]>
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 | ||
a77e28c7 KU |
1085 | static void free_rq_clone(struct request *clone) |
1086 | { | |
1087 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1088 | struct mapped_device *md = tio->md; |
a77e28c7 KU |
1089 | |
1090 | blk_rq_unprep_clone(clone); | |
bfebd1cd | 1091 | |
02233342 | 1092 | if (clone->q->mq_ops) |
e5863d9a | 1093 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1094 | else if (!md->queue->mq_ops) |
1095 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd MS |
1096 | free_clone_request(md, clone); |
1097 | ||
1098 | if (!md->queue->mq_ops) | |
1099 | free_rq_tio(tio); | |
a77e28c7 KU |
1100 | } |
1101 | ||
980691e5 KU |
1102 | /* |
1103 | * Complete the clone and the original request. | |
466d89a6 KB |
1104 | * Must be called without clone's queue lock held, |
1105 | * see end_clone_request() for more details. | |
980691e5 KU |
1106 | */ |
1107 | static void dm_end_request(struct request *clone, int error) | |
1108 | { | |
1109 | int rw = rq_data_dir(clone); | |
1110 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1111 | struct mapped_device *md = tio->md; | |
1112 | struct request *rq = tio->orig; | |
1113 | ||
29e4013d | 1114 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1115 | rq->errors = clone->errors; |
1116 | rq->resid_len = clone->resid_len; | |
1117 | ||
1118 | if (rq->sense) | |
1119 | /* | |
1120 | * We are using the sense buffer of the original | |
1121 | * request. | |
1122 | * So setting the length of the sense data is enough. | |
1123 | */ | |
1124 | rq->sense_len = clone->sense_len; | |
1125 | } | |
1126 | ||
1127 | free_rq_clone(clone); | |
bfebd1cd MS |
1128 | if (!rq->q->mq_ops) |
1129 | blk_end_request_all(rq, error); | |
1130 | else | |
1131 | blk_mq_end_request(rq, error); | |
29e4013d | 1132 | rq_completed(md, rw, true); |
980691e5 KU |
1133 | } |
1134 | ||
cec47e3d KU |
1135 | static void dm_unprep_request(struct request *rq) |
1136 | { | |
bfebd1cd | 1137 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1138 | struct request *clone = tio->clone; |
cec47e3d | 1139 | |
bfebd1cd MS |
1140 | if (!rq->q->mq_ops) { |
1141 | rq->special = NULL; | |
1142 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1143 | } | |
cec47e3d | 1144 | |
e5863d9a MS |
1145 | if (clone) |
1146 | free_rq_clone(clone); | |
cec47e3d KU |
1147 | } |
1148 | ||
1149 | /* | |
1150 | * Requeue the original request of a clone. | |
1151 | */ | |
bfebd1cd | 1152 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1153 | { |
cec47e3d KU |
1154 | struct request_queue *q = rq->q; |
1155 | unsigned long flags; | |
1156 | ||
cec47e3d | 1157 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d KU |
1158 | blk_requeue_request(q, rq); |
1159 | spin_unlock_irqrestore(q->queue_lock, flags); | |
bfebd1cd MS |
1160 | } |
1161 | ||
1162 | static void dm_requeue_unmapped_original_request(struct mapped_device *md, | |
1163 | struct request *rq) | |
1164 | { | |
1165 | int rw = rq_data_dir(rq); | |
1166 | ||
1167 | dm_unprep_request(rq); | |
1168 | ||
1169 | if (!rq->q->mq_ops) | |
1170 | old_requeue_request(rq); | |
1171 | else { | |
1172 | blk_mq_requeue_request(rq); | |
1173 | blk_mq_kick_requeue_list(rq->q); | |
1174 | } | |
cec47e3d | 1175 | |
466d89a6 KB |
1176 | rq_completed(md, rw, false); |
1177 | } | |
1178 | ||
1179 | static void dm_requeue_unmapped_request(struct request *clone) | |
1180 | { | |
1181 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1182 | ||
1183 | dm_requeue_unmapped_original_request(tio->md, tio->orig); | |
cec47e3d | 1184 | } |
cec47e3d | 1185 | |
bfebd1cd | 1186 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1187 | { |
1188 | unsigned long flags; | |
1189 | ||
bfebd1cd MS |
1190 | if (blk_queue_stopped(q)) |
1191 | return; | |
1192 | ||
cec47e3d | 1193 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1194 | blk_stop_queue(q); |
cec47e3d KU |
1195 | spin_unlock_irqrestore(q->queue_lock, flags); |
1196 | } | |
1197 | ||
bfebd1cd | 1198 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1199 | { |
bfebd1cd MS |
1200 | if (!q->mq_ops) |
1201 | old_stop_queue(q); | |
1202 | else | |
1203 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1204 | } |
1205 | ||
bfebd1cd | 1206 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1207 | { |
1208 | unsigned long flags; | |
1209 | ||
1210 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1211 | if (blk_queue_stopped(q)) |
1212 | blk_start_queue(q); | |
cec47e3d KU |
1213 | spin_unlock_irqrestore(q->queue_lock, flags); |
1214 | } | |
1215 | ||
bfebd1cd MS |
1216 | static void start_queue(struct request_queue *q) |
1217 | { | |
1218 | if (!q->mq_ops) | |
1219 | old_start_queue(q); | |
1220 | else | |
1221 | blk_mq_start_stopped_hw_queues(q, true); | |
1222 | } | |
1223 | ||
11a68244 | 1224 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1225 | { |
11a68244 | 1226 | int r = error; |
cec47e3d | 1227 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1228 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1229 | |
ba1cbad9 MS |
1230 | if (tio->ti) { |
1231 | rq_end_io = tio->ti->type->rq_end_io; | |
1232 | ||
1233 | if (mapped && rq_end_io) | |
1234 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1235 | } | |
cec47e3d | 1236 | |
7eee4ae2 MS |
1237 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1238 | !clone->q->limits.max_write_same_sectors)) | |
1239 | disable_write_same(tio->md); | |
1240 | ||
11a68244 | 1241 | if (r <= 0) |
cec47e3d | 1242 | /* The target wants to complete the I/O */ |
11a68244 KU |
1243 | dm_end_request(clone, r); |
1244 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1245 | /* The target will handle the I/O */ |
1246 | return; | |
11a68244 | 1247 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1248 | /* The target wants to requeue the I/O */ |
1249 | dm_requeue_unmapped_request(clone); | |
1250 | else { | |
11a68244 | 1251 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1252 | BUG(); |
1253 | } | |
1254 | } | |
1255 | ||
11a68244 KU |
1256 | /* |
1257 | * Request completion handler for request-based dm | |
1258 | */ | |
1259 | static void dm_softirq_done(struct request *rq) | |
1260 | { | |
1261 | bool mapped = true; | |
bfebd1cd | 1262 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1263 | struct request *clone = tio->clone; |
bfebd1cd | 1264 | int rw; |
11a68244 | 1265 | |
e5863d9a | 1266 | if (!clone) { |
bfebd1cd MS |
1267 | rw = rq_data_dir(rq); |
1268 | if (!rq->q->mq_ops) { | |
1269 | blk_end_request_all(rq, tio->error); | |
1270 | rq_completed(tio->md, rw, false); | |
1271 | free_rq_tio(tio); | |
1272 | } else { | |
1273 | blk_mq_end_request(rq, tio->error); | |
1274 | rq_completed(tio->md, rw, false); | |
1275 | } | |
e5863d9a MS |
1276 | return; |
1277 | } | |
11a68244 KU |
1278 | |
1279 | if (rq->cmd_flags & REQ_FAILED) | |
1280 | mapped = false; | |
1281 | ||
1282 | dm_done(clone, tio->error, mapped); | |
1283 | } | |
1284 | ||
cec47e3d KU |
1285 | /* |
1286 | * Complete the clone and the original request with the error status | |
1287 | * through softirq context. | |
1288 | */ | |
466d89a6 | 1289 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1290 | { |
bfebd1cd | 1291 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1292 | |
1293 | tio->error = error; | |
cec47e3d KU |
1294 | blk_complete_request(rq); |
1295 | } | |
1296 | ||
1297 | /* | |
1298 | * Complete the not-mapped clone and the original request with the error status | |
1299 | * through softirq context. | |
1300 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1301 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1302 | */ |
466d89a6 | 1303 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1304 | { |
cec47e3d | 1305 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1306 | dm_complete_request(rq, error); |
cec47e3d | 1307 | } |
cec47e3d KU |
1308 | |
1309 | /* | |
bfebd1cd | 1310 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1311 | */ |
1312 | static void end_clone_request(struct request *clone, int error) | |
1313 | { | |
466d89a6 KB |
1314 | struct dm_rq_target_io *tio = clone->end_io_data; |
1315 | ||
e5863d9a MS |
1316 | if (!clone->q->mq_ops) { |
1317 | /* | |
1318 | * For just cleaning up the information of the queue in which | |
1319 | * the clone was dispatched. | |
1320 | * The clone is *NOT* freed actually here because it is alloced | |
1321 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1322 | */ | |
1323 | __blk_put_request(clone->q, clone); | |
1324 | } | |
cec47e3d KU |
1325 | |
1326 | /* | |
1327 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1328 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1329 | * - another request may be submitted by the upper level driver |
1330 | * of the stacking during the completion | |
1331 | * - the submission which requires queue lock may be done | |
466d89a6 | 1332 | * against this clone's queue |
cec47e3d | 1333 | */ |
466d89a6 | 1334 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1335 | } |
1336 | ||
56a67df7 MS |
1337 | /* |
1338 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1339 | * target boundary. | |
1340 | */ | |
1341 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1342 | { | |
1343 | sector_t target_offset = dm_target_offset(ti, sector); | |
1344 | ||
1345 | return ti->len - target_offset; | |
1346 | } | |
1347 | ||
1348 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1349 | { |
56a67df7 | 1350 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1351 | sector_t offset, max_len; |
1da177e4 LT |
1352 | |
1353 | /* | |
542f9038 | 1354 | * Does the target need to split even further? |
1da177e4 | 1355 | */ |
542f9038 MS |
1356 | if (ti->max_io_len) { |
1357 | offset = dm_target_offset(ti, sector); | |
1358 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1359 | max_len = sector_div(offset, ti->max_io_len); | |
1360 | else | |
1361 | max_len = offset & (ti->max_io_len - 1); | |
1362 | max_len = ti->max_io_len - max_len; | |
1363 | ||
1364 | if (len > max_len) | |
1365 | len = max_len; | |
1da177e4 LT |
1366 | } |
1367 | ||
1368 | return len; | |
1369 | } | |
1370 | ||
542f9038 MS |
1371 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1372 | { | |
1373 | if (len > UINT_MAX) { | |
1374 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1375 | (unsigned long long)len, UINT_MAX); | |
1376 | ti->error = "Maximum size of target IO is too large"; | |
1377 | return -EINVAL; | |
1378 | } | |
1379 | ||
1380 | ti->max_io_len = (uint32_t) len; | |
1381 | ||
1382 | return 0; | |
1383 | } | |
1384 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1385 | ||
1dd40c3e MP |
1386 | /* |
1387 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1388 | * allowed for all bio types except REQ_FLUSH. | |
1389 | * | |
1390 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1391 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1392 | * sent in a next bio. | |
1393 | * | |
1394 | * A diagram that explains the arithmetics: | |
1395 | * +--------------------+---------------+-------+ | |
1396 | * | 1 | 2 | 3 | | |
1397 | * +--------------------+---------------+-------+ | |
1398 | * | |
1399 | * <-------------- *tio->len_ptr ---------------> | |
1400 | * <------- bi_size -------> | |
1401 | * <-- n_sectors --> | |
1402 | * | |
1403 | * Region 1 was already iterated over with bio_advance or similar function. | |
1404 | * (it may be empty if the target doesn't use bio_advance) | |
1405 | * Region 2 is the remaining bio size that the target wants to process. | |
1406 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1407 | * to make it empty) | |
1408 | * The target requires that region 3 is to be sent in the next bio. | |
1409 | * | |
1410 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1411 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1412 | * copies of the bio. | |
1413 | */ | |
1414 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1415 | { | |
1416 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1417 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1418 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1419 | BUG_ON(bi_size > *tio->len_ptr); | |
1420 | BUG_ON(n_sectors > bi_size); | |
1421 | *tio->len_ptr -= bi_size - n_sectors; | |
1422 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1423 | } | |
1424 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1425 | ||
bd2a49b8 | 1426 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1427 | { |
1428 | int r; | |
2056a782 | 1429 | sector_t sector; |
9faf400f | 1430 | struct mapped_device *md; |
dba14160 | 1431 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1432 | struct dm_target *ti = tio->ti; |
1da177e4 | 1433 | |
1da177e4 | 1434 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1435 | |
1436 | /* | |
1437 | * Map the clone. If r == 0 we don't need to do | |
1438 | * anything, the target has assumed ownership of | |
1439 | * this io. | |
1440 | */ | |
1441 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1442 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1443 | r = ti->type->map(ti, clone); |
45cbcd79 | 1444 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1445 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1446 | |
d07335e5 MS |
1447 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1448 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1449 | |
1da177e4 | 1450 | generic_make_request(clone); |
2e93ccc1 KU |
1451 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1452 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1453 | md = tio->io->md; |
1454 | dec_pending(tio->io, r); | |
9faf400f | 1455 | free_tio(md, tio); |
45cbcd79 KU |
1456 | } else if (r) { |
1457 | DMWARN("unimplemented target map return value: %d", r); | |
1458 | BUG(); | |
1da177e4 LT |
1459 | } |
1460 | } | |
1461 | ||
1462 | struct clone_info { | |
1463 | struct mapped_device *md; | |
1464 | struct dm_table *map; | |
1465 | struct bio *bio; | |
1466 | struct dm_io *io; | |
1467 | sector_t sector; | |
e0d6609a | 1468 | unsigned sector_count; |
1da177e4 LT |
1469 | }; |
1470 | ||
e0d6609a | 1471 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1472 | { |
4f024f37 KO |
1473 | bio->bi_iter.bi_sector = sector; |
1474 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1475 | } |
1476 | ||
1477 | /* | |
1478 | * Creates a bio that consists of range of complete bvecs. | |
1479 | */ | |
dba14160 | 1480 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1481 | sector_t sector, unsigned len) |
1da177e4 | 1482 | { |
dba14160 | 1483 | struct bio *clone = &tio->clone; |
1da177e4 | 1484 | |
1c3b13e6 KO |
1485 | __bio_clone_fast(clone, bio); |
1486 | ||
1487 | if (bio_integrity(bio)) | |
1488 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1489 | |
1c3b13e6 KO |
1490 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1491 | clone->bi_iter.bi_size = to_bytes(len); | |
1492 | ||
1493 | if (bio_integrity(bio)) | |
1494 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1495 | } |
1496 | ||
9015df24 | 1497 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1498 | struct dm_target *ti, |
55a62eef | 1499 | unsigned target_bio_nr) |
f9ab94ce | 1500 | { |
dba14160 MP |
1501 | struct dm_target_io *tio; |
1502 | struct bio *clone; | |
1503 | ||
99778273 | 1504 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1505 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1506 | |
1507 | tio->io = ci->io; | |
1508 | tio->ti = ti; | |
55a62eef | 1509 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1510 | |
1511 | return tio; | |
1512 | } | |
1513 | ||
14fe594d AK |
1514 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1515 | struct dm_target *ti, | |
1dd40c3e | 1516 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1517 | { |
99778273 | 1518 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1519 | struct bio *clone = &tio->clone; |
9015df24 | 1520 | |
1dd40c3e MP |
1521 | tio->len_ptr = len; |
1522 | ||
99778273 | 1523 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1524 | if (len) |
1dd40c3e | 1525 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1526 | |
bd2a49b8 | 1527 | __map_bio(tio); |
f9ab94ce MP |
1528 | } |
1529 | ||
14fe594d | 1530 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1531 | unsigned num_bios, unsigned *len) |
06a426ce | 1532 | { |
55a62eef | 1533 | unsigned target_bio_nr; |
06a426ce | 1534 | |
55a62eef | 1535 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1536 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1537 | } |
1538 | ||
14fe594d | 1539 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1540 | { |
06a426ce | 1541 | unsigned target_nr = 0; |
f9ab94ce MP |
1542 | struct dm_target *ti; |
1543 | ||
b372d360 | 1544 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1545 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1546 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1547 | |
f9ab94ce MP |
1548 | return 0; |
1549 | } | |
1550 | ||
e4c93811 | 1551 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1552 | sector_t sector, unsigned *len) |
5ae89a87 | 1553 | { |
dba14160 | 1554 | struct bio *bio = ci->bio; |
5ae89a87 | 1555 | struct dm_target_io *tio; |
b0d8ed4d AK |
1556 | unsigned target_bio_nr; |
1557 | unsigned num_target_bios = 1; | |
5ae89a87 | 1558 | |
b0d8ed4d AK |
1559 | /* |
1560 | * Does the target want to receive duplicate copies of the bio? | |
1561 | */ | |
1562 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1563 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1564 | |
b0d8ed4d | 1565 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1566 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1567 | tio->len_ptr = len; |
1568 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1569 | __map_bio(tio); |
1570 | } | |
5ae89a87 MS |
1571 | } |
1572 | ||
55a62eef | 1573 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1574 | |
55a62eef | 1575 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1576 | { |
55a62eef | 1577 | return ti->num_discard_bios; |
23508a96 MS |
1578 | } |
1579 | ||
55a62eef | 1580 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1581 | { |
55a62eef | 1582 | return ti->num_write_same_bios; |
23508a96 MS |
1583 | } |
1584 | ||
1585 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1586 | ||
1587 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1588 | { | |
55a62eef | 1589 | return ti->split_discard_bios; |
23508a96 MS |
1590 | } |
1591 | ||
14fe594d AK |
1592 | static int __send_changing_extent_only(struct clone_info *ci, |
1593 | get_num_bios_fn get_num_bios, | |
1594 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1595 | { |
1596 | struct dm_target *ti; | |
e0d6609a | 1597 | unsigned len; |
55a62eef | 1598 | unsigned num_bios; |
5ae89a87 | 1599 | |
a79245b3 MS |
1600 | do { |
1601 | ti = dm_table_find_target(ci->map, ci->sector); | |
1602 | if (!dm_target_is_valid(ti)) | |
1603 | return -EIO; | |
5ae89a87 | 1604 | |
5ae89a87 | 1605 | /* |
23508a96 MS |
1606 | * Even though the device advertised support for this type of |
1607 | * request, that does not mean every target supports it, and | |
936688d7 | 1608 | * reconfiguration might also have changed that since the |
a79245b3 | 1609 | * check was performed. |
5ae89a87 | 1610 | */ |
55a62eef AK |
1611 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1612 | if (!num_bios) | |
a79245b3 | 1613 | return -EOPNOTSUPP; |
5ae89a87 | 1614 | |
23508a96 | 1615 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1616 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1617 | else |
e0d6609a | 1618 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1619 | |
1dd40c3e | 1620 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1621 | |
1622 | ci->sector += len; | |
1623 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1624 | |
1625 | return 0; | |
1626 | } | |
1627 | ||
14fe594d | 1628 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1629 | { |
14fe594d AK |
1630 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1631 | is_split_required_for_discard); | |
23508a96 MS |
1632 | } |
1633 | ||
14fe594d | 1634 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1635 | { |
14fe594d | 1636 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1637 | } |
1638 | ||
e4c93811 AK |
1639 | /* |
1640 | * Select the correct strategy for processing a non-flush bio. | |
1641 | */ | |
14fe594d | 1642 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1643 | { |
dba14160 | 1644 | struct bio *bio = ci->bio; |
512875bd | 1645 | struct dm_target *ti; |
1c3b13e6 | 1646 | unsigned len; |
1da177e4 | 1647 | |
5ae89a87 | 1648 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1649 | return __send_discard(ci); |
23508a96 | 1650 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1651 | return __send_write_same(ci); |
5ae89a87 | 1652 | |
512875bd JN |
1653 | ti = dm_table_find_target(ci->map, ci->sector); |
1654 | if (!dm_target_is_valid(ti)) | |
1655 | return -EIO; | |
1656 | ||
1c3b13e6 | 1657 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1658 | |
1dd40c3e | 1659 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1660 | |
1c3b13e6 KO |
1661 | ci->sector += len; |
1662 | ci->sector_count -= len; | |
1da177e4 | 1663 | |
1c3b13e6 | 1664 | return 0; |
1da177e4 LT |
1665 | } |
1666 | ||
1667 | /* | |
14fe594d | 1668 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1669 | */ |
83d5e5b0 MP |
1670 | static void __split_and_process_bio(struct mapped_device *md, |
1671 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1672 | { |
1673 | struct clone_info ci; | |
512875bd | 1674 | int error = 0; |
1da177e4 | 1675 | |
83d5e5b0 | 1676 | if (unlikely(!map)) { |
6a8736d1 | 1677 | bio_io_error(bio); |
f0b9a450 MP |
1678 | return; |
1679 | } | |
692d0eb9 | 1680 | |
83d5e5b0 | 1681 | ci.map = map; |
1da177e4 | 1682 | ci.md = md; |
1da177e4 LT |
1683 | ci.io = alloc_io(md); |
1684 | ci.io->error = 0; | |
1685 | atomic_set(&ci.io->io_count, 1); | |
1686 | ci.io->bio = bio; | |
1687 | ci.io->md = md; | |
f88fb981 | 1688 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1689 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1690 | |
3eaf840e | 1691 | start_io_acct(ci.io); |
bd2a49b8 | 1692 | |
b372d360 MS |
1693 | if (bio->bi_rw & REQ_FLUSH) { |
1694 | ci.bio = &ci.md->flush_bio; | |
1695 | ci.sector_count = 0; | |
14fe594d | 1696 | error = __send_empty_flush(&ci); |
b372d360 MS |
1697 | /* dec_pending submits any data associated with flush */ |
1698 | } else { | |
6a8736d1 | 1699 | ci.bio = bio; |
d87f4c14 | 1700 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1701 | while (ci.sector_count && !error) |
14fe594d | 1702 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1703 | } |
1da177e4 LT |
1704 | |
1705 | /* drop the extra reference count */ | |
512875bd | 1706 | dec_pending(ci.io, error); |
1da177e4 LT |
1707 | } |
1708 | /*----------------------------------------------------------------- | |
1709 | * CRUD END | |
1710 | *---------------------------------------------------------------*/ | |
1711 | ||
f6fccb12 MB |
1712 | static int dm_merge_bvec(struct request_queue *q, |
1713 | struct bvec_merge_data *bvm, | |
1714 | struct bio_vec *biovec) | |
1715 | { | |
1716 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1717 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1718 | struct dm_target *ti; |
1719 | sector_t max_sectors; | |
5037108a | 1720 | int max_size = 0; |
f6fccb12 MB |
1721 | |
1722 | if (unlikely(!map)) | |
5037108a | 1723 | goto out; |
f6fccb12 MB |
1724 | |
1725 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1726 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1727 | goto out; |
f6fccb12 MB |
1728 | |
1729 | /* | |
1730 | * Find maximum amount of I/O that won't need splitting | |
1731 | */ | |
56a67df7 | 1732 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1733 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1734 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1735 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1736 | max_size = 0; |
1737 | ||
1738 | /* | |
1739 | * merge_bvec_fn() returns number of bytes | |
1740 | * it can accept at this offset | |
1741 | * max is precomputed maximal io size | |
1742 | */ | |
1743 | if (max_size && ti->type->merge) | |
1744 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1745 | /* |
1746 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1747 | * provided their merge_bvec method (we know this by looking for the |
1748 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1749 | * allow bios with multiple vector entries. So always set max_size | |
1750 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1751 | */ |
1752 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1753 | max_size = 0; |
f6fccb12 | 1754 | |
5037108a | 1755 | out: |
83d5e5b0 | 1756 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1757 | /* |
1758 | * Always allow an entire first page | |
1759 | */ | |
1760 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1761 | max_size = biovec->bv_len; | |
1762 | ||
f6fccb12 MB |
1763 | return max_size; |
1764 | } | |
1765 | ||
1da177e4 LT |
1766 | /* |
1767 | * The request function that just remaps the bio built up by | |
1768 | * dm_merge_bvec. | |
1769 | */ | |
ff36ab34 | 1770 | static void dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1771 | { |
12f03a49 | 1772 | int rw = bio_data_dir(bio); |
1da177e4 | 1773 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1774 | int srcu_idx; |
1775 | struct dm_table *map; | |
1da177e4 | 1776 | |
83d5e5b0 | 1777 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1778 | |
18c0b223 | 1779 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1780 | |
6a8736d1 TH |
1781 | /* if we're suspended, we have to queue this io for later */ |
1782 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1783 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1784 | |
6a8736d1 TH |
1785 | if (bio_rw(bio) != READA) |
1786 | queue_io(md, bio); | |
1787 | else | |
54d9a1b4 | 1788 | bio_io_error(bio); |
5a7bbad2 | 1789 | return; |
1da177e4 LT |
1790 | } |
1791 | ||
83d5e5b0 MP |
1792 | __split_and_process_bio(md, map, bio); |
1793 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1794 | return; |
cec47e3d KU |
1795 | } |
1796 | ||
fd2ed4d2 | 1797 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1798 | { |
1799 | return blk_queue_stackable(md->queue); | |
1800 | } | |
1801 | ||
466d89a6 | 1802 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1803 | { |
1804 | int r; | |
1805 | ||
466d89a6 KB |
1806 | if (blk_queue_io_stat(clone->q)) |
1807 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1808 | |
466d89a6 KB |
1809 | clone->start_time = jiffies; |
1810 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1811 | if (r) |
466d89a6 | 1812 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1813 | dm_complete_request(rq, r); |
1814 | } | |
cec47e3d | 1815 | |
cec47e3d KU |
1816 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1817 | void *data) | |
1818 | { | |
1819 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1820 | struct dm_rq_clone_bio_info *info = |
1821 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1822 | |
1823 | info->orig = bio_orig; | |
1824 | info->tio = tio; | |
1825 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1826 | |
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | static int setup_clone(struct request *clone, struct request *rq, | |
1ae49ea2 | 1831 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
cec47e3d | 1832 | { |
d0bcb878 | 1833 | int r; |
cec47e3d | 1834 | |
1ae49ea2 | 1835 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, |
29e4013d TH |
1836 | dm_rq_bio_constructor, tio); |
1837 | if (r) | |
1838 | return r; | |
cec47e3d | 1839 | |
29e4013d TH |
1840 | clone->cmd = rq->cmd; |
1841 | clone->cmd_len = rq->cmd_len; | |
1842 | clone->sense = rq->sense; | |
cec47e3d KU |
1843 | clone->end_io = end_clone_request; |
1844 | clone->end_io_data = tio; | |
1845 | ||
1ae49ea2 MS |
1846 | tio->clone = clone; |
1847 | ||
cec47e3d KU |
1848 | return 0; |
1849 | } | |
1850 | ||
6facdaff | 1851 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1852 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1853 | { |
02233342 MS |
1854 | /* |
1855 | * Do not allocate a clone if tio->clone was already set | |
1856 | * (see: dm_mq_queue_rq). | |
1857 | */ | |
1858 | bool alloc_clone = !tio->clone; | |
1859 | struct request *clone; | |
1ae49ea2 | 1860 | |
02233342 MS |
1861 | if (alloc_clone) { |
1862 | clone = alloc_clone_request(md, gfp_mask); | |
1863 | if (!clone) | |
1864 | return NULL; | |
1865 | } else | |
1866 | clone = tio->clone; | |
1ae49ea2 MS |
1867 | |
1868 | blk_rq_init(NULL, clone); | |
1869 | if (setup_clone(clone, rq, tio, gfp_mask)) { | |
1870 | /* -ENOMEM */ | |
02233342 MS |
1871 | if (alloc_clone) |
1872 | free_clone_request(md, clone); | |
1ae49ea2 MS |
1873 | return NULL; |
1874 | } | |
1875 | ||
1876 | return clone; | |
1877 | } | |
1878 | ||
2eb6e1e3 KB |
1879 | static void map_tio_request(struct kthread_work *work); |
1880 | ||
bfebd1cd MS |
1881 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1882 | struct mapped_device *md) | |
1883 | { | |
1884 | tio->md = md; | |
1885 | tio->ti = NULL; | |
1886 | tio->clone = NULL; | |
1887 | tio->orig = rq; | |
1888 | tio->error = 0; | |
1889 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1890 | if (md->kworker_task) |
1891 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1892 | } |
1893 | ||
466d89a6 KB |
1894 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1895 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1896 | { |
6facdaff | 1897 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1898 | int srcu_idx; |
1899 | struct dm_table *table; | |
6facdaff KU |
1900 | |
1901 | tio = alloc_rq_tio(md, gfp_mask); | |
1902 | if (!tio) | |
1903 | return NULL; | |
1904 | ||
bfebd1cd | 1905 | init_tio(tio, rq, md); |
6facdaff | 1906 | |
e5863d9a MS |
1907 | table = dm_get_live_table(md, &srcu_idx); |
1908 | if (!dm_table_mq_request_based(table)) { | |
1909 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1910 | dm_put_live_table(md, srcu_idx); | |
1911 | free_rq_tio(tio); | |
1912 | return NULL; | |
1913 | } | |
6facdaff | 1914 | } |
e5863d9a | 1915 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1916 | |
466d89a6 | 1917 | return tio; |
6facdaff KU |
1918 | } |
1919 | ||
cec47e3d KU |
1920 | /* |
1921 | * Called with the queue lock held. | |
1922 | */ | |
1923 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1924 | { | |
1925 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1926 | struct dm_rq_target_io *tio; |
cec47e3d | 1927 | |
cec47e3d KU |
1928 | if (unlikely(rq->special)) { |
1929 | DMWARN("Already has something in rq->special."); | |
1930 | return BLKPREP_KILL; | |
1931 | } | |
1932 | ||
466d89a6 KB |
1933 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1934 | if (!tio) | |
cec47e3d | 1935 | return BLKPREP_DEFER; |
cec47e3d | 1936 | |
466d89a6 | 1937 | rq->special = tio; |
cec47e3d KU |
1938 | rq->cmd_flags |= REQ_DONTPREP; |
1939 | ||
1940 | return BLKPREP_OK; | |
1941 | } | |
1942 | ||
9eef87da KU |
1943 | /* |
1944 | * Returns: | |
e5863d9a MS |
1945 | * 0 : the request has been processed |
1946 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1947 | * < 0 : the request was completed due to failure | |
9eef87da | 1948 | */ |
bfebd1cd | 1949 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1950 | struct mapped_device *md) |
cec47e3d | 1951 | { |
e5863d9a | 1952 | int r; |
bfebd1cd | 1953 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1954 | struct request *clone = NULL; |
1955 | ||
1956 | if (tio->clone) { | |
1957 | clone = tio->clone; | |
1958 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1959 | } else { | |
1960 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1961 | if (r < 0) { | |
1962 | /* The target wants to complete the I/O */ | |
1963 | dm_kill_unmapped_request(rq, r); | |
1964 | return r; | |
1965 | } | |
1966 | if (IS_ERR(clone)) | |
1967 | return DM_MAPIO_REQUEUE; | |
02233342 | 1968 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
e5863d9a MS |
1969 | /* -ENOMEM */ |
1970 | ti->type->release_clone_rq(clone); | |
1971 | return DM_MAPIO_REQUEUE; | |
1972 | } | |
1973 | } | |
cec47e3d | 1974 | |
cec47e3d KU |
1975 | switch (r) { |
1976 | case DM_MAPIO_SUBMITTED: | |
1977 | /* The target has taken the I/O to submit by itself later */ | |
1978 | break; | |
1979 | case DM_MAPIO_REMAPPED: | |
1980 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1981 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1982 | blk_rq_pos(rq)); |
1983 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1984 | break; |
1985 | case DM_MAPIO_REQUEUE: | |
1986 | /* The target wants to requeue the I/O */ | |
1987 | dm_requeue_unmapped_request(clone); | |
1988 | break; | |
1989 | default: | |
1990 | if (r > 0) { | |
1991 | DMWARN("unimplemented target map return value: %d", r); | |
1992 | BUG(); | |
1993 | } | |
1994 | ||
1995 | /* The target wants to complete the I/O */ | |
466d89a6 | 1996 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1997 | return r; |
cec47e3d | 1998 | } |
9eef87da | 1999 | |
e5863d9a | 2000 | return 0; |
cec47e3d KU |
2001 | } |
2002 | ||
2eb6e1e3 | 2003 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 2004 | { |
2eb6e1e3 | 2005 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
2006 | struct request *rq = tio->orig; |
2007 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 2008 | |
bfebd1cd | 2009 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
e5863d9a | 2010 | dm_requeue_unmapped_original_request(md, rq); |
2eb6e1e3 KB |
2011 | } |
2012 | ||
466d89a6 | 2013 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2014 | { |
bfebd1cd MS |
2015 | if (!orig->q->mq_ops) |
2016 | blk_start_request(orig); | |
2017 | else | |
2018 | blk_mq_start_request(orig); | |
466d89a6 | 2019 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2020 | |
0ce65797 MS |
2021 | if (md->seq_rq_merge_deadline_usecs) { |
2022 | md->last_rq_pos = rq_end_sector(orig); | |
2023 | md->last_rq_rw = rq_data_dir(orig); | |
2024 | md->last_rq_start_time = ktime_get(); | |
2025 | } | |
de3ec86d | 2026 | |
ba1cbad9 MS |
2027 | /* |
2028 | * Hold the md reference here for the in-flight I/O. | |
2029 | * We can't rely on the reference count by device opener, | |
2030 | * because the device may be closed during the request completion | |
2031 | * when all bios are completed. | |
2032 | * See the comment in rq_completed() too. | |
2033 | */ | |
2034 | dm_get(md); | |
ba1cbad9 MS |
2035 | } |
2036 | ||
0ce65797 MS |
2037 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2038 | ||
2039 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2040 | { | |
2041 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2042 | } | |
2043 | ||
2044 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2045 | const char *buf, size_t count) | |
2046 | { | |
2047 | unsigned deadline; | |
2048 | ||
17e149b8 | 2049 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2050 | return count; |
2051 | ||
2052 | if (kstrtouint(buf, 10, &deadline)) | |
2053 | return -EINVAL; | |
2054 | ||
2055 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2056 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2057 | ||
2058 | md->seq_rq_merge_deadline_usecs = deadline; | |
2059 | ||
2060 | return count; | |
2061 | } | |
2062 | ||
2063 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2064 | { | |
2065 | ktime_t kt_deadline; | |
2066 | ||
2067 | if (!md->seq_rq_merge_deadline_usecs) | |
2068 | return false; | |
2069 | ||
2070 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2071 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2072 | ||
2073 | return !ktime_after(ktime_get(), kt_deadline); | |
2074 | } | |
2075 | ||
cec47e3d KU |
2076 | /* |
2077 | * q->request_fn for request-based dm. | |
2078 | * Called with the queue lock held. | |
2079 | */ | |
2080 | static void dm_request_fn(struct request_queue *q) | |
2081 | { | |
2082 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2083 | int srcu_idx; |
2084 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2085 | struct dm_target *ti; |
466d89a6 | 2086 | struct request *rq; |
2eb6e1e3 | 2087 | struct dm_rq_target_io *tio; |
29e4013d | 2088 | sector_t pos; |
cec47e3d KU |
2089 | |
2090 | /* | |
b4324fee KU |
2091 | * For suspend, check blk_queue_stopped() and increment |
2092 | * ->pending within a single queue_lock not to increment the | |
2093 | * number of in-flight I/Os after the queue is stopped in | |
2094 | * dm_suspend(). | |
cec47e3d | 2095 | */ |
7eaceacc | 2096 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2097 | rq = blk_peek_request(q); |
2098 | if (!rq) | |
9d1deb83 | 2099 | goto out; |
cec47e3d | 2100 | |
29e4013d TH |
2101 | /* always use block 0 to find the target for flushes for now */ |
2102 | pos = 0; | |
2103 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2104 | pos = blk_rq_pos(rq); | |
2105 | ||
2106 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2107 | if (!dm_target_is_valid(ti)) { |
2108 | /* | |
466d89a6 | 2109 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2110 | * before calling dm_kill_unmapped_request |
2111 | */ | |
2112 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2113 | dm_start_request(md, rq); |
2114 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2115 | continue; |
2116 | } | |
d0bcb878 | 2117 | |
0ce65797 MS |
2118 | if (dm_request_peeked_before_merge_deadline(md) && |
2119 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2120 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2121 | goto delay_and_out; | |
2122 | ||
cec47e3d | 2123 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2124 | goto delay_and_out; |
cec47e3d | 2125 | |
466d89a6 | 2126 | dm_start_request(md, rq); |
9eef87da | 2127 | |
bfebd1cd | 2128 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2129 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2130 | tio->ti = ti; | |
2131 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2132 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2133 | } |
2134 | ||
2135 | goto out; | |
2136 | ||
7eaceacc | 2137 | delay_and_out: |
d548b34b | 2138 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2139 | out: |
83d5e5b0 | 2140 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2141 | } |
2142 | ||
1da177e4 LT |
2143 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2144 | { | |
8a57dfc6 CS |
2145 | int r = bdi_bits; |
2146 | struct mapped_device *md = congested_data; | |
2147 | struct dm_table *map; | |
1da177e4 | 2148 | |
1eb787ec | 2149 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2150 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2151 | if (map) { |
cec47e3d KU |
2152 | /* |
2153 | * Request-based dm cares about only own queue for | |
2154 | * the query about congestion status of request_queue | |
2155 | */ | |
2156 | if (dm_request_based(md)) | |
2157 | r = md->queue->backing_dev_info.state & | |
2158 | bdi_bits; | |
2159 | else | |
2160 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2161 | } |
83d5e5b0 | 2162 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2163 | } |
2164 | ||
1da177e4 LT |
2165 | return r; |
2166 | } | |
2167 | ||
2168 | /*----------------------------------------------------------------- | |
2169 | * An IDR is used to keep track of allocated minor numbers. | |
2170 | *---------------------------------------------------------------*/ | |
2b06cfff | 2171 | static void free_minor(int minor) |
1da177e4 | 2172 | { |
f32c10b0 | 2173 | spin_lock(&_minor_lock); |
1da177e4 | 2174 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2175 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2176 | } |
2177 | ||
2178 | /* | |
2179 | * See if the device with a specific minor # is free. | |
2180 | */ | |
cf13ab8e | 2181 | static int specific_minor(int minor) |
1da177e4 | 2182 | { |
c9d76be6 | 2183 | int r; |
1da177e4 LT |
2184 | |
2185 | if (minor >= (1 << MINORBITS)) | |
2186 | return -EINVAL; | |
2187 | ||
c9d76be6 | 2188 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2189 | spin_lock(&_minor_lock); |
1da177e4 | 2190 | |
c9d76be6 | 2191 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2192 | |
f32c10b0 | 2193 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2194 | idr_preload_end(); |
2195 | if (r < 0) | |
2196 | return r == -ENOSPC ? -EBUSY : r; | |
2197 | return 0; | |
1da177e4 LT |
2198 | } |
2199 | ||
cf13ab8e | 2200 | static int next_free_minor(int *minor) |
1da177e4 | 2201 | { |
c9d76be6 | 2202 | int r; |
62f75c2f | 2203 | |
c9d76be6 | 2204 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2205 | spin_lock(&_minor_lock); |
1da177e4 | 2206 | |
c9d76be6 | 2207 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2208 | |
f32c10b0 | 2209 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2210 | idr_preload_end(); |
2211 | if (r < 0) | |
2212 | return r; | |
2213 | *minor = r; | |
2214 | return 0; | |
1da177e4 LT |
2215 | } |
2216 | ||
83d5cde4 | 2217 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2218 | |
53d5914f MP |
2219 | static void dm_wq_work(struct work_struct *work); |
2220 | ||
4a0b4ddf MS |
2221 | static void dm_init_md_queue(struct mapped_device *md) |
2222 | { | |
2223 | /* | |
2224 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2225 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2226 | * The type is decided at the first table loading time. |
2227 | * To prevent problematic device stacking, clear the queue flag | |
2228 | * for request stacking support until then. | |
2229 | * | |
2230 | * This queue is new, so no concurrency on the queue_flags. | |
2231 | */ | |
2232 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
bfebd1cd | 2233 | } |
4a0b4ddf | 2234 | |
bfebd1cd MS |
2235 | static void dm_init_old_md_queue(struct mapped_device *md) |
2236 | { | |
17e149b8 | 2237 | md->use_blk_mq = false; |
bfebd1cd MS |
2238 | dm_init_md_queue(md); |
2239 | ||
2240 | /* | |
2241 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2242 | */ | |
4a0b4ddf MS |
2243 | md->queue->queuedata = md; |
2244 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2245 | md->queue->backing_dev_info.congested_data = md; | |
ff36ab34 | 2246 | |
4a0b4ddf | 2247 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2248 | } |
2249 | ||
1da177e4 LT |
2250 | /* |
2251 | * Allocate and initialise a blank device with a given minor. | |
2252 | */ | |
2b06cfff | 2253 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2254 | { |
2255 | int r; | |
cf13ab8e | 2256 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2257 | void *old_md; |
1da177e4 LT |
2258 | |
2259 | if (!md) { | |
2260 | DMWARN("unable to allocate device, out of memory."); | |
2261 | return NULL; | |
2262 | } | |
2263 | ||
10da4f79 | 2264 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2265 | goto bad_module_get; |
10da4f79 | 2266 | |
1da177e4 | 2267 | /* get a minor number for the dev */ |
2b06cfff | 2268 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2269 | r = next_free_minor(&minor); |
2b06cfff | 2270 | else |
cf13ab8e | 2271 | r = specific_minor(minor); |
1da177e4 | 2272 | if (r < 0) |
6ed7ade8 | 2273 | goto bad_minor; |
1da177e4 | 2274 | |
83d5e5b0 MP |
2275 | r = init_srcu_struct(&md->io_barrier); |
2276 | if (r < 0) | |
2277 | goto bad_io_barrier; | |
2278 | ||
17e149b8 | 2279 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2280 | md->type = DM_TYPE_NONE; |
e61290a4 | 2281 | mutex_init(&md->suspend_lock); |
a5664dad | 2282 | mutex_init(&md->type_lock); |
86f1152b | 2283 | mutex_init(&md->table_devices_lock); |
022c2611 | 2284 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2285 | atomic_set(&md->holders, 1); |
5c6bd75d | 2286 | atomic_set(&md->open_count, 0); |
1da177e4 | 2287 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2288 | atomic_set(&md->uevent_seq, 0); |
2289 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2290 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2291 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2292 | |
4a0b4ddf | 2293 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2294 | if (!md->queue) |
6ed7ade8 | 2295 | goto bad_queue; |
1da177e4 | 2296 | |
4a0b4ddf | 2297 | dm_init_md_queue(md); |
9faf400f | 2298 | |
1da177e4 LT |
2299 | md->disk = alloc_disk(1); |
2300 | if (!md->disk) | |
6ed7ade8 | 2301 | goto bad_disk; |
1da177e4 | 2302 | |
316d315b NK |
2303 | atomic_set(&md->pending[0], 0); |
2304 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2305 | init_waitqueue_head(&md->wait); |
53d5914f | 2306 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2307 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2308 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2309 | md->kworker_task = NULL; |
f0b04115 | 2310 | |
1da177e4 LT |
2311 | md->disk->major = _major; |
2312 | md->disk->first_minor = minor; | |
2313 | md->disk->fops = &dm_blk_dops; | |
2314 | md->disk->queue = md->queue; | |
2315 | md->disk->private_data = md; | |
2316 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2317 | add_disk(md->disk); | |
7e51f257 | 2318 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2319 | |
670368a8 | 2320 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2321 | if (!md->wq) |
2322 | goto bad_thread; | |
2323 | ||
32a926da MP |
2324 | md->bdev = bdget_disk(md->disk, 0); |
2325 | if (!md->bdev) | |
2326 | goto bad_bdev; | |
2327 | ||
6a8736d1 TH |
2328 | bio_init(&md->flush_bio); |
2329 | md->flush_bio.bi_bdev = md->bdev; | |
2330 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2331 | ||
fd2ed4d2 MP |
2332 | dm_stats_init(&md->stats); |
2333 | ||
ba61fdd1 | 2334 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2335 | spin_lock(&_minor_lock); |
ba61fdd1 | 2336 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2337 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2338 | |
2339 | BUG_ON(old_md != MINOR_ALLOCED); | |
2340 | ||
1da177e4 LT |
2341 | return md; |
2342 | ||
32a926da MP |
2343 | bad_bdev: |
2344 | destroy_workqueue(md->wq); | |
304f3f6a | 2345 | bad_thread: |
03022c54 | 2346 | del_gendisk(md->disk); |
304f3f6a | 2347 | put_disk(md->disk); |
6ed7ade8 | 2348 | bad_disk: |
1312f40e | 2349 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2350 | bad_queue: |
83d5e5b0 MP |
2351 | cleanup_srcu_struct(&md->io_barrier); |
2352 | bad_io_barrier: | |
1da177e4 | 2353 | free_minor(minor); |
6ed7ade8 | 2354 | bad_minor: |
10da4f79 | 2355 | module_put(THIS_MODULE); |
6ed7ade8 | 2356 | bad_module_get: |
1da177e4 LT |
2357 | kfree(md); |
2358 | return NULL; | |
2359 | } | |
2360 | ||
ae9da83f JN |
2361 | static void unlock_fs(struct mapped_device *md); |
2362 | ||
1da177e4 LT |
2363 | static void free_dev(struct mapped_device *md) |
2364 | { | |
f331c029 | 2365 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2366 | |
32a926da | 2367 | unlock_fs(md); |
304f3f6a | 2368 | destroy_workqueue(md->wq); |
2eb6e1e3 KB |
2369 | |
2370 | if (md->kworker_task) | |
2371 | kthread_stop(md->kworker_task); | |
e6ee8c0b KU |
2372 | if (md->io_pool) |
2373 | mempool_destroy(md->io_pool); | |
1ae49ea2 MS |
2374 | if (md->rq_pool) |
2375 | mempool_destroy(md->rq_pool); | |
e6ee8c0b KU |
2376 | if (md->bs) |
2377 | bioset_free(md->bs); | |
63a4f065 | 2378 | |
83d5e5b0 | 2379 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2380 | free_table_devices(&md->table_devices); |
63a4f065 | 2381 | dm_stats_cleanup(&md->stats); |
fba9f90e JM |
2382 | |
2383 | spin_lock(&_minor_lock); | |
2384 | md->disk->private_data = NULL; | |
2385 | spin_unlock(&_minor_lock); | |
63a4f065 MS |
2386 | if (blk_get_integrity(md->disk)) |
2387 | blk_integrity_unregister(md->disk); | |
2388 | del_gendisk(md->disk); | |
1da177e4 | 2389 | put_disk(md->disk); |
1312f40e | 2390 | blk_cleanup_queue(md->queue); |
17e149b8 | 2391 | if (md->use_blk_mq) |
bfebd1cd | 2392 | blk_mq_free_tag_set(&md->tag_set); |
63a4f065 MS |
2393 | bdput(md->bdev); |
2394 | free_minor(minor); | |
2395 | ||
10da4f79 | 2396 | module_put(THIS_MODULE); |
1da177e4 LT |
2397 | kfree(md); |
2398 | } | |
2399 | ||
e6ee8c0b KU |
2400 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2401 | { | |
c0820cf5 | 2402 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2403 | |
17e149b8 | 2404 | if (md->bs) { |
16245bdc JN |
2405 | /* The md already has necessary mempools. */ |
2406 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2407 | /* | |
2408 | * Reload bioset because front_pad may have changed | |
2409 | * because a different table was loaded. | |
2410 | */ | |
2411 | bioset_free(md->bs); | |
2412 | md->bs = p->bs; | |
2413 | p->bs = NULL; | |
16245bdc | 2414 | } |
466d89a6 KB |
2415 | /* |
2416 | * There's no need to reload with request-based dm | |
2417 | * because the size of front_pad doesn't change. | |
2418 | * Note for future: If you are to reload bioset, | |
2419 | * prep-ed requests in the queue may refer | |
2420 | * to bio from the old bioset, so you must walk | |
2421 | * through the queue to unprep. | |
2422 | */ | |
e6ee8c0b | 2423 | goto out; |
c0820cf5 | 2424 | } |
e6ee8c0b | 2425 | |
1ae49ea2 | 2426 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
e6ee8c0b KU |
2427 | |
2428 | md->io_pool = p->io_pool; | |
2429 | p->io_pool = NULL; | |
1ae49ea2 MS |
2430 | md->rq_pool = p->rq_pool; |
2431 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2432 | md->bs = p->bs; |
2433 | p->bs = NULL; | |
2434 | ||
2435 | out: | |
02233342 | 2436 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2437 | dm_table_free_md_mempools(t); |
2438 | } | |
2439 | ||
1da177e4 LT |
2440 | /* |
2441 | * Bind a table to the device. | |
2442 | */ | |
2443 | static void event_callback(void *context) | |
2444 | { | |
7a8c3d3b MA |
2445 | unsigned long flags; |
2446 | LIST_HEAD(uevents); | |
1da177e4 LT |
2447 | struct mapped_device *md = (struct mapped_device *) context; |
2448 | ||
7a8c3d3b MA |
2449 | spin_lock_irqsave(&md->uevent_lock, flags); |
2450 | list_splice_init(&md->uevent_list, &uevents); | |
2451 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2452 | ||
ed9e1982 | 2453 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2454 | |
1da177e4 LT |
2455 | atomic_inc(&md->event_nr); |
2456 | wake_up(&md->eventq); | |
2457 | } | |
2458 | ||
c217649b MS |
2459 | /* |
2460 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2461 | */ | |
4e90188b | 2462 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2463 | { |
4e90188b | 2464 | set_capacity(md->disk, size); |
1da177e4 | 2465 | |
db8fef4f | 2466 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2467 | } |
2468 | ||
d5b9dd04 MP |
2469 | /* |
2470 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2471 | * | |
2472 | * If this function returns 0, then the device is either a non-dm | |
2473 | * device without a merge_bvec_fn, or it is a dm device that is | |
2474 | * able to split any bios it receives that are too big. | |
2475 | */ | |
2476 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2477 | { | |
2478 | struct mapped_device *dev_md; | |
2479 | ||
2480 | if (!q->merge_bvec_fn) | |
2481 | return 0; | |
2482 | ||
ff36ab34 | 2483 | if (q->make_request_fn == dm_make_request) { |
d5b9dd04 MP |
2484 | dev_md = q->queuedata; |
2485 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2486 | return 0; | |
2487 | } | |
2488 | ||
2489 | return 1; | |
2490 | } | |
2491 | ||
2492 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2493 | struct dm_dev *dev, sector_t start, | |
2494 | sector_t len, void *data) | |
2495 | { | |
2496 | struct block_device *bdev = dev->bdev; | |
2497 | struct request_queue *q = bdev_get_queue(bdev); | |
2498 | ||
2499 | return dm_queue_merge_is_compulsory(q); | |
2500 | } | |
2501 | ||
2502 | /* | |
2503 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2504 | * on the properties of the underlying devices. | |
2505 | */ | |
2506 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2507 | { | |
2508 | unsigned i = 0; | |
2509 | struct dm_target *ti; | |
2510 | ||
2511 | while (i < dm_table_get_num_targets(table)) { | |
2512 | ti = dm_table_get_target(table, i++); | |
2513 | ||
2514 | if (ti->type->iterate_devices && | |
2515 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2516 | return 0; | |
2517 | } | |
2518 | ||
2519 | return 1; | |
2520 | } | |
2521 | ||
042d2a9b AK |
2522 | /* |
2523 | * Returns old map, which caller must destroy. | |
2524 | */ | |
2525 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2526 | struct queue_limits *limits) | |
1da177e4 | 2527 | { |
042d2a9b | 2528 | struct dm_table *old_map; |
165125e1 | 2529 | struct request_queue *q = md->queue; |
1da177e4 | 2530 | sector_t size; |
d5b9dd04 | 2531 | int merge_is_optional; |
1da177e4 LT |
2532 | |
2533 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2534 | |
2535 | /* | |
2536 | * Wipe any geometry if the size of the table changed. | |
2537 | */ | |
fd2ed4d2 | 2538 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2539 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2540 | ||
32a926da | 2541 | __set_size(md, size); |
d5816876 | 2542 | |
2ca3310e AK |
2543 | dm_table_event_callback(t, event_callback, md); |
2544 | ||
e6ee8c0b KU |
2545 | /* |
2546 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2547 | * for request-based during suspension. So stop it to prevent | |
2548 | * I/O mapping before resume. | |
2549 | * This must be done before setting the queue restrictions, | |
2550 | * because request-based dm may be run just after the setting. | |
2551 | */ | |
bfebd1cd | 2552 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2553 | stop_queue(q); |
2554 | ||
2555 | __bind_mempools(md, t); | |
2556 | ||
d5b9dd04 MP |
2557 | merge_is_optional = dm_table_merge_is_optional(t); |
2558 | ||
a12f5d48 | 2559 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2560 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2561 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2562 | ||
754c5fc7 | 2563 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2564 | if (merge_is_optional) |
2565 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2566 | else | |
2567 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2568 | if (old_map) |
2569 | dm_sync_table(md); | |
1da177e4 | 2570 | |
042d2a9b | 2571 | return old_map; |
1da177e4 LT |
2572 | } |
2573 | ||
a7940155 AK |
2574 | /* |
2575 | * Returns unbound table for the caller to free. | |
2576 | */ | |
2577 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2578 | { |
a12f5d48 | 2579 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2580 | |
2581 | if (!map) | |
a7940155 | 2582 | return NULL; |
1da177e4 LT |
2583 | |
2584 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2585 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2586 | dm_sync_table(md); |
a7940155 AK |
2587 | |
2588 | return map; | |
1da177e4 LT |
2589 | } |
2590 | ||
2591 | /* | |
2592 | * Constructor for a new device. | |
2593 | */ | |
2b06cfff | 2594 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2595 | { |
2596 | struct mapped_device *md; | |
2597 | ||
2b06cfff | 2598 | md = alloc_dev(minor); |
1da177e4 LT |
2599 | if (!md) |
2600 | return -ENXIO; | |
2601 | ||
784aae73 MB |
2602 | dm_sysfs_init(md); |
2603 | ||
1da177e4 LT |
2604 | *result = md; |
2605 | return 0; | |
2606 | } | |
2607 | ||
a5664dad MS |
2608 | /* |
2609 | * Functions to manage md->type. | |
2610 | * All are required to hold md->type_lock. | |
2611 | */ | |
2612 | void dm_lock_md_type(struct mapped_device *md) | |
2613 | { | |
2614 | mutex_lock(&md->type_lock); | |
2615 | } | |
2616 | ||
2617 | void dm_unlock_md_type(struct mapped_device *md) | |
2618 | { | |
2619 | mutex_unlock(&md->type_lock); | |
2620 | } | |
2621 | ||
2622 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2623 | { | |
00c4fc3b | 2624 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2625 | md->type = type; |
2626 | } | |
2627 | ||
2628 | unsigned dm_get_md_type(struct mapped_device *md) | |
2629 | { | |
00c4fc3b | 2630 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2631 | return md->type; |
2632 | } | |
2633 | ||
36a0456f AK |
2634 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2635 | { | |
2636 | return md->immutable_target_type; | |
2637 | } | |
2638 | ||
f84cb8a4 MS |
2639 | /* |
2640 | * The queue_limits are only valid as long as you have a reference | |
2641 | * count on 'md'. | |
2642 | */ | |
2643 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2644 | { | |
2645 | BUG_ON(!atomic_read(&md->holders)); | |
2646 | return &md->queue->limits; | |
2647 | } | |
2648 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2649 | ||
bfebd1cd MS |
2650 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2651 | { | |
2652 | /* Initialize the request-based DM worker thread */ | |
2653 | init_kthread_worker(&md->kworker); | |
2654 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2655 | "kdmwork-%s", dm_device_name(md)); | |
2656 | } | |
2657 | ||
4a0b4ddf MS |
2658 | /* |
2659 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2660 | */ | |
2661 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2662 | { | |
2663 | struct request_queue *q = NULL; | |
2664 | ||
2665 | if (md->queue->elevator) | |
bfebd1cd | 2666 | return 0; |
4a0b4ddf MS |
2667 | |
2668 | /* Fully initialize the queue */ | |
2669 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2670 | if (!q) | |
bfebd1cd | 2671 | return -EINVAL; |
4a0b4ddf | 2672 | |
0ce65797 MS |
2673 | /* disable dm_request_fn's merge heuristic by default */ |
2674 | md->seq_rq_merge_deadline_usecs = 0; | |
2675 | ||
4a0b4ddf | 2676 | md->queue = q; |
bfebd1cd | 2677 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2678 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2679 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2680 | |
bfebd1cd | 2681 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2682 | |
4a0b4ddf MS |
2683 | elv_register_queue(md->queue); |
2684 | ||
bfebd1cd MS |
2685 | return 0; |
2686 | } | |
2687 | ||
2688 | static int dm_mq_init_request(void *data, struct request *rq, | |
2689 | unsigned int hctx_idx, unsigned int request_idx, | |
2690 | unsigned int numa_node) | |
2691 | { | |
2692 | struct mapped_device *md = data; | |
2693 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2694 | ||
2695 | /* | |
2696 | * Must initialize md member of tio, otherwise it won't | |
2697 | * be available in dm_mq_queue_rq. | |
2698 | */ | |
2699 | tio->md = md; | |
2700 | ||
2701 | return 0; | |
2702 | } | |
2703 | ||
2704 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2705 | const struct blk_mq_queue_data *bd) | |
2706 | { | |
2707 | struct request *rq = bd->rq; | |
2708 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2709 | struct mapped_device *md = tio->md; | |
2710 | int srcu_idx; | |
2711 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2712 | struct dm_target *ti; | |
2713 | sector_t pos; | |
2714 | ||
2715 | /* always use block 0 to find the target for flushes for now */ | |
2716 | pos = 0; | |
2717 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2718 | pos = blk_rq_pos(rq); | |
2719 | ||
2720 | ti = dm_table_find_target(map, pos); | |
2721 | if (!dm_target_is_valid(ti)) { | |
2722 | dm_put_live_table(md, srcu_idx); | |
2723 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2724 | /* | |
2725 | * Must perform setup, that rq_completed() requires, | |
2726 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2727 | */ | |
2728 | dm_start_request(md, rq); | |
2729 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2730 | } | |
2731 | dm_put_live_table(md, srcu_idx); | |
2732 | ||
2733 | if (ti->type->busy && ti->type->busy(ti)) | |
2734 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2735 | ||
2736 | dm_start_request(md, rq); | |
2737 | ||
2738 | /* Init tio using md established in .init_request */ | |
2739 | init_tio(tio, rq, md); | |
2740 | ||
02233342 MS |
2741 | /* |
2742 | * Establish tio->ti before queuing work (map_tio_request) | |
2743 | * or making direct call to map_request(). | |
2744 | */ | |
bfebd1cd | 2745 | tio->ti = ti; |
02233342 MS |
2746 | |
2747 | /* Clone the request if underlying devices aren't blk-mq */ | |
2748 | if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) { | |
2749 | /* clone request is allocated at the end of the pdu */ | |
2750 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
2751 | if (!clone_rq(rq, md, tio, GFP_ATOMIC)) | |
2752 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2753 | queue_kthread_work(&md->kworker, &tio->work); | |
2754 | } else { | |
2755 | /* Direct call is fine since .queue_rq allows allocations */ | |
2756 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) | |
2757 | dm_requeue_unmapped_original_request(md, rq); | |
2758 | } | |
bfebd1cd MS |
2759 | |
2760 | return BLK_MQ_RQ_QUEUE_OK; | |
2761 | } | |
2762 | ||
2763 | static struct blk_mq_ops dm_mq_ops = { | |
2764 | .queue_rq = dm_mq_queue_rq, | |
2765 | .map_queue = blk_mq_map_queue, | |
2766 | .complete = dm_softirq_done, | |
2767 | .init_request = dm_mq_init_request, | |
2768 | }; | |
2769 | ||
2770 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2771 | { | |
02233342 | 2772 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2773 | struct request_queue *q; |
2774 | int err; | |
2775 | ||
2776 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2777 | md->tag_set.ops = &dm_mq_ops; | |
2778 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2779 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2780 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2781 | md->tag_set.nr_hw_queues = 1; | |
02233342 MS |
2782 | if (md_type == DM_TYPE_REQUEST_BASED) { |
2783 | /* make the memory for non-blk-mq clone part of the pdu */ | |
2784 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request); | |
2785 | } else | |
2786 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd MS |
2787 | md->tag_set.driver_data = md; |
2788 | ||
2789 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2790 | if (err) | |
2791 | return err; | |
2792 | ||
2793 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2794 | if (IS_ERR(q)) { | |
2795 | err = PTR_ERR(q); | |
2796 | goto out_tag_set; | |
2797 | } | |
2798 | md->queue = q; | |
2799 | dm_init_md_queue(md); | |
2800 | ||
2801 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2802 | blk_mq_register_disk(md->disk); | |
2803 | ||
02233342 MS |
2804 | if (md_type == DM_TYPE_REQUEST_BASED) |
2805 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2806 | |
2807 | return 0; | |
2808 | ||
2809 | out_tag_set: | |
2810 | blk_mq_free_tag_set(&md->tag_set); | |
2811 | return err; | |
4a0b4ddf MS |
2812 | } |
2813 | ||
17e149b8 MS |
2814 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2815 | { | |
2816 | if (type == DM_TYPE_BIO_BASED) | |
2817 | return type; | |
2818 | ||
2819 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2820 | } | |
2821 | ||
4a0b4ddf MS |
2822 | /* |
2823 | * Setup the DM device's queue based on md's type | |
2824 | */ | |
2825 | int dm_setup_md_queue(struct mapped_device *md) | |
2826 | { | |
bfebd1cd | 2827 | int r; |
17e149b8 | 2828 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2829 | |
2830 | switch (md_type) { | |
2831 | case DM_TYPE_REQUEST_BASED: | |
2832 | r = dm_init_request_based_queue(md); | |
2833 | if (r) { | |
ff36ab34 | 2834 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2835 | return r; |
ff36ab34 | 2836 | } |
bfebd1cd MS |
2837 | break; |
2838 | case DM_TYPE_MQ_REQUEST_BASED: | |
2839 | r = dm_init_request_based_blk_mq_queue(md); | |
2840 | if (r) { | |
2841 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2842 | return r; | |
2843 | } | |
2844 | break; | |
2845 | case DM_TYPE_BIO_BASED: | |
2846 | dm_init_old_md_queue(md); | |
ff36ab34 MS |
2847 | blk_queue_make_request(md->queue, dm_make_request); |
2848 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); | |
bfebd1cd | 2849 | break; |
4a0b4ddf MS |
2850 | } |
2851 | ||
2852 | return 0; | |
2853 | } | |
2854 | ||
2bec1f4a | 2855 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2856 | { |
2857 | struct mapped_device *md; | |
1da177e4 LT |
2858 | unsigned minor = MINOR(dev); |
2859 | ||
2860 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2861 | return NULL; | |
2862 | ||
f32c10b0 | 2863 | spin_lock(&_minor_lock); |
1da177e4 LT |
2864 | |
2865 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2866 | if (md) { |
2867 | if ((md == MINOR_ALLOCED || | |
2868 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2869 | dm_deleting_md(md) || | |
2870 | test_bit(DMF_FREEING, &md->flags))) { | |
2871 | md = NULL; | |
2872 | goto out; | |
2873 | } | |
2874 | dm_get(md); | |
fba9f90e | 2875 | } |
1da177e4 | 2876 | |
fba9f90e | 2877 | out: |
f32c10b0 | 2878 | spin_unlock(&_minor_lock); |
1da177e4 | 2879 | |
637842cf DT |
2880 | return md; |
2881 | } | |
3cf2e4ba | 2882 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2883 | |
9ade92a9 | 2884 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2885 | { |
9ade92a9 | 2886 | return md->interface_ptr; |
1da177e4 LT |
2887 | } |
2888 | ||
2889 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2890 | { | |
2891 | md->interface_ptr = ptr; | |
2892 | } | |
2893 | ||
2894 | void dm_get(struct mapped_device *md) | |
2895 | { | |
2896 | atomic_inc(&md->holders); | |
3f77316d | 2897 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2898 | } |
2899 | ||
09ee96b2 MP |
2900 | int dm_hold(struct mapped_device *md) |
2901 | { | |
2902 | spin_lock(&_minor_lock); | |
2903 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2904 | spin_unlock(&_minor_lock); | |
2905 | return -EBUSY; | |
2906 | } | |
2907 | dm_get(md); | |
2908 | spin_unlock(&_minor_lock); | |
2909 | return 0; | |
2910 | } | |
2911 | EXPORT_SYMBOL_GPL(dm_hold); | |
2912 | ||
72d94861 AK |
2913 | const char *dm_device_name(struct mapped_device *md) |
2914 | { | |
2915 | return md->name; | |
2916 | } | |
2917 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2918 | ||
3f77316d | 2919 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2920 | { |
1134e5ae | 2921 | struct dm_table *map; |
83d5e5b0 | 2922 | int srcu_idx; |
1da177e4 | 2923 | |
3f77316d | 2924 | might_sleep(); |
fba9f90e | 2925 | |
83d5e5b0 | 2926 | map = dm_get_live_table(md, &srcu_idx); |
63a4f065 MS |
2927 | |
2928 | spin_lock(&_minor_lock); | |
3f77316d KU |
2929 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2930 | set_bit(DMF_FREEING, &md->flags); | |
2931 | spin_unlock(&_minor_lock); | |
2932 | ||
02233342 | 2933 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2934 | flush_kthread_worker(&md->kworker); |
2935 | ||
ab7c7bb6 MP |
2936 | /* |
2937 | * Take suspend_lock so that presuspend and postsuspend methods | |
2938 | * do not race with internal suspend. | |
2939 | */ | |
2940 | mutex_lock(&md->suspend_lock); | |
3f77316d KU |
2941 | if (!dm_suspended_md(md)) { |
2942 | dm_table_presuspend_targets(map); | |
2943 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2944 | } |
ab7c7bb6 | 2945 | mutex_unlock(&md->suspend_lock); |
3f77316d | 2946 | |
83d5e5b0 MP |
2947 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2948 | dm_put_live_table(md, srcu_idx); | |
2949 | ||
3f77316d KU |
2950 | /* |
2951 | * Rare, but there may be I/O requests still going to complete, | |
2952 | * for example. Wait for all references to disappear. | |
2953 | * No one should increment the reference count of the mapped_device, | |
2954 | * after the mapped_device state becomes DMF_FREEING. | |
2955 | */ | |
2956 | if (wait) | |
2957 | while (atomic_read(&md->holders)) | |
2958 | msleep(1); | |
2959 | else if (atomic_read(&md->holders)) | |
2960 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2961 | dm_device_name(md), atomic_read(&md->holders)); | |
2962 | ||
2963 | dm_sysfs_exit(md); | |
3f77316d KU |
2964 | dm_table_destroy(__unbind(md)); |
2965 | free_dev(md); | |
2966 | } | |
2967 | ||
2968 | void dm_destroy(struct mapped_device *md) | |
2969 | { | |
2970 | __dm_destroy(md, true); | |
2971 | } | |
2972 | ||
2973 | void dm_destroy_immediate(struct mapped_device *md) | |
2974 | { | |
2975 | __dm_destroy(md, false); | |
2976 | } | |
2977 | ||
2978 | void dm_put(struct mapped_device *md) | |
2979 | { | |
2980 | atomic_dec(&md->holders); | |
1da177e4 | 2981 | } |
79eb885c | 2982 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2983 | |
401600df | 2984 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2985 | { |
2986 | int r = 0; | |
b44ebeb0 MP |
2987 | DECLARE_WAITQUEUE(wait, current); |
2988 | ||
b44ebeb0 | 2989 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2990 | |
2991 | while (1) { | |
401600df | 2992 | set_current_state(interruptible); |
46125c1c | 2993 | |
b4324fee | 2994 | if (!md_in_flight(md)) |
46125c1c MB |
2995 | break; |
2996 | ||
401600df MP |
2997 | if (interruptible == TASK_INTERRUPTIBLE && |
2998 | signal_pending(current)) { | |
46125c1c MB |
2999 | r = -EINTR; |
3000 | break; | |
3001 | } | |
3002 | ||
3003 | io_schedule(); | |
3004 | } | |
3005 | set_current_state(TASK_RUNNING); | |
3006 | ||
b44ebeb0 MP |
3007 | remove_wait_queue(&md->wait, &wait); |
3008 | ||
46125c1c MB |
3009 | return r; |
3010 | } | |
3011 | ||
1da177e4 LT |
3012 | /* |
3013 | * Process the deferred bios | |
3014 | */ | |
ef208587 | 3015 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 3016 | { |
ef208587 MP |
3017 | struct mapped_device *md = container_of(work, struct mapped_device, |
3018 | work); | |
6d6f10df | 3019 | struct bio *c; |
83d5e5b0 MP |
3020 | int srcu_idx; |
3021 | struct dm_table *map; | |
1da177e4 | 3022 | |
83d5e5b0 | 3023 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 3024 | |
3b00b203 | 3025 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
3026 | spin_lock_irq(&md->deferred_lock); |
3027 | c = bio_list_pop(&md->deferred); | |
3028 | spin_unlock_irq(&md->deferred_lock); | |
3029 | ||
6a8736d1 | 3030 | if (!c) |
df12ee99 | 3031 | break; |
022c2611 | 3032 | |
e6ee8c0b KU |
3033 | if (dm_request_based(md)) |
3034 | generic_make_request(c); | |
6a8736d1 | 3035 | else |
83d5e5b0 | 3036 | __split_and_process_bio(md, map, c); |
022c2611 | 3037 | } |
73d410c0 | 3038 | |
83d5e5b0 | 3039 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
3040 | } |
3041 | ||
9a1fb464 | 3042 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 3043 | { |
3b00b203 | 3044 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 3045 | smp_mb__after_atomic(); |
53d5914f | 3046 | queue_work(md->wq, &md->work); |
304f3f6a MB |
3047 | } |
3048 | ||
1da177e4 | 3049 | /* |
042d2a9b | 3050 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 3051 | */ |
042d2a9b | 3052 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3053 | { |
87eb5b21 | 3054 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3055 | struct queue_limits limits; |
042d2a9b | 3056 | int r; |
1da177e4 | 3057 | |
e61290a4 | 3058 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3059 | |
3060 | /* device must be suspended */ | |
4f186f8b | 3061 | if (!dm_suspended_md(md)) |
93c534ae | 3062 | goto out; |
1da177e4 | 3063 | |
3ae70656 MS |
3064 | /* |
3065 | * If the new table has no data devices, retain the existing limits. | |
3066 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3067 | * then new I/O is queued based on these limits, and then some paths | |
3068 | * reappear. | |
3069 | */ | |
3070 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3071 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3072 | if (live_map) |
3073 | limits = md->queue->limits; | |
83d5e5b0 | 3074 | dm_put_live_table_fast(md); |
3ae70656 MS |
3075 | } |
3076 | ||
87eb5b21 MC |
3077 | if (!live_map) { |
3078 | r = dm_calculate_queue_limits(table, &limits); | |
3079 | if (r) { | |
3080 | map = ERR_PTR(r); | |
3081 | goto out; | |
3082 | } | |
042d2a9b | 3083 | } |
754c5fc7 | 3084 | |
042d2a9b | 3085 | map = __bind(md, table, &limits); |
1da177e4 | 3086 | |
93c534ae | 3087 | out: |
e61290a4 | 3088 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3089 | return map; |
1da177e4 LT |
3090 | } |
3091 | ||
3092 | /* | |
3093 | * Functions to lock and unlock any filesystem running on the | |
3094 | * device. | |
3095 | */ | |
2ca3310e | 3096 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3097 | { |
e39e2e95 | 3098 | int r; |
1da177e4 LT |
3099 | |
3100 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3101 | |
db8fef4f | 3102 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3103 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3104 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3105 | md->frozen_sb = NULL; |
3106 | return r; | |
dfbe03f6 AK |
3107 | } |
3108 | ||
aa8d7c2f AK |
3109 | set_bit(DMF_FROZEN, &md->flags); |
3110 | ||
1da177e4 LT |
3111 | return 0; |
3112 | } | |
3113 | ||
2ca3310e | 3114 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3115 | { |
aa8d7c2f AK |
3116 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3117 | return; | |
3118 | ||
db8fef4f | 3119 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3120 | md->frozen_sb = NULL; |
aa8d7c2f | 3121 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3122 | } |
3123 | ||
3124 | /* | |
ffcc3936 MS |
3125 | * If __dm_suspend returns 0, the device is completely quiescent |
3126 | * now. There is no request-processing activity. All new requests | |
3127 | * are being added to md->deferred list. | |
cec47e3d | 3128 | * |
ffcc3936 | 3129 | * Caller must hold md->suspend_lock |
cec47e3d | 3130 | */ |
ffcc3936 MS |
3131 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3132 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3133 | { |
ffcc3936 MS |
3134 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3135 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3136 | int r; | |
1da177e4 | 3137 | |
2e93ccc1 KU |
3138 | /* |
3139 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3140 | * This flag is cleared before dm_suspend returns. | |
3141 | */ | |
3142 | if (noflush) | |
3143 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3144 | ||
d67ee213 MS |
3145 | /* |
3146 | * This gets reverted if there's an error later and the targets | |
3147 | * provide the .presuspend_undo hook. | |
3148 | */ | |
cf222b37 AK |
3149 | dm_table_presuspend_targets(map); |
3150 | ||
32a926da | 3151 | /* |
9f518b27 KU |
3152 | * Flush I/O to the device. |
3153 | * Any I/O submitted after lock_fs() may not be flushed. | |
3154 | * noflush takes precedence over do_lockfs. | |
3155 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3156 | */ |
3157 | if (!noflush && do_lockfs) { | |
3158 | r = lock_fs(md); | |
d67ee213 MS |
3159 | if (r) { |
3160 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3161 | return r; |
d67ee213 | 3162 | } |
aa8d7c2f | 3163 | } |
1da177e4 LT |
3164 | |
3165 | /* | |
3b00b203 MP |
3166 | * Here we must make sure that no processes are submitting requests |
3167 | * to target drivers i.e. no one may be executing | |
3168 | * __split_and_process_bio. This is called from dm_request and | |
3169 | * dm_wq_work. | |
3170 | * | |
3171 | * To get all processes out of __split_and_process_bio in dm_request, | |
3172 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3173 | * __split_and_process_bio from dm_request and quiesce the thread |
3174 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3175 | * flush_workqueue(md->wq). | |
1da177e4 | 3176 | */ |
1eb787ec | 3177 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3178 | if (map) |
3179 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3180 | |
d0bcb878 | 3181 | /* |
29e4013d TH |
3182 | * Stop md->queue before flushing md->wq in case request-based |
3183 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3184 | */ |
2eb6e1e3 | 3185 | if (dm_request_based(md)) { |
9f518b27 | 3186 | stop_queue(md->queue); |
02233342 MS |
3187 | if (md->kworker_task) |
3188 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3189 | } |
cec47e3d | 3190 | |
d0bcb878 KU |
3191 | flush_workqueue(md->wq); |
3192 | ||
1da177e4 | 3193 | /* |
3b00b203 MP |
3194 | * At this point no more requests are entering target request routines. |
3195 | * We call dm_wait_for_completion to wait for all existing requests | |
3196 | * to finish. | |
1da177e4 | 3197 | */ |
ffcc3936 | 3198 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3199 | |
6d6f10df | 3200 | if (noflush) |
022c2611 | 3201 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3202 | if (map) |
3203 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3204 | |
1da177e4 | 3205 | /* were we interrupted ? */ |
46125c1c | 3206 | if (r < 0) { |
9a1fb464 | 3207 | dm_queue_flush(md); |
73d410c0 | 3208 | |
cec47e3d | 3209 | if (dm_request_based(md)) |
9f518b27 | 3210 | start_queue(md->queue); |
cec47e3d | 3211 | |
2ca3310e | 3212 | unlock_fs(md); |
d67ee213 | 3213 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3214 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3215 | } |
1da177e4 | 3216 | |
ffcc3936 MS |
3217 | return r; |
3218 | } | |
3219 | ||
3220 | /* | |
3221 | * We need to be able to change a mapping table under a mounted | |
3222 | * filesystem. For example we might want to move some data in | |
3223 | * the background. Before the table can be swapped with | |
3224 | * dm_bind_table, dm_suspend must be called to flush any in | |
3225 | * flight bios and ensure that any further io gets deferred. | |
3226 | */ | |
3227 | /* | |
3228 | * Suspend mechanism in request-based dm. | |
3229 | * | |
3230 | * 1. Flush all I/Os by lock_fs() if needed. | |
3231 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3232 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3233 | * | |
3234 | * To abort suspend, start the request_queue. | |
3235 | */ | |
3236 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3237 | { | |
3238 | struct dm_table *map = NULL; | |
3239 | int r = 0; | |
3240 | ||
3241 | retry: | |
3242 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3243 | ||
3244 | if (dm_suspended_md(md)) { | |
3245 | r = -EINVAL; | |
3246 | goto out_unlock; | |
3247 | } | |
3248 | ||
3249 | if (dm_suspended_internally_md(md)) { | |
3250 | /* already internally suspended, wait for internal resume */ | |
3251 | mutex_unlock(&md->suspend_lock); | |
3252 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3253 | if (r) | |
3254 | return r; | |
3255 | goto retry; | |
3256 | } | |
3257 | ||
a12f5d48 | 3258 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3259 | |
3260 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3261 | if (r) | |
3262 | goto out_unlock; | |
3b00b203 | 3263 | |
2ca3310e | 3264 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3265 | |
4d4471cb KU |
3266 | dm_table_postsuspend_targets(map); |
3267 | ||
d287483d | 3268 | out_unlock: |
e61290a4 | 3269 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3270 | return r; |
1da177e4 LT |
3271 | } |
3272 | ||
ffcc3936 MS |
3273 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3274 | { | |
3275 | if (map) { | |
3276 | int r = dm_table_resume_targets(map); | |
3277 | if (r) | |
3278 | return r; | |
3279 | } | |
3280 | ||
3281 | dm_queue_flush(md); | |
3282 | ||
3283 | /* | |
3284 | * Flushing deferred I/Os must be done after targets are resumed | |
3285 | * so that mapping of targets can work correctly. | |
3286 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3287 | */ | |
3288 | if (dm_request_based(md)) | |
3289 | start_queue(md->queue); | |
3290 | ||
3291 | unlock_fs(md); | |
3292 | ||
3293 | return 0; | |
3294 | } | |
3295 | ||
1da177e4 LT |
3296 | int dm_resume(struct mapped_device *md) |
3297 | { | |
cf222b37 | 3298 | int r = -EINVAL; |
cf222b37 | 3299 | struct dm_table *map = NULL; |
1da177e4 | 3300 | |
ffcc3936 MS |
3301 | retry: |
3302 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3303 | ||
4f186f8b | 3304 | if (!dm_suspended_md(md)) |
cf222b37 | 3305 | goto out; |
cf222b37 | 3306 | |
ffcc3936 MS |
3307 | if (dm_suspended_internally_md(md)) { |
3308 | /* already internally suspended, wait for internal resume */ | |
3309 | mutex_unlock(&md->suspend_lock); | |
3310 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3311 | if (r) | |
3312 | return r; | |
3313 | goto retry; | |
3314 | } | |
3315 | ||
a12f5d48 | 3316 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3317 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3318 | goto out; |
1da177e4 | 3319 | |
ffcc3936 | 3320 | r = __dm_resume(md, map); |
8757b776 MB |
3321 | if (r) |
3322 | goto out; | |
2ca3310e | 3323 | |
2ca3310e AK |
3324 | clear_bit(DMF_SUSPENDED, &md->flags); |
3325 | ||
cf222b37 AK |
3326 | r = 0; |
3327 | out: | |
e61290a4 | 3328 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3329 | |
cf222b37 | 3330 | return r; |
1da177e4 LT |
3331 | } |
3332 | ||
fd2ed4d2 MP |
3333 | /* |
3334 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3335 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3336 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3337 | */ |
3338 | ||
ffcc3936 | 3339 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3340 | { |
ffcc3936 MS |
3341 | struct dm_table *map = NULL; |
3342 | ||
96b26c8c | 3343 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3344 | return; /* nested internal suspend */ |
3345 | ||
3346 | if (dm_suspended_md(md)) { | |
3347 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3348 | return; /* nest suspend */ | |
3349 | } | |
3350 | ||
a12f5d48 | 3351 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3352 | |
3353 | /* | |
3354 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3355 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3356 | * would require changing .presuspend to return an error -- avoid this | |
3357 | * until there is a need for more elaborate variants of internal suspend. | |
3358 | */ | |
3359 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3360 | ||
3361 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3362 | ||
3363 | dm_table_postsuspend_targets(map); | |
3364 | } | |
3365 | ||
3366 | static void __dm_internal_resume(struct mapped_device *md) | |
3367 | { | |
96b26c8c MP |
3368 | BUG_ON(!md->internal_suspend_count); |
3369 | ||
3370 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3371 | return; /* resume from nested internal suspend */ |
3372 | ||
fd2ed4d2 | 3373 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3374 | goto done; /* resume from nested suspend */ |
3375 | ||
3376 | /* | |
3377 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3378 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3379 | */ | |
3380 | (void) __dm_resume(md, NULL); | |
3381 | ||
3382 | done: | |
3383 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3384 | smp_mb__after_atomic(); | |
3385 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3386 | } | |
3387 | ||
3388 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3389 | { | |
3390 | mutex_lock(&md->suspend_lock); | |
3391 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3392 | mutex_unlock(&md->suspend_lock); | |
3393 | } | |
3394 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3395 | ||
3396 | void dm_internal_resume(struct mapped_device *md) | |
3397 | { | |
3398 | mutex_lock(&md->suspend_lock); | |
3399 | __dm_internal_resume(md); | |
3400 | mutex_unlock(&md->suspend_lock); | |
3401 | } | |
3402 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3403 | ||
3404 | /* | |
3405 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3406 | * which prevents interaction with userspace-driven suspend. | |
3407 | */ | |
3408 | ||
3409 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3410 | { | |
3411 | mutex_lock(&md->suspend_lock); | |
3412 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3413 | return; |
3414 | ||
3415 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3416 | synchronize_srcu(&md->io_barrier); | |
3417 | flush_workqueue(md->wq); | |
3418 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3419 | } | |
b735fede | 3420 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3421 | |
ffcc3936 | 3422 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3423 | { |
ffcc3936 | 3424 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3425 | goto done; |
3426 | ||
3427 | dm_queue_flush(md); | |
3428 | ||
3429 | done: | |
3430 | mutex_unlock(&md->suspend_lock); | |
3431 | } | |
b735fede | 3432 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3433 | |
1da177e4 LT |
3434 | /*----------------------------------------------------------------- |
3435 | * Event notification. | |
3436 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3437 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3438 | unsigned cookie) |
69267a30 | 3439 | { |
60935eb2 MB |
3440 | char udev_cookie[DM_COOKIE_LENGTH]; |
3441 | char *envp[] = { udev_cookie, NULL }; | |
3442 | ||
3443 | if (!cookie) | |
3abf85b5 | 3444 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3445 | else { |
3446 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3447 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3448 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3449 | action, envp); | |
60935eb2 | 3450 | } |
69267a30 AK |
3451 | } |
3452 | ||
7a8c3d3b MA |
3453 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3454 | { | |
3455 | return atomic_add_return(1, &md->uevent_seq); | |
3456 | } | |
3457 | ||
1da177e4 LT |
3458 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3459 | { | |
3460 | return atomic_read(&md->event_nr); | |
3461 | } | |
3462 | ||
3463 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3464 | { | |
3465 | return wait_event_interruptible(md->eventq, | |
3466 | (event_nr != atomic_read(&md->event_nr))); | |
3467 | } | |
3468 | ||
7a8c3d3b MA |
3469 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3470 | { | |
3471 | unsigned long flags; | |
3472 | ||
3473 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3474 | list_add(elist, &md->uevent_list); | |
3475 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3476 | } | |
3477 | ||
1da177e4 LT |
3478 | /* |
3479 | * The gendisk is only valid as long as you have a reference | |
3480 | * count on 'md'. | |
3481 | */ | |
3482 | struct gendisk *dm_disk(struct mapped_device *md) | |
3483 | { | |
3484 | return md->disk; | |
3485 | } | |
3486 | ||
784aae73 MB |
3487 | struct kobject *dm_kobject(struct mapped_device *md) |
3488 | { | |
2995fa78 | 3489 | return &md->kobj_holder.kobj; |
784aae73 MB |
3490 | } |
3491 | ||
784aae73 MB |
3492 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3493 | { | |
3494 | struct mapped_device *md; | |
3495 | ||
2995fa78 | 3496 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3497 | |
4d89b7b4 | 3498 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3499 | dm_deleting_md(md)) |
4d89b7b4 MB |
3500 | return NULL; |
3501 | ||
784aae73 MB |
3502 | dm_get(md); |
3503 | return md; | |
3504 | } | |
3505 | ||
4f186f8b | 3506 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3507 | { |
3508 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3509 | } | |
3510 | ||
ffcc3936 MS |
3511 | int dm_suspended_internally_md(struct mapped_device *md) |
3512 | { | |
3513 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3514 | } | |
3515 | ||
2c140a24 MP |
3516 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3517 | { | |
3518 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3519 | } | |
3520 | ||
64dbce58 KU |
3521 | int dm_suspended(struct dm_target *ti) |
3522 | { | |
ecdb2e25 | 3523 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3524 | } |
3525 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3526 | ||
2e93ccc1 KU |
3527 | int dm_noflush_suspending(struct dm_target *ti) |
3528 | { | |
ecdb2e25 | 3529 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3530 | } |
3531 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3532 | ||
17e149b8 MS |
3533 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3534 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3535 | { |
5f015204 | 3536 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
17e149b8 | 3537 | struct kmem_cache *cachep = NULL; |
e5863d9a | 3538 | unsigned int pool_size = 0; |
5f015204 | 3539 | unsigned int front_pad; |
e6ee8c0b KU |
3540 | |
3541 | if (!pools) | |
3542 | return NULL; | |
3543 | ||
17e149b8 MS |
3544 | type = filter_md_type(type, md); |
3545 | ||
e5863d9a MS |
3546 | switch (type) { |
3547 | case DM_TYPE_BIO_BASED: | |
5f015204 | 3548 | cachep = _io_cache; |
e8603136 | 3549 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 | 3550 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
e5863d9a MS |
3551 | break; |
3552 | case DM_TYPE_REQUEST_BASED: | |
17e149b8 | 3553 | cachep = _rq_tio_cache; |
f4790826 | 3554 | pool_size = dm_get_reserved_rq_based_ios(); |
1ae49ea2 MS |
3555 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); |
3556 | if (!pools->rq_pool) | |
3557 | goto out; | |
e5863d9a MS |
3558 | /* fall through to setup remaining rq-based pools */ |
3559 | case DM_TYPE_MQ_REQUEST_BASED: | |
e5863d9a MS |
3560 | if (!pool_size) |
3561 | pool_size = dm_get_reserved_rq_based_ios(); | |
5f015204 JN |
3562 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3563 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3564 | WARN_ON(per_bio_data_size != 0); | |
e5863d9a MS |
3565 | break; |
3566 | default: | |
17e149b8 | 3567 | BUG(); |
e5863d9a | 3568 | } |
e6ee8c0b | 3569 | |
17e149b8 MS |
3570 | if (cachep) { |
3571 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3572 | if (!pools->io_pool) | |
3573 | goto out; | |
3574 | } | |
e6ee8c0b | 3575 | |
3d8aab2d | 3576 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3577 | if (!pools->bs) |
5f015204 | 3578 | goto out; |
e6ee8c0b | 3579 | |
a91a2785 | 3580 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3581 | goto out; |
a91a2785 | 3582 | |
e6ee8c0b KU |
3583 | return pools; |
3584 | ||
5f015204 JN |
3585 | out: |
3586 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3587 | |
3588 | return NULL; | |
3589 | } | |
3590 | ||
3591 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3592 | { | |
3593 | if (!pools) | |
3594 | return; | |
3595 | ||
3596 | if (pools->io_pool) | |
3597 | mempool_destroy(pools->io_pool); | |
3598 | ||
1ae49ea2 MS |
3599 | if (pools->rq_pool) |
3600 | mempool_destroy(pools->rq_pool); | |
3601 | ||
e6ee8c0b KU |
3602 | if (pools->bs) |
3603 | bioset_free(pools->bs); | |
3604 | ||
3605 | kfree(pools); | |
3606 | } | |
3607 | ||
83d5cde4 | 3608 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3609 | .open = dm_blk_open, |
3610 | .release = dm_blk_close, | |
aa129a22 | 3611 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3612 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3613 | .owner = THIS_MODULE |
3614 | }; | |
3615 | ||
1da177e4 LT |
3616 | /* |
3617 | * module hooks | |
3618 | */ | |
3619 | module_init(dm_init); | |
3620 | module_exit(dm_exit); | |
3621 | ||
3622 | module_param(major, uint, 0); | |
3623 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3624 | |
e8603136 MS |
3625 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3626 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3627 | ||
f4790826 MS |
3628 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3629 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3630 | ||
17e149b8 MS |
3631 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3632 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3633 | ||
1da177e4 LT |
3634 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3635 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3636 | MODULE_LICENSE("GPL"); |