Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
ffcc3936 | 22 | #include <linux/wait.h> |
2eb6e1e3 | 23 | #include <linux/kthread.h> |
0ce65797 | 24 | #include <linux/ktime.h> |
de3ec86d | 25 | #include <linux/elevator.h> /* for rq_end_sector() */ |
bfebd1cd | 26 | #include <linux/blk-mq.h> |
55782138 LZ |
27 | |
28 | #include <trace/events/block.h> | |
1da177e4 | 29 | |
72d94861 AK |
30 | #define DM_MSG_PREFIX "core" |
31 | ||
71a16736 NK |
32 | #ifdef CONFIG_PRINTK |
33 | /* | |
34 | * ratelimit state to be used in DMXXX_LIMIT(). | |
35 | */ | |
36 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
37 | DEFAULT_RATELIMIT_INTERVAL, | |
38 | DEFAULT_RATELIMIT_BURST); | |
39 | EXPORT_SYMBOL(dm_ratelimit_state); | |
40 | #endif | |
41 | ||
60935eb2 MB |
42 | /* |
43 | * Cookies are numeric values sent with CHANGE and REMOVE | |
44 | * uevents while resuming, removing or renaming the device. | |
45 | */ | |
46 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
47 | #define DM_COOKIE_LENGTH 24 | |
48 | ||
1da177e4 LT |
49 | static const char *_name = DM_NAME; |
50 | ||
51 | static unsigned int major = 0; | |
52 | static unsigned int _major = 0; | |
53 | ||
d15b774c AK |
54 | static DEFINE_IDR(_minor_idr); |
55 | ||
f32c10b0 | 56 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
57 | |
58 | static void do_deferred_remove(struct work_struct *w); | |
59 | ||
60 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
61 | ||
acfe0ad7 MP |
62 | static struct workqueue_struct *deferred_remove_workqueue; |
63 | ||
1da177e4 | 64 | /* |
8fbf26ad | 65 | * For bio-based dm. |
1da177e4 LT |
66 | * One of these is allocated per bio. |
67 | */ | |
68 | struct dm_io { | |
69 | struct mapped_device *md; | |
70 | int error; | |
1da177e4 | 71 | atomic_t io_count; |
6ae2fa67 | 72 | struct bio *bio; |
3eaf840e | 73 | unsigned long start_time; |
f88fb981 | 74 | spinlock_t endio_lock; |
fd2ed4d2 | 75 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
76 | }; |
77 | ||
8fbf26ad KU |
78 | /* |
79 | * For request-based dm. | |
80 | * One of these is allocated per request. | |
81 | */ | |
82 | struct dm_rq_target_io { | |
83 | struct mapped_device *md; | |
84 | struct dm_target *ti; | |
1ae49ea2 | 85 | struct request *orig, *clone; |
2eb6e1e3 | 86 | struct kthread_work work; |
8fbf26ad KU |
87 | int error; |
88 | union map_info info; | |
e262f347 MP |
89 | struct dm_stats_aux stats_aux; |
90 | unsigned long duration_jiffies; | |
91 | unsigned n_sectors; | |
8fbf26ad KU |
92 | }; |
93 | ||
94 | /* | |
94818742 KO |
95 | * For request-based dm - the bio clones we allocate are embedded in these |
96 | * structs. | |
97 | * | |
98 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
99 | * the bioset is created - this means the bio has to come at the end of the | |
100 | * struct. | |
8fbf26ad KU |
101 | */ |
102 | struct dm_rq_clone_bio_info { | |
103 | struct bio *orig; | |
cec47e3d | 104 | struct dm_rq_target_io *tio; |
94818742 | 105 | struct bio clone; |
8fbf26ad KU |
106 | }; |
107 | ||
cec47e3d KU |
108 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
109 | { | |
110 | if (rq && rq->end_io_data) | |
111 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
112 | return NULL; | |
113 | } | |
114 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
115 | ||
ba61fdd1 JM |
116 | #define MINOR_ALLOCED ((void *)-1) |
117 | ||
1da177e4 LT |
118 | /* |
119 | * Bits for the md->flags field. | |
120 | */ | |
1eb787ec | 121 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 122 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 123 | #define DMF_FROZEN 2 |
fba9f90e | 124 | #define DMF_FREEING 3 |
5c6bd75d | 125 | #define DMF_DELETING 4 |
2e93ccc1 | 126 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 127 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 128 | #define DMF_DEFERRED_REMOVE 7 |
ffcc3936 | 129 | #define DMF_SUSPENDED_INTERNALLY 8 |
1da177e4 | 130 | |
83d5e5b0 MP |
131 | /* |
132 | * A dummy definition to make RCU happy. | |
133 | * struct dm_table should never be dereferenced in this file. | |
134 | */ | |
135 | struct dm_table { | |
136 | int undefined__; | |
137 | }; | |
138 | ||
304f3f6a MB |
139 | /* |
140 | * Work processed by per-device workqueue. | |
141 | */ | |
1da177e4 | 142 | struct mapped_device { |
83d5e5b0 | 143 | struct srcu_struct io_barrier; |
e61290a4 | 144 | struct mutex suspend_lock; |
1da177e4 | 145 | atomic_t holders; |
5c6bd75d | 146 | atomic_t open_count; |
1da177e4 | 147 | |
2a7faeb1 MP |
148 | /* |
149 | * The current mapping. | |
150 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
151 | * dereference. | |
152 | */ | |
6fa99520 | 153 | struct dm_table __rcu *map; |
2a7faeb1 | 154 | |
86f1152b BM |
155 | struct list_head table_devices; |
156 | struct mutex table_devices_lock; | |
157 | ||
1da177e4 LT |
158 | unsigned long flags; |
159 | ||
165125e1 | 160 | struct request_queue *queue; |
a5664dad | 161 | unsigned type; |
4a0b4ddf | 162 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
163 | struct mutex type_lock; |
164 | ||
36a0456f AK |
165 | struct target_type *immutable_target_type; |
166 | ||
1da177e4 | 167 | struct gendisk *disk; |
7e51f257 | 168 | char name[16]; |
1da177e4 LT |
169 | |
170 | void *interface_ptr; | |
171 | ||
172 | /* | |
173 | * A list of ios that arrived while we were suspended. | |
174 | */ | |
316d315b | 175 | atomic_t pending[2]; |
1da177e4 | 176 | wait_queue_head_t wait; |
53d5914f | 177 | struct work_struct work; |
74859364 | 178 | struct bio_list deferred; |
022c2611 | 179 | spinlock_t deferred_lock; |
1da177e4 | 180 | |
af7e466a | 181 | /* |
29e4013d | 182 | * Processing queue (flush) |
304f3f6a MB |
183 | */ |
184 | struct workqueue_struct *wq; | |
185 | ||
1da177e4 LT |
186 | /* |
187 | * io objects are allocated from here. | |
188 | */ | |
189 | mempool_t *io_pool; | |
1ae49ea2 | 190 | mempool_t *rq_pool; |
1da177e4 | 191 | |
9faf400f SB |
192 | struct bio_set *bs; |
193 | ||
1da177e4 LT |
194 | /* |
195 | * Event handling. | |
196 | */ | |
197 | atomic_t event_nr; | |
198 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
199 | atomic_t uevent_seq; |
200 | struct list_head uevent_list; | |
201 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
202 | |
203 | /* | |
204 | * freeze/thaw support require holding onto a super block | |
205 | */ | |
206 | struct super_block *frozen_sb; | |
db8fef4f | 207 | struct block_device *bdev; |
3ac51e74 DW |
208 | |
209 | /* forced geometry settings */ | |
210 | struct hd_geometry geometry; | |
784aae73 | 211 | |
2995fa78 MP |
212 | /* kobject and completion */ |
213 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 214 | |
d87f4c14 TH |
215 | /* zero-length flush that will be cloned and submitted to targets */ |
216 | struct bio flush_bio; | |
fd2ed4d2 | 217 | |
96b26c8c MP |
218 | /* the number of internal suspends */ |
219 | unsigned internal_suspend_count; | |
220 | ||
fd2ed4d2 | 221 | struct dm_stats stats; |
2eb6e1e3 KB |
222 | |
223 | struct kthread_worker kworker; | |
224 | struct task_struct *kworker_task; | |
de3ec86d MS |
225 | |
226 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 227 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 228 | int last_rq_rw; |
0ce65797 MS |
229 | sector_t last_rq_pos; |
230 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
231 | |
232 | /* for blk-mq request-based DM support */ | |
233 | struct blk_mq_tag_set tag_set; | |
17e149b8 | 234 | bool use_blk_mq; |
1da177e4 LT |
235 | }; |
236 | ||
17e149b8 MS |
237 | #ifdef CONFIG_DM_MQ_DEFAULT |
238 | static bool use_blk_mq = true; | |
239 | #else | |
240 | static bool use_blk_mq = false; | |
241 | #endif | |
242 | ||
243 | bool dm_use_blk_mq(struct mapped_device *md) | |
244 | { | |
245 | return md->use_blk_mq; | |
246 | } | |
247 | ||
e6ee8c0b KU |
248 | /* |
249 | * For mempools pre-allocation at the table loading time. | |
250 | */ | |
251 | struct dm_md_mempools { | |
252 | mempool_t *io_pool; | |
1ae49ea2 | 253 | mempool_t *rq_pool; |
e6ee8c0b KU |
254 | struct bio_set *bs; |
255 | }; | |
256 | ||
86f1152b BM |
257 | struct table_device { |
258 | struct list_head list; | |
259 | atomic_t count; | |
260 | struct dm_dev dm_dev; | |
261 | }; | |
262 | ||
6cfa5857 MS |
263 | #define RESERVED_BIO_BASED_IOS 16 |
264 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 265 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 266 | static struct kmem_cache *_io_cache; |
8fbf26ad | 267 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 268 | static struct kmem_cache *_rq_cache; |
94818742 | 269 | |
e8603136 MS |
270 | /* |
271 | * Bio-based DM's mempools' reserved IOs set by the user. | |
272 | */ | |
273 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
274 | ||
f4790826 MS |
275 | /* |
276 | * Request-based DM's mempools' reserved IOs set by the user. | |
277 | */ | |
278 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
279 | ||
09c2d531 | 280 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
281 | unsigned def, unsigned max) |
282 | { | |
09c2d531 MS |
283 | unsigned param = ACCESS_ONCE(*module_param); |
284 | unsigned modified_param = 0; | |
f4790826 | 285 | |
09c2d531 MS |
286 | if (!param) |
287 | modified_param = def; | |
288 | else if (param > max) | |
289 | modified_param = max; | |
f4790826 | 290 | |
09c2d531 MS |
291 | if (modified_param) { |
292 | (void)cmpxchg(module_param, param, modified_param); | |
293 | param = modified_param; | |
f4790826 MS |
294 | } |
295 | ||
09c2d531 | 296 | return param; |
f4790826 MS |
297 | } |
298 | ||
e8603136 MS |
299 | unsigned dm_get_reserved_bio_based_ios(void) |
300 | { | |
09c2d531 | 301 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
302 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
303 | } | |
304 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
305 | ||
f4790826 MS |
306 | unsigned dm_get_reserved_rq_based_ios(void) |
307 | { | |
09c2d531 | 308 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
309 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
310 | } | |
311 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
312 | ||
1da177e4 LT |
313 | static int __init local_init(void) |
314 | { | |
51157b4a | 315 | int r = -ENOMEM; |
1da177e4 | 316 | |
1da177e4 | 317 | /* allocate a slab for the dm_ios */ |
028867ac | 318 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 319 | if (!_io_cache) |
51157b4a | 320 | return r; |
1da177e4 | 321 | |
8fbf26ad KU |
322 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
323 | if (!_rq_tio_cache) | |
dba14160 | 324 | goto out_free_io_cache; |
8fbf26ad | 325 | |
1ae49ea2 MS |
326 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
327 | __alignof__(struct request), 0, NULL); | |
328 | if (!_rq_cache) | |
329 | goto out_free_rq_tio_cache; | |
330 | ||
51e5b2bd | 331 | r = dm_uevent_init(); |
51157b4a | 332 | if (r) |
1ae49ea2 | 333 | goto out_free_rq_cache; |
51e5b2bd | 334 | |
acfe0ad7 MP |
335 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
336 | if (!deferred_remove_workqueue) { | |
337 | r = -ENOMEM; | |
338 | goto out_uevent_exit; | |
339 | } | |
340 | ||
1da177e4 LT |
341 | _major = major; |
342 | r = register_blkdev(_major, _name); | |
51157b4a | 343 | if (r < 0) |
acfe0ad7 | 344 | goto out_free_workqueue; |
1da177e4 LT |
345 | |
346 | if (!_major) | |
347 | _major = r; | |
348 | ||
349 | return 0; | |
51157b4a | 350 | |
acfe0ad7 MP |
351 | out_free_workqueue: |
352 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
353 | out_uevent_exit: |
354 | dm_uevent_exit(); | |
1ae49ea2 MS |
355 | out_free_rq_cache: |
356 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
357 | out_free_rq_tio_cache: |
358 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
359 | out_free_io_cache: |
360 | kmem_cache_destroy(_io_cache); | |
361 | ||
362 | return r; | |
1da177e4 LT |
363 | } |
364 | ||
365 | static void local_exit(void) | |
366 | { | |
2c140a24 | 367 | flush_scheduled_work(); |
acfe0ad7 | 368 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 369 | |
1ae49ea2 | 370 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 371 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 372 | kmem_cache_destroy(_io_cache); |
00d59405 | 373 | unregister_blkdev(_major, _name); |
51e5b2bd | 374 | dm_uevent_exit(); |
1da177e4 LT |
375 | |
376 | _major = 0; | |
377 | ||
378 | DMINFO("cleaned up"); | |
379 | } | |
380 | ||
b9249e55 | 381 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
382 | local_init, |
383 | dm_target_init, | |
384 | dm_linear_init, | |
385 | dm_stripe_init, | |
952b3557 | 386 | dm_io_init, |
945fa4d2 | 387 | dm_kcopyd_init, |
1da177e4 | 388 | dm_interface_init, |
fd2ed4d2 | 389 | dm_statistics_init, |
1da177e4 LT |
390 | }; |
391 | ||
b9249e55 | 392 | static void (*_exits[])(void) = { |
1da177e4 LT |
393 | local_exit, |
394 | dm_target_exit, | |
395 | dm_linear_exit, | |
396 | dm_stripe_exit, | |
952b3557 | 397 | dm_io_exit, |
945fa4d2 | 398 | dm_kcopyd_exit, |
1da177e4 | 399 | dm_interface_exit, |
fd2ed4d2 | 400 | dm_statistics_exit, |
1da177e4 LT |
401 | }; |
402 | ||
403 | static int __init dm_init(void) | |
404 | { | |
405 | const int count = ARRAY_SIZE(_inits); | |
406 | ||
407 | int r, i; | |
408 | ||
409 | for (i = 0; i < count; i++) { | |
410 | r = _inits[i](); | |
411 | if (r) | |
412 | goto bad; | |
413 | } | |
414 | ||
415 | return 0; | |
416 | ||
417 | bad: | |
418 | while (i--) | |
419 | _exits[i](); | |
420 | ||
421 | return r; | |
422 | } | |
423 | ||
424 | static void __exit dm_exit(void) | |
425 | { | |
426 | int i = ARRAY_SIZE(_exits); | |
427 | ||
428 | while (i--) | |
429 | _exits[i](); | |
d15b774c AK |
430 | |
431 | /* | |
432 | * Should be empty by this point. | |
433 | */ | |
d15b774c | 434 | idr_destroy(&_minor_idr); |
1da177e4 LT |
435 | } |
436 | ||
437 | /* | |
438 | * Block device functions | |
439 | */ | |
432a212c MA |
440 | int dm_deleting_md(struct mapped_device *md) |
441 | { | |
442 | return test_bit(DMF_DELETING, &md->flags); | |
443 | } | |
444 | ||
fe5f9f2c | 445 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
446 | { |
447 | struct mapped_device *md; | |
448 | ||
fba9f90e JM |
449 | spin_lock(&_minor_lock); |
450 | ||
fe5f9f2c | 451 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
452 | if (!md) |
453 | goto out; | |
454 | ||
5c6bd75d | 455 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 456 | dm_deleting_md(md)) { |
fba9f90e JM |
457 | md = NULL; |
458 | goto out; | |
459 | } | |
460 | ||
1da177e4 | 461 | dm_get(md); |
5c6bd75d | 462 | atomic_inc(&md->open_count); |
fba9f90e JM |
463 | out: |
464 | spin_unlock(&_minor_lock); | |
465 | ||
466 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
467 | } |
468 | ||
db2a144b | 469 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 470 | { |
63a4f065 | 471 | struct mapped_device *md; |
6e9624b8 | 472 | |
4a1aeb98 MB |
473 | spin_lock(&_minor_lock); |
474 | ||
63a4f065 MS |
475 | md = disk->private_data; |
476 | if (WARN_ON(!md)) | |
477 | goto out; | |
478 | ||
2c140a24 MP |
479 | if (atomic_dec_and_test(&md->open_count) && |
480 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 481 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 482 | |
1da177e4 | 483 | dm_put(md); |
63a4f065 | 484 | out: |
4a1aeb98 | 485 | spin_unlock(&_minor_lock); |
1da177e4 LT |
486 | } |
487 | ||
5c6bd75d AK |
488 | int dm_open_count(struct mapped_device *md) |
489 | { | |
490 | return atomic_read(&md->open_count); | |
491 | } | |
492 | ||
493 | /* | |
494 | * Guarantees nothing is using the device before it's deleted. | |
495 | */ | |
2c140a24 | 496 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
497 | { |
498 | int r = 0; | |
499 | ||
500 | spin_lock(&_minor_lock); | |
501 | ||
2c140a24 | 502 | if (dm_open_count(md)) { |
5c6bd75d | 503 | r = -EBUSY; |
2c140a24 MP |
504 | if (mark_deferred) |
505 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
506 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
507 | r = -EEXIST; | |
5c6bd75d AK |
508 | else |
509 | set_bit(DMF_DELETING, &md->flags); | |
510 | ||
511 | spin_unlock(&_minor_lock); | |
512 | ||
513 | return r; | |
514 | } | |
515 | ||
2c140a24 MP |
516 | int dm_cancel_deferred_remove(struct mapped_device *md) |
517 | { | |
518 | int r = 0; | |
519 | ||
520 | spin_lock(&_minor_lock); | |
521 | ||
522 | if (test_bit(DMF_DELETING, &md->flags)) | |
523 | r = -EBUSY; | |
524 | else | |
525 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
526 | ||
527 | spin_unlock(&_minor_lock); | |
528 | ||
529 | return r; | |
530 | } | |
531 | ||
532 | static void do_deferred_remove(struct work_struct *w) | |
533 | { | |
534 | dm_deferred_remove(); | |
535 | } | |
536 | ||
fd2ed4d2 MP |
537 | sector_t dm_get_size(struct mapped_device *md) |
538 | { | |
539 | return get_capacity(md->disk); | |
540 | } | |
541 | ||
9974fa2c MS |
542 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
543 | { | |
544 | return md->queue; | |
545 | } | |
546 | ||
fd2ed4d2 MP |
547 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
548 | { | |
549 | return &md->stats; | |
550 | } | |
551 | ||
3ac51e74 DW |
552 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
553 | { | |
554 | struct mapped_device *md = bdev->bd_disk->private_data; | |
555 | ||
556 | return dm_get_geometry(md, geo); | |
557 | } | |
558 | ||
fe5f9f2c | 559 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
560 | unsigned int cmd, unsigned long arg) |
561 | { | |
fe5f9f2c | 562 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 563 | int srcu_idx; |
6c182cd8 | 564 | struct dm_table *map; |
aa129a22 MB |
565 | struct dm_target *tgt; |
566 | int r = -ENOTTY; | |
567 | ||
6c182cd8 | 568 | retry: |
83d5e5b0 MP |
569 | map = dm_get_live_table(md, &srcu_idx); |
570 | ||
aa129a22 MB |
571 | if (!map || !dm_table_get_size(map)) |
572 | goto out; | |
573 | ||
574 | /* We only support devices that have a single target */ | |
575 | if (dm_table_get_num_targets(map) != 1) | |
576 | goto out; | |
577 | ||
578 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
579 | if (!tgt->type->ioctl) |
580 | goto out; | |
aa129a22 | 581 | |
4f186f8b | 582 | if (dm_suspended_md(md)) { |
aa129a22 MB |
583 | r = -EAGAIN; |
584 | goto out; | |
585 | } | |
586 | ||
4d341d82 | 587 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
588 | |
589 | out: | |
83d5e5b0 | 590 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 591 | |
6c182cd8 HR |
592 | if (r == -ENOTCONN) { |
593 | msleep(10); | |
594 | goto retry; | |
595 | } | |
596 | ||
aa129a22 MB |
597 | return r; |
598 | } | |
599 | ||
028867ac | 600 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
601 | { |
602 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
603 | } | |
604 | ||
028867ac | 605 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
606 | { |
607 | mempool_free(io, md->io_pool); | |
608 | } | |
609 | ||
028867ac | 610 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 611 | { |
dba14160 | 612 | bio_put(&tio->clone); |
1da177e4 LT |
613 | } |
614 | ||
08885643 KU |
615 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
616 | gfp_t gfp_mask) | |
cec47e3d | 617 | { |
5f015204 | 618 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
619 | } |
620 | ||
621 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
622 | { | |
5f015204 | 623 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
624 | } |
625 | ||
1ae49ea2 MS |
626 | static struct request *alloc_clone_request(struct mapped_device *md, |
627 | gfp_t gfp_mask) | |
628 | { | |
629 | return mempool_alloc(md->rq_pool, gfp_mask); | |
630 | } | |
631 | ||
632 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
633 | { | |
634 | mempool_free(rq, md->rq_pool); | |
635 | } | |
636 | ||
90abb8c4 KU |
637 | static int md_in_flight(struct mapped_device *md) |
638 | { | |
639 | return atomic_read(&md->pending[READ]) + | |
640 | atomic_read(&md->pending[WRITE]); | |
641 | } | |
642 | ||
3eaf840e JNN |
643 | static void start_io_acct(struct dm_io *io) |
644 | { | |
645 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 646 | struct bio *bio = io->bio; |
c9959059 | 647 | int cpu; |
fd2ed4d2 | 648 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
649 | |
650 | io->start_time = jiffies; | |
651 | ||
074a7aca TH |
652 | cpu = part_stat_lock(); |
653 | part_round_stats(cpu, &dm_disk(md)->part0); | |
654 | part_stat_unlock(); | |
1e9bb880 SL |
655 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
656 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
657 | |
658 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 659 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 660 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
661 | } |
662 | ||
d221d2e7 | 663 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
664 | { |
665 | struct mapped_device *md = io->md; | |
666 | struct bio *bio = io->bio; | |
667 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 668 | int pending; |
3eaf840e JNN |
669 | int rw = bio_data_dir(bio); |
670 | ||
18c0b223 | 671 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 672 | |
fd2ed4d2 | 673 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 674 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
675 | bio_sectors(bio), true, duration, &io->stats_aux); |
676 | ||
af7e466a MP |
677 | /* |
678 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 679 | * a flush. |
af7e466a | 680 | */ |
1e9bb880 SL |
681 | pending = atomic_dec_return(&md->pending[rw]); |
682 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 683 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 684 | |
d221d2e7 MP |
685 | /* nudge anyone waiting on suspend queue */ |
686 | if (!pending) | |
687 | wake_up(&md->wait); | |
3eaf840e JNN |
688 | } |
689 | ||
1da177e4 LT |
690 | /* |
691 | * Add the bio to the list of deferred io. | |
692 | */ | |
92c63902 | 693 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 694 | { |
05447420 | 695 | unsigned long flags; |
1da177e4 | 696 | |
05447420 | 697 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 698 | bio_list_add(&md->deferred, bio); |
05447420 | 699 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 700 | queue_work(md->wq, &md->work); |
1da177e4 LT |
701 | } |
702 | ||
703 | /* | |
704 | * Everyone (including functions in this file), should use this | |
705 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 706 | * dm_put_live_table() when finished. |
1da177e4 | 707 | */ |
83d5e5b0 | 708 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 709 | { |
83d5e5b0 MP |
710 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
711 | ||
712 | return srcu_dereference(md->map, &md->io_barrier); | |
713 | } | |
1da177e4 | 714 | |
83d5e5b0 MP |
715 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
716 | { | |
717 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
718 | } | |
719 | ||
720 | void dm_sync_table(struct mapped_device *md) | |
721 | { | |
722 | synchronize_srcu(&md->io_barrier); | |
723 | synchronize_rcu_expedited(); | |
724 | } | |
725 | ||
726 | /* | |
727 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
728 | * The caller must not block between these two functions. | |
729 | */ | |
730 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
731 | { | |
732 | rcu_read_lock(); | |
733 | return rcu_dereference(md->map); | |
734 | } | |
1da177e4 | 735 | |
83d5e5b0 MP |
736 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
737 | { | |
738 | rcu_read_unlock(); | |
1da177e4 LT |
739 | } |
740 | ||
86f1152b BM |
741 | /* |
742 | * Open a table device so we can use it as a map destination. | |
743 | */ | |
744 | static int open_table_device(struct table_device *td, dev_t dev, | |
745 | struct mapped_device *md) | |
746 | { | |
747 | static char *_claim_ptr = "I belong to device-mapper"; | |
748 | struct block_device *bdev; | |
749 | ||
750 | int r; | |
751 | ||
752 | BUG_ON(td->dm_dev.bdev); | |
753 | ||
754 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
755 | if (IS_ERR(bdev)) | |
756 | return PTR_ERR(bdev); | |
757 | ||
758 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
759 | if (r) { | |
760 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
761 | return r; | |
762 | } | |
763 | ||
764 | td->dm_dev.bdev = bdev; | |
765 | return 0; | |
766 | } | |
767 | ||
768 | /* | |
769 | * Close a table device that we've been using. | |
770 | */ | |
771 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
772 | { | |
773 | if (!td->dm_dev.bdev) | |
774 | return; | |
775 | ||
776 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
777 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
778 | td->dm_dev.bdev = NULL; | |
779 | } | |
780 | ||
781 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
782 | fmode_t mode) { | |
783 | struct table_device *td; | |
784 | ||
785 | list_for_each_entry(td, l, list) | |
786 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
787 | return td; | |
788 | ||
789 | return NULL; | |
790 | } | |
791 | ||
792 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
793 | struct dm_dev **result) { | |
794 | int r; | |
795 | struct table_device *td; | |
796 | ||
797 | mutex_lock(&md->table_devices_lock); | |
798 | td = find_table_device(&md->table_devices, dev, mode); | |
799 | if (!td) { | |
800 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
801 | if (!td) { | |
802 | mutex_unlock(&md->table_devices_lock); | |
803 | return -ENOMEM; | |
804 | } | |
805 | ||
806 | td->dm_dev.mode = mode; | |
807 | td->dm_dev.bdev = NULL; | |
808 | ||
809 | if ((r = open_table_device(td, dev, md))) { | |
810 | mutex_unlock(&md->table_devices_lock); | |
811 | kfree(td); | |
812 | return r; | |
813 | } | |
814 | ||
815 | format_dev_t(td->dm_dev.name, dev); | |
816 | ||
817 | atomic_set(&td->count, 0); | |
818 | list_add(&td->list, &md->table_devices); | |
819 | } | |
820 | atomic_inc(&td->count); | |
821 | mutex_unlock(&md->table_devices_lock); | |
822 | ||
823 | *result = &td->dm_dev; | |
824 | return 0; | |
825 | } | |
826 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
827 | ||
828 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
829 | { | |
830 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
831 | ||
832 | mutex_lock(&md->table_devices_lock); | |
833 | if (atomic_dec_and_test(&td->count)) { | |
834 | close_table_device(td, md); | |
835 | list_del(&td->list); | |
836 | kfree(td); | |
837 | } | |
838 | mutex_unlock(&md->table_devices_lock); | |
839 | } | |
840 | EXPORT_SYMBOL(dm_put_table_device); | |
841 | ||
842 | static void free_table_devices(struct list_head *devices) | |
843 | { | |
844 | struct list_head *tmp, *next; | |
845 | ||
846 | list_for_each_safe(tmp, next, devices) { | |
847 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
848 | ||
849 | DMWARN("dm_destroy: %s still exists with %d references", | |
850 | td->dm_dev.name, atomic_read(&td->count)); | |
851 | kfree(td); | |
852 | } | |
853 | } | |
854 | ||
3ac51e74 DW |
855 | /* |
856 | * Get the geometry associated with a dm device | |
857 | */ | |
858 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
859 | { | |
860 | *geo = md->geometry; | |
861 | ||
862 | return 0; | |
863 | } | |
864 | ||
865 | /* | |
866 | * Set the geometry of a device. | |
867 | */ | |
868 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
869 | { | |
870 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
871 | ||
872 | if (geo->start > sz) { | |
873 | DMWARN("Start sector is beyond the geometry limits."); | |
874 | return -EINVAL; | |
875 | } | |
876 | ||
877 | md->geometry = *geo; | |
878 | ||
879 | return 0; | |
880 | } | |
881 | ||
1da177e4 LT |
882 | /*----------------------------------------------------------------- |
883 | * CRUD START: | |
884 | * A more elegant soln is in the works that uses the queue | |
885 | * merge fn, unfortunately there are a couple of changes to | |
886 | * the block layer that I want to make for this. So in the | |
887 | * interests of getting something for people to use I give | |
888 | * you this clearly demarcated crap. | |
889 | *---------------------------------------------------------------*/ | |
890 | ||
2e93ccc1 KU |
891 | static int __noflush_suspending(struct mapped_device *md) |
892 | { | |
893 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
894 | } | |
895 | ||
1da177e4 LT |
896 | /* |
897 | * Decrements the number of outstanding ios that a bio has been | |
898 | * cloned into, completing the original io if necc. | |
899 | */ | |
858119e1 | 900 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 901 | { |
2e93ccc1 | 902 | unsigned long flags; |
b35f8caa MB |
903 | int io_error; |
904 | struct bio *bio; | |
905 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
906 | |
907 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
908 | if (unlikely(error)) { |
909 | spin_lock_irqsave(&io->endio_lock, flags); | |
910 | if (!(io->error > 0 && __noflush_suspending(md))) | |
911 | io->error = error; | |
912 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
913 | } | |
1da177e4 LT |
914 | |
915 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
916 | if (io->error == DM_ENDIO_REQUEUE) { |
917 | /* | |
918 | * Target requested pushing back the I/O. | |
2e93ccc1 | 919 | */ |
022c2611 | 920 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
921 | if (__noflush_suspending(md)) |
922 | bio_list_add_head(&md->deferred, io->bio); | |
923 | else | |
2e93ccc1 KU |
924 | /* noflush suspend was interrupted. */ |
925 | io->error = -EIO; | |
022c2611 | 926 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
927 | } |
928 | ||
b35f8caa MB |
929 | io_error = io->error; |
930 | bio = io->bio; | |
6a8736d1 TH |
931 | end_io_acct(io); |
932 | free_io(md, io); | |
933 | ||
934 | if (io_error == DM_ENDIO_REQUEUE) | |
935 | return; | |
2e93ccc1 | 936 | |
4f024f37 | 937 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 938 | /* |
6a8736d1 TH |
939 | * Preflush done for flush with data, reissue |
940 | * without REQ_FLUSH. | |
af7e466a | 941 | */ |
6a8736d1 TH |
942 | bio->bi_rw &= ~REQ_FLUSH; |
943 | queue_io(md, bio); | |
af7e466a | 944 | } else { |
b372d360 | 945 | /* done with normal IO or empty flush */ |
0a82a8d1 | 946 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 947 | bio_endio(bio, io_error); |
b35f8caa | 948 | } |
1da177e4 LT |
949 | } |
950 | } | |
951 | ||
7eee4ae2 MS |
952 | static void disable_write_same(struct mapped_device *md) |
953 | { | |
954 | struct queue_limits *limits = dm_get_queue_limits(md); | |
955 | ||
956 | /* device doesn't really support WRITE SAME, disable it */ | |
957 | limits->max_write_same_sectors = 0; | |
958 | } | |
959 | ||
6712ecf8 | 960 | static void clone_endio(struct bio *bio, int error) |
1da177e4 | 961 | { |
5164bece | 962 | int r = error; |
bfc6d41c | 963 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 964 | struct dm_io *io = tio->io; |
9faf400f | 965 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
966 | dm_endio_fn endio = tio->ti->type->end_io; |
967 | ||
1da177e4 LT |
968 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
969 | error = -EIO; | |
970 | ||
971 | if (endio) { | |
7de3ee57 | 972 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
973 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
974 | /* | |
975 | * error and requeue request are handled | |
976 | * in dec_pending(). | |
977 | */ | |
1da177e4 | 978 | error = r; |
45cbcd79 KU |
979 | else if (r == DM_ENDIO_INCOMPLETE) |
980 | /* The target will handle the io */ | |
6712ecf8 | 981 | return; |
45cbcd79 KU |
982 | else if (r) { |
983 | DMWARN("unimplemented target endio return value: %d", r); | |
984 | BUG(); | |
985 | } | |
1da177e4 LT |
986 | } |
987 | ||
7eee4ae2 MS |
988 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
989 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
990 | disable_write_same(md); | |
991 | ||
9faf400f | 992 | free_tio(md, tio); |
b35f8caa | 993 | dec_pending(io, error); |
1da177e4 LT |
994 | } |
995 | ||
bfebd1cd MS |
996 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
997 | { | |
998 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
999 | } | |
1000 | ||
e262f347 MP |
1001 | static void rq_end_stats(struct mapped_device *md, struct request *orig) |
1002 | { | |
1003 | if (unlikely(dm_stats_used(&md->stats))) { | |
1004 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1005 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
1006 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1007 | tio->n_sectors, true, tio->duration_jiffies, | |
1008 | &tio->stats_aux); | |
1009 | } | |
1010 | } | |
1011 | ||
cec47e3d KU |
1012 | /* |
1013 | * Don't touch any member of the md after calling this function because | |
1014 | * the md may be freed in dm_put() at the end of this function. | |
1015 | * Or do dm_get() before calling this function and dm_put() later. | |
1016 | */ | |
466d89a6 | 1017 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1018 | { |
9a0e609e MS |
1019 | int nr_requests_pending; |
1020 | ||
b4324fee | 1021 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1022 | |
1023 | /* nudge anyone waiting on suspend queue */ | |
9a0e609e MS |
1024 | nr_requests_pending = md_in_flight(md); |
1025 | if (!nr_requests_pending) | |
cec47e3d KU |
1026 | wake_up(&md->wait); |
1027 | ||
a8c32a5c JA |
1028 | /* |
1029 | * Run this off this callpath, as drivers could invoke end_io while | |
1030 | * inside their request_fn (and holding the queue lock). Calling | |
1031 | * back into ->request_fn() could deadlock attempting to grab the | |
1032 | * queue lock again. | |
1033 | */ | |
9a0e609e | 1034 | if (run_queue) { |
bfebd1cd MS |
1035 | if (md->queue->mq_ops) |
1036 | blk_mq_run_hw_queues(md->queue, true); | |
1037 | else if (!nr_requests_pending || | |
1038 | (nr_requests_pending >= md->queue->nr_congestion_on)) | |
9a0e609e MS |
1039 | blk_run_queue_async(md->queue); |
1040 | } | |
cec47e3d KU |
1041 | |
1042 | /* | |
1043 | * dm_put() must be at the end of this function. See the comment above | |
1044 | */ | |
1045 | dm_put(md); | |
1046 | } | |
1047 | ||
e5d8de32 | 1048 | static void free_rq_clone(struct request *clone) |
a77e28c7 KU |
1049 | { |
1050 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1051 | struct mapped_device *md = tio->md; |
a77e28c7 | 1052 | |
aa6df8dd MS |
1053 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1054 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1055 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1056 | else if (!md->queue->mq_ops) |
1057 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd | 1058 | free_clone_request(md, clone); |
aa6df8dd MS |
1059 | /* |
1060 | * NOTE: for the blk-mq queue stacked on request_fn queue(s) case: | |
1061 | * no need to call free_clone_request() because we leverage blk-mq by | |
1062 | * allocating the clone at the end of the blk-mq pdu (see: clone_rq) | |
1063 | */ | |
bfebd1cd MS |
1064 | |
1065 | if (!md->queue->mq_ops) | |
1066 | free_rq_tio(tio); | |
a77e28c7 KU |
1067 | } |
1068 | ||
980691e5 KU |
1069 | /* |
1070 | * Complete the clone and the original request. | |
466d89a6 KB |
1071 | * Must be called without clone's queue lock held, |
1072 | * see end_clone_request() for more details. | |
980691e5 KU |
1073 | */ |
1074 | static void dm_end_request(struct request *clone, int error) | |
1075 | { | |
1076 | int rw = rq_data_dir(clone); | |
1077 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1078 | struct mapped_device *md = tio->md; | |
1079 | struct request *rq = tio->orig; | |
1080 | ||
29e4013d | 1081 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1082 | rq->errors = clone->errors; |
1083 | rq->resid_len = clone->resid_len; | |
1084 | ||
1085 | if (rq->sense) | |
1086 | /* | |
1087 | * We are using the sense buffer of the original | |
1088 | * request. | |
1089 | * So setting the length of the sense data is enough. | |
1090 | */ | |
1091 | rq->sense_len = clone->sense_len; | |
1092 | } | |
1093 | ||
e5d8de32 | 1094 | free_rq_clone(clone); |
e262f347 | 1095 | rq_end_stats(md, rq); |
bfebd1cd MS |
1096 | if (!rq->q->mq_ops) |
1097 | blk_end_request_all(rq, error); | |
1098 | else | |
1099 | blk_mq_end_request(rq, error); | |
29e4013d | 1100 | rq_completed(md, rw, true); |
980691e5 KU |
1101 | } |
1102 | ||
cec47e3d KU |
1103 | static void dm_unprep_request(struct request *rq) |
1104 | { | |
bfebd1cd | 1105 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1106 | struct request *clone = tio->clone; |
cec47e3d | 1107 | |
bfebd1cd MS |
1108 | if (!rq->q->mq_ops) { |
1109 | rq->special = NULL; | |
1110 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1111 | } | |
cec47e3d | 1112 | |
e5863d9a | 1113 | if (clone) |
e5d8de32 | 1114 | free_rq_clone(clone); |
cec47e3d KU |
1115 | } |
1116 | ||
1117 | /* | |
1118 | * Requeue the original request of a clone. | |
1119 | */ | |
bfebd1cd | 1120 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1121 | { |
cec47e3d KU |
1122 | struct request_queue *q = rq->q; |
1123 | unsigned long flags; | |
1124 | ||
cec47e3d | 1125 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1126 | blk_requeue_request(q, rq); |
4ae9944d | 1127 | blk_run_queue_async(q); |
cec47e3d | 1128 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1129 | } |
1130 | ||
2d76fff1 MS |
1131 | static void dm_requeue_original_request(struct mapped_device *md, |
1132 | struct request *rq) | |
bfebd1cd MS |
1133 | { |
1134 | int rw = rq_data_dir(rq); | |
1135 | ||
1136 | dm_unprep_request(rq); | |
1137 | ||
e262f347 | 1138 | rq_end_stats(md, rq); |
bfebd1cd MS |
1139 | if (!rq->q->mq_ops) |
1140 | old_requeue_request(rq); | |
1141 | else { | |
1142 | blk_mq_requeue_request(rq); | |
1143 | blk_mq_kick_requeue_list(rq->q); | |
1144 | } | |
cec47e3d | 1145 | |
466d89a6 KB |
1146 | rq_completed(md, rw, false); |
1147 | } | |
1148 | ||
bfebd1cd | 1149 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1150 | { |
1151 | unsigned long flags; | |
1152 | ||
bfebd1cd MS |
1153 | if (blk_queue_stopped(q)) |
1154 | return; | |
1155 | ||
cec47e3d | 1156 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1157 | blk_stop_queue(q); |
cec47e3d KU |
1158 | spin_unlock_irqrestore(q->queue_lock, flags); |
1159 | } | |
1160 | ||
bfebd1cd | 1161 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1162 | { |
bfebd1cd MS |
1163 | if (!q->mq_ops) |
1164 | old_stop_queue(q); | |
1165 | else | |
1166 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1167 | } |
1168 | ||
bfebd1cd | 1169 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1170 | { |
1171 | unsigned long flags; | |
1172 | ||
1173 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1174 | if (blk_queue_stopped(q)) |
1175 | blk_start_queue(q); | |
cec47e3d KU |
1176 | spin_unlock_irqrestore(q->queue_lock, flags); |
1177 | } | |
1178 | ||
bfebd1cd MS |
1179 | static void start_queue(struct request_queue *q) |
1180 | { | |
1181 | if (!q->mq_ops) | |
1182 | old_start_queue(q); | |
1183 | else | |
1184 | blk_mq_start_stopped_hw_queues(q, true); | |
1185 | } | |
1186 | ||
11a68244 | 1187 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1188 | { |
11a68244 | 1189 | int r = error; |
cec47e3d | 1190 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1191 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1192 | |
ba1cbad9 MS |
1193 | if (tio->ti) { |
1194 | rq_end_io = tio->ti->type->rq_end_io; | |
1195 | ||
1196 | if (mapped && rq_end_io) | |
1197 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1198 | } | |
cec47e3d | 1199 | |
7eee4ae2 MS |
1200 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1201 | !clone->q->limits.max_write_same_sectors)) | |
1202 | disable_write_same(tio->md); | |
1203 | ||
11a68244 | 1204 | if (r <= 0) |
cec47e3d | 1205 | /* The target wants to complete the I/O */ |
11a68244 KU |
1206 | dm_end_request(clone, r); |
1207 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1208 | /* The target will handle the I/O */ |
1209 | return; | |
11a68244 | 1210 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1211 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1212 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1213 | else { |
11a68244 | 1214 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1215 | BUG(); |
1216 | } | |
1217 | } | |
1218 | ||
11a68244 KU |
1219 | /* |
1220 | * Request completion handler for request-based dm | |
1221 | */ | |
1222 | static void dm_softirq_done(struct request *rq) | |
1223 | { | |
1224 | bool mapped = true; | |
bfebd1cd | 1225 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1226 | struct request *clone = tio->clone; |
bfebd1cd | 1227 | int rw; |
11a68244 | 1228 | |
e5863d9a | 1229 | if (!clone) { |
e262f347 | 1230 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1231 | rw = rq_data_dir(rq); |
1232 | if (!rq->q->mq_ops) { | |
1233 | blk_end_request_all(rq, tio->error); | |
1234 | rq_completed(tio->md, rw, false); | |
1235 | free_rq_tio(tio); | |
1236 | } else { | |
1237 | blk_mq_end_request(rq, tio->error); | |
1238 | rq_completed(tio->md, rw, false); | |
1239 | } | |
e5863d9a MS |
1240 | return; |
1241 | } | |
11a68244 KU |
1242 | |
1243 | if (rq->cmd_flags & REQ_FAILED) | |
1244 | mapped = false; | |
1245 | ||
1246 | dm_done(clone, tio->error, mapped); | |
1247 | } | |
1248 | ||
cec47e3d KU |
1249 | /* |
1250 | * Complete the clone and the original request with the error status | |
1251 | * through softirq context. | |
1252 | */ | |
466d89a6 | 1253 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1254 | { |
bfebd1cd | 1255 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1256 | |
1257 | tio->error = error; | |
cec47e3d KU |
1258 | blk_complete_request(rq); |
1259 | } | |
1260 | ||
1261 | /* | |
1262 | * Complete the not-mapped clone and the original request with the error status | |
1263 | * through softirq context. | |
1264 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1265 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1266 | */ |
466d89a6 | 1267 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1268 | { |
cec47e3d | 1269 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1270 | dm_complete_request(rq, error); |
cec47e3d | 1271 | } |
cec47e3d KU |
1272 | |
1273 | /* | |
bfebd1cd | 1274 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1275 | */ |
1276 | static void end_clone_request(struct request *clone, int error) | |
1277 | { | |
466d89a6 KB |
1278 | struct dm_rq_target_io *tio = clone->end_io_data; |
1279 | ||
e5863d9a MS |
1280 | if (!clone->q->mq_ops) { |
1281 | /* | |
1282 | * For just cleaning up the information of the queue in which | |
1283 | * the clone was dispatched. | |
1284 | * The clone is *NOT* freed actually here because it is alloced | |
1285 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1286 | */ | |
1287 | __blk_put_request(clone->q, clone); | |
1288 | } | |
cec47e3d KU |
1289 | |
1290 | /* | |
1291 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1292 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1293 | * - another request may be submitted by the upper level driver |
1294 | * of the stacking during the completion | |
1295 | * - the submission which requires queue lock may be done | |
466d89a6 | 1296 | * against this clone's queue |
cec47e3d | 1297 | */ |
466d89a6 | 1298 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1299 | } |
1300 | ||
56a67df7 MS |
1301 | /* |
1302 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1303 | * target boundary. | |
1304 | */ | |
1305 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1306 | { | |
1307 | sector_t target_offset = dm_target_offset(ti, sector); | |
1308 | ||
1309 | return ti->len - target_offset; | |
1310 | } | |
1311 | ||
1312 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1313 | { |
56a67df7 | 1314 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1315 | sector_t offset, max_len; |
1da177e4 LT |
1316 | |
1317 | /* | |
542f9038 | 1318 | * Does the target need to split even further? |
1da177e4 | 1319 | */ |
542f9038 MS |
1320 | if (ti->max_io_len) { |
1321 | offset = dm_target_offset(ti, sector); | |
1322 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1323 | max_len = sector_div(offset, ti->max_io_len); | |
1324 | else | |
1325 | max_len = offset & (ti->max_io_len - 1); | |
1326 | max_len = ti->max_io_len - max_len; | |
1327 | ||
1328 | if (len > max_len) | |
1329 | len = max_len; | |
1da177e4 LT |
1330 | } |
1331 | ||
1332 | return len; | |
1333 | } | |
1334 | ||
542f9038 MS |
1335 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1336 | { | |
1337 | if (len > UINT_MAX) { | |
1338 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1339 | (unsigned long long)len, UINT_MAX); | |
1340 | ti->error = "Maximum size of target IO is too large"; | |
1341 | return -EINVAL; | |
1342 | } | |
1343 | ||
1344 | ti->max_io_len = (uint32_t) len; | |
1345 | ||
1346 | return 0; | |
1347 | } | |
1348 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1349 | ||
1dd40c3e MP |
1350 | /* |
1351 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1352 | * allowed for all bio types except REQ_FLUSH. | |
1353 | * | |
1354 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1355 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1356 | * sent in a next bio. | |
1357 | * | |
1358 | * A diagram that explains the arithmetics: | |
1359 | * +--------------------+---------------+-------+ | |
1360 | * | 1 | 2 | 3 | | |
1361 | * +--------------------+---------------+-------+ | |
1362 | * | |
1363 | * <-------------- *tio->len_ptr ---------------> | |
1364 | * <------- bi_size -------> | |
1365 | * <-- n_sectors --> | |
1366 | * | |
1367 | * Region 1 was already iterated over with bio_advance or similar function. | |
1368 | * (it may be empty if the target doesn't use bio_advance) | |
1369 | * Region 2 is the remaining bio size that the target wants to process. | |
1370 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1371 | * to make it empty) | |
1372 | * The target requires that region 3 is to be sent in the next bio. | |
1373 | * | |
1374 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1375 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1376 | * copies of the bio. | |
1377 | */ | |
1378 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1379 | { | |
1380 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1381 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1382 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1383 | BUG_ON(bi_size > *tio->len_ptr); | |
1384 | BUG_ON(n_sectors > bi_size); | |
1385 | *tio->len_ptr -= bi_size - n_sectors; | |
1386 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1387 | } | |
1388 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1389 | ||
bd2a49b8 | 1390 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1391 | { |
1392 | int r; | |
2056a782 | 1393 | sector_t sector; |
9faf400f | 1394 | struct mapped_device *md; |
dba14160 | 1395 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1396 | struct dm_target *ti = tio->ti; |
1da177e4 | 1397 | |
1da177e4 | 1398 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1399 | |
1400 | /* | |
1401 | * Map the clone. If r == 0 we don't need to do | |
1402 | * anything, the target has assumed ownership of | |
1403 | * this io. | |
1404 | */ | |
1405 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1406 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1407 | r = ti->type->map(ti, clone); |
45cbcd79 | 1408 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1409 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1410 | |
d07335e5 MS |
1411 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1412 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1413 | |
1da177e4 | 1414 | generic_make_request(clone); |
2e93ccc1 KU |
1415 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1416 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1417 | md = tio->io->md; |
1418 | dec_pending(tio->io, r); | |
9faf400f | 1419 | free_tio(md, tio); |
45cbcd79 KU |
1420 | } else if (r) { |
1421 | DMWARN("unimplemented target map return value: %d", r); | |
1422 | BUG(); | |
1da177e4 LT |
1423 | } |
1424 | } | |
1425 | ||
1426 | struct clone_info { | |
1427 | struct mapped_device *md; | |
1428 | struct dm_table *map; | |
1429 | struct bio *bio; | |
1430 | struct dm_io *io; | |
1431 | sector_t sector; | |
e0d6609a | 1432 | unsigned sector_count; |
1da177e4 LT |
1433 | }; |
1434 | ||
e0d6609a | 1435 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1436 | { |
4f024f37 KO |
1437 | bio->bi_iter.bi_sector = sector; |
1438 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1439 | } |
1440 | ||
1441 | /* | |
1442 | * Creates a bio that consists of range of complete bvecs. | |
1443 | */ | |
dba14160 | 1444 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1445 | sector_t sector, unsigned len) |
1da177e4 | 1446 | { |
dba14160 | 1447 | struct bio *clone = &tio->clone; |
1da177e4 | 1448 | |
1c3b13e6 KO |
1449 | __bio_clone_fast(clone, bio); |
1450 | ||
1451 | if (bio_integrity(bio)) | |
1452 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1453 | |
1c3b13e6 KO |
1454 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1455 | clone->bi_iter.bi_size = to_bytes(len); | |
1456 | ||
1457 | if (bio_integrity(bio)) | |
1458 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1459 | } |
1460 | ||
9015df24 | 1461 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1462 | struct dm_target *ti, |
55a62eef | 1463 | unsigned target_bio_nr) |
f9ab94ce | 1464 | { |
dba14160 MP |
1465 | struct dm_target_io *tio; |
1466 | struct bio *clone; | |
1467 | ||
99778273 | 1468 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1469 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1470 | |
1471 | tio->io = ci->io; | |
1472 | tio->ti = ti; | |
55a62eef | 1473 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1474 | |
1475 | return tio; | |
1476 | } | |
1477 | ||
14fe594d AK |
1478 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1479 | struct dm_target *ti, | |
1dd40c3e | 1480 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1481 | { |
99778273 | 1482 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1483 | struct bio *clone = &tio->clone; |
9015df24 | 1484 | |
1dd40c3e MP |
1485 | tio->len_ptr = len; |
1486 | ||
99778273 | 1487 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1488 | if (len) |
1dd40c3e | 1489 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1490 | |
bd2a49b8 | 1491 | __map_bio(tio); |
f9ab94ce MP |
1492 | } |
1493 | ||
14fe594d | 1494 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1495 | unsigned num_bios, unsigned *len) |
06a426ce | 1496 | { |
55a62eef | 1497 | unsigned target_bio_nr; |
06a426ce | 1498 | |
55a62eef | 1499 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1500 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1501 | } |
1502 | ||
14fe594d | 1503 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1504 | { |
06a426ce | 1505 | unsigned target_nr = 0; |
f9ab94ce MP |
1506 | struct dm_target *ti; |
1507 | ||
b372d360 | 1508 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1509 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1510 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1511 | |
f9ab94ce MP |
1512 | return 0; |
1513 | } | |
1514 | ||
e4c93811 | 1515 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1516 | sector_t sector, unsigned *len) |
5ae89a87 | 1517 | { |
dba14160 | 1518 | struct bio *bio = ci->bio; |
5ae89a87 | 1519 | struct dm_target_io *tio; |
b0d8ed4d AK |
1520 | unsigned target_bio_nr; |
1521 | unsigned num_target_bios = 1; | |
5ae89a87 | 1522 | |
b0d8ed4d AK |
1523 | /* |
1524 | * Does the target want to receive duplicate copies of the bio? | |
1525 | */ | |
1526 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1527 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1528 | |
b0d8ed4d | 1529 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1530 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1531 | tio->len_ptr = len; |
1532 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1533 | __map_bio(tio); |
1534 | } | |
5ae89a87 MS |
1535 | } |
1536 | ||
55a62eef | 1537 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1538 | |
55a62eef | 1539 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1540 | { |
55a62eef | 1541 | return ti->num_discard_bios; |
23508a96 MS |
1542 | } |
1543 | ||
55a62eef | 1544 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1545 | { |
55a62eef | 1546 | return ti->num_write_same_bios; |
23508a96 MS |
1547 | } |
1548 | ||
1549 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1550 | ||
1551 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1552 | { | |
55a62eef | 1553 | return ti->split_discard_bios; |
23508a96 MS |
1554 | } |
1555 | ||
14fe594d AK |
1556 | static int __send_changing_extent_only(struct clone_info *ci, |
1557 | get_num_bios_fn get_num_bios, | |
1558 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1559 | { |
1560 | struct dm_target *ti; | |
e0d6609a | 1561 | unsigned len; |
55a62eef | 1562 | unsigned num_bios; |
5ae89a87 | 1563 | |
a79245b3 MS |
1564 | do { |
1565 | ti = dm_table_find_target(ci->map, ci->sector); | |
1566 | if (!dm_target_is_valid(ti)) | |
1567 | return -EIO; | |
5ae89a87 | 1568 | |
5ae89a87 | 1569 | /* |
23508a96 MS |
1570 | * Even though the device advertised support for this type of |
1571 | * request, that does not mean every target supports it, and | |
936688d7 | 1572 | * reconfiguration might also have changed that since the |
a79245b3 | 1573 | * check was performed. |
5ae89a87 | 1574 | */ |
55a62eef AK |
1575 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1576 | if (!num_bios) | |
a79245b3 | 1577 | return -EOPNOTSUPP; |
5ae89a87 | 1578 | |
23508a96 | 1579 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1580 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1581 | else |
e0d6609a | 1582 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1583 | |
1dd40c3e | 1584 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1585 | |
1586 | ci->sector += len; | |
1587 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1588 | |
1589 | return 0; | |
1590 | } | |
1591 | ||
14fe594d | 1592 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1593 | { |
14fe594d AK |
1594 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1595 | is_split_required_for_discard); | |
23508a96 MS |
1596 | } |
1597 | ||
14fe594d | 1598 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1599 | { |
14fe594d | 1600 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1601 | } |
1602 | ||
e4c93811 AK |
1603 | /* |
1604 | * Select the correct strategy for processing a non-flush bio. | |
1605 | */ | |
14fe594d | 1606 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1607 | { |
dba14160 | 1608 | struct bio *bio = ci->bio; |
512875bd | 1609 | struct dm_target *ti; |
1c3b13e6 | 1610 | unsigned len; |
1da177e4 | 1611 | |
5ae89a87 | 1612 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1613 | return __send_discard(ci); |
23508a96 | 1614 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1615 | return __send_write_same(ci); |
5ae89a87 | 1616 | |
512875bd JN |
1617 | ti = dm_table_find_target(ci->map, ci->sector); |
1618 | if (!dm_target_is_valid(ti)) | |
1619 | return -EIO; | |
1620 | ||
1c3b13e6 | 1621 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1622 | |
1dd40c3e | 1623 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1624 | |
1c3b13e6 KO |
1625 | ci->sector += len; |
1626 | ci->sector_count -= len; | |
1da177e4 | 1627 | |
1c3b13e6 | 1628 | return 0; |
1da177e4 LT |
1629 | } |
1630 | ||
1631 | /* | |
14fe594d | 1632 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1633 | */ |
83d5e5b0 MP |
1634 | static void __split_and_process_bio(struct mapped_device *md, |
1635 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1636 | { |
1637 | struct clone_info ci; | |
512875bd | 1638 | int error = 0; |
1da177e4 | 1639 | |
83d5e5b0 | 1640 | if (unlikely(!map)) { |
6a8736d1 | 1641 | bio_io_error(bio); |
f0b9a450 MP |
1642 | return; |
1643 | } | |
692d0eb9 | 1644 | |
83d5e5b0 | 1645 | ci.map = map; |
1da177e4 | 1646 | ci.md = md; |
1da177e4 LT |
1647 | ci.io = alloc_io(md); |
1648 | ci.io->error = 0; | |
1649 | atomic_set(&ci.io->io_count, 1); | |
1650 | ci.io->bio = bio; | |
1651 | ci.io->md = md; | |
f88fb981 | 1652 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1653 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1654 | |
3eaf840e | 1655 | start_io_acct(ci.io); |
bd2a49b8 | 1656 | |
b372d360 MS |
1657 | if (bio->bi_rw & REQ_FLUSH) { |
1658 | ci.bio = &ci.md->flush_bio; | |
1659 | ci.sector_count = 0; | |
14fe594d | 1660 | error = __send_empty_flush(&ci); |
b372d360 MS |
1661 | /* dec_pending submits any data associated with flush */ |
1662 | } else { | |
6a8736d1 | 1663 | ci.bio = bio; |
d87f4c14 | 1664 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1665 | while (ci.sector_count && !error) |
14fe594d | 1666 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1667 | } |
1da177e4 LT |
1668 | |
1669 | /* drop the extra reference count */ | |
512875bd | 1670 | dec_pending(ci.io, error); |
1da177e4 LT |
1671 | } |
1672 | /*----------------------------------------------------------------- | |
1673 | * CRUD END | |
1674 | *---------------------------------------------------------------*/ | |
1675 | ||
f6fccb12 MB |
1676 | static int dm_merge_bvec(struct request_queue *q, |
1677 | struct bvec_merge_data *bvm, | |
1678 | struct bio_vec *biovec) | |
1679 | { | |
1680 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1681 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 | 1682 | struct dm_target *ti; |
1c220c69 | 1683 | sector_t max_sectors, max_size = 0; |
f6fccb12 MB |
1684 | |
1685 | if (unlikely(!map)) | |
5037108a | 1686 | goto out; |
f6fccb12 MB |
1687 | |
1688 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1689 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1690 | goto out; |
f6fccb12 MB |
1691 | |
1692 | /* | |
1693 | * Find maximum amount of I/O that won't need splitting | |
1694 | */ | |
56a67df7 | 1695 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1696 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1697 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
1c220c69 JT |
1698 | |
1699 | /* | |
1700 | * FIXME: this stop-gap fix _must_ be cleaned up (by passing a sector_t | |
1701 | * to the targets' merge function since it holds sectors not bytes). | |
1702 | * Just doing this as an interim fix for stable@ because the more | |
1703 | * comprehensive cleanup of switching to sector_t will impact every | |
1704 | * DM target that implements a ->merge hook. | |
1705 | */ | |
1706 | if (max_size > INT_MAX) | |
1707 | max_size = INT_MAX; | |
f6fccb12 MB |
1708 | |
1709 | /* | |
1710 | * merge_bvec_fn() returns number of bytes | |
1711 | * it can accept at this offset | |
1712 | * max is precomputed maximal io size | |
1713 | */ | |
1714 | if (max_size && ti->type->merge) | |
1c220c69 | 1715 | max_size = ti->type->merge(ti, bvm, biovec, (int) max_size); |
8cbeb67a MP |
1716 | /* |
1717 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1718 | * provided their merge_bvec method (we know this by looking for the |
1719 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1720 | * allow bios with multiple vector entries. So always set max_size | |
1721 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1722 | */ |
1723 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1724 | max_size = 0; |
f6fccb12 | 1725 | |
5037108a | 1726 | out: |
83d5e5b0 | 1727 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1728 | /* |
1729 | * Always allow an entire first page | |
1730 | */ | |
1731 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1732 | max_size = biovec->bv_len; | |
1733 | ||
f6fccb12 MB |
1734 | return max_size; |
1735 | } | |
1736 | ||
1da177e4 LT |
1737 | /* |
1738 | * The request function that just remaps the bio built up by | |
1739 | * dm_merge_bvec. | |
1740 | */ | |
ff36ab34 | 1741 | static void dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1742 | { |
12f03a49 | 1743 | int rw = bio_data_dir(bio); |
1da177e4 | 1744 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1745 | int srcu_idx; |
1746 | struct dm_table *map; | |
1da177e4 | 1747 | |
83d5e5b0 | 1748 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1749 | |
18c0b223 | 1750 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1751 | |
6a8736d1 TH |
1752 | /* if we're suspended, we have to queue this io for later */ |
1753 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1754 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1755 | |
6a8736d1 TH |
1756 | if (bio_rw(bio) != READA) |
1757 | queue_io(md, bio); | |
1758 | else | |
54d9a1b4 | 1759 | bio_io_error(bio); |
5a7bbad2 | 1760 | return; |
1da177e4 LT |
1761 | } |
1762 | ||
83d5e5b0 MP |
1763 | __split_and_process_bio(md, map, bio); |
1764 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1765 | return; |
cec47e3d KU |
1766 | } |
1767 | ||
fd2ed4d2 | 1768 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1769 | { |
1770 | return blk_queue_stackable(md->queue); | |
1771 | } | |
1772 | ||
466d89a6 | 1773 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1774 | { |
1775 | int r; | |
1776 | ||
466d89a6 KB |
1777 | if (blk_queue_io_stat(clone->q)) |
1778 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1779 | |
466d89a6 KB |
1780 | clone->start_time = jiffies; |
1781 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1782 | if (r) |
466d89a6 | 1783 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1784 | dm_complete_request(rq, r); |
1785 | } | |
cec47e3d | 1786 | |
5f1b670d CH |
1787 | static void setup_clone(struct request *clone, struct request *rq, |
1788 | struct dm_rq_target_io *tio) | |
cec47e3d | 1789 | { |
5f1b670d | 1790 | blk_rq_prep_clone(clone, rq); |
cec47e3d KU |
1791 | clone->end_io = end_clone_request; |
1792 | clone->end_io_data = tio; | |
1ae49ea2 | 1793 | tio->clone = clone; |
cec47e3d KU |
1794 | } |
1795 | ||
6facdaff | 1796 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1797 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1798 | { |
02233342 MS |
1799 | /* |
1800 | * Do not allocate a clone if tio->clone was already set | |
1801 | * (see: dm_mq_queue_rq). | |
1802 | */ | |
1803 | bool alloc_clone = !tio->clone; | |
1804 | struct request *clone; | |
1ae49ea2 | 1805 | |
02233342 MS |
1806 | if (alloc_clone) { |
1807 | clone = alloc_clone_request(md, gfp_mask); | |
1808 | if (!clone) | |
1809 | return NULL; | |
1810 | } else | |
1811 | clone = tio->clone; | |
1ae49ea2 MS |
1812 | |
1813 | blk_rq_init(NULL, clone); | |
5f1b670d | 1814 | setup_clone(clone, rq, tio); |
1ae49ea2 MS |
1815 | |
1816 | return clone; | |
1817 | } | |
1818 | ||
2eb6e1e3 KB |
1819 | static void map_tio_request(struct kthread_work *work); |
1820 | ||
bfebd1cd MS |
1821 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1822 | struct mapped_device *md) | |
1823 | { | |
1824 | tio->md = md; | |
1825 | tio->ti = NULL; | |
1826 | tio->clone = NULL; | |
1827 | tio->orig = rq; | |
1828 | tio->error = 0; | |
1829 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1830 | if (md->kworker_task) |
1831 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1832 | } |
1833 | ||
466d89a6 KB |
1834 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1835 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1836 | { |
6facdaff | 1837 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1838 | int srcu_idx; |
1839 | struct dm_table *table; | |
6facdaff KU |
1840 | |
1841 | tio = alloc_rq_tio(md, gfp_mask); | |
1842 | if (!tio) | |
1843 | return NULL; | |
1844 | ||
bfebd1cd | 1845 | init_tio(tio, rq, md); |
6facdaff | 1846 | |
e5863d9a MS |
1847 | table = dm_get_live_table(md, &srcu_idx); |
1848 | if (!dm_table_mq_request_based(table)) { | |
1849 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1850 | dm_put_live_table(md, srcu_idx); | |
1851 | free_rq_tio(tio); | |
1852 | return NULL; | |
1853 | } | |
6facdaff | 1854 | } |
e5863d9a | 1855 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1856 | |
466d89a6 | 1857 | return tio; |
6facdaff KU |
1858 | } |
1859 | ||
cec47e3d KU |
1860 | /* |
1861 | * Called with the queue lock held. | |
1862 | */ | |
1863 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1864 | { | |
1865 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1866 | struct dm_rq_target_io *tio; |
cec47e3d | 1867 | |
cec47e3d KU |
1868 | if (unlikely(rq->special)) { |
1869 | DMWARN("Already has something in rq->special."); | |
1870 | return BLKPREP_KILL; | |
1871 | } | |
1872 | ||
466d89a6 KB |
1873 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1874 | if (!tio) | |
cec47e3d | 1875 | return BLKPREP_DEFER; |
cec47e3d | 1876 | |
466d89a6 | 1877 | rq->special = tio; |
cec47e3d KU |
1878 | rq->cmd_flags |= REQ_DONTPREP; |
1879 | ||
1880 | return BLKPREP_OK; | |
1881 | } | |
1882 | ||
9eef87da KU |
1883 | /* |
1884 | * Returns: | |
e5863d9a MS |
1885 | * 0 : the request has been processed |
1886 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1887 | * < 0 : the request was completed due to failure | |
9eef87da | 1888 | */ |
bfebd1cd | 1889 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1890 | struct mapped_device *md) |
cec47e3d | 1891 | { |
e5863d9a | 1892 | int r; |
bfebd1cd | 1893 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1894 | struct request *clone = NULL; |
1895 | ||
1896 | if (tio->clone) { | |
1897 | clone = tio->clone; | |
1898 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1899 | } else { | |
1900 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1901 | if (r < 0) { | |
1902 | /* The target wants to complete the I/O */ | |
1903 | dm_kill_unmapped_request(rq, r); | |
1904 | return r; | |
1905 | } | |
3a140755 JN |
1906 | if (r != DM_MAPIO_REMAPPED) |
1907 | return r; | |
5f1b670d | 1908 | setup_clone(clone, rq, tio); |
e5863d9a | 1909 | } |
cec47e3d | 1910 | |
cec47e3d KU |
1911 | switch (r) { |
1912 | case DM_MAPIO_SUBMITTED: | |
1913 | /* The target has taken the I/O to submit by itself later */ | |
1914 | break; | |
1915 | case DM_MAPIO_REMAPPED: | |
1916 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1917 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1918 | blk_rq_pos(rq)); |
1919 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1920 | break; |
1921 | case DM_MAPIO_REQUEUE: | |
1922 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 1923 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
1924 | break; |
1925 | default: | |
1926 | if (r > 0) { | |
1927 | DMWARN("unimplemented target map return value: %d", r); | |
1928 | BUG(); | |
1929 | } | |
1930 | ||
1931 | /* The target wants to complete the I/O */ | |
466d89a6 | 1932 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1933 | return r; |
cec47e3d | 1934 | } |
9eef87da | 1935 | |
e5863d9a | 1936 | return 0; |
cec47e3d KU |
1937 | } |
1938 | ||
2eb6e1e3 | 1939 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 1940 | { |
2eb6e1e3 | 1941 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
1942 | struct request *rq = tio->orig; |
1943 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 1944 | |
bfebd1cd | 1945 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 1946 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
1947 | } |
1948 | ||
466d89a6 | 1949 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 1950 | { |
bfebd1cd MS |
1951 | if (!orig->q->mq_ops) |
1952 | blk_start_request(orig); | |
1953 | else | |
1954 | blk_mq_start_request(orig); | |
466d89a6 | 1955 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 1956 | |
0ce65797 MS |
1957 | if (md->seq_rq_merge_deadline_usecs) { |
1958 | md->last_rq_pos = rq_end_sector(orig); | |
1959 | md->last_rq_rw = rq_data_dir(orig); | |
1960 | md->last_rq_start_time = ktime_get(); | |
1961 | } | |
de3ec86d | 1962 | |
e262f347 MP |
1963 | if (unlikely(dm_stats_used(&md->stats))) { |
1964 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1965 | tio->duration_jiffies = jiffies; | |
1966 | tio->n_sectors = blk_rq_sectors(orig); | |
1967 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1968 | tio->n_sectors, false, 0, &tio->stats_aux); | |
1969 | } | |
1970 | ||
ba1cbad9 MS |
1971 | /* |
1972 | * Hold the md reference here for the in-flight I/O. | |
1973 | * We can't rely on the reference count by device opener, | |
1974 | * because the device may be closed during the request completion | |
1975 | * when all bios are completed. | |
1976 | * See the comment in rq_completed() too. | |
1977 | */ | |
1978 | dm_get(md); | |
ba1cbad9 MS |
1979 | } |
1980 | ||
0ce65797 MS |
1981 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
1982 | ||
1983 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
1984 | { | |
1985 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
1986 | } | |
1987 | ||
1988 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
1989 | const char *buf, size_t count) | |
1990 | { | |
1991 | unsigned deadline; | |
1992 | ||
17e149b8 | 1993 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
1994 | return count; |
1995 | ||
1996 | if (kstrtouint(buf, 10, &deadline)) | |
1997 | return -EINVAL; | |
1998 | ||
1999 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2000 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2001 | ||
2002 | md->seq_rq_merge_deadline_usecs = deadline; | |
2003 | ||
2004 | return count; | |
2005 | } | |
2006 | ||
2007 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2008 | { | |
2009 | ktime_t kt_deadline; | |
2010 | ||
2011 | if (!md->seq_rq_merge_deadline_usecs) | |
2012 | return false; | |
2013 | ||
2014 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2015 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2016 | ||
2017 | return !ktime_after(ktime_get(), kt_deadline); | |
2018 | } | |
2019 | ||
cec47e3d KU |
2020 | /* |
2021 | * q->request_fn for request-based dm. | |
2022 | * Called with the queue lock held. | |
2023 | */ | |
2024 | static void dm_request_fn(struct request_queue *q) | |
2025 | { | |
2026 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2027 | int srcu_idx; |
2028 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2029 | struct dm_target *ti; |
466d89a6 | 2030 | struct request *rq; |
2eb6e1e3 | 2031 | struct dm_rq_target_io *tio; |
29e4013d | 2032 | sector_t pos; |
cec47e3d KU |
2033 | |
2034 | /* | |
b4324fee KU |
2035 | * For suspend, check blk_queue_stopped() and increment |
2036 | * ->pending within a single queue_lock not to increment the | |
2037 | * number of in-flight I/Os after the queue is stopped in | |
2038 | * dm_suspend(). | |
cec47e3d | 2039 | */ |
7eaceacc | 2040 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2041 | rq = blk_peek_request(q); |
2042 | if (!rq) | |
9d1deb83 | 2043 | goto out; |
cec47e3d | 2044 | |
29e4013d TH |
2045 | /* always use block 0 to find the target for flushes for now */ |
2046 | pos = 0; | |
2047 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2048 | pos = blk_rq_pos(rq); | |
2049 | ||
2050 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2051 | if (!dm_target_is_valid(ti)) { |
2052 | /* | |
466d89a6 | 2053 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2054 | * before calling dm_kill_unmapped_request |
2055 | */ | |
2056 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2057 | dm_start_request(md, rq); |
2058 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2059 | continue; |
2060 | } | |
d0bcb878 | 2061 | |
0ce65797 MS |
2062 | if (dm_request_peeked_before_merge_deadline(md) && |
2063 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2064 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2065 | goto delay_and_out; | |
2066 | ||
cec47e3d | 2067 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2068 | goto delay_and_out; |
cec47e3d | 2069 | |
466d89a6 | 2070 | dm_start_request(md, rq); |
9eef87da | 2071 | |
bfebd1cd | 2072 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2073 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2074 | tio->ti = ti; | |
2075 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2076 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2077 | } |
2078 | ||
2079 | goto out; | |
2080 | ||
7eaceacc | 2081 | delay_and_out: |
d548b34b | 2082 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2083 | out: |
83d5e5b0 | 2084 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2085 | } |
2086 | ||
1da177e4 LT |
2087 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2088 | { | |
8a57dfc6 CS |
2089 | int r = bdi_bits; |
2090 | struct mapped_device *md = congested_data; | |
2091 | struct dm_table *map; | |
1da177e4 | 2092 | |
1eb787ec | 2093 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2094 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2095 | if (map) { |
cec47e3d KU |
2096 | /* |
2097 | * Request-based dm cares about only own queue for | |
2098 | * the query about congestion status of request_queue | |
2099 | */ | |
2100 | if (dm_request_based(md)) | |
2101 | r = md->queue->backing_dev_info.state & | |
2102 | bdi_bits; | |
2103 | else | |
2104 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2105 | } |
83d5e5b0 | 2106 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2107 | } |
2108 | ||
1da177e4 LT |
2109 | return r; |
2110 | } | |
2111 | ||
2112 | /*----------------------------------------------------------------- | |
2113 | * An IDR is used to keep track of allocated minor numbers. | |
2114 | *---------------------------------------------------------------*/ | |
2b06cfff | 2115 | static void free_minor(int minor) |
1da177e4 | 2116 | { |
f32c10b0 | 2117 | spin_lock(&_minor_lock); |
1da177e4 | 2118 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2119 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2120 | } |
2121 | ||
2122 | /* | |
2123 | * See if the device with a specific minor # is free. | |
2124 | */ | |
cf13ab8e | 2125 | static int specific_minor(int minor) |
1da177e4 | 2126 | { |
c9d76be6 | 2127 | int r; |
1da177e4 LT |
2128 | |
2129 | if (minor >= (1 << MINORBITS)) | |
2130 | return -EINVAL; | |
2131 | ||
c9d76be6 | 2132 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2133 | spin_lock(&_minor_lock); |
1da177e4 | 2134 | |
c9d76be6 | 2135 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2136 | |
f32c10b0 | 2137 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2138 | idr_preload_end(); |
2139 | if (r < 0) | |
2140 | return r == -ENOSPC ? -EBUSY : r; | |
2141 | return 0; | |
1da177e4 LT |
2142 | } |
2143 | ||
cf13ab8e | 2144 | static int next_free_minor(int *minor) |
1da177e4 | 2145 | { |
c9d76be6 | 2146 | int r; |
62f75c2f | 2147 | |
c9d76be6 | 2148 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2149 | spin_lock(&_minor_lock); |
1da177e4 | 2150 | |
c9d76be6 | 2151 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2152 | |
f32c10b0 | 2153 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2154 | idr_preload_end(); |
2155 | if (r < 0) | |
2156 | return r; | |
2157 | *minor = r; | |
2158 | return 0; | |
1da177e4 LT |
2159 | } |
2160 | ||
83d5cde4 | 2161 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2162 | |
53d5914f MP |
2163 | static void dm_wq_work(struct work_struct *work); |
2164 | ||
4a0b4ddf MS |
2165 | static void dm_init_md_queue(struct mapped_device *md) |
2166 | { | |
2167 | /* | |
2168 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2169 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2170 | * The type is decided at the first table loading time. |
2171 | * To prevent problematic device stacking, clear the queue flag | |
2172 | * for request stacking support until then. | |
2173 | * | |
2174 | * This queue is new, so no concurrency on the queue_flags. | |
2175 | */ | |
2176 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
bfebd1cd | 2177 | } |
4a0b4ddf | 2178 | |
bfebd1cd MS |
2179 | static void dm_init_old_md_queue(struct mapped_device *md) |
2180 | { | |
17e149b8 | 2181 | md->use_blk_mq = false; |
bfebd1cd MS |
2182 | dm_init_md_queue(md); |
2183 | ||
2184 | /* | |
2185 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2186 | */ | |
4a0b4ddf MS |
2187 | md->queue->queuedata = md; |
2188 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2189 | md->queue->backing_dev_info.congested_data = md; | |
ff36ab34 | 2190 | |
4a0b4ddf | 2191 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2192 | } |
2193 | ||
0f20972f MS |
2194 | static void cleanup_mapped_device(struct mapped_device *md) |
2195 | { | |
2196 | cleanup_srcu_struct(&md->io_barrier); | |
2197 | ||
2198 | if (md->wq) | |
2199 | destroy_workqueue(md->wq); | |
2200 | if (md->kworker_task) | |
2201 | kthread_stop(md->kworker_task); | |
2202 | if (md->io_pool) | |
2203 | mempool_destroy(md->io_pool); | |
2204 | if (md->rq_pool) | |
2205 | mempool_destroy(md->rq_pool); | |
2206 | if (md->bs) | |
2207 | bioset_free(md->bs); | |
2208 | ||
2209 | if (md->disk) { | |
2210 | spin_lock(&_minor_lock); | |
2211 | md->disk->private_data = NULL; | |
2212 | spin_unlock(&_minor_lock); | |
2213 | if (blk_get_integrity(md->disk)) | |
2214 | blk_integrity_unregister(md->disk); | |
2215 | del_gendisk(md->disk); | |
2216 | put_disk(md->disk); | |
2217 | } | |
2218 | ||
2219 | if (md->queue) | |
2220 | blk_cleanup_queue(md->queue); | |
2221 | ||
2222 | if (md->bdev) { | |
2223 | bdput(md->bdev); | |
2224 | md->bdev = NULL; | |
2225 | } | |
2226 | } | |
2227 | ||
1da177e4 LT |
2228 | /* |
2229 | * Allocate and initialise a blank device with a given minor. | |
2230 | */ | |
2b06cfff | 2231 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2232 | { |
2233 | int r; | |
cf13ab8e | 2234 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2235 | void *old_md; |
1da177e4 LT |
2236 | |
2237 | if (!md) { | |
2238 | DMWARN("unable to allocate device, out of memory."); | |
2239 | return NULL; | |
2240 | } | |
2241 | ||
10da4f79 | 2242 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2243 | goto bad_module_get; |
10da4f79 | 2244 | |
1da177e4 | 2245 | /* get a minor number for the dev */ |
2b06cfff | 2246 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2247 | r = next_free_minor(&minor); |
2b06cfff | 2248 | else |
cf13ab8e | 2249 | r = specific_minor(minor); |
1da177e4 | 2250 | if (r < 0) |
6ed7ade8 | 2251 | goto bad_minor; |
1da177e4 | 2252 | |
83d5e5b0 MP |
2253 | r = init_srcu_struct(&md->io_barrier); |
2254 | if (r < 0) | |
2255 | goto bad_io_barrier; | |
2256 | ||
17e149b8 | 2257 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2258 | md->type = DM_TYPE_NONE; |
e61290a4 | 2259 | mutex_init(&md->suspend_lock); |
a5664dad | 2260 | mutex_init(&md->type_lock); |
86f1152b | 2261 | mutex_init(&md->table_devices_lock); |
022c2611 | 2262 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2263 | atomic_set(&md->holders, 1); |
5c6bd75d | 2264 | atomic_set(&md->open_count, 0); |
1da177e4 | 2265 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2266 | atomic_set(&md->uevent_seq, 0); |
2267 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2268 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2269 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2270 | |
4a0b4ddf | 2271 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2272 | if (!md->queue) |
0f20972f | 2273 | goto bad; |
1da177e4 | 2274 | |
4a0b4ddf | 2275 | dm_init_md_queue(md); |
9faf400f | 2276 | |
1da177e4 LT |
2277 | md->disk = alloc_disk(1); |
2278 | if (!md->disk) | |
0f20972f | 2279 | goto bad; |
1da177e4 | 2280 | |
316d315b NK |
2281 | atomic_set(&md->pending[0], 0); |
2282 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2283 | init_waitqueue_head(&md->wait); |
53d5914f | 2284 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2285 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2286 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2287 | md->kworker_task = NULL; |
f0b04115 | 2288 | |
1da177e4 LT |
2289 | md->disk->major = _major; |
2290 | md->disk->first_minor = minor; | |
2291 | md->disk->fops = &dm_blk_dops; | |
2292 | md->disk->queue = md->queue; | |
2293 | md->disk->private_data = md; | |
2294 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2295 | add_disk(md->disk); | |
7e51f257 | 2296 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2297 | |
670368a8 | 2298 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2299 | if (!md->wq) |
0f20972f | 2300 | goto bad; |
304f3f6a | 2301 | |
32a926da MP |
2302 | md->bdev = bdget_disk(md->disk, 0); |
2303 | if (!md->bdev) | |
0f20972f | 2304 | goto bad; |
32a926da | 2305 | |
6a8736d1 TH |
2306 | bio_init(&md->flush_bio); |
2307 | md->flush_bio.bi_bdev = md->bdev; | |
2308 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2309 | ||
fd2ed4d2 MP |
2310 | dm_stats_init(&md->stats); |
2311 | ||
ba61fdd1 | 2312 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2313 | spin_lock(&_minor_lock); |
ba61fdd1 | 2314 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2315 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2316 | |
2317 | BUG_ON(old_md != MINOR_ALLOCED); | |
2318 | ||
1da177e4 LT |
2319 | return md; |
2320 | ||
0f20972f MS |
2321 | bad: |
2322 | cleanup_mapped_device(md); | |
83d5e5b0 | 2323 | bad_io_barrier: |
1da177e4 | 2324 | free_minor(minor); |
6ed7ade8 | 2325 | bad_minor: |
10da4f79 | 2326 | module_put(THIS_MODULE); |
6ed7ade8 | 2327 | bad_module_get: |
1da177e4 LT |
2328 | kfree(md); |
2329 | return NULL; | |
2330 | } | |
2331 | ||
ae9da83f JN |
2332 | static void unlock_fs(struct mapped_device *md); |
2333 | ||
1da177e4 LT |
2334 | static void free_dev(struct mapped_device *md) |
2335 | { | |
f331c029 | 2336 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2337 | |
32a926da | 2338 | unlock_fs(md); |
2eb6e1e3 | 2339 | |
0f20972f MS |
2340 | cleanup_mapped_device(md); |
2341 | if (md->use_blk_mq) | |
2342 | blk_mq_free_tag_set(&md->tag_set); | |
63a4f065 | 2343 | |
86f1152b | 2344 | free_table_devices(&md->table_devices); |
63a4f065 | 2345 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2346 | free_minor(minor); |
2347 | ||
10da4f79 | 2348 | module_put(THIS_MODULE); |
1da177e4 LT |
2349 | kfree(md); |
2350 | } | |
2351 | ||
e6ee8c0b KU |
2352 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2353 | { | |
c0820cf5 | 2354 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2355 | |
4e6e36c3 MS |
2356 | if (md->bs) { |
2357 | /* The md already has necessary mempools. */ | |
2358 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
2359 | /* |
2360 | * Reload bioset because front_pad may have changed | |
2361 | * because a different table was loaded. | |
2362 | */ | |
2363 | bioset_free(md->bs); | |
2364 | md->bs = p->bs; | |
2365 | p->bs = NULL; | |
16245bdc | 2366 | } |
4e6e36c3 MS |
2367 | /* |
2368 | * There's no need to reload with request-based dm | |
2369 | * because the size of front_pad doesn't change. | |
2370 | * Note for future: If you are to reload bioset, | |
2371 | * prep-ed requests in the queue may refer | |
2372 | * to bio from the old bioset, so you must walk | |
2373 | * through the queue to unprep. | |
2374 | */ | |
2375 | goto out; | |
c0820cf5 | 2376 | } |
e6ee8c0b | 2377 | |
e6ee8c0b KU |
2378 | md->io_pool = p->io_pool; |
2379 | p->io_pool = NULL; | |
1ae49ea2 MS |
2380 | md->rq_pool = p->rq_pool; |
2381 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2382 | md->bs = p->bs; |
2383 | p->bs = NULL; | |
4e6e36c3 | 2384 | |
e6ee8c0b | 2385 | out: |
02233342 | 2386 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2387 | dm_table_free_md_mempools(t); |
2388 | } | |
2389 | ||
1da177e4 LT |
2390 | /* |
2391 | * Bind a table to the device. | |
2392 | */ | |
2393 | static void event_callback(void *context) | |
2394 | { | |
7a8c3d3b MA |
2395 | unsigned long flags; |
2396 | LIST_HEAD(uevents); | |
1da177e4 LT |
2397 | struct mapped_device *md = (struct mapped_device *) context; |
2398 | ||
7a8c3d3b MA |
2399 | spin_lock_irqsave(&md->uevent_lock, flags); |
2400 | list_splice_init(&md->uevent_list, &uevents); | |
2401 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2402 | ||
ed9e1982 | 2403 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2404 | |
1da177e4 LT |
2405 | atomic_inc(&md->event_nr); |
2406 | wake_up(&md->eventq); | |
2407 | } | |
2408 | ||
c217649b MS |
2409 | /* |
2410 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2411 | */ | |
4e90188b | 2412 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2413 | { |
4e90188b | 2414 | set_capacity(md->disk, size); |
1da177e4 | 2415 | |
db8fef4f | 2416 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2417 | } |
2418 | ||
d5b9dd04 MP |
2419 | /* |
2420 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2421 | * | |
2422 | * If this function returns 0, then the device is either a non-dm | |
2423 | * device without a merge_bvec_fn, or it is a dm device that is | |
2424 | * able to split any bios it receives that are too big. | |
2425 | */ | |
2426 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2427 | { | |
2428 | struct mapped_device *dev_md; | |
2429 | ||
2430 | if (!q->merge_bvec_fn) | |
2431 | return 0; | |
2432 | ||
ff36ab34 | 2433 | if (q->make_request_fn == dm_make_request) { |
d5b9dd04 MP |
2434 | dev_md = q->queuedata; |
2435 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2436 | return 0; | |
2437 | } | |
2438 | ||
2439 | return 1; | |
2440 | } | |
2441 | ||
2442 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2443 | struct dm_dev *dev, sector_t start, | |
2444 | sector_t len, void *data) | |
2445 | { | |
2446 | struct block_device *bdev = dev->bdev; | |
2447 | struct request_queue *q = bdev_get_queue(bdev); | |
2448 | ||
2449 | return dm_queue_merge_is_compulsory(q); | |
2450 | } | |
2451 | ||
2452 | /* | |
2453 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2454 | * on the properties of the underlying devices. | |
2455 | */ | |
2456 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2457 | { | |
2458 | unsigned i = 0; | |
2459 | struct dm_target *ti; | |
2460 | ||
2461 | while (i < dm_table_get_num_targets(table)) { | |
2462 | ti = dm_table_get_target(table, i++); | |
2463 | ||
2464 | if (ti->type->iterate_devices && | |
2465 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2466 | return 0; | |
2467 | } | |
2468 | ||
2469 | return 1; | |
2470 | } | |
2471 | ||
042d2a9b AK |
2472 | /* |
2473 | * Returns old map, which caller must destroy. | |
2474 | */ | |
2475 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2476 | struct queue_limits *limits) | |
1da177e4 | 2477 | { |
042d2a9b | 2478 | struct dm_table *old_map; |
165125e1 | 2479 | struct request_queue *q = md->queue; |
1da177e4 | 2480 | sector_t size; |
d5b9dd04 | 2481 | int merge_is_optional; |
1da177e4 LT |
2482 | |
2483 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2484 | |
2485 | /* | |
2486 | * Wipe any geometry if the size of the table changed. | |
2487 | */ | |
fd2ed4d2 | 2488 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2489 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2490 | ||
32a926da | 2491 | __set_size(md, size); |
d5816876 | 2492 | |
2ca3310e AK |
2493 | dm_table_event_callback(t, event_callback, md); |
2494 | ||
e6ee8c0b KU |
2495 | /* |
2496 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2497 | * for request-based during suspension. So stop it to prevent | |
2498 | * I/O mapping before resume. | |
2499 | * This must be done before setting the queue restrictions, | |
2500 | * because request-based dm may be run just after the setting. | |
2501 | */ | |
bfebd1cd | 2502 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2503 | stop_queue(q); |
2504 | ||
2505 | __bind_mempools(md, t); | |
2506 | ||
d5b9dd04 MP |
2507 | merge_is_optional = dm_table_merge_is_optional(t); |
2508 | ||
a12f5d48 | 2509 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2510 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2511 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2512 | ||
754c5fc7 | 2513 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2514 | if (merge_is_optional) |
2515 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2516 | else | |
2517 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2518 | if (old_map) |
2519 | dm_sync_table(md); | |
1da177e4 | 2520 | |
042d2a9b | 2521 | return old_map; |
1da177e4 LT |
2522 | } |
2523 | ||
a7940155 AK |
2524 | /* |
2525 | * Returns unbound table for the caller to free. | |
2526 | */ | |
2527 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2528 | { |
a12f5d48 | 2529 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2530 | |
2531 | if (!map) | |
a7940155 | 2532 | return NULL; |
1da177e4 LT |
2533 | |
2534 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2535 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2536 | dm_sync_table(md); |
a7940155 AK |
2537 | |
2538 | return map; | |
1da177e4 LT |
2539 | } |
2540 | ||
2541 | /* | |
2542 | * Constructor for a new device. | |
2543 | */ | |
2b06cfff | 2544 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2545 | { |
2546 | struct mapped_device *md; | |
2547 | ||
2b06cfff | 2548 | md = alloc_dev(minor); |
1da177e4 LT |
2549 | if (!md) |
2550 | return -ENXIO; | |
2551 | ||
784aae73 MB |
2552 | dm_sysfs_init(md); |
2553 | ||
1da177e4 LT |
2554 | *result = md; |
2555 | return 0; | |
2556 | } | |
2557 | ||
a5664dad MS |
2558 | /* |
2559 | * Functions to manage md->type. | |
2560 | * All are required to hold md->type_lock. | |
2561 | */ | |
2562 | void dm_lock_md_type(struct mapped_device *md) | |
2563 | { | |
2564 | mutex_lock(&md->type_lock); | |
2565 | } | |
2566 | ||
2567 | void dm_unlock_md_type(struct mapped_device *md) | |
2568 | { | |
2569 | mutex_unlock(&md->type_lock); | |
2570 | } | |
2571 | ||
2572 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2573 | { | |
00c4fc3b | 2574 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2575 | md->type = type; |
2576 | } | |
2577 | ||
2578 | unsigned dm_get_md_type(struct mapped_device *md) | |
2579 | { | |
00c4fc3b | 2580 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2581 | return md->type; |
2582 | } | |
2583 | ||
36a0456f AK |
2584 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2585 | { | |
2586 | return md->immutable_target_type; | |
2587 | } | |
2588 | ||
f84cb8a4 MS |
2589 | /* |
2590 | * The queue_limits are only valid as long as you have a reference | |
2591 | * count on 'md'. | |
2592 | */ | |
2593 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2594 | { | |
2595 | BUG_ON(!atomic_read(&md->holders)); | |
2596 | return &md->queue->limits; | |
2597 | } | |
2598 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2599 | ||
bfebd1cd MS |
2600 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2601 | { | |
2602 | /* Initialize the request-based DM worker thread */ | |
2603 | init_kthread_worker(&md->kworker); | |
2604 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2605 | "kdmwork-%s", dm_device_name(md)); | |
2606 | } | |
2607 | ||
4a0b4ddf MS |
2608 | /* |
2609 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2610 | */ | |
2611 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2612 | { | |
2613 | struct request_queue *q = NULL; | |
2614 | ||
4a0b4ddf MS |
2615 | /* Fully initialize the queue */ |
2616 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2617 | if (!q) | |
bfebd1cd | 2618 | return -EINVAL; |
4a0b4ddf | 2619 | |
0ce65797 MS |
2620 | /* disable dm_request_fn's merge heuristic by default */ |
2621 | md->seq_rq_merge_deadline_usecs = 0; | |
2622 | ||
4a0b4ddf | 2623 | md->queue = q; |
bfebd1cd | 2624 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2625 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2626 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2627 | |
bfebd1cd | 2628 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2629 | |
4a0b4ddf MS |
2630 | elv_register_queue(md->queue); |
2631 | ||
bfebd1cd MS |
2632 | return 0; |
2633 | } | |
2634 | ||
2635 | static int dm_mq_init_request(void *data, struct request *rq, | |
2636 | unsigned int hctx_idx, unsigned int request_idx, | |
2637 | unsigned int numa_node) | |
2638 | { | |
2639 | struct mapped_device *md = data; | |
2640 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2641 | ||
2642 | /* | |
2643 | * Must initialize md member of tio, otherwise it won't | |
2644 | * be available in dm_mq_queue_rq. | |
2645 | */ | |
2646 | tio->md = md; | |
2647 | ||
2648 | return 0; | |
2649 | } | |
2650 | ||
2651 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2652 | const struct blk_mq_queue_data *bd) | |
2653 | { | |
2654 | struct request *rq = bd->rq; | |
2655 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2656 | struct mapped_device *md = tio->md; | |
2657 | int srcu_idx; | |
2658 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2659 | struct dm_target *ti; | |
2660 | sector_t pos; | |
2661 | ||
2662 | /* always use block 0 to find the target for flushes for now */ | |
2663 | pos = 0; | |
2664 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2665 | pos = blk_rq_pos(rq); | |
2666 | ||
2667 | ti = dm_table_find_target(map, pos); | |
2668 | if (!dm_target_is_valid(ti)) { | |
2669 | dm_put_live_table(md, srcu_idx); | |
2670 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2671 | /* | |
2672 | * Must perform setup, that rq_completed() requires, | |
2673 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2674 | */ | |
2675 | dm_start_request(md, rq); | |
2676 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2677 | } | |
2678 | dm_put_live_table(md, srcu_idx); | |
2679 | ||
2680 | if (ti->type->busy && ti->type->busy(ti)) | |
2681 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2682 | ||
2683 | dm_start_request(md, rq); | |
2684 | ||
2685 | /* Init tio using md established in .init_request */ | |
2686 | init_tio(tio, rq, md); | |
2687 | ||
02233342 MS |
2688 | /* |
2689 | * Establish tio->ti before queuing work (map_tio_request) | |
2690 | * or making direct call to map_request(). | |
2691 | */ | |
bfebd1cd | 2692 | tio->ti = ti; |
02233342 MS |
2693 | |
2694 | /* Clone the request if underlying devices aren't blk-mq */ | |
2695 | if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) { | |
2696 | /* clone request is allocated at the end of the pdu */ | |
2697 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
45714fbe | 2698 | (void) clone_rq(rq, md, tio, GFP_ATOMIC); |
02233342 MS |
2699 | queue_kthread_work(&md->kworker, &tio->work); |
2700 | } else { | |
2701 | /* Direct call is fine since .queue_rq allows allocations */ | |
45714fbe MS |
2702 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { |
2703 | /* Undo dm_start_request() before requeuing */ | |
e262f347 | 2704 | rq_end_stats(md, rq); |
45714fbe MS |
2705 | rq_completed(md, rq_data_dir(rq), false); |
2706 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2707 | } | |
02233342 | 2708 | } |
bfebd1cd MS |
2709 | |
2710 | return BLK_MQ_RQ_QUEUE_OK; | |
2711 | } | |
2712 | ||
2713 | static struct blk_mq_ops dm_mq_ops = { | |
2714 | .queue_rq = dm_mq_queue_rq, | |
2715 | .map_queue = blk_mq_map_queue, | |
2716 | .complete = dm_softirq_done, | |
2717 | .init_request = dm_mq_init_request, | |
2718 | }; | |
2719 | ||
2720 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2721 | { | |
02233342 | 2722 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2723 | struct request_queue *q; |
2724 | int err; | |
2725 | ||
2726 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2727 | md->tag_set.ops = &dm_mq_ops; | |
2728 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2729 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2730 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2731 | md->tag_set.nr_hw_queues = 1; | |
02233342 MS |
2732 | if (md_type == DM_TYPE_REQUEST_BASED) { |
2733 | /* make the memory for non-blk-mq clone part of the pdu */ | |
2734 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request); | |
2735 | } else | |
2736 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd MS |
2737 | md->tag_set.driver_data = md; |
2738 | ||
2739 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2740 | if (err) | |
2741 | return err; | |
2742 | ||
2743 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2744 | if (IS_ERR(q)) { | |
2745 | err = PTR_ERR(q); | |
2746 | goto out_tag_set; | |
2747 | } | |
2748 | md->queue = q; | |
2749 | dm_init_md_queue(md); | |
2750 | ||
2751 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2752 | blk_mq_register_disk(md->disk); | |
2753 | ||
02233342 MS |
2754 | if (md_type == DM_TYPE_REQUEST_BASED) |
2755 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2756 | |
2757 | return 0; | |
2758 | ||
2759 | out_tag_set: | |
2760 | blk_mq_free_tag_set(&md->tag_set); | |
2761 | return err; | |
4a0b4ddf MS |
2762 | } |
2763 | ||
4e6e36c3 MS |
2764 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2765 | { | |
2766 | if (type == DM_TYPE_BIO_BASED) | |
2767 | return type; | |
2768 | ||
2769 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2770 | } | |
2771 | ||
4a0b4ddf MS |
2772 | /* |
2773 | * Setup the DM device's queue based on md's type | |
2774 | */ | |
2775 | int dm_setup_md_queue(struct mapped_device *md) | |
2776 | { | |
bfebd1cd | 2777 | int r; |
17e149b8 | 2778 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2779 | |
2780 | switch (md_type) { | |
2781 | case DM_TYPE_REQUEST_BASED: | |
2782 | r = dm_init_request_based_queue(md); | |
2783 | if (r) { | |
ff36ab34 | 2784 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2785 | return r; |
ff36ab34 | 2786 | } |
bfebd1cd MS |
2787 | break; |
2788 | case DM_TYPE_MQ_REQUEST_BASED: | |
2789 | r = dm_init_request_based_blk_mq_queue(md); | |
2790 | if (r) { | |
2791 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2792 | return r; | |
2793 | } | |
2794 | break; | |
2795 | case DM_TYPE_BIO_BASED: | |
2796 | dm_init_old_md_queue(md); | |
ff36ab34 MS |
2797 | blk_queue_make_request(md->queue, dm_make_request); |
2798 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); | |
bfebd1cd | 2799 | break; |
4a0b4ddf MS |
2800 | } |
2801 | ||
2802 | return 0; | |
2803 | } | |
2804 | ||
2bec1f4a | 2805 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2806 | { |
2807 | struct mapped_device *md; | |
1da177e4 LT |
2808 | unsigned minor = MINOR(dev); |
2809 | ||
2810 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2811 | return NULL; | |
2812 | ||
f32c10b0 | 2813 | spin_lock(&_minor_lock); |
1da177e4 LT |
2814 | |
2815 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2816 | if (md) { |
2817 | if ((md == MINOR_ALLOCED || | |
2818 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2819 | dm_deleting_md(md) || | |
2820 | test_bit(DMF_FREEING, &md->flags))) { | |
2821 | md = NULL; | |
2822 | goto out; | |
2823 | } | |
2824 | dm_get(md); | |
fba9f90e | 2825 | } |
1da177e4 | 2826 | |
fba9f90e | 2827 | out: |
f32c10b0 | 2828 | spin_unlock(&_minor_lock); |
1da177e4 | 2829 | |
637842cf DT |
2830 | return md; |
2831 | } | |
3cf2e4ba | 2832 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2833 | |
9ade92a9 | 2834 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2835 | { |
9ade92a9 | 2836 | return md->interface_ptr; |
1da177e4 LT |
2837 | } |
2838 | ||
2839 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2840 | { | |
2841 | md->interface_ptr = ptr; | |
2842 | } | |
2843 | ||
2844 | void dm_get(struct mapped_device *md) | |
2845 | { | |
2846 | atomic_inc(&md->holders); | |
3f77316d | 2847 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2848 | } |
2849 | ||
09ee96b2 MP |
2850 | int dm_hold(struct mapped_device *md) |
2851 | { | |
2852 | spin_lock(&_minor_lock); | |
2853 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2854 | spin_unlock(&_minor_lock); | |
2855 | return -EBUSY; | |
2856 | } | |
2857 | dm_get(md); | |
2858 | spin_unlock(&_minor_lock); | |
2859 | return 0; | |
2860 | } | |
2861 | EXPORT_SYMBOL_GPL(dm_hold); | |
2862 | ||
72d94861 AK |
2863 | const char *dm_device_name(struct mapped_device *md) |
2864 | { | |
2865 | return md->name; | |
2866 | } | |
2867 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2868 | ||
3f77316d | 2869 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2870 | { |
1134e5ae | 2871 | struct dm_table *map; |
83d5e5b0 | 2872 | int srcu_idx; |
1da177e4 | 2873 | |
3f77316d | 2874 | might_sleep(); |
fba9f90e | 2875 | |
83d5e5b0 | 2876 | map = dm_get_live_table(md, &srcu_idx); |
63a4f065 MS |
2877 | |
2878 | spin_lock(&_minor_lock); | |
3f77316d KU |
2879 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2880 | set_bit(DMF_FREEING, &md->flags); | |
2881 | spin_unlock(&_minor_lock); | |
2882 | ||
02233342 | 2883 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2884 | flush_kthread_worker(&md->kworker); |
2885 | ||
ab7c7bb6 MP |
2886 | /* |
2887 | * Take suspend_lock so that presuspend and postsuspend methods | |
2888 | * do not race with internal suspend. | |
2889 | */ | |
2890 | mutex_lock(&md->suspend_lock); | |
3f77316d KU |
2891 | if (!dm_suspended_md(md)) { |
2892 | dm_table_presuspend_targets(map); | |
2893 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2894 | } |
ab7c7bb6 | 2895 | mutex_unlock(&md->suspend_lock); |
3f77316d | 2896 | |
83d5e5b0 MP |
2897 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2898 | dm_put_live_table(md, srcu_idx); | |
2899 | ||
3f77316d KU |
2900 | /* |
2901 | * Rare, but there may be I/O requests still going to complete, | |
2902 | * for example. Wait for all references to disappear. | |
2903 | * No one should increment the reference count of the mapped_device, | |
2904 | * after the mapped_device state becomes DMF_FREEING. | |
2905 | */ | |
2906 | if (wait) | |
2907 | while (atomic_read(&md->holders)) | |
2908 | msleep(1); | |
2909 | else if (atomic_read(&md->holders)) | |
2910 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2911 | dm_device_name(md), atomic_read(&md->holders)); | |
2912 | ||
2913 | dm_sysfs_exit(md); | |
3f77316d KU |
2914 | dm_table_destroy(__unbind(md)); |
2915 | free_dev(md); | |
2916 | } | |
2917 | ||
2918 | void dm_destroy(struct mapped_device *md) | |
2919 | { | |
2920 | __dm_destroy(md, true); | |
2921 | } | |
2922 | ||
2923 | void dm_destroy_immediate(struct mapped_device *md) | |
2924 | { | |
2925 | __dm_destroy(md, false); | |
2926 | } | |
2927 | ||
2928 | void dm_put(struct mapped_device *md) | |
2929 | { | |
2930 | atomic_dec(&md->holders); | |
1da177e4 | 2931 | } |
79eb885c | 2932 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2933 | |
401600df | 2934 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2935 | { |
2936 | int r = 0; | |
b44ebeb0 MP |
2937 | DECLARE_WAITQUEUE(wait, current); |
2938 | ||
b44ebeb0 | 2939 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2940 | |
2941 | while (1) { | |
401600df | 2942 | set_current_state(interruptible); |
46125c1c | 2943 | |
b4324fee | 2944 | if (!md_in_flight(md)) |
46125c1c MB |
2945 | break; |
2946 | ||
401600df MP |
2947 | if (interruptible == TASK_INTERRUPTIBLE && |
2948 | signal_pending(current)) { | |
46125c1c MB |
2949 | r = -EINTR; |
2950 | break; | |
2951 | } | |
2952 | ||
2953 | io_schedule(); | |
2954 | } | |
2955 | set_current_state(TASK_RUNNING); | |
2956 | ||
b44ebeb0 MP |
2957 | remove_wait_queue(&md->wait, &wait); |
2958 | ||
46125c1c MB |
2959 | return r; |
2960 | } | |
2961 | ||
1da177e4 LT |
2962 | /* |
2963 | * Process the deferred bios | |
2964 | */ | |
ef208587 | 2965 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2966 | { |
ef208587 MP |
2967 | struct mapped_device *md = container_of(work, struct mapped_device, |
2968 | work); | |
6d6f10df | 2969 | struct bio *c; |
83d5e5b0 MP |
2970 | int srcu_idx; |
2971 | struct dm_table *map; | |
1da177e4 | 2972 | |
83d5e5b0 | 2973 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2974 | |
3b00b203 | 2975 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2976 | spin_lock_irq(&md->deferred_lock); |
2977 | c = bio_list_pop(&md->deferred); | |
2978 | spin_unlock_irq(&md->deferred_lock); | |
2979 | ||
6a8736d1 | 2980 | if (!c) |
df12ee99 | 2981 | break; |
022c2611 | 2982 | |
e6ee8c0b KU |
2983 | if (dm_request_based(md)) |
2984 | generic_make_request(c); | |
6a8736d1 | 2985 | else |
83d5e5b0 | 2986 | __split_and_process_bio(md, map, c); |
022c2611 | 2987 | } |
73d410c0 | 2988 | |
83d5e5b0 | 2989 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2990 | } |
2991 | ||
9a1fb464 | 2992 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2993 | { |
3b00b203 | 2994 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2995 | smp_mb__after_atomic(); |
53d5914f | 2996 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2997 | } |
2998 | ||
1da177e4 | 2999 | /* |
042d2a9b | 3000 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 3001 | */ |
042d2a9b | 3002 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3003 | { |
87eb5b21 | 3004 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3005 | struct queue_limits limits; |
042d2a9b | 3006 | int r; |
1da177e4 | 3007 | |
e61290a4 | 3008 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3009 | |
3010 | /* device must be suspended */ | |
4f186f8b | 3011 | if (!dm_suspended_md(md)) |
93c534ae | 3012 | goto out; |
1da177e4 | 3013 | |
3ae70656 MS |
3014 | /* |
3015 | * If the new table has no data devices, retain the existing limits. | |
3016 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3017 | * then new I/O is queued based on these limits, and then some paths | |
3018 | * reappear. | |
3019 | */ | |
3020 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3021 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3022 | if (live_map) |
3023 | limits = md->queue->limits; | |
83d5e5b0 | 3024 | dm_put_live_table_fast(md); |
3ae70656 MS |
3025 | } |
3026 | ||
87eb5b21 MC |
3027 | if (!live_map) { |
3028 | r = dm_calculate_queue_limits(table, &limits); | |
3029 | if (r) { | |
3030 | map = ERR_PTR(r); | |
3031 | goto out; | |
3032 | } | |
042d2a9b | 3033 | } |
754c5fc7 | 3034 | |
042d2a9b | 3035 | map = __bind(md, table, &limits); |
1da177e4 | 3036 | |
93c534ae | 3037 | out: |
e61290a4 | 3038 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3039 | return map; |
1da177e4 LT |
3040 | } |
3041 | ||
3042 | /* | |
3043 | * Functions to lock and unlock any filesystem running on the | |
3044 | * device. | |
3045 | */ | |
2ca3310e | 3046 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3047 | { |
e39e2e95 | 3048 | int r; |
1da177e4 LT |
3049 | |
3050 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3051 | |
db8fef4f | 3052 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3053 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3054 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3055 | md->frozen_sb = NULL; |
3056 | return r; | |
dfbe03f6 AK |
3057 | } |
3058 | ||
aa8d7c2f AK |
3059 | set_bit(DMF_FROZEN, &md->flags); |
3060 | ||
1da177e4 LT |
3061 | return 0; |
3062 | } | |
3063 | ||
2ca3310e | 3064 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3065 | { |
aa8d7c2f AK |
3066 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3067 | return; | |
3068 | ||
db8fef4f | 3069 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3070 | md->frozen_sb = NULL; |
aa8d7c2f | 3071 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3072 | } |
3073 | ||
3074 | /* | |
ffcc3936 MS |
3075 | * If __dm_suspend returns 0, the device is completely quiescent |
3076 | * now. There is no request-processing activity. All new requests | |
3077 | * are being added to md->deferred list. | |
cec47e3d | 3078 | * |
ffcc3936 | 3079 | * Caller must hold md->suspend_lock |
cec47e3d | 3080 | */ |
ffcc3936 MS |
3081 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3082 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3083 | { |
ffcc3936 MS |
3084 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3085 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3086 | int r; | |
1da177e4 | 3087 | |
2e93ccc1 KU |
3088 | /* |
3089 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3090 | * This flag is cleared before dm_suspend returns. | |
3091 | */ | |
3092 | if (noflush) | |
3093 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3094 | ||
d67ee213 MS |
3095 | /* |
3096 | * This gets reverted if there's an error later and the targets | |
3097 | * provide the .presuspend_undo hook. | |
3098 | */ | |
cf222b37 AK |
3099 | dm_table_presuspend_targets(map); |
3100 | ||
32a926da | 3101 | /* |
9f518b27 KU |
3102 | * Flush I/O to the device. |
3103 | * Any I/O submitted after lock_fs() may not be flushed. | |
3104 | * noflush takes precedence over do_lockfs. | |
3105 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3106 | */ |
3107 | if (!noflush && do_lockfs) { | |
3108 | r = lock_fs(md); | |
d67ee213 MS |
3109 | if (r) { |
3110 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3111 | return r; |
d67ee213 | 3112 | } |
aa8d7c2f | 3113 | } |
1da177e4 LT |
3114 | |
3115 | /* | |
3b00b203 MP |
3116 | * Here we must make sure that no processes are submitting requests |
3117 | * to target drivers i.e. no one may be executing | |
3118 | * __split_and_process_bio. This is called from dm_request and | |
3119 | * dm_wq_work. | |
3120 | * | |
3121 | * To get all processes out of __split_and_process_bio in dm_request, | |
3122 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3123 | * __split_and_process_bio from dm_request and quiesce the thread |
3124 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3125 | * flush_workqueue(md->wq). | |
1da177e4 | 3126 | */ |
1eb787ec | 3127 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3128 | if (map) |
3129 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3130 | |
d0bcb878 | 3131 | /* |
29e4013d TH |
3132 | * Stop md->queue before flushing md->wq in case request-based |
3133 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3134 | */ |
2eb6e1e3 | 3135 | if (dm_request_based(md)) { |
9f518b27 | 3136 | stop_queue(md->queue); |
02233342 MS |
3137 | if (md->kworker_task) |
3138 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3139 | } |
cec47e3d | 3140 | |
d0bcb878 KU |
3141 | flush_workqueue(md->wq); |
3142 | ||
1da177e4 | 3143 | /* |
3b00b203 MP |
3144 | * At this point no more requests are entering target request routines. |
3145 | * We call dm_wait_for_completion to wait for all existing requests | |
3146 | * to finish. | |
1da177e4 | 3147 | */ |
ffcc3936 | 3148 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3149 | |
6d6f10df | 3150 | if (noflush) |
022c2611 | 3151 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3152 | if (map) |
3153 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3154 | |
1da177e4 | 3155 | /* were we interrupted ? */ |
46125c1c | 3156 | if (r < 0) { |
9a1fb464 | 3157 | dm_queue_flush(md); |
73d410c0 | 3158 | |
cec47e3d | 3159 | if (dm_request_based(md)) |
9f518b27 | 3160 | start_queue(md->queue); |
cec47e3d | 3161 | |
2ca3310e | 3162 | unlock_fs(md); |
d67ee213 | 3163 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3164 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3165 | } |
1da177e4 | 3166 | |
ffcc3936 MS |
3167 | return r; |
3168 | } | |
3169 | ||
3170 | /* | |
3171 | * We need to be able to change a mapping table under a mounted | |
3172 | * filesystem. For example we might want to move some data in | |
3173 | * the background. Before the table can be swapped with | |
3174 | * dm_bind_table, dm_suspend must be called to flush any in | |
3175 | * flight bios and ensure that any further io gets deferred. | |
3176 | */ | |
3177 | /* | |
3178 | * Suspend mechanism in request-based dm. | |
3179 | * | |
3180 | * 1. Flush all I/Os by lock_fs() if needed. | |
3181 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3182 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3183 | * | |
3184 | * To abort suspend, start the request_queue. | |
3185 | */ | |
3186 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3187 | { | |
3188 | struct dm_table *map = NULL; | |
3189 | int r = 0; | |
3190 | ||
3191 | retry: | |
3192 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3193 | ||
3194 | if (dm_suspended_md(md)) { | |
3195 | r = -EINVAL; | |
3196 | goto out_unlock; | |
3197 | } | |
3198 | ||
3199 | if (dm_suspended_internally_md(md)) { | |
3200 | /* already internally suspended, wait for internal resume */ | |
3201 | mutex_unlock(&md->suspend_lock); | |
3202 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3203 | if (r) | |
3204 | return r; | |
3205 | goto retry; | |
3206 | } | |
3207 | ||
a12f5d48 | 3208 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3209 | |
3210 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3211 | if (r) | |
3212 | goto out_unlock; | |
3b00b203 | 3213 | |
2ca3310e | 3214 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3215 | |
4d4471cb KU |
3216 | dm_table_postsuspend_targets(map); |
3217 | ||
d287483d | 3218 | out_unlock: |
e61290a4 | 3219 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3220 | return r; |
1da177e4 LT |
3221 | } |
3222 | ||
ffcc3936 MS |
3223 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3224 | { | |
3225 | if (map) { | |
3226 | int r = dm_table_resume_targets(map); | |
3227 | if (r) | |
3228 | return r; | |
3229 | } | |
3230 | ||
3231 | dm_queue_flush(md); | |
3232 | ||
3233 | /* | |
3234 | * Flushing deferred I/Os must be done after targets are resumed | |
3235 | * so that mapping of targets can work correctly. | |
3236 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3237 | */ | |
3238 | if (dm_request_based(md)) | |
3239 | start_queue(md->queue); | |
3240 | ||
3241 | unlock_fs(md); | |
3242 | ||
3243 | return 0; | |
3244 | } | |
3245 | ||
1da177e4 LT |
3246 | int dm_resume(struct mapped_device *md) |
3247 | { | |
cf222b37 | 3248 | int r = -EINVAL; |
cf222b37 | 3249 | struct dm_table *map = NULL; |
1da177e4 | 3250 | |
ffcc3936 MS |
3251 | retry: |
3252 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3253 | ||
4f186f8b | 3254 | if (!dm_suspended_md(md)) |
cf222b37 | 3255 | goto out; |
cf222b37 | 3256 | |
ffcc3936 MS |
3257 | if (dm_suspended_internally_md(md)) { |
3258 | /* already internally suspended, wait for internal resume */ | |
3259 | mutex_unlock(&md->suspend_lock); | |
3260 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3261 | if (r) | |
3262 | return r; | |
3263 | goto retry; | |
3264 | } | |
3265 | ||
a12f5d48 | 3266 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3267 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3268 | goto out; |
1da177e4 | 3269 | |
ffcc3936 | 3270 | r = __dm_resume(md, map); |
8757b776 MB |
3271 | if (r) |
3272 | goto out; | |
2ca3310e | 3273 | |
2ca3310e AK |
3274 | clear_bit(DMF_SUSPENDED, &md->flags); |
3275 | ||
cf222b37 AK |
3276 | r = 0; |
3277 | out: | |
e61290a4 | 3278 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3279 | |
cf222b37 | 3280 | return r; |
1da177e4 LT |
3281 | } |
3282 | ||
fd2ed4d2 MP |
3283 | /* |
3284 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3285 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3286 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3287 | */ |
3288 | ||
ffcc3936 | 3289 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3290 | { |
ffcc3936 MS |
3291 | struct dm_table *map = NULL; |
3292 | ||
96b26c8c | 3293 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3294 | return; /* nested internal suspend */ |
3295 | ||
3296 | if (dm_suspended_md(md)) { | |
3297 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3298 | return; /* nest suspend */ | |
3299 | } | |
3300 | ||
a12f5d48 | 3301 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3302 | |
3303 | /* | |
3304 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3305 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3306 | * would require changing .presuspend to return an error -- avoid this | |
3307 | * until there is a need for more elaborate variants of internal suspend. | |
3308 | */ | |
3309 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3310 | ||
3311 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3312 | ||
3313 | dm_table_postsuspend_targets(map); | |
3314 | } | |
3315 | ||
3316 | static void __dm_internal_resume(struct mapped_device *md) | |
3317 | { | |
96b26c8c MP |
3318 | BUG_ON(!md->internal_suspend_count); |
3319 | ||
3320 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3321 | return; /* resume from nested internal suspend */ |
3322 | ||
fd2ed4d2 | 3323 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3324 | goto done; /* resume from nested suspend */ |
3325 | ||
3326 | /* | |
3327 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3328 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3329 | */ | |
3330 | (void) __dm_resume(md, NULL); | |
3331 | ||
3332 | done: | |
3333 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3334 | smp_mb__after_atomic(); | |
3335 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3336 | } | |
3337 | ||
3338 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3339 | { | |
3340 | mutex_lock(&md->suspend_lock); | |
3341 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3342 | mutex_unlock(&md->suspend_lock); | |
3343 | } | |
3344 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3345 | ||
3346 | void dm_internal_resume(struct mapped_device *md) | |
3347 | { | |
3348 | mutex_lock(&md->suspend_lock); | |
3349 | __dm_internal_resume(md); | |
3350 | mutex_unlock(&md->suspend_lock); | |
3351 | } | |
3352 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3353 | ||
3354 | /* | |
3355 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3356 | * which prevents interaction with userspace-driven suspend. | |
3357 | */ | |
3358 | ||
3359 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3360 | { | |
3361 | mutex_lock(&md->suspend_lock); | |
3362 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3363 | return; |
3364 | ||
3365 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3366 | synchronize_srcu(&md->io_barrier); | |
3367 | flush_workqueue(md->wq); | |
3368 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3369 | } | |
b735fede | 3370 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3371 | |
ffcc3936 | 3372 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3373 | { |
ffcc3936 | 3374 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3375 | goto done; |
3376 | ||
3377 | dm_queue_flush(md); | |
3378 | ||
3379 | done: | |
3380 | mutex_unlock(&md->suspend_lock); | |
3381 | } | |
b735fede | 3382 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3383 | |
1da177e4 LT |
3384 | /*----------------------------------------------------------------- |
3385 | * Event notification. | |
3386 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3387 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3388 | unsigned cookie) |
69267a30 | 3389 | { |
60935eb2 MB |
3390 | char udev_cookie[DM_COOKIE_LENGTH]; |
3391 | char *envp[] = { udev_cookie, NULL }; | |
3392 | ||
3393 | if (!cookie) | |
3abf85b5 | 3394 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3395 | else { |
3396 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3397 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3398 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3399 | action, envp); | |
60935eb2 | 3400 | } |
69267a30 AK |
3401 | } |
3402 | ||
7a8c3d3b MA |
3403 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3404 | { | |
3405 | return atomic_add_return(1, &md->uevent_seq); | |
3406 | } | |
3407 | ||
1da177e4 LT |
3408 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3409 | { | |
3410 | return atomic_read(&md->event_nr); | |
3411 | } | |
3412 | ||
3413 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3414 | { | |
3415 | return wait_event_interruptible(md->eventq, | |
3416 | (event_nr != atomic_read(&md->event_nr))); | |
3417 | } | |
3418 | ||
7a8c3d3b MA |
3419 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3420 | { | |
3421 | unsigned long flags; | |
3422 | ||
3423 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3424 | list_add(elist, &md->uevent_list); | |
3425 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3426 | } | |
3427 | ||
1da177e4 LT |
3428 | /* |
3429 | * The gendisk is only valid as long as you have a reference | |
3430 | * count on 'md'. | |
3431 | */ | |
3432 | struct gendisk *dm_disk(struct mapped_device *md) | |
3433 | { | |
3434 | return md->disk; | |
3435 | } | |
65ff5b7d | 3436 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3437 | |
784aae73 MB |
3438 | struct kobject *dm_kobject(struct mapped_device *md) |
3439 | { | |
2995fa78 | 3440 | return &md->kobj_holder.kobj; |
784aae73 MB |
3441 | } |
3442 | ||
784aae73 MB |
3443 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3444 | { | |
3445 | struct mapped_device *md; | |
3446 | ||
2995fa78 | 3447 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3448 | |
4d89b7b4 | 3449 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3450 | dm_deleting_md(md)) |
4d89b7b4 MB |
3451 | return NULL; |
3452 | ||
784aae73 MB |
3453 | dm_get(md); |
3454 | return md; | |
3455 | } | |
3456 | ||
4f186f8b | 3457 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3458 | { |
3459 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3460 | } | |
3461 | ||
ffcc3936 MS |
3462 | int dm_suspended_internally_md(struct mapped_device *md) |
3463 | { | |
3464 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3465 | } | |
3466 | ||
2c140a24 MP |
3467 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3468 | { | |
3469 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3470 | } | |
3471 | ||
64dbce58 KU |
3472 | int dm_suspended(struct dm_target *ti) |
3473 | { | |
ecdb2e25 | 3474 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3475 | } |
3476 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3477 | ||
2e93ccc1 KU |
3478 | int dm_noflush_suspending(struct dm_target *ti) |
3479 | { | |
ecdb2e25 | 3480 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3481 | } |
3482 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3483 | ||
5f1b670d CH |
3484 | struct dm_md_mempools *dm_alloc_bio_mempools(unsigned integrity, |
3485 | unsigned per_bio_data_size) | |
e6ee8c0b | 3486 | { |
5f1b670d CH |
3487 | struct dm_md_mempools *pools; |
3488 | unsigned int pool_size = dm_get_reserved_bio_based_ios(); | |
5f015204 | 3489 | unsigned int front_pad; |
e6ee8c0b | 3490 | |
5f1b670d | 3491 | pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
e6ee8c0b | 3492 | if (!pools) |
4e6e36c3 | 3493 | return NULL; |
e6ee8c0b | 3494 | |
5f1b670d CH |
3495 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + |
3496 | offsetof(struct dm_target_io, clone); | |
17e149b8 | 3497 | |
5f1b670d CH |
3498 | pools->io_pool = mempool_create_slab_pool(pool_size, _io_cache); |
3499 | if (!pools->io_pool) | |
3500 | goto out; | |
e6ee8c0b | 3501 | |
3d8aab2d | 3502 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3503 | if (!pools->bs) |
5f015204 | 3504 | goto out; |
e6ee8c0b | 3505 | |
a91a2785 | 3506 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3507 | goto out; |
a91a2785 | 3508 | |
e6ee8c0b | 3509 | return pools; |
5f015204 JN |
3510 | out: |
3511 | dm_free_md_mempools(pools); | |
4e6e36c3 | 3512 | return NULL; |
5f1b670d CH |
3513 | } |
3514 | ||
3515 | struct dm_md_mempools *dm_alloc_rq_mempools(struct mapped_device *md, | |
3516 | unsigned type) | |
3517 | { | |
4e6e36c3 | 3518 | unsigned int pool_size = dm_get_reserved_rq_based_ios(); |
5f1b670d | 3519 | struct dm_md_mempools *pools; |
e6ee8c0b | 3520 | |
5f1b670d CH |
3521 | pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3522 | if (!pools) | |
4e6e36c3 | 3523 | return NULL; |
5f1b670d | 3524 | |
4e6e36c3 MS |
3525 | if (filter_md_type(type, md) == DM_TYPE_REQUEST_BASED) { |
3526 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3527 | if (!pools->rq_pool) | |
3528 | goto out; | |
3529 | } | |
e6ee8c0b | 3530 | |
5f1b670d CH |
3531 | pools->io_pool = mempool_create_slab_pool(pool_size, _rq_tio_cache); |
3532 | if (!pools->io_pool) | |
3533 | goto out; | |
3534 | ||
3535 | return pools; | |
3536 | out: | |
3537 | dm_free_md_mempools(pools); | |
4e6e36c3 | 3538 | return NULL; |
e6ee8c0b KU |
3539 | } |
3540 | ||
3541 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3542 | { | |
3543 | if (!pools) | |
3544 | return; | |
3545 | ||
3546 | if (pools->io_pool) | |
3547 | mempool_destroy(pools->io_pool); | |
3548 | ||
1ae49ea2 MS |
3549 | if (pools->rq_pool) |
3550 | mempool_destroy(pools->rq_pool); | |
3551 | ||
e6ee8c0b KU |
3552 | if (pools->bs) |
3553 | bioset_free(pools->bs); | |
3554 | ||
3555 | kfree(pools); | |
3556 | } | |
3557 | ||
83d5cde4 | 3558 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3559 | .open = dm_blk_open, |
3560 | .release = dm_blk_close, | |
aa129a22 | 3561 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3562 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3563 | .owner = THIS_MODULE |
3564 | }; | |
3565 | ||
1da177e4 LT |
3566 | /* |
3567 | * module hooks | |
3568 | */ | |
3569 | module_init(dm_init); | |
3570 | module_exit(dm_exit); | |
3571 | ||
3572 | module_param(major, uint, 0); | |
3573 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3574 | |
e8603136 MS |
3575 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3576 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3577 | ||
f4790826 MS |
3578 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3579 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3580 | ||
17e149b8 MS |
3581 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3582 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3583 | ||
1da177e4 LT |
3584 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3585 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3586 | MODULE_LICENSE("GPL"); |