811 | ||
812 | format_dev_t(td->dm_dev.name, dev); | |
813 | ||
814 | atomic_set(&td->count, 0); | |
815 | list_add(&td->list, &md->table_devices); | |
816 | } | |
817 | atomic_inc(&td->count); | |
818 | mutex_unlock(&md->table_devices_lock); | |
819 | ||
820 | *result = &td->dm_dev; | |
821 | return 0; | |
822 | } | |
823 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
824 | ||
825 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
826 | { | |
827 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
828 | ||
829 | mutex_lock(&md->table_devices_lock); | |
830 | if (atomic_dec_and_test(&td->count)) { | |
831 | close_table_device(td, md); | |
832 | list_del(&td->list); | |
833 | kfree(td); | |
834 | } | |
835 | mutex_unlock(&md->table_devices_lock); | |
836 | } | |
837 | EXPORT_SYMBOL(dm_put_table_device); | |
838 | ||
839 | static void free_table_devices(struct list_head *devices) | |
840 | { | |
841 | struct list_head *tmp, *next; | |
842 | ||
843 | list_for_each_safe(tmp, next, devices) { | |
844 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
845 | ||
846 | DMWARN("dm_destroy: %s still exists with %d references", | |
847 | td->dm_dev.name, atomic_read(&td->count)); | |
848 | kfree(td); | |
849 | } | |
850 | } | |
851 | ||
3ac51e74 DW |
852 | /* |
853 | * Get the geometry associated with a dm device | |
854 | */ | |
855 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
856 | { | |
857 | *geo = md->geometry; | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
862 | /* | |
863 | * Set the geometry of a device. | |
864 | */ | |
865 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
866 | { | |
867 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
868 | ||
869 | if (geo->start > sz) { | |
870 | DMWARN("Start sector is beyond the geometry limits."); | |
871 | return -EINVAL; | |
872 | } | |
873 | ||
874 | md->geometry = *geo; | |
875 | ||
876 | return 0; | |
877 | } | |
878 | ||
1da177e4 LT |
879 | /*----------------------------------------------------------------- |
880 | * CRUD START: | |
881 | * A more elegant soln is in the works that uses the queue | |
882 | * merge fn, unfortunately there are a couple of changes to | |
883 | * the block layer that I want to make for this. So in the | |
884 | * interests of getting something for people to use I give | |
885 | * you this clearly demarcated crap. | |
886 | *---------------------------------------------------------------*/ | |
887 | ||
2e93ccc1 KU |
888 | static int __noflush_suspending(struct mapped_device *md) |
889 | { | |
890 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
891 | } | |
892 | ||
1da177e4 LT |
893 | /* |
894 | * Decrements the number of outstanding ios that a bio has been | |
895 | * cloned into, completing the original io if necc. | |
896 | */ | |
858119e1 | 897 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 898 | { |
2e93ccc1 | 899 | unsigned long flags; |
b35f8caa MB |
900 | int io_error; |
901 | struct bio *bio; | |
902 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
903 | |
904 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
905 | if (unlikely(error)) { |
906 | spin_lock_irqsave(&io->endio_lock, flags); | |
907 | if (!(io->error > 0 && __noflush_suspending(md))) | |
908 | io->error = error; | |
909 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
910 | } | |
1da177e4 LT |
911 | |
912 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
913 | if (io->error == DM_ENDIO_REQUEUE) { |
914 | /* | |
915 | * Target requested pushing back the I/O. | |
2e93ccc1 | 916 | */ |
022c2611 | 917 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
918 | if (__noflush_suspending(md)) |
919 | bio_list_add_head(&md->deferred, io->bio); | |
920 | else | |
2e93ccc1 KU |
921 | /* noflush suspend was interrupted. */ |
922 | io->error = -EIO; | |
022c2611 | 923 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
924 | } |
925 | ||
b35f8caa MB |
926 | io_error = io->error; |
927 | bio = io->bio; | |
6a8736d1 TH |
928 | end_io_acct(io); |
929 | free_io(md, io); | |
930 | ||
931 | if (io_error == DM_ENDIO_REQUEUE) | |
932 | return; | |
2e93ccc1 | 933 | |
4f024f37 | 934 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 935 | /* |
6a8736d1 TH |
936 | * Preflush done for flush with data, reissue |
937 | * without REQ_FLUSH. | |
af7e466a | 938 | */ |
6a8736d1 TH |
939 | bio->bi_rw &= ~REQ_FLUSH; |
940 | queue_io(md, bio); | |
af7e466a | 941 | } else { |
b372d360 | 942 | /* done with normal IO or empty flush */ |
0a82a8d1 | 943 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 944 | bio_endio(bio, io_error); |
b35f8caa | 945 | } |
1da177e4 LT |
946 | } |
947 | } | |
948 | ||
7eee4ae2 MS |
949 | static void disable_write_same(struct mapped_device *md) |
950 | { | |
951 | struct queue_limits *limits = dm_get_queue_limits(md); | |
952 | ||
953 | /* device doesn't really support WRITE SAME, disable it */ | |
954 | limits->max_write_same_sectors = 0; | |
955 | } | |
956 | ||
6712ecf8 | 957 | static void clone_endio(struct bio *bio, int error) |
1da177e4 | 958 | { |
5164bece | 959 | int r = error; |
bfc6d41c | 960 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 961 | struct dm_io *io = tio->io; |
9faf400f | 962 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
963 | dm_endio_fn endio = tio->ti->type->end_io; |
964 | ||
1da177e4 LT |
965 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
966 | error = -EIO; | |
967 | ||
968 | if (endio) { | |
7de3ee57 | 969 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
970 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
971 | /* | |
972 | * error and requeue request are handled | |
973 | * in dec_pending(). | |
974 | */ | |
1da177e4 | 975 | error = r; |
45cbcd79 KU |
976 | else if (r == DM_ENDIO_INCOMPLETE) |
977 | /* The target will handle the io */ | |
6712ecf8 | 978 | return; |
45cbcd79 KU |
979 | else if (r) { |
980 | DMWARN("unimplemented target endio return value: %d", r); | |
981 | BUG(); | |
982 | } | |
1da177e4 LT |
983 | } |
984 | ||
7eee4ae2 MS |
985 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
986 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
987 | disable_write_same(md); | |
988 | ||
9faf400f | 989 | free_tio(md, tio); |
b35f8caa | 990 | dec_pending(io, error); |
1da177e4 LT |
991 | } |
992 | ||
cec47e3d KU |
993 | /* |
994 | * Partial completion handling for request-based dm | |
995 | */ | |
996 | static void end_clone_bio(struct bio *clone, int error) | |
997 | { | |
bfc6d41c MP |
998 | struct dm_rq_clone_bio_info *info = |
999 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1000 | struct dm_rq_target_io *tio = info->tio; |
1001 | struct bio *bio = info->orig; | |
4f024f37 | 1002 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
1003 | |
1004 | bio_put(clone); | |
1005 | ||
1006 | if (tio->error) | |
1007 | /* | |
1008 | * An error has already been detected on the request. | |
1009 | * Once error occurred, just let clone->end_io() handle | |
1010 | * the remainder. | |
1011 | */ | |
1012 | return; | |
1013 | else if (error) { | |
1014 | /* | |
1015 | * Don't notice the error to the upper layer yet. | |
1016 | * The error handling decision is made by the target driver, | |
1017 | * when the request is completed. | |
1018 | */ | |
1019 | tio->error = error; | |
1020 | return; | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * I/O for the bio successfully completed. | |
1025 | * Notice the data completion to the upper layer. | |
1026 | */ | |
1027 | ||
1028 | /* | |
1029 | * bios are processed from the head of the list. | |
1030 | * So the completing bio should always be rq->bio. | |
1031 | * If it's not, something wrong is happening. | |
1032 | */ | |
1033 | if (tio->orig->bio != bio) | |
1034 | DMERR("bio completion is going in the middle of the request"); | |
1035 | ||
1036 | /* | |
1037 | * Update the original request. | |
1038 | * Do not use blk_end_request() here, because it may complete | |
1039 | * the original request before the clone, and break the ordering. | |
1040 | */ | |
1041 | blk_update_request(tio->orig, 0, nr_bytes); | |
1042 | } | |
1043 | ||
bfebd1cd MS |
1044 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1045 | { | |
1046 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1047 | } | |
1048 | ||
cec47e3d KU |
1049 | /* |
1050 | * Don't touch any member of the md after calling this function because | |
1051 | * the md may be freed in dm_put() at the end of this function. | |
1052 | * Or do dm_get() before calling this function and dm_put() later. | |
1053 | */ | |
466d89a6 | 1054 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1055 | { |
9a0e609e MS |
1056 | int nr_requests_pending; |
1057 | ||
b4324fee | 1058 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1059 | |
1060 | /* nudge anyone waiting on suspend queue */ | |
9a0e609e MS |
1061 | nr_requests_pending = md_in_flight(md); |
1062 | if (!nr_requests_pending) | |
cec47e3d KU |
1063 | wake_up(&md->wait); |
1064 | ||
a8c32a5c JA |
1065 | /* |
1066 | * Run this off this callpath, as drivers could invoke end_io while | |
1067 | * inside their request_fn (and holding the queue lock). Calling | |
1068 | * back into ->request_fn() could deadlock attempting to grab the | |
1069 | * queue lock again. | |
1070 | */ | |
9a0e609e | 1071 | if (run_queue) { |
bfebd1cd MS |
1072 | if (md->queue->mq_ops) |
1073 | blk_mq_run_hw_queues(md->queue, true); | |
1074 | else if (!nr_requests_pending || | |
1075 | (nr_requests_pending >= md->queue->nr_congestion_on)) | |
9a0e609e MS |
1076 | blk_run_queue_async(md->queue); |
1077 | } | |
cec47e3d KU |
1078 | |
1079 | /* | |
1080 | * dm_put() must be at the end of this function. See the comment above | |
1081 | */ | |
1082 | dm_put(md); | |
1083 | } | |
1084 | ||
aa6df8dd | 1085 | static void free_rq_clone(struct request *clone, bool must_be_mapped) |
a77e28c7 KU |
1086 | { |
1087 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1088 | struct mapped_device *md = tio->md; |
a77e28c7 | 1089 | |
aa6df8dd MS |
1090 | WARN_ON_ONCE(must_be_mapped && !clone->q); |
1091 | ||
a77e28c7 | 1092 | blk_rq_unprep_clone(clone); |
bfebd1cd | 1093 | |
aa6df8dd MS |
1094 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1095 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1096 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1097 | else if (!md->queue->mq_ops) |
1098 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd | 1099 | free_clone_request(md, clone); |
aa6df8dd MS |
1100 | /* |
1101 | * NOTE: for the blk-mq queue stacked on request_fn queue(s) case: | |
1102 | * no need to call free_clone_request() because we leverage blk-mq by | |
1103 | * allocating the clone at the end of the blk-mq pdu (see: clone_rq) | |
1104 | */ | |
bfebd1cd MS |
1105 | |
1106 | if (!md->queue->mq_ops) | |
1107 | free_rq_tio(tio); | |
a77e28c7 KU |
1108 | } |
1109 | ||
980691e5 KU |
1110 | /* |
1111 | * Complete the clone and the original request. | |
466d89a6 KB |
1112 | * Must be called without clone's queue lock held, |
1113 | * see end_clone_request() for more details. | |
980691e5 KU |
1114 | */ |
1115 | static void dm_end_request(struct request *clone, int error) | |
1116 | { | |
1117 | int rw = rq_data_dir(clone); | |
1118 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1119 | struct mapped_device *md = tio->md; | |
1120 | struct request *rq = tio->orig; | |
1121 | ||
29e4013d | 1122 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1123 | rq->errors = clone->errors; |
1124 | rq->resid_len = clone->resid_len; | |
1125 | ||
1126 | if (rq->sense) | |
1127 | /* | |
1128 | * We are using the sense buffer of the original | |
1129 | * request. | |
1130 | * So setting the length of the sense data is enough. | |
1131 | */ | |
1132 | rq->sense_len = clone->sense_len; | |
1133 | } | |
1134 | ||
aa6df8dd | 1135 | free_rq_clone(clone, true); |
bfebd1cd MS |
1136 | if (!rq->q->mq_ops) |
1137 | blk_end_request_all(rq, error); | |
1138 | else | |
1139 | blk_mq_end_request(rq, error); | |
29e4013d | 1140 | rq_completed(md, rw, true); |
980691e5 KU |
1141 | } |
1142 | ||
cec47e3d KU |
1143 | static void dm_unprep_request(struct request *rq) |
1144 | { | |
bfebd1cd | 1145 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1146 | struct request *clone = tio->clone; |
cec47e3d | 1147 | |
bfebd1cd MS |
1148 | if (!rq->q->mq_ops) { |
1149 | rq->special = NULL; | |
1150 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1151 | } | |
cec47e3d | 1152 | |
e5863d9a | 1153 | if (clone) |
aa6df8dd | 1154 | free_rq_clone(clone, false); |
cec47e3d KU |
1155 | } |
1156 | ||
1157 | /* | |
1158 | * Requeue the original request of a clone. | |
1159 | */ | |
bfebd1cd | 1160 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1161 | { |
cec47e3d KU |
1162 | struct request_queue *q = rq->q; |
1163 | unsigned long flags; | |
1164 | ||
cec47e3d | 1165 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d KU |
1166 | blk_requeue_request(q, rq); |
1167 | spin_unlock_irqrestore(q->queue_lock, flags); | |
bfebd1cd MS |
1168 | } |
1169 | ||
1170 | static void dm_requeue_unmapped_original_request(struct mapped_device *md, | |
1171 | struct request *rq) | |
1172 | { | |
1173 | int rw = rq_data_dir(rq); | |
1174 | ||
1175 | dm_unprep_request(rq); | |
1176 | ||
1177 | if (!rq->q->mq_ops) | |
1178 | old_requeue_request(rq); | |
1179 | else { | |
1180 | blk_mq_requeue_request(rq); | |
1181 | blk_mq_kick_requeue_list(rq->q); | |
1182 | } | |
cec47e3d | 1183 | |
466d89a6 KB |
1184 | rq_completed(md, rw, false); |
1185 | } | |
1186 | ||
1187 | static void dm_requeue_unmapped_request(struct request *clone) | |
1188 | { | |
1189 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1190 | ||
1191 | dm_requeue_unmapped_original_request(tio->md, tio->orig); | |
cec47e3d | 1192 | } |
cec47e3d | 1193 | |
bfebd1cd | 1194 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1195 | { |
1196 | unsigned long flags; | |
1197 | ||
bfebd1cd MS |
1198 | if (blk_queue_stopped(q)) |
1199 | return; | |
1200 | ||
cec47e3d | 1201 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1202 | blk_stop_queue(q); |
cec47e3d KU |
1203 | spin_unlock_irqrestore(q->queue_lock, flags); |
1204 | } | |
1205 | ||
bfebd1cd | 1206 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1207 | { |
bfebd1cd MS |
1208 | if (!q->mq_ops) |
1209 | old_stop_queue(q); | |
1210 | else | |
1211 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1212 | } |
1213 | ||
bfebd1cd | 1214 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1215 | { |
1216 | unsigned long flags; | |
1217 | ||
1218 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1219 | if (blk_queue_stopped(q)) |
1220 | blk_start_queue(q); | |
cec47e3d KU |
1221 | spin_unlock_irqrestore(q->queue_lock, flags); |
1222 | } | |
1223 | ||
bfebd1cd MS |
1224 | static void start_queue(struct request_queue *q) |
1225 | { | |
1226 | if (!q->mq_ops) | |
1227 | old_start_queue(q); | |
1228 | else | |
1229 | blk_mq_start_stopped_hw_queues(q, true); | |
1230 | } | |
1231 | ||
11a68244 | 1232 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1233 | { |
11a68244 | 1234 | int r = error; |
cec47e3d | 1235 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1236 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1237 | |
ba1cbad9 MS |
1238 | if (tio->ti) { |
1239 | rq_end_io = tio->ti->type->rq_end_io; | |
1240 | ||
1241 | if (mapped && rq_end_io) | |
1242 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1243 | } | |
cec47e3d | 1244 | |
7eee4ae2 MS |
1245 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1246 | !clone->q->limits.max_write_same_sectors)) | |
1247 | disable_write_same(tio->md); | |
1248 | ||
11a68244 | 1249 | if (r <= 0) |
cec47e3d | 1250 | /* The target wants to complete the I/O */ |
11a68244 KU |
1251 | dm_end_request(clone, r); |
1252 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1253 | /* The target will handle the I/O */ |
1254 | return; | |
11a68244 | 1255 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1256 | /* The target wants to requeue the I/O */ |
1257 | dm_requeue_unmapped_request(clone); | |
1258 | else { | |
11a68244 | 1259 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1260 | BUG(); |
1261 | } | |
1262 | } | |
1263 | ||
11a68244 KU |
1264 | /* |
1265 | * Request completion handler for request-based dm | |
1266 | */ | |
1267 | static void dm_softirq_done(struct request *rq) | |
1268 | { | |
1269 | bool mapped = true; | |
bfebd1cd | 1270 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1271 | struct request *clone = tio->clone; |
bfebd1cd | 1272 | int rw; |
11a68244 | 1273 | |
e5863d9a | 1274 | if (!clone) { |
bfebd1cd MS |
1275 | rw = rq_data_dir(rq); |
1276 | if (!rq->q->mq_ops) { | |
1277 | blk_end_request_all(rq, tio->error); | |
1278 | rq_completed(tio->md, rw, false); | |
1279 | free_rq_tio(tio); | |
1280 | } else { | |
1281 | blk_mq_end_request(rq, tio->error); | |
1282 | rq_completed(tio->md, rw, false); | |
1283 | } | |
e5863d9a MS |
1284 | return; |
1285 | } | |
11a68244 KU |
1286 | |
1287 | if (rq->cmd_flags & REQ_FAILED) | |
1288 | mapped = false; | |
1289 | ||
1290 | dm_done(clone, tio->error, mapped); | |
1291 | } | |
1292 | ||
cec47e3d KU |
1293 | /* |
1294 | * Complete the clone and the original request with the error status | |
1295 | * through softirq context. | |
1296 | */ | |
466d89a6 | 1297 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1298 | { |
bfebd1cd | 1299 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1300 | |
1301 | tio->error = error; | |
cec47e3d KU |
1302 | blk_complete_request(rq); |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Complete the not-mapped clone and the original request with the error status | |
1307 | * through softirq context. | |
1308 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1309 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1310 | */ |
466d89a6 | 1311 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1312 | { |
cec47e3d | 1313 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1314 | dm_complete_request(rq, error); |
cec47e3d | 1315 | } |
cec47e3d KU |
1316 | |
1317 | /* | |
bfebd1cd | 1318 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1319 | */ |
1320 | static void end_clone_request(struct request *clone, int error) | |
1321 | { | |
466d89a6 KB |
1322 | struct dm_rq_target_io *tio = clone->end_io_data; |
1323 | ||
e5863d9a MS |
1324 | if (!clone->q->mq_ops) { |
1325 | /* | |
1326 | * For just cleaning up the information of the queue in which | |
1327 | * the clone was dispatched. | |
1328 | * The clone is *NOT* freed actually here because it is alloced | |
1329 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1330 | */ | |
1331 | __blk_put_request(clone->q, clone); | |
1332 | } | |
cec47e3d KU |
1333 | |
1334 | /* | |
1335 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1336 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1337 | * - another request may be submitted by the upper level driver |
1338 | * of the stacking during the completion | |
1339 | * - the submission which requires queue lock may be done | |
466d89a6 | 1340 | * against this clone's queue |
cec47e3d | 1341 | */ |
466d89a6 | 1342 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1343 | } |
1344 | ||
56a67df7 MS |
1345 | /* |
1346 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1347 | * target boundary. | |
1348 | */ | |
1349 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1350 | { | |
1351 | sector_t target_offset = dm_target_offset(ti, sector); | |
1352 | ||
1353 | return ti->len - target_offset; | |
1354 | } | |
1355 | ||
1356 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1357 | { |
56a67df7 | 1358 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1359 | sector_t offset, max_len; |
1da177e4 LT |
1360 | |
1361 | /* | |
542f9038 | 1362 | * Does the target need to split even further? |
1da177e4 | 1363 | */ |
542f9038 MS |
1364 | if (ti->max_io_len) { |
1365 | offset = dm_target_offset(ti, sector); | |
1366 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1367 | max_len = sector_div(offset, ti->max_io_len); | |
1368 | else | |
1369 | max_len = offset & (ti->max_io_len - 1); | |
1370 | max_len = ti->max_io_len - max_len; | |
1371 | ||
1372 | if (len > max_len) | |
1373 | len = max_len; | |
1da177e4 LT |
1374 | } |
1375 | ||
1376 | return len; | |
1377 | } | |
1378 | ||
542f9038 MS |
1379 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1380 | { | |
1381 | if (len > UINT_MAX) { | |
1382 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1383 | (unsigned long long)len, UINT_MAX); | |
1384 | ti->error = "Maximum size of target IO is too large"; | |
1385 | return -EINVAL; | |
1386 | } | |
1387 | ||
1388 | ti->max_io_len = (uint32_t) len; | |
1389 | ||
1390 | return 0; | |
1391 | } | |
1392 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1393 | ||
1dd40c3e MP |
1394 | /* |
1395 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1396 | * allowed for all bio types except REQ_FLUSH. | |
1397 | * | |
1398 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1399 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1400 | * sent in a next bio. | |
1401 | * | |
1402 | * A diagram that explains the arithmetics: | |
1403 | * +--------------------+---------------+-------+ | |
1404 | * | 1 | 2 | 3 | | |
1405 | * +--------------------+---------------+-------+ | |
1406 | * | |
1407 | * <-------------- *tio->len_ptr ---------------> | |
1408 | * <------- bi_size -------> | |
1409 | * <-- n_sectors --> | |
1410 | * | |
1411 | * Region 1 was already iterated over with bio_advance or similar function. | |
1412 | * (it may be empty if the target doesn't use bio_advance) | |
1413 | * Region 2 is the remaining bio size that the target wants to process. | |
1414 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1415 | * to make it empty) | |
1416 | * The target requires that region 3 is to be sent in the next bio. | |
1417 | * | |
1418 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1419 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1420 | * copies of the bio. | |
1421 | */ | |
1422 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1423 | { | |
1424 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1425 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1426 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1427 | BUG_ON(bi_size > *tio->len_ptr); | |
1428 | BUG_ON(n_sectors > bi_size); | |
1429 | *tio->len_ptr -= bi_size - n_sectors; | |
1430 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1431 | } | |
1432 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1433 | ||
bd2a49b8 | 1434 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1435 | { |
1436 | int r; | |
2056a782 | 1437 | sector_t sector; |
9faf400f | 1438 | struct mapped_device *md; |
dba14160 | 1439 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1440 | struct dm_target *ti = tio->ti; |
1da177e4 | 1441 | |
1da177e4 | 1442 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1443 | |
1444 | /* | |
1445 | * Map the clone. If r == 0 we don't need to do | |
1446 | * anything, the target has assumed ownership of | |
1447 | * this io. | |
1448 | */ | |
1449 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1450 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1451 | r = ti->type->map(ti, clone); |
45cbcd79 | 1452 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1453 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1454 | |
d07335e5 MS |
1455 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1456 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1457 | |
1da177e4 | 1458 | generic_make_request(clone); |
2e93ccc1 KU |
1459 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1460 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1461 | md = tio->io->md; |
1462 | dec_pending(tio->io, r); | |
9faf400f | 1463 | free_tio(md, tio); |
45cbcd79 KU |
1464 | } else if (r) { |
1465 | DMWARN("unimplemented target map return value: %d", r); | |
1466 | BUG(); | |
1da177e4 LT |
1467 | } |
1468 | } | |
1469 | ||
1470 | struct clone_info { | |
1471 | struct mapped_device *md; | |
1472 | struct dm_table *map; | |
1473 | struct bio *bio; | |
1474 | struct dm_io *io; | |
1475 | sector_t sector; | |
e0d6609a | 1476 | unsigned sector_count; |
1da177e4 LT |
1477 | }; |
1478 | ||
e0d6609a | 1479 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1480 | { |
4f024f37 KO |
1481 | bio->bi_iter.bi_sector = sector; |
1482 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1483 | } |
1484 | ||
1485 | /* | |
1486 | * Creates a bio that consists of range of complete bvecs. | |
1487 | */ | |
dba14160 | 1488 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1489 | sector_t sector, unsigned len) |
1da177e4 | 1490 | { |
dba14160 | 1491 | struct bio *clone = &tio->clone; |
1da177e4 | 1492 | |
1c3b13e6 KO |
1493 | __bio_clone_fast(clone, bio); |
1494 | ||
1495 | if (bio_integrity(bio)) | |
1496 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1497 | |
1c3b13e6 KO |
1498 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1499 | clone->bi_iter.bi_size = to_bytes(len); | |
1500 | ||
1501 | if (bio_integrity(bio)) | |
1502 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1503 | } |
1504 | ||
9015df24 | 1505 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1506 | struct dm_target *ti, |
55a62eef | 1507 | unsigned target_bio_nr) |
f9ab94ce | 1508 | { |
dba14160 MP |
1509 | struct dm_target_io *tio; |
1510 | struct bio *clone; | |
1511 | ||
99778273 | 1512 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1513 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1514 | |
1515 | tio->io = ci->io; | |
1516 | tio->ti = ti; | |
55a62eef | 1517 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1518 | |
1519 | return tio; | |
1520 | } | |
1521 | ||
14fe594d AK |
1522 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1523 | struct dm_target *ti, | |
1dd40c3e | 1524 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1525 | { |
99778273 | 1526 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1527 | struct bio *clone = &tio->clone; |
9015df24 | 1528 | |
1dd40c3e MP |
1529 | tio->len_ptr = len; |
1530 | ||
99778273 | 1531 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1532 | if (len) |
1dd40c3e | 1533 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1534 | |
bd2a49b8 | 1535 | __map_bio(tio); |
f9ab94ce MP |
1536 | } |
1537 | ||
14fe594d | 1538 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1539 | unsigned num_bios, unsigned *len) |
06a426ce | 1540 | { |
55a62eef | 1541 | unsigned target_bio_nr; |
06a426ce | 1542 | |
55a62eef | 1543 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1544 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1545 | } |
1546 | ||
14fe594d | 1547 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1548 | { |
06a426ce | 1549 | unsigned target_nr = 0; |
f9ab94ce MP |
1550 | struct dm_target *ti; |
1551 | ||
b372d360 | 1552 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1553 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1554 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1555 | |
f9ab94ce MP |
1556 | return 0; |
1557 | } | |
1558 | ||
e4c93811 | 1559 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1560 | sector_t sector, unsigned *len) |
5ae89a87 | 1561 | { |
dba14160 | 1562 | struct bio *bio = ci->bio; |
5ae89a87 | 1563 | struct dm_target_io *tio; |
b0d8ed4d AK |
1564 | unsigned target_bio_nr; |
1565 | unsigned num_target_bios = 1; | |
5ae89a87 | 1566 | |
b0d8ed4d AK |
1567 | /* |
1568 | * Does the target want to receive duplicate copies of the bio? | |
1569 | */ | |
1570 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1571 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1572 | |
b0d8ed4d | 1573 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1574 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1575 | tio->len_ptr = len; |
1576 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1577 | __map_bio(tio); |
1578 | } | |
5ae89a87 MS |
1579 | } |
1580 | ||
55a62eef | 1581 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1582 | |
55a62eef | 1583 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1584 | { |
55a62eef | 1585 | return ti->num_discard_bios; |
23508a96 MS |
1586 | } |
1587 | ||
55a62eef | 1588 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1589 | { |
55a62eef | 1590 | return ti->num_write_same_bios; |
23508a96 MS |
1591 | } |
1592 | ||
1593 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1594 | ||
1595 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1596 | { | |
55a62eef | 1597 | return ti->split_discard_bios; |
23508a96 MS |
1598 | } |
1599 | ||
14fe594d AK |
1600 | static int __send_changing_extent_only(struct clone_info *ci, |
1601 | get_num_bios_fn get_num_bios, | |
1602 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1603 | { |
1604 | struct dm_target *ti; | |
e0d6609a | 1605 | unsigned len; |
55a62eef | 1606 | unsigned num_bios; |
5ae89a87 | 1607 | |
a79245b3 MS |
1608 | do { |
1609 | ti = dm_table_find_target(ci->map, ci->sector); | |
1610 | if (!dm_target_is_valid(ti)) | |
1611 | return -EIO; | |
5ae89a87 | 1612 | |
5ae89a87 | 1613 | /* |
23508a96 MS |
1614 | * Even though the device advertised support for this type of |
1615 | * request, that does not mean every target supports it, and | |
936688d7 | 1616 | * reconfiguration might also have changed that since the |
a79245b3 | 1617 | * check was performed. |
5ae89a87 | 1618 | */ |
55a62eef AK |
1619 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1620 | if (!num_bios) | |
a79245b3 | 1621 | return -EOPNOTSUPP; |
5ae89a87 | 1622 | |
23508a96 | 1623 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1624 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1625 | else |
e0d6609a | 1626 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1627 | |
1dd40c3e | 1628 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1629 | |
1630 | ci->sector += len; | |
1631 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1632 | |
1633 | return 0; | |
1634 | } | |
1635 | ||
14fe594d | 1636 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1637 | { |
14fe594d AK |
1638 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1639 | is_split_required_for_discard); | |
23508a96 MS |
1640 | } |
1641 | ||
14fe594d | 1642 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1643 | { |
14fe594d | 1644 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1645 | } |
1646 | ||
e4c93811 AK |
1647 | /* |
1648 | * Select the correct strategy for processing a non-flush bio. | |
1649 | */ | |
14fe594d | 1650 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1651 | { |
dba14160 | 1652 | struct bio *bio = ci->bio; |
512875bd | 1653 | struct dm_target *ti; |
1c3b13e6 | 1654 | unsigned len; |
1da177e4 | 1655 | |
5ae89a87 | 1656 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1657 | return __send_discard(ci); |
23508a96 | 1658 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1659 | return __send_write_same(ci); |
5ae89a87 | 1660 | |
512875bd JN |
1661 | ti = dm_table_find_target(ci->map, ci->sector); |
1662 | if (!dm_target_is_valid(ti)) | |
1663 | return -EIO; | |
1664 | ||
1c3b13e6 | 1665 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1666 | |
1dd40c3e | 1667 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1668 | |
1c3b13e6 KO |
1669 | ci->sector += len; |
1670 | ci->sector_count -= len; | |
1da177e4 | 1671 | |
1c3b13e6 | 1672 | return 0; |
1da177e4 LT |
1673 | } |
1674 | ||
1675 | /* | |
14fe594d | 1676 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1677 | */ |
83d5e5b0 MP |
1678 | static void __split_and_process_bio(struct mapped_device *md, |
1679 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1680 | { |
1681 | struct clone_info ci; | |
512875bd | 1682 | int error = 0; |
1da177e4 | 1683 | |
83d5e5b0 | 1684 | if (unlikely(!map)) { |
6a8736d1 | 1685 | bio_io_error(bio); |
f0b9a450 MP |
1686 | return; |
1687 | } | |
692d0eb9 | 1688 | |
83d5e5b0 | 1689 | ci.map = map; |
1da177e4 | 1690 | ci.md = md; |
1da177e4 LT |
1691 | ci.io = alloc_io(md); |
1692 | ci.io->error = 0; | |
1693 | atomic_set(&ci.io->io_count, 1); | |
1694 | ci.io->bio = bio; | |
1695 | ci.io->md = md; | |
f88fb981 | 1696 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1697 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1698 | |
3eaf840e | 1699 | start_io_acct(ci.io); |
bd2a49b8 | 1700 | |
b372d360 MS |
1701 | if (bio->bi_rw & REQ_FLUSH) { |
1702 | ci.bio = &ci.md->flush_bio; | |
1703 | ci.sector_count = 0; | |
14fe594d | 1704 | error = __send_empty_flush(&ci); |
b372d360 MS |
1705 | /* dec_pending submits any data associated with flush */ |
1706 | } else { | |
6a8736d1 | 1707 | ci.bio = bio; |
d87f4c14 | 1708 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1709 | while (ci.sector_count && !error) |
14fe594d | 1710 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1711 | } |
1da177e4 LT |
1712 | |
1713 | /* drop the extra reference count */ | |
512875bd | 1714 | dec_pending(ci.io, error); |
1da177e4 LT |
1715 | } |
1716 | /*----------------------------------------------------------------- | |
1717 | * CRUD END | |
1718 | *---------------------------------------------------------------*/ | |
1719 | ||
f6fccb12 MB |
1720 | static int dm_merge_bvec(struct request_queue *q, |
1721 | struct bvec_merge_data *bvm, | |
1722 | struct bio_vec *biovec) | |
1723 | { | |
1724 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1725 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1726 | struct dm_target *ti; |
1727 | sector_t max_sectors; | |
5037108a | 1728 | int max_size = 0; |
f6fccb12 MB |
1729 | |
1730 | if (unlikely(!map)) | |
5037108a | 1731 | goto out; |
f6fccb12 MB |
1732 | |
1733 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1734 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1735 | goto out; |
f6fccb12 MB |
1736 | |
1737 | /* | |
1738 | * Find maximum amount of I/O that won't need splitting | |
1739 | */ | |
56a67df7 | 1740 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1741 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1742 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1743 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1744 | max_size = 0; |
1745 | ||
1746 | /* | |
1747 | * merge_bvec_fn() returns number of bytes | |
1748 | * it can accept at this offset | |
1749 | * max is precomputed maximal io size | |
1750 | */ | |
1751 | if (max_size && ti->type->merge) | |
1752 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1753 | /* |
1754 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1755 | * provided their merge_bvec method (we know this by looking for the |
1756 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1757 | * allow bios with multiple vector entries. So always set max_size | |
1758 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1759 | */ |
1760 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1761 | max_size = 0; |
f6fccb12 | 1762 | |
5037108a | 1763 | out: |
83d5e5b0 | 1764 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1765 | /* |
1766 | * Always allow an entire first page | |
1767 | */ | |
1768 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1769 | max_size = biovec->bv_len; | |
1770 | ||
f6fccb12 MB |
1771 | return max_size; |
1772 | } | |
1773 | ||
1da177e4 LT |
1774 | /* |
1775 | * The request function that just remaps the bio built up by | |
1776 | * dm_merge_bvec. | |
1777 | */ | |
ff36ab34 | 1778 | static void dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1779 | { |
12f03a49 | 1780 | int rw = bio_data_dir(bio); |
1da177e4 | 1781 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1782 | int srcu_idx; |
1783 | struct dm_table *map; | |
1da177e4 | 1784 | |
83d5e5b0 | 1785 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1786 | |
18c0b223 | 1787 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1788 | |
6a8736d1 TH |
1789 | /* if we're suspended, we have to queue this io for later */ |
1790 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1791 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1792 | |
6a8736d1 TH |
1793 | if (bio_rw(bio) != READA) |
1794 | queue_io(md, bio); | |
1795 | else | |
54d9a1b4 | 1796 | bio_io_error(bio); |
5a7bbad2 | 1797 | return; |
1da177e4 LT |
1798 | } |
1799 | ||
83d5e5b0 MP |
1800 | __split_and_process_bio(md, map, bio); |
1801 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1802 | return; |
cec47e3d KU |
1803 | } |
1804 | ||
fd2ed4d2 | 1805 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1806 | { |
1807 | return blk_queue_stackable(md->queue); | |
1808 | } | |
1809 | ||
466d89a6 | 1810 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1811 | { |
1812 | int r; | |
1813 | ||
466d89a6 KB |
1814 | if (blk_queue_io_stat(clone->q)) |
1815 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1816 | |
466d89a6 KB |
1817 | clone->start_time = jiffies; |
1818 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1819 | if (r) |
466d89a6 | 1820 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1821 | dm_complete_request(rq, r); |
1822 | } | |
cec47e3d | 1823 | |
cec47e3d KU |
1824 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1825 | void *data) | |
1826 | { | |
1827 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1828 | struct dm_rq_clone_bio_info *info = |
1829 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1830 | |
1831 | info->orig = bio_orig; | |
1832 | info->tio = tio; | |
1833 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1834 | |
1835 | return 0; | |
1836 | } | |
1837 | ||
1838 | static int setup_clone(struct request *clone, struct request *rq, | |
1ae49ea2 | 1839 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
cec47e3d | 1840 | { |
d0bcb878 | 1841 | int r; |
cec47e3d | 1842 | |
1ae49ea2 | 1843 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, |
29e4013d TH |
1844 | dm_rq_bio_constructor, tio); |
1845 | if (r) | |
1846 | return r; | |
cec47e3d | 1847 | |
29e4013d TH |
1848 | clone->cmd = rq->cmd; |
1849 | clone->cmd_len = rq->cmd_len; | |
1850 | clone->sense = rq->sense; | |
cec47e3d KU |
1851 | clone->end_io = end_clone_request; |
1852 | clone->end_io_data = tio; | |
1853 | ||
1ae49ea2 MS |
1854 | tio->clone = clone; |
1855 | ||
cec47e3d KU |
1856 | return 0; |
1857 | } | |
1858 | ||
6facdaff | 1859 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1860 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1861 | { |
02233342 MS |
1862 | /* |
1863 | * Do not allocate a clone if tio->clone was already set | |
1864 | * (see: dm_mq_queue_rq). | |
1865 | */ | |
1866 | bool alloc_clone = !tio->clone; | |
1867 | struct request *clone; | |
1ae49ea2 | 1868 | |
02233342 MS |
1869 | if (alloc_clone) { |
1870 | clone = alloc_clone_request(md, gfp_mask); | |
1871 | if (!clone) | |
1872 | return NULL; | |
1873 | } else | |
1874 | clone = tio->clone; | |
1ae49ea2 MS |
1875 | |
1876 | blk_rq_init(NULL, clone); | |
1877 | if (setup_clone(clone, rq, tio, gfp_mask)) { | |
1878 | /* -ENOMEM */ | |
02233342 MS |
1879 | if (alloc_clone) |
1880 | free_clone_request(md, clone); | |
1ae49ea2 MS |
1881 | return NULL; |
1882 | } | |
1883 | ||
1884 | return clone; | |
1885 | } | |
1886 | ||
2eb6e1e3 KB |
1887 | static void map_tio_request(struct kthread_work *work); |
1888 | ||
bfebd1cd MS |
1889 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1890 | struct mapped_device *md) | |
1891 | { | |
1892 | tio->md = md; | |
1893 | tio->ti = NULL; | |
1894 | tio->clone = NULL; | |
1895 | tio->orig = rq; | |
1896 | tio->error = 0; | |
1897 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1898 | if (md->kworker_task) |
1899 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1900 | } |
1901 | ||
466d89a6 KB |
1902 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1903 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1904 | { |
6facdaff | 1905 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1906 | int srcu_idx; |
1907 | struct dm_table *table; | |
6facdaff KU |
1908 | |
1909 | tio = alloc_rq_tio(md, gfp_mask); | |
1910 | if (!tio) | |
1911 | return NULL; | |
1912 | ||
bfebd1cd | 1913 | init_tio(tio, rq, md); |
6facdaff | 1914 | |
e5863d9a MS |
1915 | table = dm_get_live_table(md, &srcu_idx); |
1916 | if (!dm_table_mq_request_based(table)) { | |
1917 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1918 | dm_put_live_table(md, srcu_idx); | |
1919 | free_rq_tio(tio); | |
1920 | return NULL; | |
1921 | } | |
6facdaff | 1922 | } |
e5863d9a | 1923 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1924 | |
466d89a6 | 1925 | return tio; |
6facdaff KU |
1926 | } |
1927 | ||
cec47e3d KU |
1928 | /* |
1929 | * Called with the queue lock held. | |
1930 | */ | |
1931 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1932 | { | |
1933 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1934 | struct dm_rq_target_io *tio; |
cec47e3d | 1935 | |
cec47e3d KU |
1936 | if (unlikely(rq->special)) { |
1937 | DMWARN("Already has something in rq->special."); | |
1938 | return BLKPREP_KILL; | |
1939 | } | |
1940 | ||
466d89a6 KB |
1941 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1942 | if (!tio) | |
cec47e3d | 1943 | return BLKPREP_DEFER; |
cec47e3d | 1944 | |
466d89a6 | 1945 | rq->special = tio; |
cec47e3d KU |
1946 | rq->cmd_flags |= REQ_DONTPREP; |
1947 | ||
1948 | return BLKPREP_OK; | |
1949 | } | |
1950 | ||
9eef87da KU |
1951 | /* |
1952 | * Returns: | |
e5863d9a MS |
1953 | * 0 : the request has been processed |
1954 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1955 | * < 0 : the request was completed due to failure | |
9eef87da | 1956 | */ |
bfebd1cd | 1957 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1958 | struct mapped_device *md) |
cec47e3d | 1959 | { |
e5863d9a | 1960 | int r; |
bfebd1cd | 1961 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1962 | struct request *clone = NULL; |
1963 | ||
1964 | if (tio->clone) { | |
1965 | clone = tio->clone; | |
1966 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1967 | } else { | |
1968 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1969 | if (r < 0) { | |
1970 | /* The target wants to complete the I/O */ | |
1971 | dm_kill_unmapped_request(rq, r); | |
1972 | return r; | |
1973 | } | |
1974 | if (IS_ERR(clone)) | |
1975 | return DM_MAPIO_REQUEUE; | |
02233342 | 1976 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
e5863d9a MS |
1977 | /* -ENOMEM */ |
1978 | ti->type->release_clone_rq(clone); | |
1979 | return DM_MAPIO_REQUEUE; | |
1980 | } | |
1981 | } | |
cec47e3d | 1982 | |
cec47e3d KU |
1983 | switch (r) { |
1984 | case DM_MAPIO_SUBMITTED: | |
1985 | /* The target has taken the I/O to submit by itself later */ | |
1986 | break; | |
1987 | case DM_MAPIO_REMAPPED: | |
1988 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1989 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1990 | blk_rq_pos(rq)); |
1991 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1992 | break; |
1993 | case DM_MAPIO_REQUEUE: | |
1994 | /* The target wants to requeue the I/O */ | |
1995 | dm_requeue_unmapped_request(clone); | |
1996 | break; | |
1997 | default: | |
1998 | if (r > 0) { | |
1999 | DMWARN("unimplemented target map return value: %d", r); | |
2000 | BUG(); | |
2001 | } | |
2002 | ||
2003 | /* The target wants to complete the I/O */ | |
466d89a6 | 2004 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 2005 | return r; |
cec47e3d | 2006 | } |
9eef87da | 2007 | |
e5863d9a | 2008 | return 0; |
cec47e3d KU |
2009 | } |
2010 | ||
2eb6e1e3 | 2011 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 2012 | { |
2eb6e1e3 | 2013 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
2014 | struct request *rq = tio->orig; |
2015 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 2016 | |
bfebd1cd | 2017 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
e5863d9a | 2018 | dm_requeue_unmapped_original_request(md, rq); |
2eb6e1e3 KB |
2019 | } |
2020 | ||
466d89a6 | 2021 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2022 | { |
bfebd1cd MS |
2023 | if (!orig->q->mq_ops) |
2024 | blk_start_request(orig); | |
2025 | else | |
2026 | blk_mq_start_request(orig); | |
466d89a6 | 2027 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2028 | |
0ce65797 MS |
2029 | if (md->seq_rq_merge_deadline_usecs) { |
2030 | md->last_rq_pos = rq_end_sector(orig); | |
2031 | md->last_rq_rw = rq_data_dir(orig); | |
2032 | md->last_rq_start_time = ktime_get(); | |
2033 | } | |
de3ec86d | 2034 | |
ba1cbad9 MS |
2035 | /* |
2036 | * Hold the md reference here for the in-flight I/O. | |
2037 | * We can't rely on the reference count by device opener, | |
2038 | * because the device may be closed during the request completion | |
2039 | * when all bios are completed. | |
2040 | * See the comment in rq_completed() too. | |
2041 | */ | |
2042 | dm_get(md); | |
ba1cbad9 MS |
2043 | } |
2044 | ||
0ce65797 MS |
2045 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2046 | ||
2047 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2048 | { | |
2049 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2050 | } | |
2051 | ||
2052 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2053 | const char *buf, size_t count) | |
2054 | { | |
2055 | unsigned deadline; | |
2056 | ||
17e149b8 | 2057 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2058 | return count; |
2059 | ||
2060 | if (kstrtouint(buf, 10, &deadline)) | |
2061 | return -EINVAL; | |
2062 | ||
2063 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2064 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2065 | ||
2066 | md->seq_rq_merge_deadline_usecs = deadline; | |
2067 | ||
2068 | return count; | |
2069 | } | |
2070 | ||
2071 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2072 | { | |
2073 | ktime_t kt_deadline; | |
2074 | ||
2075 | if (!md->seq_rq_merge_deadline_usecs) | |
2076 | return false; | |
2077 | ||
2078 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2079 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2080 | ||
2081 | return !ktime_after(ktime_get(), kt_deadline); | |
2082 | } | |
2083 | ||
cec47e3d KU |
2084 | /* |
2085 | * q->request_fn for request-based dm. | |
2086 | * Called with the queue lock held. | |
2087 | */ | |
2088 | static void dm_request_fn(struct request_queue *q) | |
2089 | { | |
2090 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2091 | int srcu_idx; |
2092 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2093 | struct dm_target *ti; |
466d89a6 | 2094 | struct request *rq; |
2eb6e1e3 | 2095 | struct dm_rq_target_io *tio; |
29e4013d | 2096 | sector_t pos; |
cec47e3d KU |
2097 | |
2098 | /* | |
b4324fee KU |
2099 | * For suspend, check blk_queue_stopped() and increment |
2100 | * ->pending within a single queue_lock not to increment the | |
2101 | * number of in-flight I/Os after the queue is stopped in | |
2102 | * dm_suspend(). | |
cec47e3d | 2103 | */ |
7eaceacc | 2104 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2105 | rq = blk_peek_request(q); |
2106 | if (!rq) | |
9d1deb83 | 2107 | goto out; |
cec47e3d | 2108 | |
29e4013d TH |
2109 | /* always use block 0 to find the target for flushes for now */ |
2110 | pos = 0; | |
2111 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2112 | pos = blk_rq_pos(rq); | |
2113 | ||
2114 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2115 | if (!dm_target_is_valid(ti)) { |
2116 | /* | |
466d89a6 | 2117 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2118 | * before calling dm_kill_unmapped_request |
2119 | */ | |
2120 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2121 | dm_start_request(md, rq); |
2122 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2123 | continue; |
2124 | } | |
d0bcb878 | 2125 | |
0ce65797 MS |
2126 | if (dm_request_peeked_before_merge_deadline(md) && |
2127 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2128 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2129 | goto delay_and_out; | |
2130 | ||
cec47e3d | 2131 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2132 | goto delay_and_out; |
cec47e3d | 2133 | |
466d89a6 | 2134 | dm_start_request(md, rq); |
9eef87da | 2135 | |
bfebd1cd | 2136 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2137 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2138 | tio->ti = ti; | |
2139 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2140 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2141 | } |
2142 | ||
2143 | goto out; | |
2144 | ||
7eaceacc | 2145 | delay_and_out: |
d548b34b | 2146 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2147 | out: |
83d5e5b0 | 2148 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2149 | } |
2150 | ||
1da177e4 LT |
2151 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2152 | { | |
8a57dfc6 CS |
2153 | int r = bdi_bits; |
2154 | struct mapped_device *md = congested_data; | |
2155 | struct dm_table *map; | |
1da177e4 | 2156 | |
1eb787ec | 2157 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2158 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2159 | if (map) { |
cec47e3d KU |
2160 | /* |
2161 | * Request-based dm cares about only own queue for | |
2162 | * the query about congestion status of request_queue | |
2163 | */ | |
2164 | if (dm_request_based(md)) | |
2165 | r = md->queue->backing_dev_info.state & | |
2166 | bdi_bits; | |
2167 | else | |
2168 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2169 | } |
83d5e5b0 | 2170 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2171 | } |
2172 | ||
1da177e4 LT |
2173 | return r; |
2174 | } | |
2175 | ||
2176 | /*----------------------------------------------------------------- | |
2177 | * An IDR is used to keep track of allocated minor numbers. | |
2178 | *---------------------------------------------------------------*/ | |
2b06cfff | 2179 | static void free_minor(int minor) |
1da177e4 | 2180 | { |
f32c10b0 | 2181 | spin_lock(&_minor_lock); |
1da177e4 | 2182 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2183 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2184 | } |
2185 | ||
2186 | /* | |
2187 | * See if the device with a specific minor # is free. | |
2188 | */ | |
cf13ab8e | 2189 | static int specific_minor(int minor) |
1da177e4 | 2190 | { |
c9d76be6 | 2191 | int r; |
1da177e4 LT |
2192 | |
2193 | if (minor >= (1 << MINORBITS)) | |
2194 | return -EINVAL; | |
2195 | ||
c9d76be6 | 2196 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2197 | spin_lock(&_minor_lock); |
1da177e4 | 2198 | |
c9d76be6 | 2199 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2200 | |
f32c10b0 | 2201 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2202 | idr_preload_end(); |
2203 | if (r < 0) | |
2204 | return r == -ENOSPC ? -EBUSY : r; | |
2205 | return 0; | |
1da177e4 LT |
2206 | } |
2207 | ||
cf13ab8e | 2208 | static int next_free_minor(int *minor) |
1da177e4 | 2209 | { |
c9d76be6 | 2210 | int r; |
62f75c2f | 2211 | |
c9d76be6 | 2212 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2213 | spin_lock(&_minor_lock); |
1da177e4 | 2214 | |
c9d76be6 | 2215 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2216 | |
f32c10b0 | 2217 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2218 | idr_preload_end(); |
2219 | if (r < 0) | |
2220 | return r; | |
2221 | *minor = r; | |
2222 | return 0; | |
1da177e4 LT |
2223 | } |
2224 | ||
83d5cde4 | 2225 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2226 | |
53d5914f MP |
2227 | static void dm_wq_work(struct work_struct *work); |
2228 | ||
4a0b4ddf MS |
2229 | static void dm_init_md_queue(struct mapped_device *md) |
2230 | { | |
2231 | /* | |
2232 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2233 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2234 | * The type is decided at the first table loading time. |
2235 | * To prevent problematic device stacking, clear the queue flag | |
2236 | * for request stacking support until then. | |
2237 | * | |
2238 | * This queue is new, so no concurrency on the queue_flags. | |
2239 | */ | |
2240 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
bfebd1cd | 2241 | } |
4a0b4ddf | 2242 | |
bfebd1cd MS |
2243 | static void dm_init_old_md_queue(struct mapped_device *md) |
2244 | { | |
17e149b8 | 2245 | md->use_blk_mq = false; |
bfebd1cd MS |
2246 | dm_init_md_queue(md); |
2247 | ||
2248 | /* | |
2249 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2250 | */ | |
4a0b4ddf MS |
2251 | md->queue->queuedata = md; |
2252 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2253 | md->queue->backing_dev_info.congested_data = md; | |
ff36ab34 | 2254 | |
4a0b4ddf | 2255 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2256 | } |
2257 | ||
1da177e4 LT |
2258 | /* |
2259 | * Allocate and initialise a blank device with a given minor. | |
2260 | */ | |
2b06cfff | 2261 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2262 | { |
2263 | int r; | |
cf13ab8e | 2264 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2265 | void *old_md; |
1da177e4 LT |
2266 | |
2267 | if (!md) { | |
2268 | DMWARN("unable to allocate device, out of memory."); | |
2269 | return NULL; | |
2270 | } | |
2271 | ||
10da4f79 | 2272 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2273 | goto bad_module_get; |
10da4f79 | 2274 | |
1da177e4 | 2275 | /* get a minor number for the dev */ |
2b06cfff | 2276 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2277 | r = next_free_minor(&minor); |
2b06cfff | 2278 | else |
cf13ab8e | 2279 | r = specific_minor(minor); |
1da177e4 | 2280 | if (r < 0) |
6ed7ade8 | 2281 | goto bad_minor; |
1da177e4 | 2282 | |
83d5e5b0 MP |
2283 | r = init_srcu_struct(&md->io_barrier); |
2284 | if (r < 0) | |
2285 | goto bad_io_barrier; | |
2286 | ||
17e149b8 | 2287 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2288 | md->type = DM_TYPE_NONE; |
e61290a4 | 2289 | mutex_init(&md->suspend_lock); |
a5664dad | 2290 | mutex_init(&md->type_lock); |
86f1152b | 2291 | mutex_init(&md->table_devices_lock); |
022c2611 | 2292 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2293 | atomic_set(&md->holders, 1); |
5c6bd75d | 2294 | atomic_set(&md->open_count, 0); |
1da177e4 | 2295 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2296 | atomic_set(&md->uevent_seq, 0); |
2297 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2298 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2299 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2300 | |
4a0b4ddf | 2301 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2302 | if (!md->queue) |
6ed7ade8 | 2303 | goto bad_queue; |
1da177e4 | 2304 | |
4a0b4ddf | 2305 | dm_init_md_queue(md); |
9faf400f | 2306 | |
1da177e4 LT |
2307 | md->disk = alloc_disk(1); |
2308 | if (!md->disk) | |
6ed7ade8 | 2309 | goto bad_disk; |
1da177e4 | 2310 | |
316d315b NK |
2311 | atomic_set(&md->pending[0], 0); |
2312 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2313 | init_waitqueue_head(&md->wait); |
53d5914f | 2314 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2315 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2316 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2317 | md->kworker_task = NULL; |
f0b04115 | 2318 | |
1da177e4 LT |
2319 | md->disk->major = _major; |
2320 | md->disk->first_minor = minor; | |
2321 | md->disk->fops = &dm_blk_dops; | |
2322 | md->disk->queue = md->queue; | |
2323 | md->disk->private_data = md; | |
2324 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2325 | add_disk(md->disk); | |
7e51f257 | 2326 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2327 | |
670368a8 | 2328 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2329 | if (!md->wq) |
2330 | goto bad_thread; | |
2331 | ||
32a926da MP |
2332 | md->bdev = bdget_disk(md->disk, 0); |
2333 | if (!md->bdev) | |
2334 | goto bad_bdev; | |
2335 | ||
6a8736d1 TH |
2336 | bio_init(&md->flush_bio); |
2337 | md->flush_bio.bi_bdev = md->bdev; | |
2338 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2339 | ||
fd2ed4d2 MP |
2340 | dm_stats_init(&md->stats); |
2341 | ||
ba61fdd1 | 2342 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2343 | spin_lock(&_minor_lock); |
ba61fdd1 | 2344 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2345 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2346 | |
2347 | BUG_ON(old_md != MINOR_ALLOCED); | |
2348 | ||
1da177e4 LT |
2349 | return md; |
2350 | ||
32a926da MP |
2351 | bad_bdev: |
2352 | destroy_workqueue(md->wq); | |
304f3f6a | 2353 | bad_thread: |
03022c54 | 2354 | del_gendisk(md->disk); |
304f3f6a | 2355 | put_disk(md->disk); |
6ed7ade8 | 2356 | bad_disk: |
1312f40e | 2357 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2358 | bad_queue: |
83d5e5b0 MP |
2359 | cleanup_srcu_struct(&md->io_barrier); |
2360 | bad_io_barrier: | |
1da177e4 | 2361 | free_minor(minor); |
6ed7ade8 | 2362 | bad_minor: |
10da4f79 | 2363 | module_put(THIS_MODULE); |
6ed7ade8 | 2364 | bad_module_get: |
1da177e4 LT |
2365 | kfree(md); |
2366 | return NULL; | |
2367 | } | |
2368 | ||
ae9da83f JN |
2369 | static void unlock_fs(struct mapped_device *md); |
2370 | ||
1da177e4 LT |
2371 | static void free_dev(struct mapped_device *md) |
2372 | { | |
f331c029 | 2373 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2374 | |
32a926da | 2375 | unlock_fs(md); |
304f3f6a | 2376 | destroy_workqueue(md->wq); |
2eb6e1e3 KB |
2377 | |
2378 | if (md->kworker_task) | |
2379 | kthread_stop(md->kworker_task); | |
e6ee8c0b KU |
2380 | if (md->io_pool) |
2381 | mempool_destroy(md->io_pool); | |
1ae49ea2 MS |
2382 | if (md->rq_pool) |
2383 | mempool_destroy(md->rq_pool); | |
e6ee8c0b KU |
2384 | if (md->bs) |
2385 | bioset_free(md->bs); | |
63a4f065 | 2386 | |
83d5e5b0 | 2387 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2388 | free_table_devices(&md->table_devices); |
63a4f065 | 2389 | dm_stats_cleanup(&md->stats); |
fba9f90e JM |
2390 | |
2391 | spin_lock(&_minor_lock); | |
2392 | md->disk->private_data = NULL; | |
2393 | spin_unlock(&_minor_lock); | |
63a4f065 MS |
2394 | if (blk_get_integrity(md->disk)) |
2395 | blk_integrity_unregister(md->disk); | |
2396 | del_gendisk(md->disk); | |
1da177e4 | 2397 | put_disk(md->disk); |
1312f40e | 2398 | blk_cleanup_queue(md->queue); |
17e149b8 | 2399 | if (md->use_blk_mq) |
bfebd1cd | 2400 | blk_mq_free_tag_set(&md->tag_set); |
63a4f065 MS |
2401 | bdput(md->bdev); |
2402 | free_minor(minor); | |
2403 | ||
10da4f79 | 2404 | module_put(THIS_MODULE); |
1da177e4 LT |
2405 | kfree(md); |
2406 | } | |
2407 | ||
e6ee8c0b KU |
2408 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2409 | { | |
c0820cf5 | 2410 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2411 | |
17e149b8 | 2412 | if (md->bs) { |
16245bdc JN |
2413 | /* The md already has necessary mempools. */ |
2414 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2415 | /* | |
2416 | * Reload bioset because front_pad may have changed | |
2417 | * because a different table was loaded. | |
2418 | */ | |
2419 | bioset_free(md->bs); | |
2420 | md->bs = p->bs; | |
2421 | p->bs = NULL; | |
16245bdc | 2422 | } |
466d89a6 KB |
2423 | /* |
2424 | * There's no need to reload with request-based dm | |
2425 | * because the size of front_pad doesn't change. | |
2426 | * Note for future: If you are to reload bioset, | |
2427 | * prep-ed requests in the queue may refer | |
2428 | * to bio from the old bioset, so you must walk | |
2429 | * through the queue to unprep. | |
2430 | */ | |
e6ee8c0b | 2431 | goto out; |
c0820cf5 | 2432 | } |
e6ee8c0b | 2433 | |
1ae49ea2 | 2434 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
e6ee8c0b KU |
2435 | |
2436 | md->io_pool = p->io_pool; | |
2437 | p->io_pool = NULL; | |
1ae49ea2 MS |
2438 | md->rq_pool = p->rq_pool; |
2439 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2440 | md->bs = p->bs; |
2441 | p->bs = NULL; | |
2442 | ||
2443 | out: | |
02233342 | 2444 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2445 | dm_table_free_md_mempools(t); |
2446 | } | |
2447 | ||
1da177e4 LT |
2448 | /* |
2449 | * Bind a table to the device. | |
2450 | */ | |
2451 | static void event_callback(void *context) | |
2452 | { | |
7a8c3d3b MA |
2453 | unsigned long flags; |
2454 | LIST_HEAD(uevents); | |
1da177e4 LT |
2455 | struct mapped_device *md = (struct mapped_device *) context; |
2456 | ||
7a8c3d3b MA |
2457 | spin_lock_irqsave(&md->uevent_lock, flags); |
2458 | list_splice_init(&md->uevent_list, &uevents); | |
2459 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2460 | ||
ed9e1982 | 2461 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2462 | |
1da177e4 LT |
2463 | atomic_inc(&md->event_nr); |
2464 | wake_up(&md->eventq); | |
2465 | } | |
2466 | ||
c217649b MS |
2467 | /* |
2468 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2469 | */ | |
4e90188b | 2470 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2471 | { |
4e90188b | 2472 | set_capacity(md->disk, size); |
1da177e4 | 2473 | |
db8fef4f | 2474 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2475 | } |
2476 | ||
d5b9dd04 MP |
2477 | /* |
2478 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2479 | * | |
2480 | * If this function returns 0, then the device is either a non-dm | |
2481 | * device without a merge_bvec_fn, or it is a dm device that is | |
2482 | * able to split any bios it receives that are too big. | |
2483 | */ | |
2484 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2485 | { | |
2486 | struct mapped_device *dev_md; | |
2487 | ||
2488 | if (!q->merge_bvec_fn) | |
2489 | return 0; | |
2490 | ||
ff36ab34 | 2491 | if (q->make_request_fn == dm_make_request) { |
d5b9dd04 MP |
2492 | dev_md = q->queuedata; |
2493 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2494 | return 0; | |
2495 | } | |
2496 | ||
2497 | return 1; | |
2498 | } | |
2499 | ||
2500 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2501 | struct dm_dev *dev, sector_t start, | |
2502 | sector_t len, void *data) | |
2503 | { | |
2504 | struct block_device *bdev = dev->bdev; | |
2505 | struct request_queue *q = bdev_get_queue(bdev); | |
2506 | ||
2507 | return dm_queue_merge_is_compulsory(q); | |
2508 | } | |
2509 | ||
2510 | /* | |
2511 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2512 | * on the properties of the underlying devices. | |
2513 | */ | |
2514 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2515 | { | |
2516 | unsigned i = 0; | |
2517 | struct dm_target *ti; | |
2518 | ||
2519 | while (i < dm_table_get_num_targets(table)) { | |
2520 | ti = dm_table_get_target(table, i++); | |
2521 | ||
2522 | if (ti->type->iterate_devices && | |
2523 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2524 | return 0; | |
2525 | } | |
2526 | ||
2527 | return 1; | |
2528 | } | |
2529 | ||
042d2a9b AK |
2530 | /* |
2531 | * Returns old map, which caller must destroy. | |
2532 | */ | |
2533 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2534 | struct queue_limits *limits) | |
1da177e4 | 2535 | { |
042d2a9b | 2536 | struct dm_table *old_map; |
165125e1 | 2537 | struct request_queue *q = md->queue; |
1da177e4 | 2538 | sector_t size; |
d5b9dd04 | 2539 | int merge_is_optional; |
1da177e4 LT |
2540 | |
2541 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2542 | |
2543 | /* | |
2544 | * Wipe any geometry if the size of the table changed. | |
2545 | */ | |
fd2ed4d2 | 2546 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2547 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2548 | ||
32a926da | 2549 | __set_size(md, size); |
d5816876 | 2550 | |
2ca3310e AK |
2551 | dm_table_event_callback(t, event_callback, md); |
2552 | ||
e6ee8c0b KU |
2553 | /* |
2554 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2555 | * for request-based during suspension. So stop it to prevent | |
2556 | * I/O mapping before resume. | |
2557 | * This must be done before setting the queue restrictions, | |
2558 | * because request-based dm may be run just after the setting. | |
2559 | */ | |
bfebd1cd | 2560 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2561 | stop_queue(q); |
2562 | ||
2563 | __bind_mempools(md, t); | |
2564 | ||
d5b9dd04 MP |
2565 | merge_is_optional = dm_table_merge_is_optional(t); |
2566 | ||
a12f5d48 | 2567 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2568 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2569 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2570 | ||
754c5fc7 | 2571 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2572 | if (merge_is_optional) |
2573 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2574 | else | |
2575 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2576 | if (old_map) |
2577 | dm_sync_table(md); | |
1da177e4 | 2578 | |
042d2a9b | 2579 | return old_map; |
1da177e4 LT |
2580 | } |
2581 | ||
a7940155 AK |
2582 | /* |
2583 | * Returns unbound table for the caller to free. | |
2584 | */ | |
2585 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2586 | { |
a12f5d48 | 2587 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2588 | |
2589 | if (!map) | |
a7940155 | 2590 | return NULL; |
1da177e4 LT |
2591 | |
2592 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2593 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2594 | dm_sync_table(md); |
a7940155 AK |
2595 | |
2596 | return map; | |
1da177e4 LT |
2597 | } |
2598 | ||
2599 | /* | |
2600 | * Constructor for a new device. | |
2601 | */ | |
2b06cfff | 2602 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2603 | { |
2604 | struct mapped_device *md; | |
2605 | ||
2b06cfff | 2606 | md = alloc_dev(minor); |
1da177e4 LT |
2607 | if (!md) |
2608 | return -ENXIO; | |
2609 | ||
784aae73 MB |
2610 | dm_sysfs_init(md); |
2611 | ||
1da177e4 LT |
2612 | *result = md; |
2613 | return 0; | |
2614 | } | |
2615 | ||
a5664dad MS |
2616 | /* |
2617 | * Functions to manage md->type. | |
2618 | * All are required to hold md->type_lock. | |
2619 | */ | |
2620 | void dm_lock_md_type(struct mapped_device *md) | |
2621 | { | |
2622 | mutex_lock(&md->type_lock); | |
2623 | } | |
2624 | ||
2625 | void dm_unlock_md_type(struct mapped_device *md) | |
2626 | { | |
2627 | mutex_unlock(&md->type_lock); | |
2628 | } | |
2629 | ||
2630 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2631 | { | |
00c4fc3b | 2632 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2633 | md->type = type; |
2634 | } | |
2635 | ||
2636 | unsigned dm_get_md_type(struct mapped_device *md) | |
2637 | { | |
00c4fc3b | 2638 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2639 | return md->type; |
2640 | } | |
2641 | ||
36a0456f AK |
2642 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2643 | { | |
2644 | return md->immutable_target_type; | |
2645 | } | |
2646 | ||
f84cb8a4 MS |
2647 | /* |
2648 | * The queue_limits are only valid as long as you have a reference | |
2649 | * count on 'md'. | |
2650 | */ | |
2651 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2652 | { | |
2653 | BUG_ON(!atomic_read(&md->holders)); | |
2654 | return &md->queue->limits; | |
2655 | } | |
2656 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2657 | ||
bfebd1cd MS |
2658 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2659 | { | |
2660 | /* Initialize the request-based DM worker thread */ | |
2661 | init_kthread_worker(&md->kworker); | |
2662 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2663 | "kdmwork-%s", dm_device_name(md)); | |
2664 | } | |
2665 | ||
4a0b4ddf MS |
2666 | /* |
2667 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2668 | */ | |
2669 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2670 | { | |
2671 | struct request_queue *q = NULL; | |
2672 | ||
4a0b4ddf MS |
2673 | /* Fully initialize the queue */ |
2674 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2675 | if (!q) | |
bfebd1cd | 2676 | return -EINVAL; |
4a0b4ddf | 2677 | |
0ce65797 MS |
2678 | /* disable dm_request_fn's merge heuristic by default */ |
2679 | md->seq_rq_merge_deadline_usecs = 0; | |
2680 | ||
4a0b4ddf | 2681 | md->queue = q; |
bfebd1cd | 2682 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2683 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2684 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2685 | |
bfebd1cd | 2686 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2687 | |
4a0b4ddf MS |
2688 | elv_register_queue(md->queue); |
2689 | ||
bfebd1cd MS |
2690 | return 0; |
2691 | } | |
2692 | ||
2693 | static int dm_mq_init_request(void *data, struct request *rq, | |
2694 | unsigned int hctx_idx, unsigned int request_idx, | |
2695 | unsigned int numa_node) | |
2696 | { | |
2697 | struct mapped_device *md = data; | |
2698 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2699 | ||
2700 | /* | |
2701 | * Must initialize md member of tio, otherwise it won't | |
2702 | * be available in dm_mq_queue_rq. | |
2703 | */ | |
2704 | tio->md = md; | |
2705 | ||
2706 | return 0; | |
2707 | } | |
2708 | ||
2709 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2710 | const struct blk_mq_queue_data *bd) | |
2711 | { | |
2712 | struct request *rq = bd->rq; | |
2713 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2714 | struct mapped_device *md = tio->md; | |
2715 | int srcu_idx; | |
2716 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2717 | struct dm_target *ti; | |
2718 | sector_t pos; | |
2719 | ||
2720 | /* always use block 0 to find the target for flushes for now */ | |
2721 | pos = 0; | |
2722 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2723 | pos = blk_rq_pos(rq); | |
2724 | ||
2725 | ti = dm_table_find_target(map, pos); | |
2726 | if (!dm_target_is_valid(ti)) { | |
2727 | dm_put_live_table(md, srcu_idx); | |
2728 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2729 | /* | |
2730 | * Must perform setup, that rq_completed() requires, | |
2731 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2732 | */ | |
2733 | dm_start_request(md, rq); | |
2734 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2735 | } | |
2736 | dm_put_live_table(md, srcu_idx); | |
2737 | ||
2738 | if (ti->type->busy && ti->type->busy(ti)) | |
2739 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2740 | ||
2741 | dm_start_request(md, rq); | |
2742 | ||
2743 | /* Init tio using md established in .init_request */ | |
2744 | init_tio(tio, rq, md); | |
2745 | ||
02233342 MS |
2746 | /* |
2747 | * Establish tio->ti before queuing work (map_tio_request) | |
2748 | * or making direct call to map_request(). | |
2749 | */ | |
bfebd1cd | 2750 | tio->ti = ti; |
02233342 MS |
2751 | |
2752 | /* Clone the request if underlying devices aren't blk-mq */ | |
2753 | if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) { | |
2754 | /* clone request is allocated at the end of the pdu */ | |
2755 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
2756 | if (!clone_rq(rq, md, tio, GFP_ATOMIC)) | |
2757 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2758 | queue_kthread_work(&md->kworker, &tio->work); | |
2759 | } else { | |
2760 | /* Direct call is fine since .queue_rq allows allocations */ | |
2761 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) | |
2762 | dm_requeue_unmapped_original_request(md, rq); | |
2763 | } | |
bfebd1cd MS |
2764 | |
2765 | return BLK_MQ_RQ_QUEUE_OK; | |
2766 | } | |
2767 | ||
2768 | static struct blk_mq_ops dm_mq_ops = { | |
2769 | .queue_rq = dm_mq_queue_rq, | |
2770 | .map_queue = blk_mq_map_queue, | |
2771 | .complete = dm_softirq_done, | |
2772 | .init_request = dm_mq_init_request, | |
2773 | }; | |
2774 | ||
2775 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2776 | { | |
02233342 | 2777 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2778 | struct request_queue *q; |
2779 | int err; | |
2780 | ||
2781 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2782 | md->tag_set.ops = &dm_mq_ops; | |
2783 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2784 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2785 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2786 | md->tag_set.nr_hw_queues = 1; | |
02233342 MS |
2787 | if (md_type == DM_TYPE_REQUEST_BASED) { |
2788 | /* make the memory for non-blk-mq clone part of the pdu */ | |
2789 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request); | |
2790 | } else | |
2791 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd MS |
2792 | md->tag_set.driver_data = md; |
2793 | ||
2794 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2795 | if (err) | |
2796 | return err; | |
2797 | ||
2798 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2799 | if (IS_ERR(q)) { | |
2800 | err = PTR_ERR(q); | |
2801 | goto out_tag_set; | |
2802 | } | |
2803 | md->queue = q; | |
2804 | dm_init_md_queue(md); | |
2805 | ||
2806 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2807 | blk_mq_register_disk(md->disk); | |
2808 | ||
02233342 MS |
2809 | if (md_type == DM_TYPE_REQUEST_BASED) |
2810 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2811 | |
2812 | return 0; | |
2813 | ||
2814 | out_tag_set: | |
2815 | blk_mq_free_tag_set(&md->tag_set); | |
2816 | return err; | |
4a0b4ddf MS |
2817 | } |
2818 | ||
17e149b8 MS |
2819 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2820 | { | |
2821 | if (type == DM_TYPE_BIO_BASED) | |
2822 | return type; | |
2823 | ||
2824 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2825 | } | |
2826 | ||
4a0b4ddf MS |
2827 | /* |
2828 | * Setup the DM device's queue based on md's type | |
2829 | */ | |
2830 | int dm_setup_md_queue(struct mapped_device *md) | |
2831 | { | |
bfebd1cd | 2832 | int r; |
17e149b8 | 2833 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2834 | |
2835 | switch (md_type) { | |
2836 | case DM_TYPE_REQUEST_BASED: | |
2837 | r = dm_init_request_based_queue(md); | |
2838 | if (r) { | |
ff36ab34 | 2839 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2840 | return r; |
ff36ab34 | 2841 | } |
bfebd1cd MS |
2842 | break; |
2843 | case DM_TYPE_MQ_REQUEST_BASED: | |
2844 | r = dm_init_request_based_blk_mq_queue(md); | |
2845 | if (r) { | |
2846 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2847 | return r; | |
2848 | } | |
2849 | break; | |
2850 | case DM_TYPE_BIO_BASED: | |
2851 | dm_init_old_md_queue(md); | |
ff36ab34 MS |
2852 | blk_queue_make_request(md->queue, dm_make_request); |
2853 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); | |
bfebd1cd | 2854 | break; |
4a0b4ddf MS |
2855 | } |
2856 | ||
2857 | return 0; | |
2858 | } | |
2859 | ||
2bec1f4a | 2860 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2861 | { |
2862 | struct mapped_device *md; | |
1da177e4 LT |
2863 | unsigned minor = MINOR(dev); |
2864 | ||
2865 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2866 | return NULL; | |
2867 | ||
f32c10b0 | 2868 | spin_lock(&_minor_lock); |
1da177e4 LT |
2869 | |
2870 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2871 | if (md) { |
2872 | if ((md == MINOR_ALLOCED || | |
2873 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2874 | dm_deleting_md(md) || | |
2875 | test_bit(DMF_FREEING, &md->flags))) { | |
2876 | md = NULL; | |
2877 | goto out; | |
2878 | } | |
2879 | dm_get(md); | |
fba9f90e | 2880 | } |
1da177e4 | 2881 | |
fba9f90e | 2882 | out: |
f32c10b0 | 2883 | spin_unlock(&_minor_lock); |
1da177e4 | 2884 | |
637842cf DT |
2885 | return md; |
2886 | } | |
3cf2e4ba | 2887 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2888 | |
9ade92a9 | 2889 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2890 | { |
9ade92a9 | 2891 | return md->interface_ptr; |
1da177e4 LT |
2892 | } |
2893 | ||
2894 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2895 | { | |
2896 | md->interface_ptr = ptr; | |
2897 | } | |
2898 | ||
2899 | void dm_get(struct mapped_device *md) | |
2900 | { | |
2901 | atomic_inc(&md->holders); | |
3f77316d | 2902 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2903 | } |
2904 | ||
09ee96b2 MP |
2905 | int dm_hold(struct mapped_device *md) |
2906 | { | |
2907 | spin_lock(&_minor_lock); | |
2908 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2909 | spin_unlock(&_minor_lock); | |
2910 | return -EBUSY; | |
2911 | } | |
2912 | dm_get(md); | |
2913 | spin_unlock(&_minor_lock); | |
2914 | return 0; | |
2915 | } | |
2916 | EXPORT_SYMBOL_GPL(dm_hold); | |
2917 | ||
72d94861 AK |
2918 | const char *dm_device_name(struct mapped_device *md) |
2919 | { | |
2920 | return md->name; | |
2921 | } | |
2922 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2923 | ||
3f77316d | 2924 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2925 | { |
1134e5ae | 2926 | struct dm_table *map; |
83d5e5b0 | 2927 | int srcu_idx; |
1da177e4 | 2928 | |
3f77316d | 2929 | might_sleep(); |
fba9f90e | 2930 | |
83d5e5b0 | 2931 | map = dm_get_live_table(md, &srcu_idx); |
63a4f065 MS |
2932 | |
2933 | spin_lock(&_minor_lock); | |
3f77316d KU |
2934 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2935 | set_bit(DMF_FREEING, &md->flags); | |
2936 | spin_unlock(&_minor_lock); | |
2937 | ||
02233342 | 2938 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2939 | flush_kthread_worker(&md->kworker); |
2940 | ||
ab7c7bb6 MP |
2941 | /* |
2942 | * Take suspend_lock so that presuspend and postsuspend methods | |
2943 | * do not race with internal suspend. | |
2944 | */ | |
2945 | mutex_lock(&md->suspend_lock); | |
3f77316d KU |
2946 | if (!dm_suspended_md(md)) { |
2947 | dm_table_presuspend_targets(map); | |
2948 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2949 | } |
ab7c7bb6 | 2950 | mutex_unlock(&md->suspend_lock); |
3f77316d | 2951 | |
83d5e5b0 MP |
2952 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2953 | dm_put_live_table(md, srcu_idx); | |
2954 | ||
3f77316d KU |
2955 | /* |
2956 | * Rare, but there may be I/O requests still going to complete, | |
2957 | * for example. Wait for all references to disappear. | |
2958 | * No one should increment the reference count of the mapped_device, | |
2959 | * after the mapped_device state becomes DMF_FREEING. | |
2960 | */ | |
2961 | if (wait) | |
2962 | while (atomic_read(&md->holders)) | |
2963 | msleep(1); | |
2964 | else if (atomic_read(&md->holders)) | |
2965 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2966 | dm_device_name(md), atomic_read(&md->holders)); | |
2967 | ||
2968 | dm_sysfs_exit(md); | |
3f77316d KU |
2969 | dm_table_destroy(__unbind(md)); |
2970 | free_dev(md); | |
2971 | } | |
2972 | ||
2973 | void dm_destroy(struct mapped_device *md) | |
2974 | { | |
2975 | __dm_destroy(md, true); | |
2976 | } | |
2977 | ||
2978 | void dm_destroy_immediate(struct mapped_device *md) | |
2979 | { | |
2980 | __dm_destroy(md, false); | |
2981 | } | |
2982 | ||
2983 | void dm_put(struct mapped_device *md) | |
2984 | { | |
2985 | atomic_dec(&md->holders); | |
1da177e4 | 2986 | } |
79eb885c | 2987 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2988 | |
401600df | 2989 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2990 | { |
2991 | int r = 0; | |
b44ebeb0 MP |
2992 | DECLARE_WAITQUEUE(wait, current); |
2993 | ||
b44ebeb0 | 2994 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2995 | |
2996 | while (1) { | |
401600df | 2997 | set_current_state(interruptible); |
46125c1c | 2998 | |
b4324fee | 2999 | if (!md_in_flight(md)) |
46125c1c MB |
3000 | break; |
3001 | ||
401600df MP |
3002 | if (interruptible == TASK_INTERRUPTIBLE && |
3003 | signal_pending(current)) { | |
46125c1c MB |
3004 | r = -EINTR; |
3005 | break; | |
3006 | } | |
3007 | ||
3008 | io_schedule(); | |
3009 | } | |
3010 | set_current_state(TASK_RUNNING); | |
3011 | ||
b44ebeb0 MP |
3012 | remove_wait_queue(&md->wait, &wait); |
3013 | ||
46125c1c MB |
3014 | return r; |
3015 | } | |
3016 | ||
1da177e4 LT |
3017 | /* |
3018 | * Process the deferred bios | |
3019 | */ | |
ef208587 | 3020 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 3021 | { |
ef208587 MP |
3022 | struct mapped_device *md = container_of(work, struct mapped_device, |
3023 | work); | |
6d6f10df | 3024 | struct bio *c; |
83d5e5b0 MP |
3025 | int srcu_idx; |
3026 | struct dm_table *map; | |
1da177e4 | 3027 | |
83d5e5b0 | 3028 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 3029 | |
3b00b203 | 3030 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
3031 | spin_lock_irq(&md->deferred_lock); |
3032 | c = bio_list_pop(&md->deferred); | |
3033 | spin_unlock_irq(&md->deferred_lock); | |
3034 | ||
6a8736d1 | 3035 | if (!c) |
df12ee99 | 3036 | break; |
022c2611 | 3037 | |
e6ee8c0b KU |
3038 | if (dm_request_based(md)) |
3039 | generic_make_request(c); | |
6a8736d1 | 3040 | else |
83d5e5b0 | 3041 | __split_and_process_bio(md, map, c); |
022c2611 | 3042 | } |
73d410c0 | 3043 | |
83d5e5b0 | 3044 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
3045 | } |
3046 | ||
9a1fb464 | 3047 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 3048 | { |
3b00b203 | 3049 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 3050 | smp_mb__after_atomic(); |
53d5914f | 3051 | queue_work(md->wq, &md->work); |
304f3f6a MB |
3052 | } |
3053 | ||
1da177e4 | 3054 | /* |
042d2a9b | 3055 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 3056 | */ |
042d2a9b | 3057 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3058 | { |
87eb5b21 | 3059 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3060 | struct queue_limits limits; |
042d2a9b | 3061 | int r; |
1da177e4 | 3062 | |
e61290a4 | 3063 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3064 | |
3065 | /* device must be suspended */ | |
4f186f8b | 3066 | if (!dm_suspended_md(md)) |
93c534ae | 3067 | goto out; |
1da177e4 | 3068 | |
3ae70656 MS |
3069 | /* |
3070 | * If the new table has no data devices, retain the existing limits. | |
3071 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3072 | * then new I/O is queued based on these limits, and then some paths | |
3073 | * reappear. | |
3074 | */ | |
3075 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3076 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3077 | if (live_map) |
3078 | limits = md->queue->limits; | |
83d5e5b0 | 3079 | dm_put_live_table_fast(md); |
3ae70656 MS |
3080 | } |
3081 | ||
87eb5b21 MC |
3082 | if (!live_map) { |
3083 | r = dm_calculate_queue_limits(table, &limits); | |
3084 | if (r) { | |
3085 | map = ERR_PTR(r); | |
3086 | goto out; | |
3087 | } | |
042d2a9b | 3088 | } |
754c5fc7 | 3089 | |
042d2a9b | 3090 | map = __bind(md, table, &limits); |
1da177e4 | 3091 | |
93c534ae | 3092 | out: |
e61290a4 | 3093 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3094 | return map; |
1da177e4 LT |
3095 | } |
3096 | ||
3097 | /* | |
3098 | * Functions to lock and unlock any filesystem running on the | |
3099 | * device. | |
3100 | */ | |
2ca3310e | 3101 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3102 | { |
e39e2e95 | 3103 | int r; |
1da177e4 LT |
3104 | |
3105 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3106 | |
db8fef4f | 3107 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3108 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3109 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3110 | md->frozen_sb = NULL; |
3111 | return r; | |
dfbe03f6 AK |
3112 | } |
3113 | ||
aa8d7c2f AK |
3114 | set_bit(DMF_FROZEN, &md->flags); |
3115 | ||
1da177e4 LT |
3116 | return 0; |
3117 | } | |
3118 | ||
2ca3310e | 3119 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3120 | { |
aa8d7c2f AK |
3121 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3122 | return; | |
3123 | ||
db8fef4f | 3124 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3125 | md->frozen_sb = NULL; |
aa8d7c2f | 3126 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3127 | } |
3128 | ||
3129 | /* | |
ffcc3936 MS |
3130 | * If __dm_suspend returns 0, the device is completely quiescent |
3131 | * now. There is no request-processing activity. All new requests | |
3132 | * are being added to md->deferred list. | |
cec47e3d | 3133 | * |
ffcc3936 | 3134 | * Caller must hold md->suspend_lock |
cec47e3d | 3135 | */ |
ffcc3936 MS |
3136 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3137 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3138 | { |
ffcc3936 MS |
3139 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3140 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3141 | int r; | |
1da177e4 | 3142 | |
2e93ccc1 KU |
3143 | /* |
3144 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3145 | * This flag is cleared before dm_suspend returns. | |
3146 | */ | |
3147 | if (noflush) | |
3148 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3149 | ||
d67ee213 MS |
3150 | /* |
3151 | * This gets reverted if there's an error later and the targets | |
3152 | * provide the .presuspend_undo hook. | |
3153 | */ | |
cf222b37 AK |
3154 | dm_table_presuspend_targets(map); |
3155 | ||
32a926da | 3156 | /* |
9f518b27 KU |
3157 | * Flush I/O to the device. |
3158 | * Any I/O submitted after lock_fs() may not be flushed. | |
3159 | * noflush takes precedence over do_lockfs. | |
3160 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3161 | */ |
3162 | if (!noflush && do_lockfs) { | |
3163 | r = lock_fs(md); | |
d67ee213 MS |
3164 | if (r) { |
3165 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3166 | return r; |
d67ee213 | 3167 | } |
aa8d7c2f | 3168 | } |
1da177e4 LT |
3169 | |
3170 | /* | |
3b00b203 MP |
3171 | * Here we must make sure that no processes are submitting requests |
3172 | * to target drivers i.e. no one may be executing | |
3173 | * __split_and_process_bio. This is called from dm_request and | |
3174 | * dm_wq_work. | |
3175 | * | |
3176 | * To get all processes out of __split_and_process_bio in dm_request, | |
3177 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3178 | * __split_and_process_bio from dm_request and quiesce the thread |
3179 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3180 | * flush_workqueue(md->wq). | |
1da177e4 | 3181 | */ |
1eb787ec | 3182 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3183 | if (map) |
3184 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3185 | |
d0bcb878 | 3186 | /* |
29e4013d TH |
3187 | * Stop md->queue before flushing md->wq in case request-based |
3188 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3189 | */ |
2eb6e1e3 | 3190 | if (dm_request_based(md)) { |
9f518b27 | 3191 | stop_queue(md->queue); |
02233342 MS |
3192 | if (md->kworker_task) |
3193 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3194 | } |
cec47e3d | 3195 | |
d0bcb878 KU |
3196 | flush_workqueue(md->wq); |
3197 | ||
1da177e4 | 3198 | /* |
3b00b203 MP |
3199 | * At this point no more requests are entering target request routines. |
3200 | * We call dm_wait_for_completion to wait for all existing requests | |
3201 | * to finish. | |
1da177e4 | 3202 | */ |
ffcc3936 | 3203 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3204 | |
6d6f10df | 3205 | if (noflush) |
022c2611 | 3206 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3207 | if (map) |
3208 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3209 | |
1da177e4 | 3210 | /* were we interrupted ? */ |
46125c1c | 3211 | if (r < 0) { |
9a1fb464 | 3212 | dm_queue_flush(md); |
73d410c0 | 3213 | |
cec47e3d | 3214 | if (dm_request_based(md)) |
9f518b27 | 3215 | start_queue(md->queue); |
cec47e3d | 3216 | |
2ca3310e | 3217 | unlock_fs(md); |
d67ee213 | 3218 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3219 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3220 | } |
1da177e4 | 3221 | |
ffcc3936 MS |
3222 | return r; |
3223 | } | |
3224 | ||
3225 | /* | |
3226 | * We need to be able to change a mapping table under a mounted | |
3227 | * filesystem. For example we might want to move some data in | |
3228 | * the background. Before the table can be swapped with | |
3229 | * dm_bind_table, dm_suspend must be called to flush any in | |
3230 | * flight bios and ensure that any further io gets deferred. | |
3231 | */ | |
3232 | /* | |
3233 | * Suspend mechanism in request-based dm. | |
3234 | * | |
3235 | * 1. Flush all I/Os by lock_fs() if needed. | |
3236 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3237 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3238 | * | |
3239 | * To abort suspend, start the request_queue. | |
3240 | */ | |
3241 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3242 | { | |
3243 | struct dm_table *map = NULL; | |
3244 | int r = 0; | |
3245 | ||
3246 | retry: | |
3247 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3248 | ||
3249 | if (dm_suspended_md(md)) { | |
3250 | r = -EINVAL; | |
3251 | goto out_unlock; | |
3252 | } | |
3253 | ||
3254 | if (dm_suspended_internally_md(md)) { | |
3255 | /* already internally suspended, wait for internal resume */ | |
3256 | mutex_unlock(&md->suspend_lock); | |
3257 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3258 | if (r) | |
3259 | return r; | |
3260 | goto retry; | |
3261 | } | |
3262 | ||
a12f5d48 | 3263 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3264 | |
3265 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3266 | if (r) | |
3267 | goto out_unlock; | |
3b00b203 | 3268 | |
2ca3310e | 3269 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3270 | |
4d4471cb KU |
3271 | dm_table_postsuspend_targets(map); |
3272 | ||
d287483d | 3273 | out_unlock: |
e61290a4 | 3274 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3275 | return r; |
1da177e4 LT |
3276 | } |
3277 | ||
ffcc3936 MS |
3278 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3279 | { | |
3280 | if (map) { | |
3281 | int r = dm_table_resume_targets(map); | |
3282 | if (r) | |
3283 | return r; | |
3284 | } | |
3285 | ||
3286 | dm_queue_flush(md); | |
3287 | ||
3288 | /* | |
3289 | * Flushing deferred I/Os must be done after targets are resumed | |
3290 | * so that mapping of targets can work correctly. | |
3291 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3292 | */ | |
3293 | if (dm_request_based(md)) | |
3294 | start_queue(md->queue); | |
3295 | ||
3296 | unlock_fs(md); | |
3297 | ||
3298 | return 0; | |
3299 | } | |
3300 | ||
1da177e4 LT |
3301 | int dm_resume(struct mapped_device *md) |
3302 | { | |
cf222b37 | 3303 | int r = -EINVAL; |
cf222b37 | 3304 | struct dm_table *map = NULL; |
1da177e4 | 3305 | |
ffcc3936 MS |
3306 | retry: |
3307 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3308 | ||
4f186f8b | 3309 | if (!dm_suspended_md(md)) |
cf222b37 | 3310 | goto out; |
cf222b37 | 3311 | |
ffcc3936 MS |
3312 | if (dm_suspended_internally_md(md)) { |
3313 | /* already internally suspended, wait for internal resume */ | |
3314 | mutex_unlock(&md->suspend_lock); | |
3315 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3316 | if (r) | |
3317 | return r; | |
3318 | goto retry; | |
3319 | } | |
3320 | ||
a12f5d48 | 3321 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3322 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3323 | goto out; |
1da177e4 | 3324 | |
ffcc3936 | 3325 | r = __dm_resume(md, map); |
8757b776 MB |
3326 | if (r) |
3327 | goto out; | |
2ca3310e | 3328 | |
2ca3310e AK |
3329 | clear_bit(DMF_SUSPENDED, &md->flags); |
3330 | ||
cf222b37 AK |
3331 | r = 0; |
3332 | out: | |
e61290a4 | 3333 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3334 | |
cf222b37 | 3335 | return r; |
1da177e4 LT |
3336 | } |
3337 | ||
fd2ed4d2 MP |
3338 | /* |
3339 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3340 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3341 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3342 | */ |
3343 | ||
ffcc3936 | 3344 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3345 | { |
ffcc3936 MS |
3346 | struct dm_table *map = NULL; |
3347 | ||
96b26c8c | 3348 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3349 | return; /* nested internal suspend */ |
3350 | ||
3351 | if (dm_suspended_md(md)) { | |
3352 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3353 | return; /* nest suspend */ | |
3354 | } | |
3355 | ||
a12f5d48 | 3356 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3357 | |
3358 | /* | |
3359 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3360 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3361 | * would require changing .presuspend to return an error -- avoid this | |
3362 | * until there is a need for more elaborate variants of internal suspend. | |
3363 | */ | |
3364 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3365 | ||
3366 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3367 | ||
3368 | dm_table_postsuspend_targets(map); | |
3369 | } | |
3370 | ||
3371 | static void __dm_internal_resume(struct mapped_device *md) | |
3372 | { | |
96b26c8c MP |
3373 | BUG_ON(!md->internal_suspend_count); |
3374 | ||
3375 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3376 | return; /* resume from nested internal suspend */ |
3377 | ||
fd2ed4d2 | 3378 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3379 | goto done; /* resume from nested suspend */ |
3380 | ||
3381 | /* | |
3382 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3383 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3384 | */ | |
3385 | (void) __dm_resume(md, NULL); | |
3386 | ||
3387 | done: | |
3388 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3389 | smp_mb__after_atomic(); | |
3390 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3391 | } | |
3392 | ||
3393 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3394 | { | |
3395 | mutex_lock(&md->suspend_lock); | |
3396 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3397 | mutex_unlock(&md->suspend_lock); | |
3398 | } | |
3399 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3400 | ||
3401 | void dm_internal_resume(struct mapped_device *md) | |
3402 | { | |
3403 | mutex_lock(&md->suspend_lock); | |
3404 | __dm_internal_resume(md); | |
3405 | mutex_unlock(&md->suspend_lock); | |
3406 | } | |
3407 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3408 | ||
3409 | /* | |
3410 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3411 | * which prevents interaction with userspace-driven suspend. | |
3412 | */ | |
3413 | ||
3414 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3415 | { | |
3416 | mutex_lock(&md->suspend_lock); | |
3417 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3418 | return; |
3419 | ||
3420 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3421 | synchronize_srcu(&md->io_barrier); | |
3422 | flush_workqueue(md->wq); | |
3423 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3424 | } | |
b735fede | 3425 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3426 | |
ffcc3936 | 3427 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3428 | { |
ffcc3936 | 3429 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3430 | goto done; |
3431 | ||
3432 | dm_queue_flush(md); | |
3433 | ||
3434 | done: | |
3435 | mutex_unlock(&md->suspend_lock); | |
3436 | } | |
b735fede | 3437 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3438 | |
1da177e4 LT |
3439 | /*----------------------------------------------------------------- |
3440 | * Event notification. | |
3441 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3442 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3443 | unsigned cookie) |
69267a30 | 3444 | { |
60935eb2 MB |
3445 | char udev_cookie[DM_COOKIE_LENGTH]; |
3446 | char *envp[] = { udev_cookie, NULL }; | |
3447 | ||
3448 | if (!cookie) | |
3abf85b5 | 3449 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3450 | else { |
3451 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3452 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3453 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3454 | action, envp); | |
60935eb2 | 3455 | } |
69267a30 AK |
3456 | } |
3457 | ||
7a8c3d3b MA |
3458 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3459 | { | |
3460 | return atomic_add_return(1, &md->uevent_seq); | |
3461 | } | |
3462 | ||
1da177e4 LT |
3463 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3464 | { | |
3465 | return atomic_read(&md->event_nr); | |
3466 | } | |
3467 | ||
3468 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3469 | { | |
3470 | return wait_event_interruptible(md->eventq, | |
3471 | (event_nr != atomic_read(&md->event_nr))); | |
3472 | } | |
3473 | ||
7a8c3d3b MA |
3474 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3475 | { | |
3476 | unsigned long flags; | |
3477 | ||
3478 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3479 | list_add(elist, &md->uevent_list); | |
3480 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3481 | } | |
3482 | ||
1da177e4 LT |
3483 | /* |
3484 | * The gendisk is only valid as long as you have a reference | |
3485 | * count on 'md'. | |
3486 | */ | |
3487 | struct gendisk *dm_disk(struct mapped_device *md) | |
3488 | { | |
3489 | return md->disk; | |
3490 | } | |
65ff5b7d | 3491 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3492 | |
784aae73 MB |
3493 | struct kobject *dm_kobject(struct mapped_device *md) |
3494 | { | |
2995fa78 | 3495 | return &md->kobj_holder.kobj; |
784aae73 MB |
3496 | } |
3497 | ||
784aae73 MB |
3498 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3499 | { | |
3500 | struct mapped_device *md; | |
3501 | ||
2995fa78 | 3502 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3503 | |
4d89b7b4 | 3504 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3505 | dm_deleting_md(md)) |
4d89b7b4 MB |
3506 | return NULL; |
3507 | ||
784aae73 MB |
3508 | dm_get(md); |
3509 | return md; | |
3510 | } | |
3511 | ||
4f186f8b | 3512 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3513 | { |
3514 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3515 | } | |
3516 | ||
ffcc3936 MS |
3517 | int dm_suspended_internally_md(struct mapped_device *md) |
3518 | { | |
3519 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3520 | } | |
3521 | ||
2c140a24 MP |
3522 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3523 | { | |
3524 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3525 | } | |
3526 | ||
64dbce58 KU |
3527 | int dm_suspended(struct dm_target *ti) |
3528 | { | |
ecdb2e25 | 3529 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3530 | } |
3531 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3532 | ||
2e93ccc1 KU |
3533 | int dm_noflush_suspending(struct dm_target *ti) |
3534 | { | |
ecdb2e25 | 3535 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3536 | } |
3537 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3538 | ||
17e149b8 MS |
3539 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3540 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3541 | { |
5f015204 | 3542 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
17e149b8 | 3543 | struct kmem_cache *cachep = NULL; |
e5863d9a | 3544 | unsigned int pool_size = 0; |
5f015204 | 3545 | unsigned int front_pad; |
e6ee8c0b KU |
3546 | |
3547 | if (!pools) | |
3548 | return NULL; | |
3549 | ||
17e149b8 MS |
3550 | type = filter_md_type(type, md); |
3551 | ||
e5863d9a MS |
3552 | switch (type) { |
3553 | case DM_TYPE_BIO_BASED: | |
5f015204 | 3554 | cachep = _io_cache; |
e8603136 | 3555 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 | 3556 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
e5863d9a MS |
3557 | break; |
3558 | case DM_TYPE_REQUEST_BASED: | |
17e149b8 | 3559 | cachep = _rq_tio_cache; |
f4790826 | 3560 | pool_size = dm_get_reserved_rq_based_ios(); |
1ae49ea2 MS |
3561 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); |
3562 | if (!pools->rq_pool) | |
3563 | goto out; | |
e5863d9a MS |
3564 | /* fall through to setup remaining rq-based pools */ |
3565 | case DM_TYPE_MQ_REQUEST_BASED: | |
e5863d9a MS |
3566 | if (!pool_size) |
3567 | pool_size = dm_get_reserved_rq_based_ios(); | |
5f015204 JN |
3568 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3569 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3570 | WARN_ON(per_bio_data_size != 0); | |
e5863d9a MS |
3571 | break; |
3572 | default: | |
17e149b8 | 3573 | BUG(); |
e5863d9a | 3574 | } |
e6ee8c0b | 3575 | |
17e149b8 MS |
3576 | if (cachep) { |
3577 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3578 | if (!pools->io_pool) | |
3579 | goto out; | |
3580 | } | |
e6ee8c0b | 3581 | |
3d8aab2d | 3582 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3583 | if (!pools->bs) |
5f015204 | 3584 | goto out; |
e6ee8c0b | 3585 | |
a91a2785 | 3586 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3587 | goto out; |
a91a2785 | 3588 | |
e6ee8c0b KU |
3589 | return pools; |
3590 | ||
5f015204 JN |
3591 | out: |
3592 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3593 | |
3594 | return NULL; | |
3595 | } | |
3596 | ||
3597 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3598 | { | |
3599 | if (!pools) | |
3600 | return; | |
3601 | ||
3602 | if (pools->io_pool) | |
3603 | mempool_destroy(pools->io_pool); | |
3604 | ||
1ae49ea2 MS |
3605 | if (pools->rq_pool) |
3606 | mempool_destroy(pools->rq_pool); | |
3607 | ||
e6ee8c0b KU |
3608 | if (pools->bs) |
3609 | bioset_free(pools->bs); | |
3610 | ||
3611 | kfree(pools); | |
3612 | } | |
3613 | ||
83d5cde4 | 3614 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3615 | .open = dm_blk_open, |
3616 | .release = dm_blk_close, | |
aa129a22 | 3617 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3618 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3619 | .owner = THIS_MODULE |
3620 | }; | |
3621 | ||
1da177e4 LT |
3622 | /* |
3623 | * module hooks | |
3624 | */ | |
3625 | module_init(dm_init); | |
3626 | module_exit(dm_exit); | |
3627 | ||
3628 | module_param(major, uint, 0); | |
3629 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3630 | |
e8603136 MS |
3631 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3632 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3633 | ||
f4790826 MS |
3634 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3635 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3636 | ||
17e149b8 MS |
3637 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3638 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3639 | ||
1da177e4 LT |
3640 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3641 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3642 | MODULE_LICENSE("GPL"); |