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> |
71cdb697 | 27 | #include <linux/pr.h> |
55782138 LZ |
28 | |
29 | #include <trace/events/block.h> | |
1da177e4 | 30 | |
72d94861 AK |
31 | #define DM_MSG_PREFIX "core" |
32 | ||
71a16736 NK |
33 | #ifdef CONFIG_PRINTK |
34 | /* | |
35 | * ratelimit state to be used in DMXXX_LIMIT(). | |
36 | */ | |
37 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
38 | DEFAULT_RATELIMIT_INTERVAL, | |
39 | DEFAULT_RATELIMIT_BURST); | |
40 | EXPORT_SYMBOL(dm_ratelimit_state); | |
41 | #endif | |
42 | ||
60935eb2 MB |
43 | /* |
44 | * Cookies are numeric values sent with CHANGE and REMOVE | |
45 | * uevents while resuming, removing or renaming the device. | |
46 | */ | |
47 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
48 | #define DM_COOKIE_LENGTH 24 | |
49 | ||
1da177e4 LT |
50 | static const char *_name = DM_NAME; |
51 | ||
52 | static unsigned int major = 0; | |
53 | static unsigned int _major = 0; | |
54 | ||
d15b774c AK |
55 | static DEFINE_IDR(_minor_idr); |
56 | ||
f32c10b0 | 57 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
58 | |
59 | static void do_deferred_remove(struct work_struct *w); | |
60 | ||
61 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
62 | ||
acfe0ad7 MP |
63 | static struct workqueue_struct *deferred_remove_workqueue; |
64 | ||
1da177e4 | 65 | /* |
8fbf26ad | 66 | * For bio-based dm. |
1da177e4 LT |
67 | * One of these is allocated per bio. |
68 | */ | |
69 | struct dm_io { | |
70 | struct mapped_device *md; | |
71 | int error; | |
1da177e4 | 72 | atomic_t io_count; |
6ae2fa67 | 73 | struct bio *bio; |
3eaf840e | 74 | unsigned long start_time; |
f88fb981 | 75 | spinlock_t endio_lock; |
fd2ed4d2 | 76 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
77 | }; |
78 | ||
8fbf26ad KU |
79 | /* |
80 | * For request-based dm. | |
81 | * One of these is allocated per request. | |
82 | */ | |
83 | struct dm_rq_target_io { | |
84 | struct mapped_device *md; | |
85 | struct dm_target *ti; | |
1ae49ea2 | 86 | struct request *orig, *clone; |
2eb6e1e3 | 87 | struct kthread_work work; |
8fbf26ad KU |
88 | int error; |
89 | union map_info info; | |
e262f347 MP |
90 | struct dm_stats_aux stats_aux; |
91 | unsigned long duration_jiffies; | |
92 | unsigned n_sectors; | |
8fbf26ad KU |
93 | }; |
94 | ||
95 | /* | |
94818742 KO |
96 | * For request-based dm - the bio clones we allocate are embedded in these |
97 | * structs. | |
98 | * | |
99 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
100 | * the bioset is created - this means the bio has to come at the end of the | |
101 | * struct. | |
8fbf26ad KU |
102 | */ |
103 | struct dm_rq_clone_bio_info { | |
104 | struct bio *orig; | |
cec47e3d | 105 | struct dm_rq_target_io *tio; |
94818742 | 106 | struct bio clone; |
8fbf26ad KU |
107 | }; |
108 | ||
ba61fdd1 JM |
109 | #define MINOR_ALLOCED ((void *)-1) |
110 | ||
1da177e4 LT |
111 | /* |
112 | * Bits for the md->flags field. | |
113 | */ | |
1eb787ec | 114 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 115 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 116 | #define DMF_FROZEN 2 |
fba9f90e | 117 | #define DMF_FREEING 3 |
5c6bd75d | 118 | #define DMF_DELETING 4 |
2e93ccc1 | 119 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
120 | #define DMF_DEFERRED_REMOVE 6 |
121 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 122 | |
83d5e5b0 MP |
123 | /* |
124 | * A dummy definition to make RCU happy. | |
125 | * struct dm_table should never be dereferenced in this file. | |
126 | */ | |
127 | struct dm_table { | |
128 | int undefined__; | |
129 | }; | |
130 | ||
304f3f6a MB |
131 | /* |
132 | * Work processed by per-device workqueue. | |
133 | */ | |
1da177e4 | 134 | struct mapped_device { |
83d5e5b0 | 135 | struct srcu_struct io_barrier; |
e61290a4 | 136 | struct mutex suspend_lock; |
1da177e4 | 137 | atomic_t holders; |
5c6bd75d | 138 | atomic_t open_count; |
1da177e4 | 139 | |
2a7faeb1 MP |
140 | /* |
141 | * The current mapping. | |
142 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
143 | * dereference. | |
144 | */ | |
6fa99520 | 145 | struct dm_table __rcu *map; |
2a7faeb1 | 146 | |
86f1152b BM |
147 | struct list_head table_devices; |
148 | struct mutex table_devices_lock; | |
149 | ||
1da177e4 LT |
150 | unsigned long flags; |
151 | ||
165125e1 | 152 | struct request_queue *queue; |
a5664dad | 153 | unsigned type; |
4a0b4ddf | 154 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
155 | struct mutex type_lock; |
156 | ||
16f12266 | 157 | struct dm_target *immutable_target; |
36a0456f AK |
158 | struct target_type *immutable_target_type; |
159 | ||
1da177e4 | 160 | struct gendisk *disk; |
7e51f257 | 161 | char name[16]; |
1da177e4 LT |
162 | |
163 | void *interface_ptr; | |
164 | ||
165 | /* | |
166 | * A list of ios that arrived while we were suspended. | |
167 | */ | |
316d315b | 168 | atomic_t pending[2]; |
1da177e4 | 169 | wait_queue_head_t wait; |
53d5914f | 170 | struct work_struct work; |
74859364 | 171 | struct bio_list deferred; |
022c2611 | 172 | spinlock_t deferred_lock; |
1da177e4 | 173 | |
af7e466a | 174 | /* |
29e4013d | 175 | * Processing queue (flush) |
304f3f6a MB |
176 | */ |
177 | struct workqueue_struct *wq; | |
178 | ||
1da177e4 LT |
179 | /* |
180 | * io objects are allocated from here. | |
181 | */ | |
182 | mempool_t *io_pool; | |
1ae49ea2 | 183 | mempool_t *rq_pool; |
1da177e4 | 184 | |
9faf400f SB |
185 | struct bio_set *bs; |
186 | ||
1da177e4 LT |
187 | /* |
188 | * Event handling. | |
189 | */ | |
190 | atomic_t event_nr; | |
191 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
192 | atomic_t uevent_seq; |
193 | struct list_head uevent_list; | |
194 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
195 | |
196 | /* | |
197 | * freeze/thaw support require holding onto a super block | |
198 | */ | |
199 | struct super_block *frozen_sb; | |
db8fef4f | 200 | struct block_device *bdev; |
3ac51e74 DW |
201 | |
202 | /* forced geometry settings */ | |
203 | struct hd_geometry geometry; | |
784aae73 | 204 | |
2995fa78 MP |
205 | /* kobject and completion */ |
206 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 207 | |
d87f4c14 TH |
208 | /* zero-length flush that will be cloned and submitted to targets */ |
209 | struct bio flush_bio; | |
fd2ed4d2 | 210 | |
96b26c8c MP |
211 | /* the number of internal suspends */ |
212 | unsigned internal_suspend_count; | |
213 | ||
fd2ed4d2 | 214 | struct dm_stats stats; |
2eb6e1e3 KB |
215 | |
216 | struct kthread_worker kworker; | |
217 | struct task_struct *kworker_task; | |
de3ec86d MS |
218 | |
219 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 220 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 221 | int last_rq_rw; |
0ce65797 MS |
222 | sector_t last_rq_pos; |
223 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
224 | |
225 | /* for blk-mq request-based DM support */ | |
1c357a1e | 226 | struct blk_mq_tag_set *tag_set; |
17e149b8 | 227 | bool use_blk_mq; |
1da177e4 LT |
228 | }; |
229 | ||
17e149b8 MS |
230 | #ifdef CONFIG_DM_MQ_DEFAULT |
231 | static bool use_blk_mq = true; | |
232 | #else | |
233 | static bool use_blk_mq = false; | |
234 | #endif | |
235 | ||
faad87df MS |
236 | #define DM_MQ_NR_HW_QUEUES 1 |
237 | #define DM_MQ_QUEUE_DEPTH 2048 | |
238 | ||
239 | static unsigned dm_mq_nr_hw_queues = DM_MQ_NR_HW_QUEUES; | |
240 | static unsigned dm_mq_queue_depth = DM_MQ_QUEUE_DEPTH; | |
241 | ||
17e149b8 MS |
242 | bool dm_use_blk_mq(struct mapped_device *md) |
243 | { | |
244 | return md->use_blk_mq; | |
245 | } | |
246 | ||
e6ee8c0b KU |
247 | /* |
248 | * For mempools pre-allocation at the table loading time. | |
249 | */ | |
250 | struct dm_md_mempools { | |
251 | mempool_t *io_pool; | |
1ae49ea2 | 252 | mempool_t *rq_pool; |
e6ee8c0b KU |
253 | struct bio_set *bs; |
254 | }; | |
255 | ||
86f1152b BM |
256 | struct table_device { |
257 | struct list_head list; | |
258 | atomic_t count; | |
259 | struct dm_dev dm_dev; | |
260 | }; | |
261 | ||
6cfa5857 MS |
262 | #define RESERVED_BIO_BASED_IOS 16 |
263 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 264 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 265 | static struct kmem_cache *_io_cache; |
8fbf26ad | 266 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 267 | static struct kmem_cache *_rq_cache; |
94818742 | 268 | |
e8603136 MS |
269 | /* |
270 | * Bio-based DM's mempools' reserved IOs set by the user. | |
271 | */ | |
272 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
273 | ||
f4790826 MS |
274 | /* |
275 | * Request-based DM's mempools' reserved IOs set by the user. | |
276 | */ | |
277 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
278 | ||
09c2d531 | 279 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
280 | unsigned def, unsigned max) |
281 | { | |
09c2d531 MS |
282 | unsigned param = ACCESS_ONCE(*module_param); |
283 | unsigned modified_param = 0; | |
f4790826 | 284 | |
09c2d531 MS |
285 | if (!param) |
286 | modified_param = def; | |
287 | else if (param > max) | |
288 | modified_param = max; | |
f4790826 | 289 | |
09c2d531 MS |
290 | if (modified_param) { |
291 | (void)cmpxchg(module_param, param, modified_param); | |
292 | param = modified_param; | |
f4790826 MS |
293 | } |
294 | ||
09c2d531 | 295 | return param; |
f4790826 MS |
296 | } |
297 | ||
e8603136 MS |
298 | unsigned dm_get_reserved_bio_based_ios(void) |
299 | { | |
09c2d531 | 300 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
301 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
302 | } | |
303 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
304 | ||
f4790826 MS |
305 | unsigned dm_get_reserved_rq_based_ios(void) |
306 | { | |
09c2d531 | 307 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
308 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
309 | } | |
310 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
311 | ||
faad87df MS |
312 | static unsigned dm_get_blk_mq_nr_hw_queues(void) |
313 | { | |
314 | return __dm_get_module_param(&dm_mq_nr_hw_queues, 1, 32); | |
315 | } | |
316 | ||
317 | static unsigned dm_get_blk_mq_queue_depth(void) | |
318 | { | |
319 | return __dm_get_module_param(&dm_mq_queue_depth, | |
320 | DM_MQ_QUEUE_DEPTH, BLK_MQ_MAX_DEPTH); | |
321 | } | |
322 | ||
1da177e4 LT |
323 | static int __init local_init(void) |
324 | { | |
51157b4a | 325 | int r = -ENOMEM; |
1da177e4 | 326 | |
1da177e4 | 327 | /* allocate a slab for the dm_ios */ |
028867ac | 328 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 329 | if (!_io_cache) |
51157b4a | 330 | return r; |
1da177e4 | 331 | |
8fbf26ad KU |
332 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
333 | if (!_rq_tio_cache) | |
dba14160 | 334 | goto out_free_io_cache; |
8fbf26ad | 335 | |
1ae49ea2 MS |
336 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
337 | __alignof__(struct request), 0, NULL); | |
338 | if (!_rq_cache) | |
339 | goto out_free_rq_tio_cache; | |
340 | ||
51e5b2bd | 341 | r = dm_uevent_init(); |
51157b4a | 342 | if (r) |
1ae49ea2 | 343 | goto out_free_rq_cache; |
51e5b2bd | 344 | |
acfe0ad7 MP |
345 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
346 | if (!deferred_remove_workqueue) { | |
347 | r = -ENOMEM; | |
348 | goto out_uevent_exit; | |
349 | } | |
350 | ||
1da177e4 LT |
351 | _major = major; |
352 | r = register_blkdev(_major, _name); | |
51157b4a | 353 | if (r < 0) |
acfe0ad7 | 354 | goto out_free_workqueue; |
1da177e4 LT |
355 | |
356 | if (!_major) | |
357 | _major = r; | |
358 | ||
359 | return 0; | |
51157b4a | 360 | |
acfe0ad7 MP |
361 | out_free_workqueue: |
362 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
363 | out_uevent_exit: |
364 | dm_uevent_exit(); | |
1ae49ea2 MS |
365 | out_free_rq_cache: |
366 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
367 | out_free_rq_tio_cache: |
368 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
369 | out_free_io_cache: |
370 | kmem_cache_destroy(_io_cache); | |
371 | ||
372 | return r; | |
1da177e4 LT |
373 | } |
374 | ||
375 | static void local_exit(void) | |
376 | { | |
2c140a24 | 377 | flush_scheduled_work(); |
acfe0ad7 | 378 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 379 | |
1ae49ea2 | 380 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 381 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 382 | kmem_cache_destroy(_io_cache); |
00d59405 | 383 | unregister_blkdev(_major, _name); |
51e5b2bd | 384 | dm_uevent_exit(); |
1da177e4 LT |
385 | |
386 | _major = 0; | |
387 | ||
388 | DMINFO("cleaned up"); | |
389 | } | |
390 | ||
b9249e55 | 391 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
392 | local_init, |
393 | dm_target_init, | |
394 | dm_linear_init, | |
395 | dm_stripe_init, | |
952b3557 | 396 | dm_io_init, |
945fa4d2 | 397 | dm_kcopyd_init, |
1da177e4 | 398 | dm_interface_init, |
fd2ed4d2 | 399 | dm_statistics_init, |
1da177e4 LT |
400 | }; |
401 | ||
b9249e55 | 402 | static void (*_exits[])(void) = { |
1da177e4 LT |
403 | local_exit, |
404 | dm_target_exit, | |
405 | dm_linear_exit, | |
406 | dm_stripe_exit, | |
952b3557 | 407 | dm_io_exit, |
945fa4d2 | 408 | dm_kcopyd_exit, |
1da177e4 | 409 | dm_interface_exit, |
fd2ed4d2 | 410 | dm_statistics_exit, |
1da177e4 LT |
411 | }; |
412 | ||
413 | static int __init dm_init(void) | |
414 | { | |
415 | const int count = ARRAY_SIZE(_inits); | |
416 | ||
417 | int r, i; | |
418 | ||
419 | for (i = 0; i < count; i++) { | |
420 | r = _inits[i](); | |
421 | if (r) | |
422 | goto bad; | |
423 | } | |
424 | ||
425 | return 0; | |
426 | ||
427 | bad: | |
428 | while (i--) | |
429 | _exits[i](); | |
430 | ||
431 | return r; | |
432 | } | |
433 | ||
434 | static void __exit dm_exit(void) | |
435 | { | |
436 | int i = ARRAY_SIZE(_exits); | |
437 | ||
438 | while (i--) | |
439 | _exits[i](); | |
d15b774c AK |
440 | |
441 | /* | |
442 | * Should be empty by this point. | |
443 | */ | |
d15b774c | 444 | idr_destroy(&_minor_idr); |
1da177e4 LT |
445 | } |
446 | ||
447 | /* | |
448 | * Block device functions | |
449 | */ | |
432a212c MA |
450 | int dm_deleting_md(struct mapped_device *md) |
451 | { | |
452 | return test_bit(DMF_DELETING, &md->flags); | |
453 | } | |
454 | ||
fe5f9f2c | 455 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
456 | { |
457 | struct mapped_device *md; | |
458 | ||
fba9f90e JM |
459 | spin_lock(&_minor_lock); |
460 | ||
fe5f9f2c | 461 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
462 | if (!md) |
463 | goto out; | |
464 | ||
5c6bd75d | 465 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 466 | dm_deleting_md(md)) { |
fba9f90e JM |
467 | md = NULL; |
468 | goto out; | |
469 | } | |
470 | ||
1da177e4 | 471 | dm_get(md); |
5c6bd75d | 472 | atomic_inc(&md->open_count); |
fba9f90e JM |
473 | out: |
474 | spin_unlock(&_minor_lock); | |
475 | ||
476 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
477 | } |
478 | ||
db2a144b | 479 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 480 | { |
63a4f065 | 481 | struct mapped_device *md; |
6e9624b8 | 482 | |
4a1aeb98 MB |
483 | spin_lock(&_minor_lock); |
484 | ||
63a4f065 MS |
485 | md = disk->private_data; |
486 | if (WARN_ON(!md)) | |
487 | goto out; | |
488 | ||
2c140a24 MP |
489 | if (atomic_dec_and_test(&md->open_count) && |
490 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 491 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 492 | |
1da177e4 | 493 | dm_put(md); |
63a4f065 | 494 | out: |
4a1aeb98 | 495 | spin_unlock(&_minor_lock); |
1da177e4 LT |
496 | } |
497 | ||
5c6bd75d AK |
498 | int dm_open_count(struct mapped_device *md) |
499 | { | |
500 | return atomic_read(&md->open_count); | |
501 | } | |
502 | ||
503 | /* | |
504 | * Guarantees nothing is using the device before it's deleted. | |
505 | */ | |
2c140a24 | 506 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
507 | { |
508 | int r = 0; | |
509 | ||
510 | spin_lock(&_minor_lock); | |
511 | ||
2c140a24 | 512 | if (dm_open_count(md)) { |
5c6bd75d | 513 | r = -EBUSY; |
2c140a24 MP |
514 | if (mark_deferred) |
515 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
516 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
517 | r = -EEXIST; | |
5c6bd75d AK |
518 | else |
519 | set_bit(DMF_DELETING, &md->flags); | |
520 | ||
521 | spin_unlock(&_minor_lock); | |
522 | ||
523 | return r; | |
524 | } | |
525 | ||
2c140a24 MP |
526 | int dm_cancel_deferred_remove(struct mapped_device *md) |
527 | { | |
528 | int r = 0; | |
529 | ||
530 | spin_lock(&_minor_lock); | |
531 | ||
532 | if (test_bit(DMF_DELETING, &md->flags)) | |
533 | r = -EBUSY; | |
534 | else | |
535 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
536 | ||
537 | spin_unlock(&_minor_lock); | |
538 | ||
539 | return r; | |
540 | } | |
541 | ||
542 | static void do_deferred_remove(struct work_struct *w) | |
543 | { | |
544 | dm_deferred_remove(); | |
545 | } | |
546 | ||
fd2ed4d2 MP |
547 | sector_t dm_get_size(struct mapped_device *md) |
548 | { | |
549 | return get_capacity(md->disk); | |
550 | } | |
551 | ||
9974fa2c MS |
552 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
553 | { | |
554 | return md->queue; | |
555 | } | |
556 | ||
fd2ed4d2 MP |
557 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
558 | { | |
559 | return &md->stats; | |
560 | } | |
561 | ||
3ac51e74 DW |
562 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
563 | { | |
564 | struct mapped_device *md = bdev->bd_disk->private_data; | |
565 | ||
566 | return dm_get_geometry(md, geo); | |
567 | } | |
568 | ||
956a4025 MS |
569 | static int dm_grab_bdev_for_ioctl(struct mapped_device *md, |
570 | struct block_device **bdev, | |
571 | fmode_t *mode) | |
aa129a22 | 572 | { |
66482026 | 573 | struct dm_target *tgt; |
6c182cd8 | 574 | struct dm_table *map; |
956a4025 | 575 | int srcu_idx, r; |
aa129a22 | 576 | |
6c182cd8 | 577 | retry: |
e56f81e0 | 578 | r = -ENOTTY; |
956a4025 | 579 | map = dm_get_live_table(md, &srcu_idx); |
aa129a22 MB |
580 | if (!map || !dm_table_get_size(map)) |
581 | goto out; | |
582 | ||
583 | /* We only support devices that have a single target */ | |
584 | if (dm_table_get_num_targets(map) != 1) | |
585 | goto out; | |
586 | ||
66482026 MS |
587 | tgt = dm_table_get_target(map, 0); |
588 | if (!tgt->type->prepare_ioctl) | |
4d341d82 | 589 | goto out; |
aa129a22 | 590 | |
4f186f8b | 591 | if (dm_suspended_md(md)) { |
aa129a22 MB |
592 | r = -EAGAIN; |
593 | goto out; | |
594 | } | |
595 | ||
66482026 | 596 | r = tgt->type->prepare_ioctl(tgt, bdev, mode); |
e56f81e0 CH |
597 | if (r < 0) |
598 | goto out; | |
aa129a22 | 599 | |
956a4025 MS |
600 | bdgrab(*bdev); |
601 | dm_put_live_table(md, srcu_idx); | |
e56f81e0 | 602 | return r; |
aa129a22 | 603 | |
aa129a22 | 604 | out: |
956a4025 | 605 | dm_put_live_table(md, srcu_idx); |
5bbbfdf6 | 606 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
607 | msleep(10); |
608 | goto retry; | |
609 | } | |
e56f81e0 CH |
610 | return r; |
611 | } | |
612 | ||
613 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
614 | unsigned int cmd, unsigned long arg) | |
615 | { | |
616 | struct mapped_device *md = bdev->bd_disk->private_data; | |
956a4025 | 617 | int r; |
e56f81e0 | 618 | |
956a4025 | 619 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
e56f81e0 CH |
620 | if (r < 0) |
621 | return r; | |
6c182cd8 | 622 | |
e56f81e0 CH |
623 | if (r > 0) { |
624 | /* | |
625 | * Target determined this ioctl is being issued against | |
626 | * a logical partition of the parent bdev; so extra | |
627 | * validation is needed. | |
628 | */ | |
629 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
630 | if (r) | |
631 | goto out; | |
632 | } | |
6c182cd8 | 633 | |
66482026 | 634 | r = __blkdev_driver_ioctl(bdev, mode, cmd, arg); |
e56f81e0 | 635 | out: |
956a4025 | 636 | bdput(bdev); |
aa129a22 MB |
637 | return r; |
638 | } | |
639 | ||
028867ac | 640 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
641 | { |
642 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
643 | } | |
644 | ||
028867ac | 645 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
646 | { |
647 | mempool_free(io, md->io_pool); | |
648 | } | |
649 | ||
028867ac | 650 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 651 | { |
dba14160 | 652 | bio_put(&tio->clone); |
1da177e4 LT |
653 | } |
654 | ||
08885643 KU |
655 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
656 | gfp_t gfp_mask) | |
cec47e3d | 657 | { |
5f015204 | 658 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
659 | } |
660 | ||
661 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
662 | { | |
5f015204 | 663 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
664 | } |
665 | ||
1ae49ea2 MS |
666 | static struct request *alloc_clone_request(struct mapped_device *md, |
667 | gfp_t gfp_mask) | |
668 | { | |
669 | return mempool_alloc(md->rq_pool, gfp_mask); | |
670 | } | |
671 | ||
672 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
673 | { | |
674 | mempool_free(rq, md->rq_pool); | |
675 | } | |
676 | ||
90abb8c4 KU |
677 | static int md_in_flight(struct mapped_device *md) |
678 | { | |
679 | return atomic_read(&md->pending[READ]) + | |
680 | atomic_read(&md->pending[WRITE]); | |
681 | } | |
682 | ||
3eaf840e JNN |
683 | static void start_io_acct(struct dm_io *io) |
684 | { | |
685 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 686 | struct bio *bio = io->bio; |
c9959059 | 687 | int cpu; |
fd2ed4d2 | 688 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
689 | |
690 | io->start_time = jiffies; | |
691 | ||
074a7aca TH |
692 | cpu = part_stat_lock(); |
693 | part_round_stats(cpu, &dm_disk(md)->part0); | |
694 | part_stat_unlock(); | |
1e9bb880 SL |
695 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
696 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
697 | |
698 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 699 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 700 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
701 | } |
702 | ||
d221d2e7 | 703 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
704 | { |
705 | struct mapped_device *md = io->md; | |
706 | struct bio *bio = io->bio; | |
707 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 708 | int pending; |
3eaf840e JNN |
709 | int rw = bio_data_dir(bio); |
710 | ||
18c0b223 | 711 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 712 | |
fd2ed4d2 | 713 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 714 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
715 | bio_sectors(bio), true, duration, &io->stats_aux); |
716 | ||
af7e466a MP |
717 | /* |
718 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 719 | * a flush. |
af7e466a | 720 | */ |
1e9bb880 SL |
721 | pending = atomic_dec_return(&md->pending[rw]); |
722 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 723 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 724 | |
d221d2e7 MP |
725 | /* nudge anyone waiting on suspend queue */ |
726 | if (!pending) | |
727 | wake_up(&md->wait); | |
3eaf840e JNN |
728 | } |
729 | ||
1da177e4 LT |
730 | /* |
731 | * Add the bio to the list of deferred io. | |
732 | */ | |
92c63902 | 733 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 734 | { |
05447420 | 735 | unsigned long flags; |
1da177e4 | 736 | |
05447420 | 737 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 738 | bio_list_add(&md->deferred, bio); |
05447420 | 739 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 740 | queue_work(md->wq, &md->work); |
1da177e4 LT |
741 | } |
742 | ||
743 | /* | |
744 | * Everyone (including functions in this file), should use this | |
745 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 746 | * dm_put_live_table() when finished. |
1da177e4 | 747 | */ |
83d5e5b0 | 748 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 749 | { |
83d5e5b0 MP |
750 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
751 | ||
752 | return srcu_dereference(md->map, &md->io_barrier); | |
753 | } | |
1da177e4 | 754 | |
83d5e5b0 MP |
755 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
756 | { | |
757 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
758 | } | |
759 | ||
760 | void dm_sync_table(struct mapped_device *md) | |
761 | { | |
762 | synchronize_srcu(&md->io_barrier); | |
763 | synchronize_rcu_expedited(); | |
764 | } | |
765 | ||
766 | /* | |
767 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
768 | * The caller must not block between these two functions. | |
769 | */ | |
770 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
771 | { | |
772 | rcu_read_lock(); | |
773 | return rcu_dereference(md->map); | |
774 | } | |
1da177e4 | 775 | |
83d5e5b0 MP |
776 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
777 | { | |
778 | rcu_read_unlock(); | |
1da177e4 LT |
779 | } |
780 | ||
86f1152b BM |
781 | /* |
782 | * Open a table device so we can use it as a map destination. | |
783 | */ | |
784 | static int open_table_device(struct table_device *td, dev_t dev, | |
785 | struct mapped_device *md) | |
786 | { | |
787 | static char *_claim_ptr = "I belong to device-mapper"; | |
788 | struct block_device *bdev; | |
789 | ||
790 | int r; | |
791 | ||
792 | BUG_ON(td->dm_dev.bdev); | |
793 | ||
794 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
795 | if (IS_ERR(bdev)) | |
796 | return PTR_ERR(bdev); | |
797 | ||
798 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
799 | if (r) { | |
800 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
801 | return r; | |
802 | } | |
803 | ||
804 | td->dm_dev.bdev = bdev; | |
805 | return 0; | |
806 | } | |
807 | ||
808 | /* | |
809 | * Close a table device that we've been using. | |
810 | */ | |
811 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
812 | { | |
813 | if (!td->dm_dev.bdev) | |
814 | return; | |
815 | ||
816 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
817 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
818 | td->dm_dev.bdev = NULL; | |
819 | } | |
820 | ||
821 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
822 | fmode_t mode) { | |
823 | struct table_device *td; | |
824 | ||
825 | list_for_each_entry(td, l, list) | |
826 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
827 | return td; | |
828 | ||
829 | return NULL; | |
830 | } | |
831 | ||
832 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
833 | struct dm_dev **result) { | |
834 | int r; | |
835 | struct table_device *td; | |
836 | ||
837 | mutex_lock(&md->table_devices_lock); | |
838 | td = find_table_device(&md->table_devices, dev, mode); | |
839 | if (!td) { | |
840 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
841 | if (!td) { | |
842 | mutex_unlock(&md->table_devices_lock); | |
843 | return -ENOMEM; | |
844 | } | |
845 | ||
846 | td->dm_dev.mode = mode; | |
847 | td->dm_dev.bdev = NULL; | |
848 | ||
849 | if ((r = open_table_device(td, dev, md))) { | |
850 | mutex_unlock(&md->table_devices_lock); | |
851 | kfree(td); | |
852 | return r; | |
853 | } | |
854 | ||
855 | format_dev_t(td->dm_dev.name, dev); | |
856 | ||
857 | atomic_set(&td->count, 0); | |
858 | list_add(&td->list, &md->table_devices); | |
859 | } | |
860 | atomic_inc(&td->count); | |
861 | mutex_unlock(&md->table_devices_lock); | |
862 | ||
863 | *result = &td->dm_dev; | |
864 | return 0; | |
865 | } | |
866 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
867 | ||
868 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
869 | { | |
870 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
871 | ||
872 | mutex_lock(&md->table_devices_lock); | |
873 | if (atomic_dec_and_test(&td->count)) { | |
874 | close_table_device(td, md); | |
875 | list_del(&td->list); | |
876 | kfree(td); | |
877 | } | |
878 | mutex_unlock(&md->table_devices_lock); | |
879 | } | |
880 | EXPORT_SYMBOL(dm_put_table_device); | |
881 | ||
882 | static void free_table_devices(struct list_head *devices) | |
883 | { | |
884 | struct list_head *tmp, *next; | |
885 | ||
886 | list_for_each_safe(tmp, next, devices) { | |
887 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
888 | ||
889 | DMWARN("dm_destroy: %s still exists with %d references", | |
890 | td->dm_dev.name, atomic_read(&td->count)); | |
891 | kfree(td); | |
892 | } | |
893 | } | |
894 | ||
3ac51e74 DW |
895 | /* |
896 | * Get the geometry associated with a dm device | |
897 | */ | |
898 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
899 | { | |
900 | *geo = md->geometry; | |
901 | ||
902 | return 0; | |
903 | } | |
904 | ||
905 | /* | |
906 | * Set the geometry of a device. | |
907 | */ | |
908 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
909 | { | |
910 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
911 | ||
912 | if (geo->start > sz) { | |
913 | DMWARN("Start sector is beyond the geometry limits."); | |
914 | return -EINVAL; | |
915 | } | |
916 | ||
917 | md->geometry = *geo; | |
918 | ||
919 | return 0; | |
920 | } | |
921 | ||
1da177e4 LT |
922 | /*----------------------------------------------------------------- |
923 | * CRUD START: | |
924 | * A more elegant soln is in the works that uses the queue | |
925 | * merge fn, unfortunately there are a couple of changes to | |
926 | * the block layer that I want to make for this. So in the | |
927 | * interests of getting something for people to use I give | |
928 | * you this clearly demarcated crap. | |
929 | *---------------------------------------------------------------*/ | |
930 | ||
2e93ccc1 KU |
931 | static int __noflush_suspending(struct mapped_device *md) |
932 | { | |
933 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
934 | } | |
935 | ||
1da177e4 LT |
936 | /* |
937 | * Decrements the number of outstanding ios that a bio has been | |
938 | * cloned into, completing the original io if necc. | |
939 | */ | |
858119e1 | 940 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 941 | { |
2e93ccc1 | 942 | unsigned long flags; |
b35f8caa MB |
943 | int io_error; |
944 | struct bio *bio; | |
945 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
946 | |
947 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
948 | if (unlikely(error)) { |
949 | spin_lock_irqsave(&io->endio_lock, flags); | |
950 | if (!(io->error > 0 && __noflush_suspending(md))) | |
951 | io->error = error; | |
952 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
953 | } | |
1da177e4 LT |
954 | |
955 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
956 | if (io->error == DM_ENDIO_REQUEUE) { |
957 | /* | |
958 | * Target requested pushing back the I/O. | |
2e93ccc1 | 959 | */ |
022c2611 | 960 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
961 | if (__noflush_suspending(md)) |
962 | bio_list_add_head(&md->deferred, io->bio); | |
963 | else | |
2e93ccc1 KU |
964 | /* noflush suspend was interrupted. */ |
965 | io->error = -EIO; | |
022c2611 | 966 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
967 | } |
968 | ||
b35f8caa MB |
969 | io_error = io->error; |
970 | bio = io->bio; | |
6a8736d1 TH |
971 | end_io_acct(io); |
972 | free_io(md, io); | |
973 | ||
974 | if (io_error == DM_ENDIO_REQUEUE) | |
975 | return; | |
2e93ccc1 | 976 | |
4f024f37 | 977 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 978 | /* |
6a8736d1 TH |
979 | * Preflush done for flush with data, reissue |
980 | * without REQ_FLUSH. | |
af7e466a | 981 | */ |
6a8736d1 TH |
982 | bio->bi_rw &= ~REQ_FLUSH; |
983 | queue_io(md, bio); | |
af7e466a | 984 | } else { |
b372d360 | 985 | /* done with normal IO or empty flush */ |
0a82a8d1 | 986 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
987 | bio->bi_error = io_error; |
988 | bio_endio(bio); | |
b35f8caa | 989 | } |
1da177e4 LT |
990 | } |
991 | } | |
992 | ||
7eee4ae2 MS |
993 | static void disable_write_same(struct mapped_device *md) |
994 | { | |
995 | struct queue_limits *limits = dm_get_queue_limits(md); | |
996 | ||
997 | /* device doesn't really support WRITE SAME, disable it */ | |
998 | limits->max_write_same_sectors = 0; | |
999 | } | |
1000 | ||
4246a0b6 | 1001 | static void clone_endio(struct bio *bio) |
1da177e4 | 1002 | { |
4246a0b6 | 1003 | int error = bio->bi_error; |
5164bece | 1004 | int r = error; |
bfc6d41c | 1005 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 1006 | struct dm_io *io = tio->io; |
9faf400f | 1007 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
1008 | dm_endio_fn endio = tio->ti->type->end_io; |
1009 | ||
1da177e4 | 1010 | if (endio) { |
7de3ee57 | 1011 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
1012 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
1013 | /* | |
1014 | * error and requeue request are handled | |
1015 | * in dec_pending(). | |
1016 | */ | |
1da177e4 | 1017 | error = r; |
45cbcd79 KU |
1018 | else if (r == DM_ENDIO_INCOMPLETE) |
1019 | /* The target will handle the io */ | |
6712ecf8 | 1020 | return; |
45cbcd79 KU |
1021 | else if (r) { |
1022 | DMWARN("unimplemented target endio return value: %d", r); | |
1023 | BUG(); | |
1024 | } | |
1da177e4 LT |
1025 | } |
1026 | ||
7eee4ae2 MS |
1027 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
1028 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
1029 | disable_write_same(md); | |
1030 | ||
9faf400f | 1031 | free_tio(md, tio); |
b35f8caa | 1032 | dec_pending(io, error); |
1da177e4 LT |
1033 | } |
1034 | ||
78d8e58a MS |
1035 | /* |
1036 | * Partial completion handling for request-based dm | |
1037 | */ | |
4246a0b6 | 1038 | static void end_clone_bio(struct bio *clone) |
78d8e58a MS |
1039 | { |
1040 | struct dm_rq_clone_bio_info *info = | |
1041 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
1042 | struct dm_rq_target_io *tio = info->tio; | |
1043 | struct bio *bio = info->orig; | |
1044 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; | |
50887bd1 | 1045 | int error = clone->bi_error; |
78d8e58a MS |
1046 | |
1047 | bio_put(clone); | |
1048 | ||
1049 | if (tio->error) | |
1050 | /* | |
1051 | * An error has already been detected on the request. | |
1052 | * Once error occurred, just let clone->end_io() handle | |
1053 | * the remainder. | |
1054 | */ | |
1055 | return; | |
50887bd1 | 1056 | else if (error) { |
78d8e58a MS |
1057 | /* |
1058 | * Don't notice the error to the upper layer yet. | |
1059 | * The error handling decision is made by the target driver, | |
1060 | * when the request is completed. | |
1061 | */ | |
50887bd1 | 1062 | tio->error = error; |
78d8e58a MS |
1063 | return; |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * I/O for the bio successfully completed. | |
1068 | * Notice the data completion to the upper layer. | |
1069 | */ | |
1070 | ||
1071 | /* | |
1072 | * bios are processed from the head of the list. | |
1073 | * So the completing bio should always be rq->bio. | |
1074 | * If it's not, something wrong is happening. | |
1075 | */ | |
1076 | if (tio->orig->bio != bio) | |
1077 | DMERR("bio completion is going in the middle of the request"); | |
1078 | ||
1079 | /* | |
1080 | * Update the original request. | |
1081 | * Do not use blk_end_request() here, because it may complete | |
1082 | * the original request before the clone, and break the ordering. | |
1083 | */ | |
1084 | blk_update_request(tio->orig, 0, nr_bytes); | |
1085 | } | |
1086 | ||
bfebd1cd MS |
1087 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1088 | { | |
1089 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1090 | } | |
1091 | ||
e262f347 MP |
1092 | static void rq_end_stats(struct mapped_device *md, struct request *orig) |
1093 | { | |
1094 | if (unlikely(dm_stats_used(&md->stats))) { | |
1095 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1096 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
1097 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1098 | tio->n_sectors, true, tio->duration_jiffies, | |
1099 | &tio->stats_aux); | |
1100 | } | |
1101 | } | |
1102 | ||
cec47e3d KU |
1103 | /* |
1104 | * Don't touch any member of the md after calling this function because | |
1105 | * the md may be freed in dm_put() at the end of this function. | |
1106 | * Or do dm_get() before calling this function and dm_put() later. | |
1107 | */ | |
466d89a6 | 1108 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1109 | { |
b4324fee | 1110 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1111 | |
1112 | /* nudge anyone waiting on suspend queue */ | |
621739b0 | 1113 | if (!md_in_flight(md)) |
cec47e3d KU |
1114 | wake_up(&md->wait); |
1115 | ||
a8c32a5c JA |
1116 | /* |
1117 | * Run this off this callpath, as drivers could invoke end_io while | |
1118 | * inside their request_fn (and holding the queue lock). Calling | |
1119 | * back into ->request_fn() could deadlock attempting to grab the | |
1120 | * queue lock again. | |
1121 | */ | |
6acfe68b MS |
1122 | if (!md->queue->mq_ops && run_queue) |
1123 | blk_run_queue_async(md->queue); | |
cec47e3d KU |
1124 | |
1125 | /* | |
1126 | * dm_put() must be at the end of this function. See the comment above | |
1127 | */ | |
1128 | dm_put(md); | |
1129 | } | |
1130 | ||
e5d8de32 | 1131 | static void free_rq_clone(struct request *clone) |
a77e28c7 KU |
1132 | { |
1133 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1134 | struct mapped_device *md = tio->md; |
a77e28c7 | 1135 | |
78d8e58a MS |
1136 | blk_rq_unprep_clone(clone); |
1137 | ||
aa6df8dd MS |
1138 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1139 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1140 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1141 | else if (!md->queue->mq_ops) |
1142 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd | 1143 | free_clone_request(md, clone); |
aa6df8dd MS |
1144 | /* |
1145 | * NOTE: for the blk-mq queue stacked on request_fn queue(s) case: | |
1146 | * no need to call free_clone_request() because we leverage blk-mq by | |
1147 | * allocating the clone at the end of the blk-mq pdu (see: clone_rq) | |
1148 | */ | |
bfebd1cd MS |
1149 | |
1150 | if (!md->queue->mq_ops) | |
1151 | free_rq_tio(tio); | |
a77e28c7 KU |
1152 | } |
1153 | ||
980691e5 KU |
1154 | /* |
1155 | * Complete the clone and the original request. | |
466d89a6 KB |
1156 | * Must be called without clone's queue lock held, |
1157 | * see end_clone_request() for more details. | |
980691e5 KU |
1158 | */ |
1159 | static void dm_end_request(struct request *clone, int error) | |
1160 | { | |
1161 | int rw = rq_data_dir(clone); | |
1162 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1163 | struct mapped_device *md = tio->md; | |
1164 | struct request *rq = tio->orig; | |
1165 | ||
29e4013d | 1166 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1167 | rq->errors = clone->errors; |
1168 | rq->resid_len = clone->resid_len; | |
1169 | ||
1170 | if (rq->sense) | |
1171 | /* | |
1172 | * We are using the sense buffer of the original | |
1173 | * request. | |
1174 | * So setting the length of the sense data is enough. | |
1175 | */ | |
1176 | rq->sense_len = clone->sense_len; | |
1177 | } | |
1178 | ||
e5d8de32 | 1179 | free_rq_clone(clone); |
e262f347 | 1180 | rq_end_stats(md, rq); |
bfebd1cd MS |
1181 | if (!rq->q->mq_ops) |
1182 | blk_end_request_all(rq, error); | |
1183 | else | |
1184 | blk_mq_end_request(rq, error); | |
29e4013d | 1185 | rq_completed(md, rw, true); |
980691e5 KU |
1186 | } |
1187 | ||
cec47e3d KU |
1188 | static void dm_unprep_request(struct request *rq) |
1189 | { | |
bfebd1cd | 1190 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1191 | struct request *clone = tio->clone; |
cec47e3d | 1192 | |
bfebd1cd MS |
1193 | if (!rq->q->mq_ops) { |
1194 | rq->special = NULL; | |
1195 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1196 | } | |
cec47e3d | 1197 | |
e5863d9a | 1198 | if (clone) |
e5d8de32 | 1199 | free_rq_clone(clone); |
4328daa2 MS |
1200 | else if (!tio->md->queue->mq_ops) |
1201 | free_rq_tio(tio); | |
cec47e3d KU |
1202 | } |
1203 | ||
1204 | /* | |
1205 | * Requeue the original request of a clone. | |
1206 | */ | |
bfebd1cd | 1207 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1208 | { |
cec47e3d KU |
1209 | struct request_queue *q = rq->q; |
1210 | unsigned long flags; | |
1211 | ||
cec47e3d | 1212 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1213 | blk_requeue_request(q, rq); |
4ae9944d | 1214 | blk_run_queue_async(q); |
cec47e3d | 1215 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1216 | } |
1217 | ||
2d76fff1 MS |
1218 | static void dm_requeue_original_request(struct mapped_device *md, |
1219 | struct request *rq) | |
bfebd1cd MS |
1220 | { |
1221 | int rw = rq_data_dir(rq); | |
1222 | ||
1223 | dm_unprep_request(rq); | |
1224 | ||
e262f347 | 1225 | rq_end_stats(md, rq); |
bfebd1cd MS |
1226 | if (!rq->q->mq_ops) |
1227 | old_requeue_request(rq); | |
1228 | else { | |
1229 | blk_mq_requeue_request(rq); | |
1230 | blk_mq_kick_requeue_list(rq->q); | |
1231 | } | |
cec47e3d | 1232 | |
466d89a6 KB |
1233 | rq_completed(md, rw, false); |
1234 | } | |
1235 | ||
bfebd1cd | 1236 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1237 | { |
1238 | unsigned long flags; | |
1239 | ||
bfebd1cd MS |
1240 | if (blk_queue_stopped(q)) |
1241 | return; | |
1242 | ||
cec47e3d | 1243 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1244 | blk_stop_queue(q); |
cec47e3d KU |
1245 | spin_unlock_irqrestore(q->queue_lock, flags); |
1246 | } | |
1247 | ||
bfebd1cd | 1248 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1249 | { |
bfebd1cd MS |
1250 | if (!q->mq_ops) |
1251 | old_stop_queue(q); | |
1252 | else | |
1253 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1254 | } |
1255 | ||
bfebd1cd | 1256 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1257 | { |
1258 | unsigned long flags; | |
1259 | ||
1260 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1261 | if (blk_queue_stopped(q)) |
1262 | blk_start_queue(q); | |
cec47e3d KU |
1263 | spin_unlock_irqrestore(q->queue_lock, flags); |
1264 | } | |
1265 | ||
bfebd1cd MS |
1266 | static void start_queue(struct request_queue *q) |
1267 | { | |
1268 | if (!q->mq_ops) | |
1269 | old_start_queue(q); | |
1270 | else | |
1271 | blk_mq_start_stopped_hw_queues(q, true); | |
1272 | } | |
1273 | ||
11a68244 | 1274 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1275 | { |
11a68244 | 1276 | int r = error; |
cec47e3d | 1277 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1278 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1279 | |
ba1cbad9 MS |
1280 | if (tio->ti) { |
1281 | rq_end_io = tio->ti->type->rq_end_io; | |
1282 | ||
1283 | if (mapped && rq_end_io) | |
1284 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1285 | } | |
cec47e3d | 1286 | |
7eee4ae2 MS |
1287 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1288 | !clone->q->limits.max_write_same_sectors)) | |
1289 | disable_write_same(tio->md); | |
1290 | ||
11a68244 | 1291 | if (r <= 0) |
cec47e3d | 1292 | /* The target wants to complete the I/O */ |
11a68244 KU |
1293 | dm_end_request(clone, r); |
1294 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1295 | /* The target will handle the I/O */ |
1296 | return; | |
11a68244 | 1297 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1298 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1299 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1300 | else { |
11a68244 | 1301 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1302 | BUG(); |
1303 | } | |
1304 | } | |
1305 | ||
11a68244 KU |
1306 | /* |
1307 | * Request completion handler for request-based dm | |
1308 | */ | |
1309 | static void dm_softirq_done(struct request *rq) | |
1310 | { | |
1311 | bool mapped = true; | |
bfebd1cd | 1312 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1313 | struct request *clone = tio->clone; |
bfebd1cd | 1314 | int rw; |
11a68244 | 1315 | |
e5863d9a | 1316 | if (!clone) { |
e262f347 | 1317 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1318 | rw = rq_data_dir(rq); |
1319 | if (!rq->q->mq_ops) { | |
1320 | blk_end_request_all(rq, tio->error); | |
1321 | rq_completed(tio->md, rw, false); | |
1322 | free_rq_tio(tio); | |
1323 | } else { | |
1324 | blk_mq_end_request(rq, tio->error); | |
1325 | rq_completed(tio->md, rw, false); | |
1326 | } | |
e5863d9a MS |
1327 | return; |
1328 | } | |
11a68244 KU |
1329 | |
1330 | if (rq->cmd_flags & REQ_FAILED) | |
1331 | mapped = false; | |
1332 | ||
1333 | dm_done(clone, tio->error, mapped); | |
1334 | } | |
1335 | ||
cec47e3d KU |
1336 | /* |
1337 | * Complete the clone and the original request with the error status | |
1338 | * through softirq context. | |
1339 | */ | |
466d89a6 | 1340 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1341 | { |
bfebd1cd | 1342 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1343 | |
1344 | tio->error = error; | |
6acfe68b MS |
1345 | if (!rq->q->mq_ops) |
1346 | blk_complete_request(rq); | |
1347 | else | |
1348 | blk_mq_complete_request(rq, error); | |
cec47e3d KU |
1349 | } |
1350 | ||
1351 | /* | |
1352 | * Complete the not-mapped clone and the original request with the error status | |
1353 | * through softirq context. | |
1354 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1355 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1356 | */ |
466d89a6 | 1357 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1358 | { |
cec47e3d | 1359 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1360 | dm_complete_request(rq, error); |
cec47e3d | 1361 | } |
cec47e3d KU |
1362 | |
1363 | /* | |
bfebd1cd | 1364 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1365 | */ |
1366 | static void end_clone_request(struct request *clone, int error) | |
1367 | { | |
466d89a6 KB |
1368 | struct dm_rq_target_io *tio = clone->end_io_data; |
1369 | ||
e5863d9a MS |
1370 | if (!clone->q->mq_ops) { |
1371 | /* | |
1372 | * For just cleaning up the information of the queue in which | |
1373 | * the clone was dispatched. | |
1374 | * The clone is *NOT* freed actually here because it is alloced | |
1375 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1376 | */ | |
1377 | __blk_put_request(clone->q, clone); | |
1378 | } | |
cec47e3d KU |
1379 | |
1380 | /* | |
1381 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1382 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1383 | * - another request may be submitted by the upper level driver |
1384 | * of the stacking during the completion | |
1385 | * - the submission which requires queue lock may be done | |
466d89a6 | 1386 | * against this clone's queue |
cec47e3d | 1387 | */ |
466d89a6 | 1388 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1389 | } |
1390 | ||
56a67df7 MS |
1391 | /* |
1392 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1393 | * target boundary. | |
1394 | */ | |
1395 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1396 | { | |
1397 | sector_t target_offset = dm_target_offset(ti, sector); | |
1398 | ||
1399 | return ti->len - target_offset; | |
1400 | } | |
1401 | ||
1402 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1403 | { |
56a67df7 | 1404 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1405 | sector_t offset, max_len; |
1da177e4 LT |
1406 | |
1407 | /* | |
542f9038 | 1408 | * Does the target need to split even further? |
1da177e4 | 1409 | */ |
542f9038 MS |
1410 | if (ti->max_io_len) { |
1411 | offset = dm_target_offset(ti, sector); | |
1412 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1413 | max_len = sector_div(offset, ti->max_io_len); | |
1414 | else | |
1415 | max_len = offset & (ti->max_io_len - 1); | |
1416 | max_len = ti->max_io_len - max_len; | |
1417 | ||
1418 | if (len > max_len) | |
1419 | len = max_len; | |
1da177e4 LT |
1420 | } |
1421 | ||
1422 | return len; | |
1423 | } | |
1424 | ||
542f9038 MS |
1425 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1426 | { | |
1427 | if (len > UINT_MAX) { | |
1428 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1429 | (unsigned long long)len, UINT_MAX); | |
1430 | ti->error = "Maximum size of target IO is too large"; | |
1431 | return -EINVAL; | |
1432 | } | |
1433 | ||
1434 | ti->max_io_len = (uint32_t) len; | |
1435 | ||
1436 | return 0; | |
1437 | } | |
1438 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1439 | ||
1dd40c3e MP |
1440 | /* |
1441 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1442 | * allowed for all bio types except REQ_FLUSH. | |
1443 | * | |
1444 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1445 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1446 | * sent in a next bio. | |
1447 | * | |
1448 | * A diagram that explains the arithmetics: | |
1449 | * +--------------------+---------------+-------+ | |
1450 | * | 1 | 2 | 3 | | |
1451 | * +--------------------+---------------+-------+ | |
1452 | * | |
1453 | * <-------------- *tio->len_ptr ---------------> | |
1454 | * <------- bi_size -------> | |
1455 | * <-- n_sectors --> | |
1456 | * | |
1457 | * Region 1 was already iterated over with bio_advance or similar function. | |
1458 | * (it may be empty if the target doesn't use bio_advance) | |
1459 | * Region 2 is the remaining bio size that the target wants to process. | |
1460 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1461 | * to make it empty) | |
1462 | * The target requires that region 3 is to be sent in the next bio. | |
1463 | * | |
1464 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1465 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1466 | * copies of the bio. | |
1467 | */ | |
1468 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1469 | { | |
1470 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1471 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1472 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1473 | BUG_ON(bi_size > *tio->len_ptr); | |
1474 | BUG_ON(n_sectors > bi_size); | |
1475 | *tio->len_ptr -= bi_size - n_sectors; | |
1476 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1477 | } | |
1478 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1479 | ||
bd2a49b8 | 1480 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1481 | { |
1482 | int r; | |
2056a782 | 1483 | sector_t sector; |
9faf400f | 1484 | struct mapped_device *md; |
dba14160 | 1485 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1486 | struct dm_target *ti = tio->ti; |
1da177e4 | 1487 | |
1da177e4 | 1488 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1489 | |
1490 | /* | |
1491 | * Map the clone. If r == 0 we don't need to do | |
1492 | * anything, the target has assumed ownership of | |
1493 | * this io. | |
1494 | */ | |
1495 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1496 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1497 | r = ti->type->map(ti, clone); |
45cbcd79 | 1498 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1499 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1500 | |
d07335e5 MS |
1501 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1502 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1503 | |
1da177e4 | 1504 | generic_make_request(clone); |
2e93ccc1 KU |
1505 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1506 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1507 | md = tio->io->md; |
1508 | dec_pending(tio->io, r); | |
9faf400f | 1509 | free_tio(md, tio); |
ab37844d | 1510 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1511 | DMWARN("unimplemented target map return value: %d", r); |
1512 | BUG(); | |
1da177e4 LT |
1513 | } |
1514 | } | |
1515 | ||
1516 | struct clone_info { | |
1517 | struct mapped_device *md; | |
1518 | struct dm_table *map; | |
1519 | struct bio *bio; | |
1520 | struct dm_io *io; | |
1521 | sector_t sector; | |
e0d6609a | 1522 | unsigned sector_count; |
1da177e4 LT |
1523 | }; |
1524 | ||
e0d6609a | 1525 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1526 | { |
4f024f37 KO |
1527 | bio->bi_iter.bi_sector = sector; |
1528 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1529 | } |
1530 | ||
1531 | /* | |
1532 | * Creates a bio that consists of range of complete bvecs. | |
1533 | */ | |
dba14160 | 1534 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1535 | sector_t sector, unsigned len) |
1da177e4 | 1536 | { |
dba14160 | 1537 | struct bio *clone = &tio->clone; |
1da177e4 | 1538 | |
1c3b13e6 KO |
1539 | __bio_clone_fast(clone, bio); |
1540 | ||
1541 | if (bio_integrity(bio)) | |
1542 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1543 | |
1c3b13e6 KO |
1544 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1545 | clone->bi_iter.bi_size = to_bytes(len); | |
1546 | ||
1547 | if (bio_integrity(bio)) | |
1548 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1549 | } |
1550 | ||
9015df24 | 1551 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1552 | struct dm_target *ti, |
55a62eef | 1553 | unsigned target_bio_nr) |
f9ab94ce | 1554 | { |
dba14160 MP |
1555 | struct dm_target_io *tio; |
1556 | struct bio *clone; | |
1557 | ||
99778273 | 1558 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1559 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1560 | |
1561 | tio->io = ci->io; | |
1562 | tio->ti = ti; | |
55a62eef | 1563 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1564 | |
1565 | return tio; | |
1566 | } | |
1567 | ||
14fe594d AK |
1568 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1569 | struct dm_target *ti, | |
1dd40c3e | 1570 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1571 | { |
99778273 | 1572 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1573 | struct bio *clone = &tio->clone; |
9015df24 | 1574 | |
1dd40c3e MP |
1575 | tio->len_ptr = len; |
1576 | ||
99778273 | 1577 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1578 | if (len) |
1dd40c3e | 1579 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1580 | |
bd2a49b8 | 1581 | __map_bio(tio); |
f9ab94ce MP |
1582 | } |
1583 | ||
14fe594d | 1584 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1585 | unsigned num_bios, unsigned *len) |
06a426ce | 1586 | { |
55a62eef | 1587 | unsigned target_bio_nr; |
06a426ce | 1588 | |
55a62eef | 1589 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1590 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1591 | } |
1592 | ||
14fe594d | 1593 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1594 | { |
06a426ce | 1595 | unsigned target_nr = 0; |
f9ab94ce MP |
1596 | struct dm_target *ti; |
1597 | ||
b372d360 | 1598 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1599 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1600 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1601 | |
f9ab94ce MP |
1602 | return 0; |
1603 | } | |
1604 | ||
e4c93811 | 1605 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1606 | sector_t sector, unsigned *len) |
5ae89a87 | 1607 | { |
dba14160 | 1608 | struct bio *bio = ci->bio; |
5ae89a87 | 1609 | struct dm_target_io *tio; |
b0d8ed4d AK |
1610 | unsigned target_bio_nr; |
1611 | unsigned num_target_bios = 1; | |
5ae89a87 | 1612 | |
b0d8ed4d AK |
1613 | /* |
1614 | * Does the target want to receive duplicate copies of the bio? | |
1615 | */ | |
1616 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1617 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1618 | |
b0d8ed4d | 1619 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1620 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1621 | tio->len_ptr = len; |
1622 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1623 | __map_bio(tio); |
1624 | } | |
5ae89a87 MS |
1625 | } |
1626 | ||
55a62eef | 1627 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1628 | |
55a62eef | 1629 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1630 | { |
55a62eef | 1631 | return ti->num_discard_bios; |
23508a96 MS |
1632 | } |
1633 | ||
55a62eef | 1634 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1635 | { |
55a62eef | 1636 | return ti->num_write_same_bios; |
23508a96 MS |
1637 | } |
1638 | ||
1639 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1640 | ||
1641 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1642 | { | |
55a62eef | 1643 | return ti->split_discard_bios; |
23508a96 MS |
1644 | } |
1645 | ||
14fe594d AK |
1646 | static int __send_changing_extent_only(struct clone_info *ci, |
1647 | get_num_bios_fn get_num_bios, | |
1648 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1649 | { |
1650 | struct dm_target *ti; | |
e0d6609a | 1651 | unsigned len; |
55a62eef | 1652 | unsigned num_bios; |
5ae89a87 | 1653 | |
a79245b3 MS |
1654 | do { |
1655 | ti = dm_table_find_target(ci->map, ci->sector); | |
1656 | if (!dm_target_is_valid(ti)) | |
1657 | return -EIO; | |
5ae89a87 | 1658 | |
5ae89a87 | 1659 | /* |
23508a96 MS |
1660 | * Even though the device advertised support for this type of |
1661 | * request, that does not mean every target supports it, and | |
936688d7 | 1662 | * reconfiguration might also have changed that since the |
a79245b3 | 1663 | * check was performed. |
5ae89a87 | 1664 | */ |
55a62eef AK |
1665 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1666 | if (!num_bios) | |
a79245b3 | 1667 | return -EOPNOTSUPP; |
5ae89a87 | 1668 | |
23508a96 | 1669 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1670 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1671 | else |
e0d6609a | 1672 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1673 | |
1dd40c3e | 1674 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1675 | |
1676 | ci->sector += len; | |
1677 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1678 | |
1679 | return 0; | |
1680 | } | |
1681 | ||
14fe594d | 1682 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1683 | { |
14fe594d AK |
1684 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1685 | is_split_required_for_discard); | |
23508a96 MS |
1686 | } |
1687 | ||
14fe594d | 1688 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1689 | { |
14fe594d | 1690 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1691 | } |
1692 | ||
e4c93811 AK |
1693 | /* |
1694 | * Select the correct strategy for processing a non-flush bio. | |
1695 | */ | |
14fe594d | 1696 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1697 | { |
dba14160 | 1698 | struct bio *bio = ci->bio; |
512875bd | 1699 | struct dm_target *ti; |
1c3b13e6 | 1700 | unsigned len; |
1da177e4 | 1701 | |
5ae89a87 | 1702 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1703 | return __send_discard(ci); |
23508a96 | 1704 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1705 | return __send_write_same(ci); |
5ae89a87 | 1706 | |
512875bd JN |
1707 | ti = dm_table_find_target(ci->map, ci->sector); |
1708 | if (!dm_target_is_valid(ti)) | |
1709 | return -EIO; | |
1710 | ||
1c3b13e6 | 1711 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1712 | |
1dd40c3e | 1713 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1714 | |
1c3b13e6 KO |
1715 | ci->sector += len; |
1716 | ci->sector_count -= len; | |
1da177e4 | 1717 | |
1c3b13e6 | 1718 | return 0; |
1da177e4 LT |
1719 | } |
1720 | ||
1721 | /* | |
14fe594d | 1722 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1723 | */ |
83d5e5b0 MP |
1724 | static void __split_and_process_bio(struct mapped_device *md, |
1725 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1726 | { |
1727 | struct clone_info ci; | |
512875bd | 1728 | int error = 0; |
1da177e4 | 1729 | |
83d5e5b0 | 1730 | if (unlikely(!map)) { |
6a8736d1 | 1731 | bio_io_error(bio); |
f0b9a450 MP |
1732 | return; |
1733 | } | |
692d0eb9 | 1734 | |
83d5e5b0 | 1735 | ci.map = map; |
1da177e4 | 1736 | ci.md = md; |
1da177e4 LT |
1737 | ci.io = alloc_io(md); |
1738 | ci.io->error = 0; | |
1739 | atomic_set(&ci.io->io_count, 1); | |
1740 | ci.io->bio = bio; | |
1741 | ci.io->md = md; | |
f88fb981 | 1742 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1743 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1744 | |
3eaf840e | 1745 | start_io_acct(ci.io); |
bd2a49b8 | 1746 | |
b372d360 MS |
1747 | if (bio->bi_rw & REQ_FLUSH) { |
1748 | ci.bio = &ci.md->flush_bio; | |
1749 | ci.sector_count = 0; | |
14fe594d | 1750 | error = __send_empty_flush(&ci); |
b372d360 MS |
1751 | /* dec_pending submits any data associated with flush */ |
1752 | } else { | |
6a8736d1 | 1753 | ci.bio = bio; |
d87f4c14 | 1754 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1755 | while (ci.sector_count && !error) |
14fe594d | 1756 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1757 | } |
1da177e4 LT |
1758 | |
1759 | /* drop the extra reference count */ | |
512875bd | 1760 | dec_pending(ci.io, error); |
1da177e4 LT |
1761 | } |
1762 | /*----------------------------------------------------------------- | |
1763 | * CRUD END | |
1764 | *---------------------------------------------------------------*/ | |
1765 | ||
1766 | /* | |
1767 | * The request function that just remaps the bio built up by | |
1768 | * dm_merge_bvec. | |
1769 | */ | |
dece1635 | 1770 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1771 | { |
12f03a49 | 1772 | int rw = bio_data_dir(bio); |
1da177e4 | 1773 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1774 | int srcu_idx; |
1775 | struct dm_table *map; | |
1da177e4 | 1776 | |
83d5e5b0 | 1777 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1778 | |
18c0b223 | 1779 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1780 | |
6a8736d1 TH |
1781 | /* if we're suspended, we have to queue this io for later */ |
1782 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1783 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1784 | |
6a8736d1 TH |
1785 | if (bio_rw(bio) != READA) |
1786 | queue_io(md, bio); | |
1787 | else | |
54d9a1b4 | 1788 | bio_io_error(bio); |
dece1635 | 1789 | return BLK_QC_T_NONE; |
1da177e4 LT |
1790 | } |
1791 | ||
83d5e5b0 MP |
1792 | __split_and_process_bio(md, map, bio); |
1793 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1794 | return BLK_QC_T_NONE; |
cec47e3d KU |
1795 | } |
1796 | ||
fd2ed4d2 | 1797 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1798 | { |
1799 | return blk_queue_stackable(md->queue); | |
1800 | } | |
1801 | ||
466d89a6 | 1802 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1803 | { |
1804 | int r; | |
1805 | ||
466d89a6 KB |
1806 | if (blk_queue_io_stat(clone->q)) |
1807 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1808 | |
466d89a6 KB |
1809 | clone->start_time = jiffies; |
1810 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1811 | if (r) |
466d89a6 | 1812 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1813 | dm_complete_request(rq, r); |
1814 | } | |
cec47e3d | 1815 | |
78d8e58a MS |
1816 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1817 | void *data) | |
cec47e3d | 1818 | { |
78d8e58a MS |
1819 | struct dm_rq_target_io *tio = data; |
1820 | struct dm_rq_clone_bio_info *info = | |
1821 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
1822 | ||
1823 | info->orig = bio_orig; | |
1824 | info->tio = tio; | |
1825 | bio->bi_end_io = end_clone_bio; | |
1826 | ||
1827 | return 0; | |
1828 | } | |
1829 | ||
1830 | static int setup_clone(struct request *clone, struct request *rq, | |
1831 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
cec47e3d | 1832 | { |
78d8e58a MS |
1833 | int r; |
1834 | ||
1835 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
1836 | dm_rq_bio_constructor, tio); | |
1837 | if (r) | |
1838 | return r; | |
1839 | ||
1840 | clone->cmd = rq->cmd; | |
1841 | clone->cmd_len = rq->cmd_len; | |
1842 | clone->sense = rq->sense; | |
cec47e3d KU |
1843 | clone->end_io = end_clone_request; |
1844 | clone->end_io_data = tio; | |
78d8e58a | 1845 | |
1ae49ea2 | 1846 | tio->clone = clone; |
78d8e58a MS |
1847 | |
1848 | return 0; | |
cec47e3d KU |
1849 | } |
1850 | ||
6facdaff | 1851 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1852 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1853 | { |
02233342 MS |
1854 | /* |
1855 | * Do not allocate a clone if tio->clone was already set | |
1856 | * (see: dm_mq_queue_rq). | |
1857 | */ | |
1858 | bool alloc_clone = !tio->clone; | |
1859 | struct request *clone; | |
1ae49ea2 | 1860 | |
02233342 MS |
1861 | if (alloc_clone) { |
1862 | clone = alloc_clone_request(md, gfp_mask); | |
1863 | if (!clone) | |
1864 | return NULL; | |
1865 | } else | |
1866 | clone = tio->clone; | |
1ae49ea2 MS |
1867 | |
1868 | blk_rq_init(NULL, clone); | |
78d8e58a MS |
1869 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
1870 | /* -ENOMEM */ | |
1871 | if (alloc_clone) | |
1872 | free_clone_request(md, clone); | |
1873 | return NULL; | |
1874 | } | |
1ae49ea2 MS |
1875 | |
1876 | return clone; | |
1877 | } | |
1878 | ||
2eb6e1e3 KB |
1879 | static void map_tio_request(struct kthread_work *work); |
1880 | ||
bfebd1cd MS |
1881 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1882 | struct mapped_device *md) | |
1883 | { | |
1884 | tio->md = md; | |
1885 | tio->ti = NULL; | |
1886 | tio->clone = NULL; | |
1887 | tio->orig = rq; | |
1888 | tio->error = 0; | |
1889 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1890 | if (md->kworker_task) |
1891 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1892 | } |
1893 | ||
466d89a6 KB |
1894 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1895 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1896 | { |
6facdaff | 1897 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1898 | int srcu_idx; |
1899 | struct dm_table *table; | |
6facdaff KU |
1900 | |
1901 | tio = alloc_rq_tio(md, gfp_mask); | |
1902 | if (!tio) | |
1903 | return NULL; | |
1904 | ||
bfebd1cd | 1905 | init_tio(tio, rq, md); |
6facdaff | 1906 | |
e5863d9a MS |
1907 | table = dm_get_live_table(md, &srcu_idx); |
1908 | if (!dm_table_mq_request_based(table)) { | |
1909 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1910 | dm_put_live_table(md, srcu_idx); | |
1911 | free_rq_tio(tio); | |
1912 | return NULL; | |
1913 | } | |
6facdaff | 1914 | } |
e5863d9a | 1915 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1916 | |
466d89a6 | 1917 | return tio; |
6facdaff KU |
1918 | } |
1919 | ||
cec47e3d KU |
1920 | /* |
1921 | * Called with the queue lock held. | |
1922 | */ | |
1923 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1924 | { | |
1925 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1926 | struct dm_rq_target_io *tio; |
cec47e3d | 1927 | |
cec47e3d KU |
1928 | if (unlikely(rq->special)) { |
1929 | DMWARN("Already has something in rq->special."); | |
1930 | return BLKPREP_KILL; | |
1931 | } | |
1932 | ||
466d89a6 KB |
1933 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1934 | if (!tio) | |
cec47e3d | 1935 | return BLKPREP_DEFER; |
cec47e3d | 1936 | |
466d89a6 | 1937 | rq->special = tio; |
cec47e3d KU |
1938 | rq->cmd_flags |= REQ_DONTPREP; |
1939 | ||
1940 | return BLKPREP_OK; | |
1941 | } | |
1942 | ||
9eef87da KU |
1943 | /* |
1944 | * Returns: | |
e5863d9a MS |
1945 | * 0 : the request has been processed |
1946 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1947 | * < 0 : the request was completed due to failure | |
9eef87da | 1948 | */ |
bfebd1cd | 1949 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1950 | struct mapped_device *md) |
cec47e3d | 1951 | { |
e5863d9a | 1952 | int r; |
bfebd1cd | 1953 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1954 | struct request *clone = NULL; |
1955 | ||
1956 | if (tio->clone) { | |
1957 | clone = tio->clone; | |
1958 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1959 | } else { | |
1960 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1961 | if (r < 0) { | |
1962 | /* The target wants to complete the I/O */ | |
1963 | dm_kill_unmapped_request(rq, r); | |
1964 | return r; | |
1965 | } | |
3a140755 JN |
1966 | if (r != DM_MAPIO_REMAPPED) |
1967 | return r; | |
78d8e58a MS |
1968 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
1969 | /* -ENOMEM */ | |
1970 | ti->type->release_clone_rq(clone); | |
1971 | return DM_MAPIO_REQUEUE; | |
1972 | } | |
e5863d9a | 1973 | } |
cec47e3d | 1974 | |
cec47e3d KU |
1975 | switch (r) { |
1976 | case DM_MAPIO_SUBMITTED: | |
1977 | /* The target has taken the I/O to submit by itself later */ | |
1978 | break; | |
1979 | case DM_MAPIO_REMAPPED: | |
1980 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1981 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1982 | blk_rq_pos(rq)); |
1983 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1984 | break; |
1985 | case DM_MAPIO_REQUEUE: | |
1986 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 1987 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
1988 | break; |
1989 | default: | |
1990 | if (r > 0) { | |
1991 | DMWARN("unimplemented target map return value: %d", r); | |
1992 | BUG(); | |
1993 | } | |
1994 | ||
1995 | /* The target wants to complete the I/O */ | |
466d89a6 | 1996 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1997 | return r; |
cec47e3d | 1998 | } |
9eef87da | 1999 | |
e5863d9a | 2000 | return 0; |
cec47e3d KU |
2001 | } |
2002 | ||
2eb6e1e3 | 2003 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 2004 | { |
2eb6e1e3 | 2005 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
2006 | struct request *rq = tio->orig; |
2007 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 2008 | |
bfebd1cd | 2009 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 2010 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
2011 | } |
2012 | ||
466d89a6 | 2013 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2014 | { |
bfebd1cd MS |
2015 | if (!orig->q->mq_ops) |
2016 | blk_start_request(orig); | |
2017 | else | |
2018 | blk_mq_start_request(orig); | |
466d89a6 | 2019 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2020 | |
0ce65797 MS |
2021 | if (md->seq_rq_merge_deadline_usecs) { |
2022 | md->last_rq_pos = rq_end_sector(orig); | |
2023 | md->last_rq_rw = rq_data_dir(orig); | |
2024 | md->last_rq_start_time = ktime_get(); | |
2025 | } | |
de3ec86d | 2026 | |
e262f347 MP |
2027 | if (unlikely(dm_stats_used(&md->stats))) { |
2028 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
2029 | tio->duration_jiffies = jiffies; | |
2030 | tio->n_sectors = blk_rq_sectors(orig); | |
2031 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
2032 | tio->n_sectors, false, 0, &tio->stats_aux); | |
2033 | } | |
2034 | ||
ba1cbad9 MS |
2035 | /* |
2036 | * Hold the md reference here for the in-flight I/O. | |
2037 | * We can't rely on the reference count by device opener, | |
2038 | * because the device may be closed during the request completion | |
2039 | * when all bios are completed. | |
2040 | * See the comment in rq_completed() too. | |
2041 | */ | |
2042 | dm_get(md); | |
ba1cbad9 MS |
2043 | } |
2044 | ||
0ce65797 MS |
2045 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2046 | ||
2047 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2048 | { | |
2049 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2050 | } | |
2051 | ||
2052 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2053 | const char *buf, size_t count) | |
2054 | { | |
2055 | unsigned deadline; | |
2056 | ||
17e149b8 | 2057 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2058 | return count; |
2059 | ||
2060 | if (kstrtouint(buf, 10, &deadline)) | |
2061 | return -EINVAL; | |
2062 | ||
2063 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2064 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2065 | ||
2066 | md->seq_rq_merge_deadline_usecs = deadline; | |
2067 | ||
2068 | return count; | |
2069 | } | |
2070 | ||
2071 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2072 | { | |
2073 | ktime_t kt_deadline; | |
2074 | ||
2075 | if (!md->seq_rq_merge_deadline_usecs) | |
2076 | return false; | |
2077 | ||
2078 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2079 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2080 | ||
2081 | return !ktime_after(ktime_get(), kt_deadline); | |
2082 | } | |
2083 | ||
cec47e3d KU |
2084 | /* |
2085 | * q->request_fn for request-based dm. | |
2086 | * Called with the queue lock held. | |
2087 | */ | |
2088 | static void dm_request_fn(struct request_queue *q) | |
2089 | { | |
2090 | struct mapped_device *md = q->queuedata; | |
c91852ff | 2091 | struct dm_target *ti = md->immutable_target; |
466d89a6 | 2092 | struct request *rq; |
2eb6e1e3 | 2093 | struct dm_rq_target_io *tio; |
c91852ff MS |
2094 | sector_t pos = 0; |
2095 | ||
2096 | if (unlikely(!ti)) { | |
2097 | int srcu_idx; | |
2098 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2099 | ||
2100 | ti = dm_table_find_target(map, pos); | |
2101 | dm_put_live_table(md, srcu_idx); | |
2102 | } | |
cec47e3d KU |
2103 | |
2104 | /* | |
b4324fee KU |
2105 | * For suspend, check blk_queue_stopped() and increment |
2106 | * ->pending within a single queue_lock not to increment the | |
2107 | * number of in-flight I/Os after the queue is stopped in | |
2108 | * dm_suspend(). | |
cec47e3d | 2109 | */ |
7eaceacc | 2110 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2111 | rq = blk_peek_request(q); |
2112 | if (!rq) | |
c91852ff | 2113 | return; |
cec47e3d | 2114 | |
29e4013d TH |
2115 | /* always use block 0 to find the target for flushes for now */ |
2116 | pos = 0; | |
2117 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2118 | pos = blk_rq_pos(rq); | |
2119 | ||
c91852ff MS |
2120 | if ((dm_request_peeked_before_merge_deadline(md) && |
2121 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
2122 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) || | |
2123 | (ti->type->busy && ti->type->busy(ti))) { | |
2124 | blk_delay_queue(q, HZ / 100); | |
2125 | return; | |
ba1cbad9 | 2126 | } |
d0bcb878 | 2127 | |
466d89a6 | 2128 | dm_start_request(md, rq); |
9eef87da | 2129 | |
bfebd1cd | 2130 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2131 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2132 | tio->ti = ti; | |
2133 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2134 | BUG_ON(!irqs_disabled()); |
cec47e3d | 2135 | } |
cec47e3d KU |
2136 | } |
2137 | ||
1da177e4 LT |
2138 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2139 | { | |
8a57dfc6 CS |
2140 | int r = bdi_bits; |
2141 | struct mapped_device *md = congested_data; | |
2142 | struct dm_table *map; | |
1da177e4 | 2143 | |
1eb787ec | 2144 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 2145 | if (dm_request_based(md)) { |
cec47e3d | 2146 | /* |
e522c039 MS |
2147 | * With request-based DM we only need to check the |
2148 | * top-level queue for congestion. | |
cec47e3d | 2149 | */ |
e522c039 MS |
2150 | r = md->queue->backing_dev_info.wb.state & bdi_bits; |
2151 | } else { | |
2152 | map = dm_get_live_table_fast(md); | |
2153 | if (map) | |
cec47e3d | 2154 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 2155 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2156 | } |
2157 | } | |
2158 | ||
1da177e4 LT |
2159 | return r; |
2160 | } | |
2161 | ||
2162 | /*----------------------------------------------------------------- | |
2163 | * An IDR is used to keep track of allocated minor numbers. | |
2164 | *---------------------------------------------------------------*/ | |
2b06cfff | 2165 | static void free_minor(int minor) |
1da177e4 | 2166 | { |
f32c10b0 | 2167 | spin_lock(&_minor_lock); |
1da177e4 | 2168 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2169 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2170 | } |
2171 | ||
2172 | /* | |
2173 | * See if the device with a specific minor # is free. | |
2174 | */ | |
cf13ab8e | 2175 | static int specific_minor(int minor) |
1da177e4 | 2176 | { |
c9d76be6 | 2177 | int r; |
1da177e4 LT |
2178 | |
2179 | if (minor >= (1 << MINORBITS)) | |
2180 | return -EINVAL; | |
2181 | ||
c9d76be6 | 2182 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2183 | spin_lock(&_minor_lock); |
1da177e4 | 2184 | |
c9d76be6 | 2185 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2186 | |
f32c10b0 | 2187 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2188 | idr_preload_end(); |
2189 | if (r < 0) | |
2190 | return r == -ENOSPC ? -EBUSY : r; | |
2191 | return 0; | |
1da177e4 LT |
2192 | } |
2193 | ||
cf13ab8e | 2194 | static int next_free_minor(int *minor) |
1da177e4 | 2195 | { |
c9d76be6 | 2196 | int r; |
62f75c2f | 2197 | |
c9d76be6 | 2198 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2199 | spin_lock(&_minor_lock); |
1da177e4 | 2200 | |
c9d76be6 | 2201 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2202 | |
f32c10b0 | 2203 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2204 | idr_preload_end(); |
2205 | if (r < 0) | |
2206 | return r; | |
2207 | *minor = r; | |
2208 | return 0; | |
1da177e4 LT |
2209 | } |
2210 | ||
83d5cde4 | 2211 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2212 | |
53d5914f MP |
2213 | static void dm_wq_work(struct work_struct *work); |
2214 | ||
4a0b4ddf MS |
2215 | static void dm_init_md_queue(struct mapped_device *md) |
2216 | { | |
2217 | /* | |
2218 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2219 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2220 | * The type is decided at the first table loading time. |
2221 | * To prevent problematic device stacking, clear the queue flag | |
2222 | * for request stacking support until then. | |
2223 | * | |
2224 | * This queue is new, so no concurrency on the queue_flags. | |
2225 | */ | |
2226 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
2227 | |
2228 | /* | |
2229 | * Initialize data that will only be used by a non-blk-mq DM queue | |
2230 | * - must do so here (in alloc_dev callchain) before queue is used | |
2231 | */ | |
2232 | md->queue->queuedata = md; | |
2233 | md->queue->backing_dev_info.congested_data = md; | |
bfebd1cd | 2234 | } |
4a0b4ddf | 2235 | |
bfebd1cd MS |
2236 | static void dm_init_old_md_queue(struct mapped_device *md) |
2237 | { | |
17e149b8 | 2238 | md->use_blk_mq = false; |
bfebd1cd MS |
2239 | dm_init_md_queue(md); |
2240 | ||
2241 | /* | |
2242 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2243 | */ | |
4a0b4ddf | 2244 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
4a0b4ddf | 2245 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2246 | } |
2247 | ||
0f20972f MS |
2248 | static void cleanup_mapped_device(struct mapped_device *md) |
2249 | { | |
0f20972f MS |
2250 | if (md->wq) |
2251 | destroy_workqueue(md->wq); | |
2252 | if (md->kworker_task) | |
2253 | kthread_stop(md->kworker_task); | |
6f65985e JL |
2254 | mempool_destroy(md->io_pool); |
2255 | mempool_destroy(md->rq_pool); | |
0f20972f MS |
2256 | if (md->bs) |
2257 | bioset_free(md->bs); | |
2258 | ||
b06075a9 MP |
2259 | cleanup_srcu_struct(&md->io_barrier); |
2260 | ||
0f20972f MS |
2261 | if (md->disk) { |
2262 | spin_lock(&_minor_lock); | |
2263 | md->disk->private_data = NULL; | |
2264 | spin_unlock(&_minor_lock); | |
0f20972f MS |
2265 | del_gendisk(md->disk); |
2266 | put_disk(md->disk); | |
2267 | } | |
2268 | ||
2269 | if (md->queue) | |
2270 | blk_cleanup_queue(md->queue); | |
2271 | ||
2272 | if (md->bdev) { | |
2273 | bdput(md->bdev); | |
2274 | md->bdev = NULL; | |
2275 | } | |
2276 | } | |
2277 | ||
1da177e4 LT |
2278 | /* |
2279 | * Allocate and initialise a blank device with a given minor. | |
2280 | */ | |
2b06cfff | 2281 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2282 | { |
2283 | int r; | |
cf13ab8e | 2284 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2285 | void *old_md; |
1da177e4 LT |
2286 | |
2287 | if (!md) { | |
2288 | DMWARN("unable to allocate device, out of memory."); | |
2289 | return NULL; | |
2290 | } | |
2291 | ||
10da4f79 | 2292 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2293 | goto bad_module_get; |
10da4f79 | 2294 | |
1da177e4 | 2295 | /* get a minor number for the dev */ |
2b06cfff | 2296 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2297 | r = next_free_minor(&minor); |
2b06cfff | 2298 | else |
cf13ab8e | 2299 | r = specific_minor(minor); |
1da177e4 | 2300 | if (r < 0) |
6ed7ade8 | 2301 | goto bad_minor; |
1da177e4 | 2302 | |
83d5e5b0 MP |
2303 | r = init_srcu_struct(&md->io_barrier); |
2304 | if (r < 0) | |
2305 | goto bad_io_barrier; | |
2306 | ||
17e149b8 | 2307 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2308 | md->type = DM_TYPE_NONE; |
e61290a4 | 2309 | mutex_init(&md->suspend_lock); |
a5664dad | 2310 | mutex_init(&md->type_lock); |
86f1152b | 2311 | mutex_init(&md->table_devices_lock); |
022c2611 | 2312 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2313 | atomic_set(&md->holders, 1); |
5c6bd75d | 2314 | atomic_set(&md->open_count, 0); |
1da177e4 | 2315 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2316 | atomic_set(&md->uevent_seq, 0); |
2317 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2318 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2319 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2320 | |
4a0b4ddf | 2321 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2322 | if (!md->queue) |
0f20972f | 2323 | goto bad; |
1da177e4 | 2324 | |
4a0b4ddf | 2325 | dm_init_md_queue(md); |
9faf400f | 2326 | |
1da177e4 LT |
2327 | md->disk = alloc_disk(1); |
2328 | if (!md->disk) | |
0f20972f | 2329 | goto bad; |
1da177e4 | 2330 | |
316d315b NK |
2331 | atomic_set(&md->pending[0], 0); |
2332 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2333 | init_waitqueue_head(&md->wait); |
53d5914f | 2334 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2335 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2336 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2337 | md->kworker_task = NULL; |
f0b04115 | 2338 | |
1da177e4 LT |
2339 | md->disk->major = _major; |
2340 | md->disk->first_minor = minor; | |
2341 | md->disk->fops = &dm_blk_dops; | |
2342 | md->disk->queue = md->queue; | |
2343 | md->disk->private_data = md; | |
2344 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2345 | add_disk(md->disk); | |
7e51f257 | 2346 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2347 | |
670368a8 | 2348 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2349 | if (!md->wq) |
0f20972f | 2350 | goto bad; |
304f3f6a | 2351 | |
32a926da MP |
2352 | md->bdev = bdget_disk(md->disk, 0); |
2353 | if (!md->bdev) | |
0f20972f | 2354 | goto bad; |
32a926da | 2355 | |
6a8736d1 TH |
2356 | bio_init(&md->flush_bio); |
2357 | md->flush_bio.bi_bdev = md->bdev; | |
2358 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2359 | ||
fd2ed4d2 MP |
2360 | dm_stats_init(&md->stats); |
2361 | ||
ba61fdd1 | 2362 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2363 | spin_lock(&_minor_lock); |
ba61fdd1 | 2364 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2365 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2366 | |
2367 | BUG_ON(old_md != MINOR_ALLOCED); | |
2368 | ||
1da177e4 LT |
2369 | return md; |
2370 | ||
0f20972f MS |
2371 | bad: |
2372 | cleanup_mapped_device(md); | |
83d5e5b0 | 2373 | bad_io_barrier: |
1da177e4 | 2374 | free_minor(minor); |
6ed7ade8 | 2375 | bad_minor: |
10da4f79 | 2376 | module_put(THIS_MODULE); |
6ed7ade8 | 2377 | bad_module_get: |
1da177e4 LT |
2378 | kfree(md); |
2379 | return NULL; | |
2380 | } | |
2381 | ||
ae9da83f JN |
2382 | static void unlock_fs(struct mapped_device *md); |
2383 | ||
1da177e4 LT |
2384 | static void free_dev(struct mapped_device *md) |
2385 | { | |
f331c029 | 2386 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2387 | |
32a926da | 2388 | unlock_fs(md); |
2eb6e1e3 | 2389 | |
0f20972f | 2390 | cleanup_mapped_device(md); |
1c357a1e MS |
2391 | if (md->tag_set) { |
2392 | blk_mq_free_tag_set(md->tag_set); | |
2393 | kfree(md->tag_set); | |
2394 | } | |
63a4f065 | 2395 | |
86f1152b | 2396 | free_table_devices(&md->table_devices); |
63a4f065 | 2397 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2398 | free_minor(minor); |
2399 | ||
10da4f79 | 2400 | module_put(THIS_MODULE); |
1da177e4 LT |
2401 | kfree(md); |
2402 | } | |
2403 | ||
e6ee8c0b KU |
2404 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2405 | { | |
c0820cf5 | 2406 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2407 | |
4e6e36c3 MS |
2408 | if (md->bs) { |
2409 | /* The md already has necessary mempools. */ | |
2410 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
2411 | /* |
2412 | * Reload bioset because front_pad may have changed | |
2413 | * because a different table was loaded. | |
2414 | */ | |
2415 | bioset_free(md->bs); | |
2416 | md->bs = p->bs; | |
2417 | p->bs = NULL; | |
16245bdc | 2418 | } |
4e6e36c3 MS |
2419 | /* |
2420 | * There's no need to reload with request-based dm | |
2421 | * because the size of front_pad doesn't change. | |
2422 | * Note for future: If you are to reload bioset, | |
2423 | * prep-ed requests in the queue may refer | |
2424 | * to bio from the old bioset, so you must walk | |
2425 | * through the queue to unprep. | |
2426 | */ | |
2427 | goto out; | |
c0820cf5 | 2428 | } |
e6ee8c0b | 2429 | |
cbc4e3c1 MS |
2430 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
2431 | ||
e6ee8c0b KU |
2432 | md->io_pool = p->io_pool; |
2433 | p->io_pool = NULL; | |
1ae49ea2 MS |
2434 | md->rq_pool = p->rq_pool; |
2435 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2436 | md->bs = p->bs; |
2437 | p->bs = NULL; | |
4e6e36c3 | 2438 | |
e6ee8c0b | 2439 | out: |
02233342 | 2440 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2441 | dm_table_free_md_mempools(t); |
2442 | } | |
2443 | ||
1da177e4 LT |
2444 | /* |
2445 | * Bind a table to the device. | |
2446 | */ | |
2447 | static void event_callback(void *context) | |
2448 | { | |
7a8c3d3b MA |
2449 | unsigned long flags; |
2450 | LIST_HEAD(uevents); | |
1da177e4 LT |
2451 | struct mapped_device *md = (struct mapped_device *) context; |
2452 | ||
7a8c3d3b MA |
2453 | spin_lock_irqsave(&md->uevent_lock, flags); |
2454 | list_splice_init(&md->uevent_list, &uevents); | |
2455 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2456 | ||
ed9e1982 | 2457 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2458 | |
1da177e4 LT |
2459 | atomic_inc(&md->event_nr); |
2460 | wake_up(&md->eventq); | |
2461 | } | |
2462 | ||
c217649b MS |
2463 | /* |
2464 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2465 | */ | |
4e90188b | 2466 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2467 | { |
4e90188b | 2468 | set_capacity(md->disk, size); |
1da177e4 | 2469 | |
db8fef4f | 2470 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2471 | } |
2472 | ||
042d2a9b AK |
2473 | /* |
2474 | * Returns old map, which caller must destroy. | |
2475 | */ | |
2476 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2477 | struct queue_limits *limits) | |
1da177e4 | 2478 | { |
042d2a9b | 2479 | struct dm_table *old_map; |
165125e1 | 2480 | struct request_queue *q = md->queue; |
1da177e4 LT |
2481 | sector_t size; |
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 | */ | |
16f12266 | 2502 | if (dm_table_request_based(t)) { |
e6ee8c0b | 2503 | stop_queue(q); |
16f12266 MS |
2504 | /* |
2505 | * Leverage the fact that request-based DM targets are | |
2506 | * immutable singletons and establish md->immutable_target | |
2507 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
2508 | */ | |
2509 | md->immutable_target = dm_table_get_immutable_target(t); | |
2510 | } | |
e6ee8c0b KU |
2511 | |
2512 | __bind_mempools(md, t); | |
2513 | ||
a12f5d48 | 2514 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2515 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2516 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2517 | ||
754c5fc7 | 2518 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2519 | if (old_map) |
2520 | dm_sync_table(md); | |
1da177e4 | 2521 | |
042d2a9b | 2522 | return old_map; |
1da177e4 LT |
2523 | } |
2524 | ||
a7940155 AK |
2525 | /* |
2526 | * Returns unbound table for the caller to free. | |
2527 | */ | |
2528 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2529 | { |
a12f5d48 | 2530 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2531 | |
2532 | if (!map) | |
a7940155 | 2533 | return NULL; |
1da177e4 LT |
2534 | |
2535 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2536 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2537 | dm_sync_table(md); |
a7940155 AK |
2538 | |
2539 | return map; | |
1da177e4 LT |
2540 | } |
2541 | ||
2542 | /* | |
2543 | * Constructor for a new device. | |
2544 | */ | |
2b06cfff | 2545 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2546 | { |
2547 | struct mapped_device *md; | |
2548 | ||
2b06cfff | 2549 | md = alloc_dev(minor); |
1da177e4 LT |
2550 | if (!md) |
2551 | return -ENXIO; | |
2552 | ||
784aae73 MB |
2553 | dm_sysfs_init(md); |
2554 | ||
1da177e4 LT |
2555 | *result = md; |
2556 | return 0; | |
2557 | } | |
2558 | ||
a5664dad MS |
2559 | /* |
2560 | * Functions to manage md->type. | |
2561 | * All are required to hold md->type_lock. | |
2562 | */ | |
2563 | void dm_lock_md_type(struct mapped_device *md) | |
2564 | { | |
2565 | mutex_lock(&md->type_lock); | |
2566 | } | |
2567 | ||
2568 | void dm_unlock_md_type(struct mapped_device *md) | |
2569 | { | |
2570 | mutex_unlock(&md->type_lock); | |
2571 | } | |
2572 | ||
2573 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2574 | { | |
00c4fc3b | 2575 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2576 | md->type = type; |
2577 | } | |
2578 | ||
2579 | unsigned dm_get_md_type(struct mapped_device *md) | |
2580 | { | |
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; | |
16f12266 | 2657 | struct dm_target *ti = md->immutable_target; |
bfebd1cd | 2658 | |
16f12266 MS |
2659 | if (unlikely(!ti)) { |
2660 | int srcu_idx; | |
2661 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
bfebd1cd | 2662 | |
16f12266 | 2663 | ti = dm_table_find_target(map, 0); |
bfebd1cd | 2664 | dm_put_live_table(md, srcu_idx); |
bfebd1cd | 2665 | } |
bfebd1cd MS |
2666 | |
2667 | if (ti->type->busy && ti->type->busy(ti)) | |
2668 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2669 | ||
2670 | dm_start_request(md, rq); | |
2671 | ||
2672 | /* Init tio using md established in .init_request */ | |
2673 | init_tio(tio, rq, md); | |
2674 | ||
02233342 MS |
2675 | /* |
2676 | * Establish tio->ti before queuing work (map_tio_request) | |
2677 | * or making direct call to map_request(). | |
2678 | */ | |
bfebd1cd | 2679 | tio->ti = ti; |
02233342 | 2680 | |
16f12266 MS |
2681 | /* |
2682 | * Both the table and md type cannot change after initial table load | |
2683 | */ | |
2684 | if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { | |
02233342 MS |
2685 | /* clone request is allocated at the end of the pdu */ |
2686 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
45714fbe | 2687 | (void) clone_rq(rq, md, tio, GFP_ATOMIC); |
02233342 MS |
2688 | queue_kthread_work(&md->kworker, &tio->work); |
2689 | } else { | |
2690 | /* Direct call is fine since .queue_rq allows allocations */ | |
45714fbe MS |
2691 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { |
2692 | /* Undo dm_start_request() before requeuing */ | |
e262f347 | 2693 | rq_end_stats(md, rq); |
45714fbe MS |
2694 | rq_completed(md, rq_data_dir(rq), false); |
2695 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2696 | } | |
02233342 | 2697 | } |
bfebd1cd MS |
2698 | |
2699 | return BLK_MQ_RQ_QUEUE_OK; | |
2700 | } | |
2701 | ||
2702 | static struct blk_mq_ops dm_mq_ops = { | |
2703 | .queue_rq = dm_mq_queue_rq, | |
2704 | .map_queue = blk_mq_map_queue, | |
2705 | .complete = dm_softirq_done, | |
2706 | .init_request = dm_mq_init_request, | |
2707 | }; | |
2708 | ||
2709 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2710 | { | |
02233342 | 2711 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2712 | struct request_queue *q; |
2713 | int err; | |
2714 | ||
1c357a1e MS |
2715 | md->tag_set = kzalloc(sizeof(struct blk_mq_tag_set), GFP_KERNEL); |
2716 | if (!md->tag_set) | |
2717 | return -ENOMEM; | |
2718 | ||
2719 | md->tag_set->ops = &dm_mq_ops; | |
2720 | md->tag_set->queue_depth = dm_get_blk_mq_queue_depth(); | |
2721 | md->tag_set->numa_node = NUMA_NO_NODE; | |
2722 | md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2723 | md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues(); | |
2724 | md->tag_set->driver_data = md; | |
2725 | ||
2726 | md->tag_set->cmd_size = sizeof(struct dm_rq_target_io); | |
02233342 | 2727 | if (md_type == DM_TYPE_REQUEST_BASED) { |
1c357a1e MS |
2728 | /* put the memory for non-blk-mq clone at the end of the pdu */ |
2729 | md->tag_set->cmd_size += sizeof(struct request); | |
2730 | } | |
bfebd1cd | 2731 | |
1c357a1e | 2732 | err = blk_mq_alloc_tag_set(md->tag_set); |
bfebd1cd | 2733 | if (err) |
1c357a1e | 2734 | goto out_kfree_tag_set; |
bfebd1cd | 2735 | |
1c357a1e | 2736 | q = blk_mq_init_allocated_queue(md->tag_set, md->queue); |
bfebd1cd MS |
2737 | if (IS_ERR(q)) { |
2738 | err = PTR_ERR(q); | |
2739 | goto out_tag_set; | |
2740 | } | |
2741 | md->queue = q; | |
2742 | dm_init_md_queue(md); | |
2743 | ||
2744 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2745 | blk_mq_register_disk(md->disk); | |
2746 | ||
02233342 MS |
2747 | if (md_type == DM_TYPE_REQUEST_BASED) |
2748 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2749 | |
2750 | return 0; | |
2751 | ||
2752 | out_tag_set: | |
1c357a1e MS |
2753 | blk_mq_free_tag_set(md->tag_set); |
2754 | out_kfree_tag_set: | |
2755 | kfree(md->tag_set); | |
2756 | ||
bfebd1cd | 2757 | return err; |
4a0b4ddf MS |
2758 | } |
2759 | ||
4e6e36c3 MS |
2760 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2761 | { | |
2762 | if (type == DM_TYPE_BIO_BASED) | |
2763 | return type; | |
2764 | ||
2765 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2766 | } | |
2767 | ||
4a0b4ddf MS |
2768 | /* |
2769 | * Setup the DM device's queue based on md's type | |
2770 | */ | |
2771 | int dm_setup_md_queue(struct mapped_device *md) | |
2772 | { | |
bfebd1cd | 2773 | int r; |
17e149b8 | 2774 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2775 | |
2776 | switch (md_type) { | |
2777 | case DM_TYPE_REQUEST_BASED: | |
2778 | r = dm_init_request_based_queue(md); | |
2779 | if (r) { | |
ff36ab34 | 2780 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2781 | return r; |
ff36ab34 | 2782 | } |
bfebd1cd MS |
2783 | break; |
2784 | case DM_TYPE_MQ_REQUEST_BASED: | |
2785 | r = dm_init_request_based_blk_mq_queue(md); | |
2786 | if (r) { | |
2787 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2788 | return r; | |
2789 | } | |
2790 | break; | |
2791 | case DM_TYPE_BIO_BASED: | |
2792 | dm_init_old_md_queue(md); | |
ff36ab34 | 2793 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
2794 | /* |
2795 | * DM handles splitting bios as needed. Free the bio_split bioset | |
2796 | * since it won't be used (saves 1 process per bio-based DM device). | |
2797 | */ | |
2798 | bioset_free(md->queue->bio_split); | |
2799 | md->queue->bio_split = NULL; | |
bfebd1cd | 2800 | break; |
4a0b4ddf MS |
2801 | } |
2802 | ||
2803 | return 0; | |
2804 | } | |
2805 | ||
2bec1f4a | 2806 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2807 | { |
2808 | struct mapped_device *md; | |
1da177e4 LT |
2809 | unsigned minor = MINOR(dev); |
2810 | ||
2811 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2812 | return NULL; | |
2813 | ||
f32c10b0 | 2814 | spin_lock(&_minor_lock); |
1da177e4 LT |
2815 | |
2816 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2817 | if (md) { |
2818 | if ((md == MINOR_ALLOCED || | |
2819 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2820 | dm_deleting_md(md) || | |
2821 | test_bit(DMF_FREEING, &md->flags))) { | |
2822 | md = NULL; | |
2823 | goto out; | |
2824 | } | |
2825 | dm_get(md); | |
fba9f90e | 2826 | } |
1da177e4 | 2827 | |
fba9f90e | 2828 | out: |
f32c10b0 | 2829 | spin_unlock(&_minor_lock); |
1da177e4 | 2830 | |
637842cf DT |
2831 | return md; |
2832 | } | |
3cf2e4ba | 2833 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2834 | |
9ade92a9 | 2835 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2836 | { |
9ade92a9 | 2837 | return md->interface_ptr; |
1da177e4 LT |
2838 | } |
2839 | ||
2840 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2841 | { | |
2842 | md->interface_ptr = ptr; | |
2843 | } | |
2844 | ||
2845 | void dm_get(struct mapped_device *md) | |
2846 | { | |
2847 | atomic_inc(&md->holders); | |
3f77316d | 2848 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2849 | } |
2850 | ||
09ee96b2 MP |
2851 | int dm_hold(struct mapped_device *md) |
2852 | { | |
2853 | spin_lock(&_minor_lock); | |
2854 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2855 | spin_unlock(&_minor_lock); | |
2856 | return -EBUSY; | |
2857 | } | |
2858 | dm_get(md); | |
2859 | spin_unlock(&_minor_lock); | |
2860 | return 0; | |
2861 | } | |
2862 | EXPORT_SYMBOL_GPL(dm_hold); | |
2863 | ||
72d94861 AK |
2864 | const char *dm_device_name(struct mapped_device *md) |
2865 | { | |
2866 | return md->name; | |
2867 | } | |
2868 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2869 | ||
3f77316d | 2870 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2871 | { |
1134e5ae | 2872 | struct dm_table *map; |
83d5e5b0 | 2873 | int srcu_idx; |
1da177e4 | 2874 | |
3f77316d | 2875 | might_sleep(); |
fba9f90e | 2876 | |
63a4f065 | 2877 | spin_lock(&_minor_lock); |
3f77316d KU |
2878 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2879 | set_bit(DMF_FREEING, &md->flags); | |
2880 | spin_unlock(&_minor_lock); | |
2881 | ||
02233342 | 2882 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2883 | flush_kthread_worker(&md->kworker); |
2884 | ||
ab7c7bb6 MP |
2885 | /* |
2886 | * Take suspend_lock so that presuspend and postsuspend methods | |
2887 | * do not race with internal suspend. | |
2888 | */ | |
2889 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2890 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2891 | if (!dm_suspended_md(md)) { |
2892 | dm_table_presuspend_targets(map); | |
2893 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2894 | } |
83d5e5b0 MP |
2895 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2896 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2897 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2898 | |
3f77316d KU |
2899 | /* |
2900 | * Rare, but there may be I/O requests still going to complete, | |
2901 | * for example. Wait for all references to disappear. | |
2902 | * No one should increment the reference count of the mapped_device, | |
2903 | * after the mapped_device state becomes DMF_FREEING. | |
2904 | */ | |
2905 | if (wait) | |
2906 | while (atomic_read(&md->holders)) | |
2907 | msleep(1); | |
2908 | else if (atomic_read(&md->holders)) | |
2909 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2910 | dm_device_name(md), atomic_read(&md->holders)); | |
2911 | ||
2912 | dm_sysfs_exit(md); | |
3f77316d KU |
2913 | dm_table_destroy(__unbind(md)); |
2914 | free_dev(md); | |
2915 | } | |
2916 | ||
2917 | void dm_destroy(struct mapped_device *md) | |
2918 | { | |
2919 | __dm_destroy(md, true); | |
2920 | } | |
2921 | ||
2922 | void dm_destroy_immediate(struct mapped_device *md) | |
2923 | { | |
2924 | __dm_destroy(md, false); | |
2925 | } | |
2926 | ||
2927 | void dm_put(struct mapped_device *md) | |
2928 | { | |
2929 | atomic_dec(&md->holders); | |
1da177e4 | 2930 | } |
79eb885c | 2931 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2932 | |
401600df | 2933 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2934 | { |
2935 | int r = 0; | |
b44ebeb0 MP |
2936 | DECLARE_WAITQUEUE(wait, current); |
2937 | ||
b44ebeb0 | 2938 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2939 | |
2940 | while (1) { | |
401600df | 2941 | set_current_state(interruptible); |
46125c1c | 2942 | |
b4324fee | 2943 | if (!md_in_flight(md)) |
46125c1c MB |
2944 | break; |
2945 | ||
401600df MP |
2946 | if (interruptible == TASK_INTERRUPTIBLE && |
2947 | signal_pending(current)) { | |
46125c1c MB |
2948 | r = -EINTR; |
2949 | break; | |
2950 | } | |
2951 | ||
2952 | io_schedule(); | |
2953 | } | |
2954 | set_current_state(TASK_RUNNING); | |
2955 | ||
b44ebeb0 MP |
2956 | remove_wait_queue(&md->wait, &wait); |
2957 | ||
46125c1c MB |
2958 | return r; |
2959 | } | |
2960 | ||
1da177e4 LT |
2961 | /* |
2962 | * Process the deferred bios | |
2963 | */ | |
ef208587 | 2964 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2965 | { |
ef208587 MP |
2966 | struct mapped_device *md = container_of(work, struct mapped_device, |
2967 | work); | |
6d6f10df | 2968 | struct bio *c; |
83d5e5b0 MP |
2969 | int srcu_idx; |
2970 | struct dm_table *map; | |
1da177e4 | 2971 | |
83d5e5b0 | 2972 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2973 | |
3b00b203 | 2974 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2975 | spin_lock_irq(&md->deferred_lock); |
2976 | c = bio_list_pop(&md->deferred); | |
2977 | spin_unlock_irq(&md->deferred_lock); | |
2978 | ||
6a8736d1 | 2979 | if (!c) |
df12ee99 | 2980 | break; |
022c2611 | 2981 | |
e6ee8c0b KU |
2982 | if (dm_request_based(md)) |
2983 | generic_make_request(c); | |
6a8736d1 | 2984 | else |
83d5e5b0 | 2985 | __split_and_process_bio(md, map, c); |
022c2611 | 2986 | } |
73d410c0 | 2987 | |
83d5e5b0 | 2988 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2989 | } |
2990 | ||
9a1fb464 | 2991 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2992 | { |
3b00b203 | 2993 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2994 | smp_mb__after_atomic(); |
53d5914f | 2995 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2996 | } |
2997 | ||
1da177e4 | 2998 | /* |
042d2a9b | 2999 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 3000 | */ |
042d2a9b | 3001 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3002 | { |
87eb5b21 | 3003 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3004 | struct queue_limits limits; |
042d2a9b | 3005 | int r; |
1da177e4 | 3006 | |
e61290a4 | 3007 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3008 | |
3009 | /* device must be suspended */ | |
4f186f8b | 3010 | if (!dm_suspended_md(md)) |
93c534ae | 3011 | goto out; |
1da177e4 | 3012 | |
3ae70656 MS |
3013 | /* |
3014 | * If the new table has no data devices, retain the existing limits. | |
3015 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3016 | * then new I/O is queued based on these limits, and then some paths | |
3017 | * reappear. | |
3018 | */ | |
3019 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3020 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3021 | if (live_map) |
3022 | limits = md->queue->limits; | |
83d5e5b0 | 3023 | dm_put_live_table_fast(md); |
3ae70656 MS |
3024 | } |
3025 | ||
87eb5b21 MC |
3026 | if (!live_map) { |
3027 | r = dm_calculate_queue_limits(table, &limits); | |
3028 | if (r) { | |
3029 | map = ERR_PTR(r); | |
3030 | goto out; | |
3031 | } | |
042d2a9b | 3032 | } |
754c5fc7 | 3033 | |
042d2a9b | 3034 | map = __bind(md, table, &limits); |
1da177e4 | 3035 | |
93c534ae | 3036 | out: |
e61290a4 | 3037 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3038 | return map; |
1da177e4 LT |
3039 | } |
3040 | ||
3041 | /* | |
3042 | * Functions to lock and unlock any filesystem running on the | |
3043 | * device. | |
3044 | */ | |
2ca3310e | 3045 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3046 | { |
e39e2e95 | 3047 | int r; |
1da177e4 LT |
3048 | |
3049 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3050 | |
db8fef4f | 3051 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3052 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3053 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3054 | md->frozen_sb = NULL; |
3055 | return r; | |
dfbe03f6 AK |
3056 | } |
3057 | ||
aa8d7c2f AK |
3058 | set_bit(DMF_FROZEN, &md->flags); |
3059 | ||
1da177e4 LT |
3060 | return 0; |
3061 | } | |
3062 | ||
2ca3310e | 3063 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3064 | { |
aa8d7c2f AK |
3065 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3066 | return; | |
3067 | ||
db8fef4f | 3068 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3069 | md->frozen_sb = NULL; |
aa8d7c2f | 3070 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3071 | } |
3072 | ||
3073 | /* | |
ffcc3936 MS |
3074 | * If __dm_suspend returns 0, the device is completely quiescent |
3075 | * now. There is no request-processing activity. All new requests | |
3076 | * are being added to md->deferred list. | |
cec47e3d | 3077 | * |
ffcc3936 | 3078 | * Caller must hold md->suspend_lock |
cec47e3d | 3079 | */ |
ffcc3936 MS |
3080 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3081 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3082 | { |
ffcc3936 MS |
3083 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3084 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3085 | int r; | |
1da177e4 | 3086 | |
2e93ccc1 KU |
3087 | /* |
3088 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3089 | * This flag is cleared before dm_suspend returns. | |
3090 | */ | |
3091 | if (noflush) | |
3092 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3093 | ||
d67ee213 MS |
3094 | /* |
3095 | * This gets reverted if there's an error later and the targets | |
3096 | * provide the .presuspend_undo hook. | |
3097 | */ | |
cf222b37 AK |
3098 | dm_table_presuspend_targets(map); |
3099 | ||
32a926da | 3100 | /* |
9f518b27 KU |
3101 | * Flush I/O to the device. |
3102 | * Any I/O submitted after lock_fs() may not be flushed. | |
3103 | * noflush takes precedence over do_lockfs. | |
3104 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3105 | */ |
3106 | if (!noflush && do_lockfs) { | |
3107 | r = lock_fs(md); | |
d67ee213 MS |
3108 | if (r) { |
3109 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3110 | return r; |
d67ee213 | 3111 | } |
aa8d7c2f | 3112 | } |
1da177e4 LT |
3113 | |
3114 | /* | |
3b00b203 MP |
3115 | * Here we must make sure that no processes are submitting requests |
3116 | * to target drivers i.e. no one may be executing | |
3117 | * __split_and_process_bio. This is called from dm_request and | |
3118 | * dm_wq_work. | |
3119 | * | |
3120 | * To get all processes out of __split_and_process_bio in dm_request, | |
3121 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3122 | * __split_and_process_bio from dm_request and quiesce the thread |
3123 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3124 | * flush_workqueue(md->wq). | |
1da177e4 | 3125 | */ |
1eb787ec | 3126 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3127 | if (map) |
3128 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3129 | |
d0bcb878 | 3130 | /* |
29e4013d TH |
3131 | * Stop md->queue before flushing md->wq in case request-based |
3132 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3133 | */ |
2eb6e1e3 | 3134 | if (dm_request_based(md)) { |
9f518b27 | 3135 | stop_queue(md->queue); |
02233342 MS |
3136 | if (md->kworker_task) |
3137 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3138 | } |
cec47e3d | 3139 | |
d0bcb878 KU |
3140 | flush_workqueue(md->wq); |
3141 | ||
1da177e4 | 3142 | /* |
3b00b203 MP |
3143 | * At this point no more requests are entering target request routines. |
3144 | * We call dm_wait_for_completion to wait for all existing requests | |
3145 | * to finish. | |
1da177e4 | 3146 | */ |
ffcc3936 | 3147 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3148 | |
6d6f10df | 3149 | if (noflush) |
022c2611 | 3150 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3151 | if (map) |
3152 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3153 | |
1da177e4 | 3154 | /* were we interrupted ? */ |
46125c1c | 3155 | if (r < 0) { |
9a1fb464 | 3156 | dm_queue_flush(md); |
73d410c0 | 3157 | |
cec47e3d | 3158 | if (dm_request_based(md)) |
9f518b27 | 3159 | start_queue(md->queue); |
cec47e3d | 3160 | |
2ca3310e | 3161 | unlock_fs(md); |
d67ee213 | 3162 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3163 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3164 | } |
1da177e4 | 3165 | |
ffcc3936 MS |
3166 | return r; |
3167 | } | |
3168 | ||
3169 | /* | |
3170 | * We need to be able to change a mapping table under a mounted | |
3171 | * filesystem. For example we might want to move some data in | |
3172 | * the background. Before the table can be swapped with | |
3173 | * dm_bind_table, dm_suspend must be called to flush any in | |
3174 | * flight bios and ensure that any further io gets deferred. | |
3175 | */ | |
3176 | /* | |
3177 | * Suspend mechanism in request-based dm. | |
3178 | * | |
3179 | * 1. Flush all I/Os by lock_fs() if needed. | |
3180 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3181 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3182 | * | |
3183 | * To abort suspend, start the request_queue. | |
3184 | */ | |
3185 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3186 | { | |
3187 | struct dm_table *map = NULL; | |
3188 | int r = 0; | |
3189 | ||
3190 | retry: | |
3191 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3192 | ||
3193 | if (dm_suspended_md(md)) { | |
3194 | r = -EINVAL; | |
3195 | goto out_unlock; | |
3196 | } | |
3197 | ||
3198 | if (dm_suspended_internally_md(md)) { | |
3199 | /* already internally suspended, wait for internal resume */ | |
3200 | mutex_unlock(&md->suspend_lock); | |
3201 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3202 | if (r) | |
3203 | return r; | |
3204 | goto retry; | |
3205 | } | |
3206 | ||
a12f5d48 | 3207 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3208 | |
3209 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3210 | if (r) | |
3211 | goto out_unlock; | |
3b00b203 | 3212 | |
2ca3310e | 3213 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3214 | |
4d4471cb KU |
3215 | dm_table_postsuspend_targets(map); |
3216 | ||
d287483d | 3217 | out_unlock: |
e61290a4 | 3218 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3219 | return r; |
1da177e4 LT |
3220 | } |
3221 | ||
ffcc3936 MS |
3222 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3223 | { | |
3224 | if (map) { | |
3225 | int r = dm_table_resume_targets(map); | |
3226 | if (r) | |
3227 | return r; | |
3228 | } | |
3229 | ||
3230 | dm_queue_flush(md); | |
3231 | ||
3232 | /* | |
3233 | * Flushing deferred I/Os must be done after targets are resumed | |
3234 | * so that mapping of targets can work correctly. | |
3235 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3236 | */ | |
3237 | if (dm_request_based(md)) | |
3238 | start_queue(md->queue); | |
3239 | ||
3240 | unlock_fs(md); | |
3241 | ||
3242 | return 0; | |
3243 | } | |
3244 | ||
1da177e4 LT |
3245 | int dm_resume(struct mapped_device *md) |
3246 | { | |
cf222b37 | 3247 | int r = -EINVAL; |
cf222b37 | 3248 | struct dm_table *map = NULL; |
1da177e4 | 3249 | |
ffcc3936 MS |
3250 | retry: |
3251 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3252 | ||
4f186f8b | 3253 | if (!dm_suspended_md(md)) |
cf222b37 | 3254 | goto out; |
cf222b37 | 3255 | |
ffcc3936 MS |
3256 | if (dm_suspended_internally_md(md)) { |
3257 | /* already internally suspended, wait for internal resume */ | |
3258 | mutex_unlock(&md->suspend_lock); | |
3259 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3260 | if (r) | |
3261 | return r; | |
3262 | goto retry; | |
3263 | } | |
3264 | ||
a12f5d48 | 3265 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3266 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3267 | goto out; |
1da177e4 | 3268 | |
ffcc3936 | 3269 | r = __dm_resume(md, map); |
8757b776 MB |
3270 | if (r) |
3271 | goto out; | |
2ca3310e | 3272 | |
2ca3310e AK |
3273 | clear_bit(DMF_SUSPENDED, &md->flags); |
3274 | ||
cf222b37 AK |
3275 | r = 0; |
3276 | out: | |
e61290a4 | 3277 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3278 | |
cf222b37 | 3279 | return r; |
1da177e4 LT |
3280 | } |
3281 | ||
fd2ed4d2 MP |
3282 | /* |
3283 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3284 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3285 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3286 | */ |
3287 | ||
ffcc3936 | 3288 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3289 | { |
ffcc3936 MS |
3290 | struct dm_table *map = NULL; |
3291 | ||
96b26c8c | 3292 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3293 | return; /* nested internal suspend */ |
3294 | ||
3295 | if (dm_suspended_md(md)) { | |
3296 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3297 | return; /* nest suspend */ | |
3298 | } | |
3299 | ||
a12f5d48 | 3300 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3301 | |
3302 | /* | |
3303 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3304 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3305 | * would require changing .presuspend to return an error -- avoid this | |
3306 | * until there is a need for more elaborate variants of internal suspend. | |
3307 | */ | |
3308 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3309 | ||
3310 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3311 | ||
3312 | dm_table_postsuspend_targets(map); | |
3313 | } | |
3314 | ||
3315 | static void __dm_internal_resume(struct mapped_device *md) | |
3316 | { | |
96b26c8c MP |
3317 | BUG_ON(!md->internal_suspend_count); |
3318 | ||
3319 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3320 | return; /* resume from nested internal suspend */ |
3321 | ||
fd2ed4d2 | 3322 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3323 | goto done; /* resume from nested suspend */ |
3324 | ||
3325 | /* | |
3326 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3327 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3328 | */ | |
3329 | (void) __dm_resume(md, NULL); | |
3330 | ||
3331 | done: | |
3332 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3333 | smp_mb__after_atomic(); | |
3334 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3335 | } | |
3336 | ||
3337 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3338 | { | |
3339 | mutex_lock(&md->suspend_lock); | |
3340 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3341 | mutex_unlock(&md->suspend_lock); | |
3342 | } | |
3343 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3344 | ||
3345 | void dm_internal_resume(struct mapped_device *md) | |
3346 | { | |
3347 | mutex_lock(&md->suspend_lock); | |
3348 | __dm_internal_resume(md); | |
3349 | mutex_unlock(&md->suspend_lock); | |
3350 | } | |
3351 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3352 | ||
3353 | /* | |
3354 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3355 | * which prevents interaction with userspace-driven suspend. | |
3356 | */ | |
3357 | ||
3358 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3359 | { | |
3360 | mutex_lock(&md->suspend_lock); | |
3361 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3362 | return; |
3363 | ||
3364 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3365 | synchronize_srcu(&md->io_barrier); | |
3366 | flush_workqueue(md->wq); | |
3367 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3368 | } | |
b735fede | 3369 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3370 | |
ffcc3936 | 3371 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3372 | { |
ffcc3936 | 3373 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3374 | goto done; |
3375 | ||
3376 | dm_queue_flush(md); | |
3377 | ||
3378 | done: | |
3379 | mutex_unlock(&md->suspend_lock); | |
3380 | } | |
b735fede | 3381 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3382 | |
1da177e4 LT |
3383 | /*----------------------------------------------------------------- |
3384 | * Event notification. | |
3385 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3386 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3387 | unsigned cookie) |
69267a30 | 3388 | { |
60935eb2 MB |
3389 | char udev_cookie[DM_COOKIE_LENGTH]; |
3390 | char *envp[] = { udev_cookie, NULL }; | |
3391 | ||
3392 | if (!cookie) | |
3abf85b5 | 3393 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3394 | else { |
3395 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3396 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3397 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3398 | action, envp); | |
60935eb2 | 3399 | } |
69267a30 AK |
3400 | } |
3401 | ||
7a8c3d3b MA |
3402 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3403 | { | |
3404 | return atomic_add_return(1, &md->uevent_seq); | |
3405 | } | |
3406 | ||
1da177e4 LT |
3407 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3408 | { | |
3409 | return atomic_read(&md->event_nr); | |
3410 | } | |
3411 | ||
3412 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3413 | { | |
3414 | return wait_event_interruptible(md->eventq, | |
3415 | (event_nr != atomic_read(&md->event_nr))); | |
3416 | } | |
3417 | ||
7a8c3d3b MA |
3418 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3419 | { | |
3420 | unsigned long flags; | |
3421 | ||
3422 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3423 | list_add(elist, &md->uevent_list); | |
3424 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3425 | } | |
3426 | ||
1da177e4 LT |
3427 | /* |
3428 | * The gendisk is only valid as long as you have a reference | |
3429 | * count on 'md'. | |
3430 | */ | |
3431 | struct gendisk *dm_disk(struct mapped_device *md) | |
3432 | { | |
3433 | return md->disk; | |
3434 | } | |
65ff5b7d | 3435 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3436 | |
784aae73 MB |
3437 | struct kobject *dm_kobject(struct mapped_device *md) |
3438 | { | |
2995fa78 | 3439 | return &md->kobj_holder.kobj; |
784aae73 MB |
3440 | } |
3441 | ||
784aae73 MB |
3442 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3443 | { | |
3444 | struct mapped_device *md; | |
3445 | ||
2995fa78 | 3446 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3447 | |
4d89b7b4 | 3448 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3449 | dm_deleting_md(md)) |
4d89b7b4 MB |
3450 | return NULL; |
3451 | ||
784aae73 MB |
3452 | dm_get(md); |
3453 | return md; | |
3454 | } | |
3455 | ||
4f186f8b | 3456 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3457 | { |
3458 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3459 | } | |
3460 | ||
ffcc3936 MS |
3461 | int dm_suspended_internally_md(struct mapped_device *md) |
3462 | { | |
3463 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3464 | } | |
3465 | ||
2c140a24 MP |
3466 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3467 | { | |
3468 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3469 | } | |
3470 | ||
64dbce58 KU |
3471 | int dm_suspended(struct dm_target *ti) |
3472 | { | |
ecdb2e25 | 3473 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3474 | } |
3475 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3476 | ||
2e93ccc1 KU |
3477 | int dm_noflush_suspending(struct dm_target *ti) |
3478 | { | |
ecdb2e25 | 3479 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3480 | } |
3481 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3482 | ||
78d8e58a MS |
3483 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3484 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3485 | { |
78d8e58a MS |
3486 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3487 | struct kmem_cache *cachep = NULL; | |
3488 | unsigned int pool_size = 0; | |
5f015204 | 3489 | unsigned int front_pad; |
e6ee8c0b KU |
3490 | |
3491 | if (!pools) | |
4e6e36c3 | 3492 | return NULL; |
e6ee8c0b | 3493 | |
78d8e58a | 3494 | type = filter_md_type(type, md); |
17e149b8 | 3495 | |
78d8e58a MS |
3496 | switch (type) { |
3497 | case DM_TYPE_BIO_BASED: | |
3498 | cachep = _io_cache; | |
3499 | pool_size = dm_get_reserved_bio_based_ios(); | |
3500 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
3501 | break; | |
3502 | case DM_TYPE_REQUEST_BASED: | |
3503 | cachep = _rq_tio_cache; | |
3504 | pool_size = dm_get_reserved_rq_based_ios(); | |
3505 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3506 | if (!pools->rq_pool) | |
3507 | goto out; | |
3508 | /* fall through to setup remaining rq-based pools */ | |
3509 | case DM_TYPE_MQ_REQUEST_BASED: | |
3510 | if (!pool_size) | |
3511 | pool_size = dm_get_reserved_rq_based_ios(); | |
3512 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
3513 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3514 | WARN_ON(per_bio_data_size != 0); | |
3515 | break; | |
3516 | default: | |
3517 | BUG(); | |
3518 | } | |
3519 | ||
3520 | if (cachep) { | |
3521 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3522 | if (!pools->io_pool) | |
3523 | goto out; | |
3524 | } | |
e6ee8c0b | 3525 | |
3d8aab2d | 3526 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3527 | if (!pools->bs) |
5f015204 | 3528 | goto out; |
e6ee8c0b | 3529 | |
a91a2785 | 3530 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3531 | goto out; |
a91a2785 | 3532 | |
e6ee8c0b | 3533 | return pools; |
5f1b670d | 3534 | |
5f1b670d CH |
3535 | out: |
3536 | dm_free_md_mempools(pools); | |
78d8e58a | 3537 | |
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 | ||
6f65985e JL |
3546 | mempool_destroy(pools->io_pool); |
3547 | mempool_destroy(pools->rq_pool); | |
1ae49ea2 | 3548 | |
e6ee8c0b KU |
3549 | if (pools->bs) |
3550 | bioset_free(pools->bs); | |
3551 | ||
3552 | kfree(pools); | |
3553 | } | |
3554 | ||
71cdb697 | 3555 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, |
956a4025 | 3556 | u32 flags) |
71cdb697 CH |
3557 | { |
3558 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3559 | const struct pr_ops *ops; | |
71cdb697 | 3560 | fmode_t mode; |
956a4025 | 3561 | int r; |
71cdb697 | 3562 | |
956a4025 | 3563 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3564 | if (r < 0) |
3565 | return r; | |
3566 | ||
3567 | ops = bdev->bd_disk->fops->pr_ops; | |
3568 | if (ops && ops->pr_register) | |
3569 | r = ops->pr_register(bdev, old_key, new_key, flags); | |
3570 | else | |
3571 | r = -EOPNOTSUPP; | |
3572 | ||
956a4025 | 3573 | bdput(bdev); |
71cdb697 CH |
3574 | return r; |
3575 | } | |
3576 | ||
3577 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3578 | u32 flags) |
71cdb697 CH |
3579 | { |
3580 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3581 | const struct pr_ops *ops; | |
71cdb697 | 3582 | fmode_t mode; |
956a4025 | 3583 | int r; |
71cdb697 | 3584 | |
956a4025 | 3585 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3586 | if (r < 0) |
3587 | return r; | |
3588 | ||
3589 | ops = bdev->bd_disk->fops->pr_ops; | |
3590 | if (ops && ops->pr_reserve) | |
3591 | r = ops->pr_reserve(bdev, key, type, flags); | |
3592 | else | |
3593 | r = -EOPNOTSUPP; | |
3594 | ||
956a4025 | 3595 | bdput(bdev); |
71cdb697 CH |
3596 | return r; |
3597 | } | |
3598 | ||
3599 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3600 | { | |
3601 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3602 | const struct pr_ops *ops; | |
71cdb697 | 3603 | fmode_t mode; |
956a4025 | 3604 | int r; |
71cdb697 | 3605 | |
956a4025 | 3606 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3607 | if (r < 0) |
3608 | return r; | |
3609 | ||
3610 | ops = bdev->bd_disk->fops->pr_ops; | |
3611 | if (ops && ops->pr_release) | |
3612 | r = ops->pr_release(bdev, key, type); | |
3613 | else | |
3614 | r = -EOPNOTSUPP; | |
3615 | ||
956a4025 | 3616 | bdput(bdev); |
71cdb697 CH |
3617 | return r; |
3618 | } | |
3619 | ||
3620 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3621 | enum pr_type type, bool abort) |
71cdb697 CH |
3622 | { |
3623 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3624 | const struct pr_ops *ops; | |
71cdb697 | 3625 | fmode_t mode; |
956a4025 | 3626 | int r; |
71cdb697 | 3627 | |
956a4025 | 3628 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3629 | if (r < 0) |
3630 | return r; | |
3631 | ||
3632 | ops = bdev->bd_disk->fops->pr_ops; | |
3633 | if (ops && ops->pr_preempt) | |
3634 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3635 | else | |
3636 | r = -EOPNOTSUPP; | |
3637 | ||
956a4025 | 3638 | bdput(bdev); |
71cdb697 CH |
3639 | return r; |
3640 | } | |
3641 | ||
3642 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3643 | { | |
3644 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3645 | const struct pr_ops *ops; | |
71cdb697 | 3646 | fmode_t mode; |
956a4025 | 3647 | int r; |
71cdb697 | 3648 | |
956a4025 | 3649 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3650 | if (r < 0) |
3651 | return r; | |
3652 | ||
3653 | ops = bdev->bd_disk->fops->pr_ops; | |
3654 | if (ops && ops->pr_clear) | |
3655 | r = ops->pr_clear(bdev, key); | |
3656 | else | |
3657 | r = -EOPNOTSUPP; | |
3658 | ||
956a4025 | 3659 | bdput(bdev); |
71cdb697 CH |
3660 | return r; |
3661 | } | |
3662 | ||
3663 | static const struct pr_ops dm_pr_ops = { | |
3664 | .pr_register = dm_pr_register, | |
3665 | .pr_reserve = dm_pr_reserve, | |
3666 | .pr_release = dm_pr_release, | |
3667 | .pr_preempt = dm_pr_preempt, | |
3668 | .pr_clear = dm_pr_clear, | |
3669 | }; | |
3670 | ||
83d5cde4 | 3671 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3672 | .open = dm_blk_open, |
3673 | .release = dm_blk_close, | |
aa129a22 | 3674 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3675 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3676 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3677 | .owner = THIS_MODULE |
3678 | }; | |
3679 | ||
1da177e4 LT |
3680 | /* |
3681 | * module hooks | |
3682 | */ | |
3683 | module_init(dm_init); | |
3684 | module_exit(dm_exit); | |
3685 | ||
3686 | module_param(major, uint, 0); | |
3687 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3688 | |
e8603136 MS |
3689 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3690 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3691 | ||
f4790826 MS |
3692 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3693 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3694 | ||
17e149b8 MS |
3695 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3696 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3697 | ||
faad87df MS |
3698 | module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR); |
3699 | MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices"); | |
3700 | ||
3701 | module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR); | |
3702 | MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices"); | |
3703 | ||
1da177e4 LT |
3704 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3705 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3706 | MODULE_LICENSE("GPL"); |