]>
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 | |
eca7ee6d | 336 | _rq_cache = kmem_cache_create("dm_old_clone_request", sizeof(struct request), |
1ae49ea2 MS |
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 | ||
eca7ee6d MS |
655 | static struct dm_rq_target_io *alloc_old_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 | ||
eca7ee6d | 661 | static void free_old_rq_tio(struct dm_rq_target_io *tio) |
cec47e3d | 662 | { |
5f015204 | 663 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
664 | } |
665 | ||
eca7ee6d MS |
666 | static struct request *alloc_old_clone_request(struct mapped_device *md, |
667 | gfp_t gfp_mask) | |
1ae49ea2 MS |
668 | { |
669 | return mempool_alloc(md->rq_pool, gfp_mask); | |
670 | } | |
671 | ||
eca7ee6d | 672 | static void free_old_clone_request(struct mapped_device *md, struct request *rq) |
1ae49ea2 MS |
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) */ | |
eca7ee6d | 1143 | free_old_clone_request(md, clone); |
bfebd1cd MS |
1144 | |
1145 | if (!md->queue->mq_ops) | |
eca7ee6d | 1146 | free_old_rq_tio(tio); |
a77e28c7 KU |
1147 | } |
1148 | ||
980691e5 KU |
1149 | /* |
1150 | * Complete the clone and the original request. | |
466d89a6 KB |
1151 | * Must be called without clone's queue lock held, |
1152 | * see end_clone_request() for more details. | |
980691e5 KU |
1153 | */ |
1154 | static void dm_end_request(struct request *clone, int error) | |
1155 | { | |
1156 | int rw = rq_data_dir(clone); | |
1157 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1158 | struct mapped_device *md = tio->md; | |
1159 | struct request *rq = tio->orig; | |
1160 | ||
29e4013d | 1161 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1162 | rq->errors = clone->errors; |
1163 | rq->resid_len = clone->resid_len; | |
1164 | ||
1165 | if (rq->sense) | |
1166 | /* | |
1167 | * We are using the sense buffer of the original | |
1168 | * request. | |
1169 | * So setting the length of the sense data is enough. | |
1170 | */ | |
1171 | rq->sense_len = clone->sense_len; | |
1172 | } | |
1173 | ||
e5d8de32 | 1174 | free_rq_clone(clone); |
e262f347 | 1175 | rq_end_stats(md, rq); |
bfebd1cd MS |
1176 | if (!rq->q->mq_ops) |
1177 | blk_end_request_all(rq, error); | |
1178 | else | |
1179 | blk_mq_end_request(rq, error); | |
29e4013d | 1180 | rq_completed(md, rw, true); |
980691e5 KU |
1181 | } |
1182 | ||
cec47e3d KU |
1183 | static void dm_unprep_request(struct request *rq) |
1184 | { | |
bfebd1cd | 1185 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1186 | struct request *clone = tio->clone; |
cec47e3d | 1187 | |
bfebd1cd MS |
1188 | if (!rq->q->mq_ops) { |
1189 | rq->special = NULL; | |
1190 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1191 | } | |
cec47e3d | 1192 | |
e5863d9a | 1193 | if (clone) |
e5d8de32 | 1194 | free_rq_clone(clone); |
4328daa2 | 1195 | else if (!tio->md->queue->mq_ops) |
eca7ee6d | 1196 | free_old_rq_tio(tio); |
cec47e3d KU |
1197 | } |
1198 | ||
1199 | /* | |
1200 | * Requeue the original request of a clone. | |
1201 | */ | |
eca7ee6d | 1202 | static void dm_old_requeue_request(struct request *rq) |
cec47e3d | 1203 | { |
cec47e3d KU |
1204 | struct request_queue *q = rq->q; |
1205 | unsigned long flags; | |
1206 | ||
cec47e3d | 1207 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1208 | blk_requeue_request(q, rq); |
4ae9944d | 1209 | blk_run_queue_async(q); |
cec47e3d | 1210 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1211 | } |
1212 | ||
818c5f3b MS |
1213 | static void dm_mq_requeue_request(struct request *rq) |
1214 | { | |
1215 | struct request_queue *q = rq->q; | |
1216 | unsigned long flags; | |
1217 | ||
1218 | blk_mq_requeue_request(rq); | |
1219 | spin_lock_irqsave(q->queue_lock, flags); | |
1220 | if (!blk_queue_stopped(q)) | |
1221 | blk_mq_kick_requeue_list(q); | |
1222 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1223 | } | |
1224 | ||
2d76fff1 MS |
1225 | static void dm_requeue_original_request(struct mapped_device *md, |
1226 | struct request *rq) | |
bfebd1cd MS |
1227 | { |
1228 | int rw = rq_data_dir(rq); | |
1229 | ||
1230 | dm_unprep_request(rq); | |
1231 | ||
e262f347 | 1232 | rq_end_stats(md, rq); |
bfebd1cd | 1233 | if (!rq->q->mq_ops) |
eca7ee6d | 1234 | dm_old_requeue_request(rq); |
818c5f3b MS |
1235 | else |
1236 | dm_mq_requeue_request(rq); | |
cec47e3d | 1237 | |
466d89a6 KB |
1238 | rq_completed(md, rw, false); |
1239 | } | |
1240 | ||
eca7ee6d | 1241 | static void dm_old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1242 | { |
1243 | unsigned long flags; | |
1244 | ||
818c5f3b MS |
1245 | spin_lock_irqsave(q->queue_lock, flags); |
1246 | if (blk_queue_stopped(q)) { | |
1247 | spin_unlock_irqrestore(q->queue_lock, flags); | |
bfebd1cd | 1248 | return; |
818c5f3b | 1249 | } |
bfebd1cd | 1250 | |
bfebd1cd | 1251 | blk_stop_queue(q); |
cec47e3d KU |
1252 | spin_unlock_irqrestore(q->queue_lock, flags); |
1253 | } | |
1254 | ||
eca7ee6d | 1255 | static void dm_stop_queue(struct request_queue *q) |
cec47e3d | 1256 | { |
bfebd1cd | 1257 | if (!q->mq_ops) |
eca7ee6d | 1258 | dm_old_stop_queue(q); |
bfebd1cd MS |
1259 | else |
1260 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1261 | } |
1262 | ||
eca7ee6d | 1263 | static void dm_old_start_queue(struct request_queue *q) |
cec47e3d KU |
1264 | { |
1265 | unsigned long flags; | |
1266 | ||
1267 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1268 | if (blk_queue_stopped(q)) |
1269 | blk_start_queue(q); | |
cec47e3d KU |
1270 | spin_unlock_irqrestore(q->queue_lock, flags); |
1271 | } | |
1272 | ||
eca7ee6d | 1273 | static void dm_start_queue(struct request_queue *q) |
bfebd1cd MS |
1274 | { |
1275 | if (!q->mq_ops) | |
eca7ee6d | 1276 | dm_old_start_queue(q); |
818c5f3b | 1277 | else { |
bfebd1cd | 1278 | blk_mq_start_stopped_hw_queues(q, true); |
818c5f3b MS |
1279 | blk_mq_kick_requeue_list(q); |
1280 | } | |
bfebd1cd MS |
1281 | } |
1282 | ||
11a68244 | 1283 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1284 | { |
11a68244 | 1285 | int r = error; |
cec47e3d | 1286 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1287 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1288 | |
ba1cbad9 MS |
1289 | if (tio->ti) { |
1290 | rq_end_io = tio->ti->type->rq_end_io; | |
1291 | ||
1292 | if (mapped && rq_end_io) | |
1293 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1294 | } | |
cec47e3d | 1295 | |
7eee4ae2 MS |
1296 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1297 | !clone->q->limits.max_write_same_sectors)) | |
1298 | disable_write_same(tio->md); | |
1299 | ||
11a68244 | 1300 | if (r <= 0) |
cec47e3d | 1301 | /* The target wants to complete the I/O */ |
11a68244 KU |
1302 | dm_end_request(clone, r); |
1303 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1304 | /* The target will handle the I/O */ |
1305 | return; | |
11a68244 | 1306 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1307 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1308 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1309 | else { |
11a68244 | 1310 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1311 | BUG(); |
1312 | } | |
1313 | } | |
1314 | ||
11a68244 KU |
1315 | /* |
1316 | * Request completion handler for request-based dm | |
1317 | */ | |
1318 | static void dm_softirq_done(struct request *rq) | |
1319 | { | |
1320 | bool mapped = true; | |
bfebd1cd | 1321 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1322 | struct request *clone = tio->clone; |
bfebd1cd | 1323 | int rw; |
11a68244 | 1324 | |
e5863d9a | 1325 | if (!clone) { |
e262f347 | 1326 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1327 | rw = rq_data_dir(rq); |
1328 | if (!rq->q->mq_ops) { | |
1329 | blk_end_request_all(rq, tio->error); | |
1330 | rq_completed(tio->md, rw, false); | |
eca7ee6d | 1331 | free_old_rq_tio(tio); |
bfebd1cd MS |
1332 | } else { |
1333 | blk_mq_end_request(rq, tio->error); | |
1334 | rq_completed(tio->md, rw, false); | |
1335 | } | |
e5863d9a MS |
1336 | return; |
1337 | } | |
11a68244 KU |
1338 | |
1339 | if (rq->cmd_flags & REQ_FAILED) | |
1340 | mapped = false; | |
1341 | ||
1342 | dm_done(clone, tio->error, mapped); | |
1343 | } | |
1344 | ||
cec47e3d KU |
1345 | /* |
1346 | * Complete the clone and the original request with the error status | |
1347 | * through softirq context. | |
1348 | */ | |
466d89a6 | 1349 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1350 | { |
bfebd1cd | 1351 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1352 | |
1353 | tio->error = error; | |
6acfe68b MS |
1354 | if (!rq->q->mq_ops) |
1355 | blk_complete_request(rq); | |
1356 | else | |
1357 | blk_mq_complete_request(rq, error); | |
cec47e3d KU |
1358 | } |
1359 | ||
1360 | /* | |
1361 | * Complete the not-mapped clone and the original request with the error status | |
1362 | * through softirq context. | |
1363 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1364 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1365 | */ |
466d89a6 | 1366 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1367 | { |
cec47e3d | 1368 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1369 | dm_complete_request(rq, error); |
cec47e3d | 1370 | } |
cec47e3d KU |
1371 | |
1372 | /* | |
eca7ee6d | 1373 | * Called with the clone's queue lock held (in the case of .request_fn) |
cec47e3d KU |
1374 | */ |
1375 | static void end_clone_request(struct request *clone, int error) | |
1376 | { | |
466d89a6 KB |
1377 | struct dm_rq_target_io *tio = clone->end_io_data; |
1378 | ||
e5863d9a MS |
1379 | if (!clone->q->mq_ops) { |
1380 | /* | |
1381 | * For just cleaning up the information of the queue in which | |
1382 | * the clone was dispatched. | |
1383 | * The clone is *NOT* freed actually here because it is alloced | |
1384 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1385 | */ | |
1386 | __blk_put_request(clone->q, clone); | |
1387 | } | |
cec47e3d KU |
1388 | |
1389 | /* | |
1390 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1391 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1392 | * - another request may be submitted by the upper level driver |
1393 | * of the stacking during the completion | |
1394 | * - the submission which requires queue lock may be done | |
466d89a6 | 1395 | * against this clone's queue |
cec47e3d | 1396 | */ |
466d89a6 | 1397 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1398 | } |
1399 | ||
56a67df7 MS |
1400 | /* |
1401 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1402 | * target boundary. | |
1403 | */ | |
1404 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1405 | { | |
1406 | sector_t target_offset = dm_target_offset(ti, sector); | |
1407 | ||
1408 | return ti->len - target_offset; | |
1409 | } | |
1410 | ||
1411 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1412 | { |
56a67df7 | 1413 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1414 | sector_t offset, max_len; |
1da177e4 LT |
1415 | |
1416 | /* | |
542f9038 | 1417 | * Does the target need to split even further? |
1da177e4 | 1418 | */ |
542f9038 MS |
1419 | if (ti->max_io_len) { |
1420 | offset = dm_target_offset(ti, sector); | |
1421 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1422 | max_len = sector_div(offset, ti->max_io_len); | |
1423 | else | |
1424 | max_len = offset & (ti->max_io_len - 1); | |
1425 | max_len = ti->max_io_len - max_len; | |
1426 | ||
1427 | if (len > max_len) | |
1428 | len = max_len; | |
1da177e4 LT |
1429 | } |
1430 | ||
1431 | return len; | |
1432 | } | |
1433 | ||
542f9038 MS |
1434 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1435 | { | |
1436 | if (len > UINT_MAX) { | |
1437 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1438 | (unsigned long long)len, UINT_MAX); | |
1439 | ti->error = "Maximum size of target IO is too large"; | |
1440 | return -EINVAL; | |
1441 | } | |
1442 | ||
1443 | ti->max_io_len = (uint32_t) len; | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1448 | ||
1dd40c3e MP |
1449 | /* |
1450 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1451 | * allowed for all bio types except REQ_FLUSH. | |
1452 | * | |
1453 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1454 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1455 | * sent in a next bio. | |
1456 | * | |
1457 | * A diagram that explains the arithmetics: | |
1458 | * +--------------------+---------------+-------+ | |
1459 | * | 1 | 2 | 3 | | |
1460 | * +--------------------+---------------+-------+ | |
1461 | * | |
1462 | * <-------------- *tio->len_ptr ---------------> | |
1463 | * <------- bi_size -------> | |
1464 | * <-- n_sectors --> | |
1465 | * | |
1466 | * Region 1 was already iterated over with bio_advance or similar function. | |
1467 | * (it may be empty if the target doesn't use bio_advance) | |
1468 | * Region 2 is the remaining bio size that the target wants to process. | |
1469 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1470 | * to make it empty) | |
1471 | * The target requires that region 3 is to be sent in the next bio. | |
1472 | * | |
1473 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1474 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1475 | * copies of the bio. | |
1476 | */ | |
1477 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1478 | { | |
1479 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1480 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1481 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1482 | BUG_ON(bi_size > *tio->len_ptr); | |
1483 | BUG_ON(n_sectors > bi_size); | |
1484 | *tio->len_ptr -= bi_size - n_sectors; | |
1485 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1486 | } | |
1487 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1488 | ||
bd2a49b8 | 1489 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1490 | { |
1491 | int r; | |
2056a782 | 1492 | sector_t sector; |
9faf400f | 1493 | struct mapped_device *md; |
dba14160 | 1494 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1495 | struct dm_target *ti = tio->ti; |
1da177e4 | 1496 | |
1da177e4 | 1497 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1498 | |
1499 | /* | |
1500 | * Map the clone. If r == 0 we don't need to do | |
1501 | * anything, the target has assumed ownership of | |
1502 | * this io. | |
1503 | */ | |
1504 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1505 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1506 | r = ti->type->map(ti, clone); |
45cbcd79 | 1507 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1508 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1509 | |
d07335e5 MS |
1510 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1511 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1512 | |
1da177e4 | 1513 | generic_make_request(clone); |
2e93ccc1 KU |
1514 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1515 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1516 | md = tio->io->md; |
1517 | dec_pending(tio->io, r); | |
9faf400f | 1518 | free_tio(md, tio); |
ab37844d | 1519 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1520 | DMWARN("unimplemented target map return value: %d", r); |
1521 | BUG(); | |
1da177e4 LT |
1522 | } |
1523 | } | |
1524 | ||
1525 | struct clone_info { | |
1526 | struct mapped_device *md; | |
1527 | struct dm_table *map; | |
1528 | struct bio *bio; | |
1529 | struct dm_io *io; | |
1530 | sector_t sector; | |
e0d6609a | 1531 | unsigned sector_count; |
1da177e4 LT |
1532 | }; |
1533 | ||
e0d6609a | 1534 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1535 | { |
4f024f37 KO |
1536 | bio->bi_iter.bi_sector = sector; |
1537 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1538 | } |
1539 | ||
1540 | /* | |
1541 | * Creates a bio that consists of range of complete bvecs. | |
1542 | */ | |
dba14160 | 1543 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1544 | sector_t sector, unsigned len) |
1da177e4 | 1545 | { |
dba14160 | 1546 | struct bio *clone = &tio->clone; |
1da177e4 | 1547 | |
1c3b13e6 KO |
1548 | __bio_clone_fast(clone, bio); |
1549 | ||
1550 | if (bio_integrity(bio)) | |
1551 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1552 | |
1c3b13e6 KO |
1553 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1554 | clone->bi_iter.bi_size = to_bytes(len); | |
1555 | ||
1556 | if (bio_integrity(bio)) | |
1557 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1558 | } |
1559 | ||
9015df24 | 1560 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1561 | struct dm_target *ti, |
55a62eef | 1562 | unsigned target_bio_nr) |
f9ab94ce | 1563 | { |
dba14160 MP |
1564 | struct dm_target_io *tio; |
1565 | struct bio *clone; | |
1566 | ||
99778273 | 1567 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1568 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1569 | |
1570 | tio->io = ci->io; | |
1571 | tio->ti = ti; | |
55a62eef | 1572 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1573 | |
1574 | return tio; | |
1575 | } | |
1576 | ||
14fe594d AK |
1577 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1578 | struct dm_target *ti, | |
1dd40c3e | 1579 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1580 | { |
99778273 | 1581 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1582 | struct bio *clone = &tio->clone; |
9015df24 | 1583 | |
1dd40c3e MP |
1584 | tio->len_ptr = len; |
1585 | ||
99778273 | 1586 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1587 | if (len) |
1dd40c3e | 1588 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1589 | |
bd2a49b8 | 1590 | __map_bio(tio); |
f9ab94ce MP |
1591 | } |
1592 | ||
14fe594d | 1593 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1594 | unsigned num_bios, unsigned *len) |
06a426ce | 1595 | { |
55a62eef | 1596 | unsigned target_bio_nr; |
06a426ce | 1597 | |
55a62eef | 1598 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1599 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1600 | } |
1601 | ||
14fe594d | 1602 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1603 | { |
06a426ce | 1604 | unsigned target_nr = 0; |
f9ab94ce MP |
1605 | struct dm_target *ti; |
1606 | ||
b372d360 | 1607 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1608 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1609 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1610 | |
f9ab94ce MP |
1611 | return 0; |
1612 | } | |
1613 | ||
e4c93811 | 1614 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1615 | sector_t sector, unsigned *len) |
5ae89a87 | 1616 | { |
dba14160 | 1617 | struct bio *bio = ci->bio; |
5ae89a87 | 1618 | struct dm_target_io *tio; |
b0d8ed4d AK |
1619 | unsigned target_bio_nr; |
1620 | unsigned num_target_bios = 1; | |
5ae89a87 | 1621 | |
b0d8ed4d AK |
1622 | /* |
1623 | * Does the target want to receive duplicate copies of the bio? | |
1624 | */ | |
1625 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1626 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1627 | |
b0d8ed4d | 1628 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1629 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1630 | tio->len_ptr = len; |
1631 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1632 | __map_bio(tio); |
1633 | } | |
5ae89a87 MS |
1634 | } |
1635 | ||
55a62eef | 1636 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1637 | |
55a62eef | 1638 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1639 | { |
55a62eef | 1640 | return ti->num_discard_bios; |
23508a96 MS |
1641 | } |
1642 | ||
55a62eef | 1643 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1644 | { |
55a62eef | 1645 | return ti->num_write_same_bios; |
23508a96 MS |
1646 | } |
1647 | ||
1648 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1649 | ||
1650 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1651 | { | |
55a62eef | 1652 | return ti->split_discard_bios; |
23508a96 MS |
1653 | } |
1654 | ||
14fe594d AK |
1655 | static int __send_changing_extent_only(struct clone_info *ci, |
1656 | get_num_bios_fn get_num_bios, | |
1657 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1658 | { |
1659 | struct dm_target *ti; | |
e0d6609a | 1660 | unsigned len; |
55a62eef | 1661 | unsigned num_bios; |
5ae89a87 | 1662 | |
a79245b3 MS |
1663 | do { |
1664 | ti = dm_table_find_target(ci->map, ci->sector); | |
1665 | if (!dm_target_is_valid(ti)) | |
1666 | return -EIO; | |
5ae89a87 | 1667 | |
5ae89a87 | 1668 | /* |
23508a96 MS |
1669 | * Even though the device advertised support for this type of |
1670 | * request, that does not mean every target supports it, and | |
936688d7 | 1671 | * reconfiguration might also have changed that since the |
a79245b3 | 1672 | * check was performed. |
5ae89a87 | 1673 | */ |
55a62eef AK |
1674 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1675 | if (!num_bios) | |
a79245b3 | 1676 | return -EOPNOTSUPP; |
5ae89a87 | 1677 | |
23508a96 | 1678 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1679 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1680 | else |
e0d6609a | 1681 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1682 | |
1dd40c3e | 1683 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1684 | |
1685 | ci->sector += len; | |
1686 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1687 | |
1688 | return 0; | |
1689 | } | |
1690 | ||
14fe594d | 1691 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1692 | { |
14fe594d AK |
1693 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1694 | is_split_required_for_discard); | |
23508a96 MS |
1695 | } |
1696 | ||
14fe594d | 1697 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1698 | { |
14fe594d | 1699 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1700 | } |
1701 | ||
e4c93811 AK |
1702 | /* |
1703 | * Select the correct strategy for processing a non-flush bio. | |
1704 | */ | |
14fe594d | 1705 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1706 | { |
dba14160 | 1707 | struct bio *bio = ci->bio; |
512875bd | 1708 | struct dm_target *ti; |
1c3b13e6 | 1709 | unsigned len; |
1da177e4 | 1710 | |
5ae89a87 | 1711 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1712 | return __send_discard(ci); |
23508a96 | 1713 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1714 | return __send_write_same(ci); |
5ae89a87 | 1715 | |
512875bd JN |
1716 | ti = dm_table_find_target(ci->map, ci->sector); |
1717 | if (!dm_target_is_valid(ti)) | |
1718 | return -EIO; | |
1719 | ||
1c3b13e6 | 1720 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1721 | |
1dd40c3e | 1722 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1723 | |
1c3b13e6 KO |
1724 | ci->sector += len; |
1725 | ci->sector_count -= len; | |
1da177e4 | 1726 | |
1c3b13e6 | 1727 | return 0; |
1da177e4 LT |
1728 | } |
1729 | ||
1730 | /* | |
14fe594d | 1731 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1732 | */ |
83d5e5b0 MP |
1733 | static void __split_and_process_bio(struct mapped_device *md, |
1734 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1735 | { |
1736 | struct clone_info ci; | |
512875bd | 1737 | int error = 0; |
1da177e4 | 1738 | |
83d5e5b0 | 1739 | if (unlikely(!map)) { |
6a8736d1 | 1740 | bio_io_error(bio); |
f0b9a450 MP |
1741 | return; |
1742 | } | |
692d0eb9 | 1743 | |
83d5e5b0 | 1744 | ci.map = map; |
1da177e4 | 1745 | ci.md = md; |
1da177e4 LT |
1746 | ci.io = alloc_io(md); |
1747 | ci.io->error = 0; | |
1748 | atomic_set(&ci.io->io_count, 1); | |
1749 | ci.io->bio = bio; | |
1750 | ci.io->md = md; | |
f88fb981 | 1751 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1752 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1753 | |
3eaf840e | 1754 | start_io_acct(ci.io); |
bd2a49b8 | 1755 | |
b372d360 MS |
1756 | if (bio->bi_rw & REQ_FLUSH) { |
1757 | ci.bio = &ci.md->flush_bio; | |
1758 | ci.sector_count = 0; | |
14fe594d | 1759 | error = __send_empty_flush(&ci); |
b372d360 MS |
1760 | /* dec_pending submits any data associated with flush */ |
1761 | } else { | |
6a8736d1 | 1762 | ci.bio = bio; |
d87f4c14 | 1763 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1764 | while (ci.sector_count && !error) |
14fe594d | 1765 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1766 | } |
1da177e4 LT |
1767 | |
1768 | /* drop the extra reference count */ | |
512875bd | 1769 | dec_pending(ci.io, error); |
1da177e4 LT |
1770 | } |
1771 | /*----------------------------------------------------------------- | |
1772 | * CRUD END | |
1773 | *---------------------------------------------------------------*/ | |
1774 | ||
1775 | /* | |
1776 | * The request function that just remaps the bio built up by | |
1777 | * dm_merge_bvec. | |
1778 | */ | |
dece1635 | 1779 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1780 | { |
12f03a49 | 1781 | int rw = bio_data_dir(bio); |
1da177e4 | 1782 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1783 | int srcu_idx; |
1784 | struct dm_table *map; | |
1da177e4 | 1785 | |
83d5e5b0 | 1786 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1787 | |
18c0b223 | 1788 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1789 | |
6a8736d1 TH |
1790 | /* if we're suspended, we have to queue this io for later */ |
1791 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1792 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1793 | |
6a8736d1 TH |
1794 | if (bio_rw(bio) != READA) |
1795 | queue_io(md, bio); | |
1796 | else | |
54d9a1b4 | 1797 | bio_io_error(bio); |
dece1635 | 1798 | return BLK_QC_T_NONE; |
1da177e4 LT |
1799 | } |
1800 | ||
83d5e5b0 MP |
1801 | __split_and_process_bio(md, map, bio); |
1802 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1803 | return BLK_QC_T_NONE; |
cec47e3d KU |
1804 | } |
1805 | ||
fd2ed4d2 | 1806 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1807 | { |
1808 | return blk_queue_stackable(md->queue); | |
1809 | } | |
1810 | ||
466d89a6 | 1811 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1812 | { |
1813 | int r; | |
1814 | ||
466d89a6 KB |
1815 | if (blk_queue_io_stat(clone->q)) |
1816 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1817 | |
466d89a6 KB |
1818 | clone->start_time = jiffies; |
1819 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1820 | if (r) |
466d89a6 | 1821 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1822 | dm_complete_request(rq, r); |
1823 | } | |
cec47e3d | 1824 | |
78d8e58a MS |
1825 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1826 | void *data) | |
cec47e3d | 1827 | { |
78d8e58a MS |
1828 | struct dm_rq_target_io *tio = data; |
1829 | struct dm_rq_clone_bio_info *info = | |
1830 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
1831 | ||
1832 | info->orig = bio_orig; | |
1833 | info->tio = tio; | |
1834 | bio->bi_end_io = end_clone_bio; | |
1835 | ||
1836 | return 0; | |
1837 | } | |
1838 | ||
1839 | static int setup_clone(struct request *clone, struct request *rq, | |
1840 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
cec47e3d | 1841 | { |
78d8e58a MS |
1842 | int r; |
1843 | ||
1844 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
1845 | dm_rq_bio_constructor, tio); | |
1846 | if (r) | |
1847 | return r; | |
1848 | ||
1849 | clone->cmd = rq->cmd; | |
1850 | clone->cmd_len = rq->cmd_len; | |
1851 | clone->sense = rq->sense; | |
cec47e3d KU |
1852 | clone->end_io = end_clone_request; |
1853 | clone->end_io_data = tio; | |
78d8e58a | 1854 | |
1ae49ea2 | 1855 | tio->clone = clone; |
78d8e58a MS |
1856 | |
1857 | return 0; | |
cec47e3d KU |
1858 | } |
1859 | ||
eca7ee6d MS |
1860 | static struct request *clone_old_rq(struct request *rq, struct mapped_device *md, |
1861 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
1ae49ea2 | 1862 | { |
02233342 | 1863 | /* |
c5248f79 | 1864 | * Create clone for use with .request_fn request_queue |
02233342 | 1865 | */ |
02233342 | 1866 | struct request *clone; |
1ae49ea2 | 1867 | |
eca7ee6d | 1868 | clone = alloc_old_clone_request(md, gfp_mask); |
c5248f79 MS |
1869 | if (!clone) |
1870 | return NULL; | |
1ae49ea2 MS |
1871 | |
1872 | blk_rq_init(NULL, clone); | |
78d8e58a MS |
1873 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
1874 | /* -ENOMEM */ | |
eca7ee6d | 1875 | free_old_clone_request(md, clone); |
78d8e58a MS |
1876 | return NULL; |
1877 | } | |
1ae49ea2 MS |
1878 | |
1879 | return clone; | |
1880 | } | |
1881 | ||
2eb6e1e3 KB |
1882 | static void map_tio_request(struct kthread_work *work); |
1883 | ||
bfebd1cd MS |
1884 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1885 | struct mapped_device *md) | |
1886 | { | |
1887 | tio->md = md; | |
1888 | tio->ti = NULL; | |
1889 | tio->clone = NULL; | |
1890 | tio->orig = rq; | |
1891 | tio->error = 0; | |
1892 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1893 | if (md->kworker_task) |
1894 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1895 | } |
1896 | ||
eca7ee6d MS |
1897 | static struct dm_rq_target_io *dm_old_prep_tio(struct request *rq, |
1898 | struct mapped_device *md, | |
1899 | gfp_t gfp_mask) | |
6facdaff | 1900 | { |
6facdaff | 1901 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1902 | int srcu_idx; |
1903 | struct dm_table *table; | |
6facdaff | 1904 | |
eca7ee6d | 1905 | tio = alloc_old_rq_tio(md, gfp_mask); |
6facdaff KU |
1906 | if (!tio) |
1907 | return NULL; | |
1908 | ||
bfebd1cd | 1909 | init_tio(tio, rq, md); |
6facdaff | 1910 | |
e5863d9a | 1911 | table = dm_get_live_table(md, &srcu_idx); |
eca7ee6d MS |
1912 | /* |
1913 | * Must clone a request if this .request_fn DM device | |
1914 | * is stacked on .request_fn device(s). | |
1915 | */ | |
e5863d9a | 1916 | if (!dm_table_mq_request_based(table)) { |
eca7ee6d | 1917 | if (!clone_old_rq(rq, md, tio, gfp_mask)) { |
e5863d9a | 1918 | dm_put_live_table(md, srcu_idx); |
eca7ee6d | 1919 | free_old_rq_tio(tio); |
e5863d9a MS |
1920 | return NULL; |
1921 | } | |
6facdaff | 1922 | } |
e5863d9a | 1923 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1924 | |
466d89a6 | 1925 | return tio; |
6facdaff KU |
1926 | } |
1927 | ||
cec47e3d KU |
1928 | /* |
1929 | * Called with the queue lock held. | |
1930 | */ | |
eca7ee6d | 1931 | static int dm_old_prep_fn(struct request_queue *q, struct request *rq) |
cec47e3d KU |
1932 | { |
1933 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1934 | struct dm_rq_target_io *tio; |
cec47e3d | 1935 | |
cec47e3d KU |
1936 | if (unlikely(rq->special)) { |
1937 | DMWARN("Already has something in rq->special."); | |
1938 | return BLKPREP_KILL; | |
1939 | } | |
1940 | ||
eca7ee6d | 1941 | tio = dm_old_prep_tio(rq, md, GFP_ATOMIC); |
466d89a6 | 1942 | if (!tio) |
cec47e3d | 1943 | return BLKPREP_DEFER; |
cec47e3d | 1944 | |
466d89a6 | 1945 | rq->special = tio; |
cec47e3d KU |
1946 | rq->cmd_flags |= REQ_DONTPREP; |
1947 | ||
1948 | return BLKPREP_OK; | |
1949 | } | |
1950 | ||
9eef87da KU |
1951 | /* |
1952 | * Returns: | |
e5863d9a MS |
1953 | * 0 : the request has been processed |
1954 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1955 | * < 0 : the request was completed due to failure | |
9eef87da | 1956 | */ |
bfebd1cd | 1957 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1958 | struct mapped_device *md) |
cec47e3d | 1959 | { |
e5863d9a | 1960 | int r; |
bfebd1cd | 1961 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1962 | struct request *clone = NULL; |
1963 | ||
1964 | if (tio->clone) { | |
1965 | clone = tio->clone; | |
1966 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1967 | } else { | |
1968 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1969 | if (r < 0) { | |
1970 | /* The target wants to complete the I/O */ | |
1971 | dm_kill_unmapped_request(rq, r); | |
1972 | return r; | |
1973 | } | |
3a140755 JN |
1974 | if (r != DM_MAPIO_REMAPPED) |
1975 | return r; | |
78d8e58a MS |
1976 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
1977 | /* -ENOMEM */ | |
1978 | ti->type->release_clone_rq(clone); | |
1979 | return DM_MAPIO_REQUEUE; | |
1980 | } | |
e5863d9a | 1981 | } |
cec47e3d | 1982 | |
cec47e3d KU |
1983 | switch (r) { |
1984 | case DM_MAPIO_SUBMITTED: | |
1985 | /* The target has taken the I/O to submit by itself later */ | |
1986 | break; | |
1987 | case DM_MAPIO_REMAPPED: | |
1988 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1989 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1990 | blk_rq_pos(rq)); |
1991 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1992 | break; |
1993 | case DM_MAPIO_REQUEUE: | |
1994 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 1995 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
1996 | break; |
1997 | default: | |
1998 | if (r > 0) { | |
1999 | DMWARN("unimplemented target map return value: %d", r); | |
2000 | BUG(); | |
2001 | } | |
2002 | ||
2003 | /* The target wants to complete the I/O */ | |
466d89a6 | 2004 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 2005 | return r; |
cec47e3d | 2006 | } |
9eef87da | 2007 | |
e5863d9a | 2008 | return 0; |
cec47e3d KU |
2009 | } |
2010 | ||
2eb6e1e3 | 2011 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 2012 | { |
2eb6e1e3 | 2013 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
2014 | struct request *rq = tio->orig; |
2015 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 2016 | |
bfebd1cd | 2017 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 2018 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
2019 | } |
2020 | ||
466d89a6 | 2021 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2022 | { |
bfebd1cd MS |
2023 | if (!orig->q->mq_ops) |
2024 | blk_start_request(orig); | |
2025 | else | |
2026 | blk_mq_start_request(orig); | |
466d89a6 | 2027 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2028 | |
0ce65797 MS |
2029 | if (md->seq_rq_merge_deadline_usecs) { |
2030 | md->last_rq_pos = rq_end_sector(orig); | |
2031 | md->last_rq_rw = rq_data_dir(orig); | |
2032 | md->last_rq_start_time = ktime_get(); | |
2033 | } | |
de3ec86d | 2034 | |
e262f347 MP |
2035 | if (unlikely(dm_stats_used(&md->stats))) { |
2036 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
2037 | tio->duration_jiffies = jiffies; | |
2038 | tio->n_sectors = blk_rq_sectors(orig); | |
2039 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
2040 | tio->n_sectors, false, 0, &tio->stats_aux); | |
2041 | } | |
2042 | ||
ba1cbad9 MS |
2043 | /* |
2044 | * Hold the md reference here for the in-flight I/O. | |
2045 | * We can't rely on the reference count by device opener, | |
2046 | * because the device may be closed during the request completion | |
2047 | * when all bios are completed. | |
2048 | * See the comment in rq_completed() too. | |
2049 | */ | |
2050 | dm_get(md); | |
ba1cbad9 MS |
2051 | } |
2052 | ||
0ce65797 MS |
2053 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2054 | ||
2055 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2056 | { | |
2057 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2058 | } | |
2059 | ||
2060 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2061 | const char *buf, size_t count) | |
2062 | { | |
2063 | unsigned deadline; | |
2064 | ||
17e149b8 | 2065 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2066 | return count; |
2067 | ||
2068 | if (kstrtouint(buf, 10, &deadline)) | |
2069 | return -EINVAL; | |
2070 | ||
2071 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2072 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2073 | ||
2074 | md->seq_rq_merge_deadline_usecs = deadline; | |
2075 | ||
2076 | return count; | |
2077 | } | |
2078 | ||
2079 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2080 | { | |
2081 | ktime_t kt_deadline; | |
2082 | ||
2083 | if (!md->seq_rq_merge_deadline_usecs) | |
2084 | return false; | |
2085 | ||
2086 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2087 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2088 | ||
2089 | return !ktime_after(ktime_get(), kt_deadline); | |
2090 | } | |
2091 | ||
cec47e3d KU |
2092 | /* |
2093 | * q->request_fn for request-based dm. | |
2094 | * Called with the queue lock held. | |
2095 | */ | |
2096 | static void dm_request_fn(struct request_queue *q) | |
2097 | { | |
2098 | struct mapped_device *md = q->queuedata; | |
c91852ff | 2099 | struct dm_target *ti = md->immutable_target; |
466d89a6 | 2100 | struct request *rq; |
2eb6e1e3 | 2101 | struct dm_rq_target_io *tio; |
c91852ff MS |
2102 | sector_t pos = 0; |
2103 | ||
2104 | if (unlikely(!ti)) { | |
2105 | int srcu_idx; | |
2106 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2107 | ||
2108 | ti = dm_table_find_target(map, pos); | |
2109 | dm_put_live_table(md, srcu_idx); | |
2110 | } | |
cec47e3d KU |
2111 | |
2112 | /* | |
b4324fee KU |
2113 | * For suspend, check blk_queue_stopped() and increment |
2114 | * ->pending within a single queue_lock not to increment the | |
2115 | * number of in-flight I/Os after the queue is stopped in | |
2116 | * dm_suspend(). | |
cec47e3d | 2117 | */ |
7eaceacc | 2118 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2119 | rq = blk_peek_request(q); |
2120 | if (!rq) | |
c91852ff | 2121 | return; |
cec47e3d | 2122 | |
29e4013d TH |
2123 | /* always use block 0 to find the target for flushes for now */ |
2124 | pos = 0; | |
2125 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2126 | pos = blk_rq_pos(rq); | |
2127 | ||
c91852ff MS |
2128 | if ((dm_request_peeked_before_merge_deadline(md) && |
2129 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
2130 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) || | |
2131 | (ti->type->busy && ti->type->busy(ti))) { | |
2132 | blk_delay_queue(q, HZ / 100); | |
2133 | return; | |
ba1cbad9 | 2134 | } |
d0bcb878 | 2135 | |
466d89a6 | 2136 | dm_start_request(md, rq); |
9eef87da | 2137 | |
bfebd1cd | 2138 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2139 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2140 | tio->ti = ti; | |
2141 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2142 | BUG_ON(!irqs_disabled()); |
cec47e3d | 2143 | } |
cec47e3d KU |
2144 | } |
2145 | ||
1da177e4 LT |
2146 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2147 | { | |
8a57dfc6 CS |
2148 | int r = bdi_bits; |
2149 | struct mapped_device *md = congested_data; | |
2150 | struct dm_table *map; | |
1da177e4 | 2151 | |
1eb787ec | 2152 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
e522c039 | 2153 | if (dm_request_based(md)) { |
cec47e3d | 2154 | /* |
e522c039 MS |
2155 | * With request-based DM we only need to check the |
2156 | * top-level queue for congestion. | |
cec47e3d | 2157 | */ |
e522c039 MS |
2158 | r = md->queue->backing_dev_info.wb.state & bdi_bits; |
2159 | } else { | |
2160 | map = dm_get_live_table_fast(md); | |
2161 | if (map) | |
cec47e3d | 2162 | r = dm_table_any_congested(map, bdi_bits); |
e522c039 | 2163 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2164 | } |
2165 | } | |
2166 | ||
1da177e4 LT |
2167 | return r; |
2168 | } | |
2169 | ||
2170 | /*----------------------------------------------------------------- | |
2171 | * An IDR is used to keep track of allocated minor numbers. | |
2172 | *---------------------------------------------------------------*/ | |
2b06cfff | 2173 | static void free_minor(int minor) |
1da177e4 | 2174 | { |
f32c10b0 | 2175 | spin_lock(&_minor_lock); |
1da177e4 | 2176 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2177 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2178 | } |
2179 | ||
2180 | /* | |
2181 | * See if the device with a specific minor # is free. | |
2182 | */ | |
cf13ab8e | 2183 | static int specific_minor(int minor) |
1da177e4 | 2184 | { |
c9d76be6 | 2185 | int r; |
1da177e4 LT |
2186 | |
2187 | if (minor >= (1 << MINORBITS)) | |
2188 | return -EINVAL; | |
2189 | ||
c9d76be6 | 2190 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2191 | spin_lock(&_minor_lock); |
1da177e4 | 2192 | |
c9d76be6 | 2193 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2194 | |
f32c10b0 | 2195 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2196 | idr_preload_end(); |
2197 | if (r < 0) | |
2198 | return r == -ENOSPC ? -EBUSY : r; | |
2199 | return 0; | |
1da177e4 LT |
2200 | } |
2201 | ||
cf13ab8e | 2202 | static int next_free_minor(int *minor) |
1da177e4 | 2203 | { |
c9d76be6 | 2204 | int r; |
62f75c2f | 2205 | |
c9d76be6 | 2206 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2207 | spin_lock(&_minor_lock); |
1da177e4 | 2208 | |
c9d76be6 | 2209 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2210 | |
f32c10b0 | 2211 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2212 | idr_preload_end(); |
2213 | if (r < 0) | |
2214 | return r; | |
2215 | *minor = r; | |
2216 | return 0; | |
1da177e4 LT |
2217 | } |
2218 | ||
83d5cde4 | 2219 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2220 | |
53d5914f MP |
2221 | static void dm_wq_work(struct work_struct *work); |
2222 | ||
4a0b4ddf MS |
2223 | static void dm_init_md_queue(struct mapped_device *md) |
2224 | { | |
2225 | /* | |
2226 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2227 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2228 | * The type is decided at the first table loading time. |
2229 | * To prevent problematic device stacking, clear the queue flag | |
2230 | * for request stacking support until then. | |
2231 | * | |
2232 | * This queue is new, so no concurrency on the queue_flags. | |
2233 | */ | |
2234 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
2235 | |
2236 | /* | |
2237 | * Initialize data that will only be used by a non-blk-mq DM queue | |
2238 | * - must do so here (in alloc_dev callchain) before queue is used | |
2239 | */ | |
2240 | md->queue->queuedata = md; | |
2241 | md->queue->backing_dev_info.congested_data = md; | |
bfebd1cd | 2242 | } |
4a0b4ddf | 2243 | |
eca7ee6d | 2244 | static void dm_init_normal_md_queue(struct mapped_device *md) |
bfebd1cd | 2245 | { |
17e149b8 | 2246 | md->use_blk_mq = false; |
bfebd1cd MS |
2247 | dm_init_md_queue(md); |
2248 | ||
2249 | /* | |
2250 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2251 | */ | |
4a0b4ddf | 2252 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
4a0b4ddf | 2253 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2254 | } |
2255 | ||
0f20972f MS |
2256 | static void cleanup_mapped_device(struct mapped_device *md) |
2257 | { | |
0f20972f MS |
2258 | if (md->wq) |
2259 | destroy_workqueue(md->wq); | |
2260 | if (md->kworker_task) | |
2261 | kthread_stop(md->kworker_task); | |
6f65985e JL |
2262 | mempool_destroy(md->io_pool); |
2263 | mempool_destroy(md->rq_pool); | |
0f20972f MS |
2264 | if (md->bs) |
2265 | bioset_free(md->bs); | |
2266 | ||
b06075a9 MP |
2267 | cleanup_srcu_struct(&md->io_barrier); |
2268 | ||
0f20972f MS |
2269 | if (md->disk) { |
2270 | spin_lock(&_minor_lock); | |
2271 | md->disk->private_data = NULL; | |
2272 | spin_unlock(&_minor_lock); | |
0f20972f MS |
2273 | del_gendisk(md->disk); |
2274 | put_disk(md->disk); | |
2275 | } | |
2276 | ||
2277 | if (md->queue) | |
2278 | blk_cleanup_queue(md->queue); | |
2279 | ||
2280 | if (md->bdev) { | |
2281 | bdput(md->bdev); | |
2282 | md->bdev = NULL; | |
2283 | } | |
2284 | } | |
2285 | ||
1da177e4 LT |
2286 | /* |
2287 | * Allocate and initialise a blank device with a given minor. | |
2288 | */ | |
2b06cfff | 2289 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2290 | { |
2291 | int r; | |
cf13ab8e | 2292 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2293 | void *old_md; |
1da177e4 LT |
2294 | |
2295 | if (!md) { | |
2296 | DMWARN("unable to allocate device, out of memory."); | |
2297 | return NULL; | |
2298 | } | |
2299 | ||
10da4f79 | 2300 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2301 | goto bad_module_get; |
10da4f79 | 2302 | |
1da177e4 | 2303 | /* get a minor number for the dev */ |
2b06cfff | 2304 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2305 | r = next_free_minor(&minor); |
2b06cfff | 2306 | else |
cf13ab8e | 2307 | r = specific_minor(minor); |
1da177e4 | 2308 | if (r < 0) |
6ed7ade8 | 2309 | goto bad_minor; |
1da177e4 | 2310 | |
83d5e5b0 MP |
2311 | r = init_srcu_struct(&md->io_barrier); |
2312 | if (r < 0) | |
2313 | goto bad_io_barrier; | |
2314 | ||
17e149b8 | 2315 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2316 | md->type = DM_TYPE_NONE; |
e61290a4 | 2317 | mutex_init(&md->suspend_lock); |
a5664dad | 2318 | mutex_init(&md->type_lock); |
86f1152b | 2319 | mutex_init(&md->table_devices_lock); |
022c2611 | 2320 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2321 | atomic_set(&md->holders, 1); |
5c6bd75d | 2322 | atomic_set(&md->open_count, 0); |
1da177e4 | 2323 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2324 | atomic_set(&md->uevent_seq, 0); |
2325 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2326 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2327 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2328 | |
4a0b4ddf | 2329 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2330 | if (!md->queue) |
0f20972f | 2331 | goto bad; |
1da177e4 | 2332 | |
4a0b4ddf | 2333 | dm_init_md_queue(md); |
9faf400f | 2334 | |
1da177e4 LT |
2335 | md->disk = alloc_disk(1); |
2336 | if (!md->disk) | |
0f20972f | 2337 | goto bad; |
1da177e4 | 2338 | |
316d315b NK |
2339 | atomic_set(&md->pending[0], 0); |
2340 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2341 | init_waitqueue_head(&md->wait); |
53d5914f | 2342 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2343 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2344 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2345 | md->kworker_task = NULL; |
f0b04115 | 2346 | |
1da177e4 LT |
2347 | md->disk->major = _major; |
2348 | md->disk->first_minor = minor; | |
2349 | md->disk->fops = &dm_blk_dops; | |
2350 | md->disk->queue = md->queue; | |
2351 | md->disk->private_data = md; | |
2352 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2353 | add_disk(md->disk); | |
7e51f257 | 2354 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2355 | |
670368a8 | 2356 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2357 | if (!md->wq) |
0f20972f | 2358 | goto bad; |
304f3f6a | 2359 | |
32a926da MP |
2360 | md->bdev = bdget_disk(md->disk, 0); |
2361 | if (!md->bdev) | |
0f20972f | 2362 | goto bad; |
32a926da | 2363 | |
6a8736d1 TH |
2364 | bio_init(&md->flush_bio); |
2365 | md->flush_bio.bi_bdev = md->bdev; | |
2366 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2367 | ||
fd2ed4d2 MP |
2368 | dm_stats_init(&md->stats); |
2369 | ||
ba61fdd1 | 2370 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2371 | spin_lock(&_minor_lock); |
ba61fdd1 | 2372 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2373 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2374 | |
2375 | BUG_ON(old_md != MINOR_ALLOCED); | |
2376 | ||
1da177e4 LT |
2377 | return md; |
2378 | ||
0f20972f MS |
2379 | bad: |
2380 | cleanup_mapped_device(md); | |
83d5e5b0 | 2381 | bad_io_barrier: |
1da177e4 | 2382 | free_minor(minor); |
6ed7ade8 | 2383 | bad_minor: |
10da4f79 | 2384 | module_put(THIS_MODULE); |
6ed7ade8 | 2385 | bad_module_get: |
1da177e4 LT |
2386 | kfree(md); |
2387 | return NULL; | |
2388 | } | |
2389 | ||
ae9da83f JN |
2390 | static void unlock_fs(struct mapped_device *md); |
2391 | ||
1da177e4 LT |
2392 | static void free_dev(struct mapped_device *md) |
2393 | { | |
f331c029 | 2394 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2395 | |
32a926da | 2396 | unlock_fs(md); |
2eb6e1e3 | 2397 | |
0f20972f | 2398 | cleanup_mapped_device(md); |
1c357a1e MS |
2399 | if (md->tag_set) { |
2400 | blk_mq_free_tag_set(md->tag_set); | |
2401 | kfree(md->tag_set); | |
2402 | } | |
63a4f065 | 2403 | |
86f1152b | 2404 | free_table_devices(&md->table_devices); |
63a4f065 | 2405 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2406 | free_minor(minor); |
2407 | ||
10da4f79 | 2408 | module_put(THIS_MODULE); |
1da177e4 LT |
2409 | kfree(md); |
2410 | } | |
2411 | ||
e6ee8c0b KU |
2412 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2413 | { | |
c0820cf5 | 2414 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2415 | |
4e6e36c3 MS |
2416 | if (md->bs) { |
2417 | /* The md already has necessary mempools. */ | |
2418 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
2419 | /* |
2420 | * Reload bioset because front_pad may have changed | |
2421 | * because a different table was loaded. | |
2422 | */ | |
2423 | bioset_free(md->bs); | |
2424 | md->bs = p->bs; | |
2425 | p->bs = NULL; | |
16245bdc | 2426 | } |
4e6e36c3 MS |
2427 | /* |
2428 | * There's no need to reload with request-based dm | |
2429 | * because the size of front_pad doesn't change. | |
2430 | * Note for future: If you are to reload bioset, | |
2431 | * prep-ed requests in the queue may refer | |
2432 | * to bio from the old bioset, so you must walk | |
2433 | * through the queue to unprep. | |
2434 | */ | |
2435 | goto out; | |
c0820cf5 | 2436 | } |
e6ee8c0b | 2437 | |
cbc4e3c1 MS |
2438 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
2439 | ||
e6ee8c0b KU |
2440 | md->io_pool = p->io_pool; |
2441 | p->io_pool = NULL; | |
1ae49ea2 MS |
2442 | md->rq_pool = p->rq_pool; |
2443 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2444 | md->bs = p->bs; |
2445 | p->bs = NULL; | |
4e6e36c3 | 2446 | |
e6ee8c0b | 2447 | out: |
02233342 | 2448 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2449 | dm_table_free_md_mempools(t); |
2450 | } | |
2451 | ||
1da177e4 LT |
2452 | /* |
2453 | * Bind a table to the device. | |
2454 | */ | |
2455 | static void event_callback(void *context) | |
2456 | { | |
7a8c3d3b MA |
2457 | unsigned long flags; |
2458 | LIST_HEAD(uevents); | |
1da177e4 LT |
2459 | struct mapped_device *md = (struct mapped_device *) context; |
2460 | ||
7a8c3d3b MA |
2461 | spin_lock_irqsave(&md->uevent_lock, flags); |
2462 | list_splice_init(&md->uevent_list, &uevents); | |
2463 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2464 | ||
ed9e1982 | 2465 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2466 | |
1da177e4 LT |
2467 | atomic_inc(&md->event_nr); |
2468 | wake_up(&md->eventq); | |
2469 | } | |
2470 | ||
c217649b MS |
2471 | /* |
2472 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2473 | */ | |
4e90188b | 2474 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2475 | { |
4e90188b | 2476 | set_capacity(md->disk, size); |
1da177e4 | 2477 | |
db8fef4f | 2478 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2479 | } |
2480 | ||
042d2a9b AK |
2481 | /* |
2482 | * Returns old map, which caller must destroy. | |
2483 | */ | |
2484 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2485 | struct queue_limits *limits) | |
1da177e4 | 2486 | { |
042d2a9b | 2487 | struct dm_table *old_map; |
165125e1 | 2488 | struct request_queue *q = md->queue; |
1da177e4 LT |
2489 | sector_t size; |
2490 | ||
2491 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2492 | |
2493 | /* | |
2494 | * Wipe any geometry if the size of the table changed. | |
2495 | */ | |
fd2ed4d2 | 2496 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2497 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2498 | ||
32a926da | 2499 | __set_size(md, size); |
d5816876 | 2500 | |
2ca3310e AK |
2501 | dm_table_event_callback(t, event_callback, md); |
2502 | ||
e6ee8c0b KU |
2503 | /* |
2504 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2505 | * for request-based during suspension. So stop it to prevent | |
2506 | * I/O mapping before resume. | |
2507 | * This must be done before setting the queue restrictions, | |
2508 | * because request-based dm may be run just after the setting. | |
2509 | */ | |
16f12266 | 2510 | if (dm_table_request_based(t)) { |
eca7ee6d | 2511 | dm_stop_queue(q); |
16f12266 MS |
2512 | /* |
2513 | * Leverage the fact that request-based DM targets are | |
2514 | * immutable singletons and establish md->immutable_target | |
2515 | * - used to optimize both dm_request_fn and dm_mq_queue_rq | |
2516 | */ | |
2517 | md->immutable_target = dm_table_get_immutable_target(t); | |
2518 | } | |
e6ee8c0b KU |
2519 | |
2520 | __bind_mempools(md, t); | |
2521 | ||
a12f5d48 | 2522 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2523 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2524 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2525 | ||
754c5fc7 | 2526 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2527 | if (old_map) |
2528 | dm_sync_table(md); | |
1da177e4 | 2529 | |
042d2a9b | 2530 | return old_map; |
1da177e4 LT |
2531 | } |
2532 | ||
a7940155 AK |
2533 | /* |
2534 | * Returns unbound table for the caller to free. | |
2535 | */ | |
2536 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2537 | { |
a12f5d48 | 2538 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2539 | |
2540 | if (!map) | |
a7940155 | 2541 | return NULL; |
1da177e4 LT |
2542 | |
2543 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2544 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2545 | dm_sync_table(md); |
a7940155 AK |
2546 | |
2547 | return map; | |
1da177e4 LT |
2548 | } |
2549 | ||
2550 | /* | |
2551 | * Constructor for a new device. | |
2552 | */ | |
2b06cfff | 2553 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2554 | { |
2555 | struct mapped_device *md; | |
2556 | ||
2b06cfff | 2557 | md = alloc_dev(minor); |
1da177e4 LT |
2558 | if (!md) |
2559 | return -ENXIO; | |
2560 | ||
784aae73 MB |
2561 | dm_sysfs_init(md); |
2562 | ||
1da177e4 LT |
2563 | *result = md; |
2564 | return 0; | |
2565 | } | |
2566 | ||
a5664dad MS |
2567 | /* |
2568 | * Functions to manage md->type. | |
2569 | * All are required to hold md->type_lock. | |
2570 | */ | |
2571 | void dm_lock_md_type(struct mapped_device *md) | |
2572 | { | |
2573 | mutex_lock(&md->type_lock); | |
2574 | } | |
2575 | ||
2576 | void dm_unlock_md_type(struct mapped_device *md) | |
2577 | { | |
2578 | mutex_unlock(&md->type_lock); | |
2579 | } | |
2580 | ||
2581 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2582 | { | |
00c4fc3b | 2583 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2584 | md->type = type; |
2585 | } | |
2586 | ||
2587 | unsigned dm_get_md_type(struct mapped_device *md) | |
2588 | { | |
2589 | return md->type; | |
2590 | } | |
2591 | ||
36a0456f AK |
2592 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2593 | { | |
2594 | return md->immutable_target_type; | |
2595 | } | |
2596 | ||
f84cb8a4 MS |
2597 | /* |
2598 | * The queue_limits are only valid as long as you have a reference | |
2599 | * count on 'md'. | |
2600 | */ | |
2601 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2602 | { | |
2603 | BUG_ON(!atomic_read(&md->holders)); | |
2604 | return &md->queue->limits; | |
2605 | } | |
2606 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2607 | ||
eca7ee6d | 2608 | static void dm_old_init_rq_based_worker_thread(struct mapped_device *md) |
bfebd1cd MS |
2609 | { |
2610 | /* Initialize the request-based DM worker thread */ | |
2611 | init_kthread_worker(&md->kworker); | |
2612 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2613 | "kdmwork-%s", dm_device_name(md)); | |
2614 | } | |
2615 | ||
4a0b4ddf | 2616 | /* |
eca7ee6d | 2617 | * Fully initialize a .request_fn request-based queue. |
4a0b4ddf | 2618 | */ |
eca7ee6d | 2619 | static int dm_old_init_request_queue(struct mapped_device *md) |
4a0b4ddf MS |
2620 | { |
2621 | struct request_queue *q = NULL; | |
2622 | ||
4a0b4ddf MS |
2623 | /* Fully initialize the queue */ |
2624 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2625 | if (!q) | |
bfebd1cd | 2626 | return -EINVAL; |
4a0b4ddf | 2627 | |
0ce65797 MS |
2628 | /* disable dm_request_fn's merge heuristic by default */ |
2629 | md->seq_rq_merge_deadline_usecs = 0; | |
2630 | ||
4a0b4ddf | 2631 | md->queue = q; |
eca7ee6d | 2632 | dm_init_normal_md_queue(md); |
4a0b4ddf | 2633 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
eca7ee6d | 2634 | blk_queue_prep_rq(md->queue, dm_old_prep_fn); |
4a0b4ddf | 2635 | |
eca7ee6d | 2636 | dm_old_init_rq_based_worker_thread(md); |
2eb6e1e3 | 2637 | |
4a0b4ddf MS |
2638 | elv_register_queue(md->queue); |
2639 | ||
bfebd1cd MS |
2640 | return 0; |
2641 | } | |
2642 | ||
2643 | static int dm_mq_init_request(void *data, struct request *rq, | |
2644 | unsigned int hctx_idx, unsigned int request_idx, | |
2645 | unsigned int numa_node) | |
2646 | { | |
2647 | struct mapped_device *md = data; | |
2648 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2649 | ||
2650 | /* | |
2651 | * Must initialize md member of tio, otherwise it won't | |
2652 | * be available in dm_mq_queue_rq. | |
2653 | */ | |
2654 | tio->md = md; | |
2655 | ||
2656 | return 0; | |
2657 | } | |
2658 | ||
2659 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2660 | const struct blk_mq_queue_data *bd) | |
2661 | { | |
2662 | struct request *rq = bd->rq; | |
2663 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2664 | struct mapped_device *md = tio->md; | |
16f12266 | 2665 | struct dm_target *ti = md->immutable_target; |
bfebd1cd | 2666 | |
16f12266 MS |
2667 | if (unlikely(!ti)) { |
2668 | int srcu_idx; | |
2669 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
bfebd1cd | 2670 | |
16f12266 | 2671 | ti = dm_table_find_target(map, 0); |
bfebd1cd | 2672 | dm_put_live_table(md, srcu_idx); |
bfebd1cd | 2673 | } |
bfebd1cd MS |
2674 | |
2675 | if (ti->type->busy && ti->type->busy(ti)) | |
2676 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2677 | ||
2678 | dm_start_request(md, rq); | |
2679 | ||
2680 | /* Init tio using md established in .init_request */ | |
2681 | init_tio(tio, rq, md); | |
2682 | ||
02233342 MS |
2683 | /* |
2684 | * Establish tio->ti before queuing work (map_tio_request) | |
2685 | * or making direct call to map_request(). | |
2686 | */ | |
bfebd1cd | 2687 | tio->ti = ti; |
02233342 | 2688 | |
c5248f79 MS |
2689 | /* Direct call is fine since .queue_rq allows allocations */ |
2690 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { | |
2691 | /* Undo dm_start_request() before requeuing */ | |
2692 | rq_end_stats(md, rq); | |
2693 | rq_completed(md, rq_data_dir(rq), false); | |
2694 | return BLK_MQ_RQ_QUEUE_BUSY; | |
02233342 | 2695 | } |
bfebd1cd MS |
2696 | |
2697 | return BLK_MQ_RQ_QUEUE_OK; | |
2698 | } | |
2699 | ||
2700 | static struct blk_mq_ops dm_mq_ops = { | |
2701 | .queue_rq = dm_mq_queue_rq, | |
2702 | .map_queue = blk_mq_map_queue, | |
2703 | .complete = dm_softirq_done, | |
2704 | .init_request = dm_mq_init_request, | |
2705 | }; | |
2706 | ||
eca7ee6d | 2707 | static int dm_mq_init_request_queue(struct mapped_device *md) |
bfebd1cd MS |
2708 | { |
2709 | struct request_queue *q; | |
2710 | int err; | |
2711 | ||
c5248f79 MS |
2712 | if (dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) { |
2713 | DMERR("request-based dm-mq may only be stacked on blk-mq device(s)"); | |
2714 | return -EINVAL; | |
2715 | } | |
2716 | ||
1c357a1e MS |
2717 | md->tag_set = kzalloc(sizeof(struct blk_mq_tag_set), GFP_KERNEL); |
2718 | if (!md->tag_set) | |
2719 | return -ENOMEM; | |
2720 | ||
2721 | md->tag_set->ops = &dm_mq_ops; | |
2722 | md->tag_set->queue_depth = dm_get_blk_mq_queue_depth(); | |
2723 | md->tag_set->numa_node = NUMA_NO_NODE; | |
2724 | md->tag_set->flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2725 | md->tag_set->nr_hw_queues = dm_get_blk_mq_nr_hw_queues(); | |
2726 | md->tag_set->driver_data = md; | |
2727 | ||
2728 | md->tag_set->cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd | 2729 | |
1c357a1e | 2730 | err = blk_mq_alloc_tag_set(md->tag_set); |
bfebd1cd | 2731 | if (err) |
1c357a1e | 2732 | goto out_kfree_tag_set; |
bfebd1cd | 2733 | |
1c357a1e | 2734 | q = blk_mq_init_allocated_queue(md->tag_set, md->queue); |
bfebd1cd MS |
2735 | if (IS_ERR(q)) { |
2736 | err = PTR_ERR(q); | |
2737 | goto out_tag_set; | |
2738 | } | |
2739 | md->queue = q; | |
2740 | dm_init_md_queue(md); | |
2741 | ||
2742 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2743 | blk_mq_register_disk(md->disk); | |
2744 | ||
bfebd1cd MS |
2745 | return 0; |
2746 | ||
2747 | out_tag_set: | |
1c357a1e MS |
2748 | blk_mq_free_tag_set(md->tag_set); |
2749 | out_kfree_tag_set: | |
2750 | kfree(md->tag_set); | |
2751 | ||
bfebd1cd | 2752 | return err; |
4a0b4ddf MS |
2753 | } |
2754 | ||
4e6e36c3 MS |
2755 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2756 | { | |
2757 | if (type == DM_TYPE_BIO_BASED) | |
2758 | return type; | |
2759 | ||
2760 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2761 | } | |
2762 | ||
4a0b4ddf MS |
2763 | /* |
2764 | * Setup the DM device's queue based on md's type | |
2765 | */ | |
2766 | int dm_setup_md_queue(struct mapped_device *md) | |
2767 | { | |
bfebd1cd | 2768 | int r; |
17e149b8 | 2769 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2770 | |
2771 | switch (md_type) { | |
2772 | case DM_TYPE_REQUEST_BASED: | |
eca7ee6d | 2773 | r = dm_old_init_request_queue(md); |
bfebd1cd | 2774 | if (r) { |
eca7ee6d | 2775 | DMERR("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2776 | return r; |
ff36ab34 | 2777 | } |
bfebd1cd MS |
2778 | break; |
2779 | case DM_TYPE_MQ_REQUEST_BASED: | |
eca7ee6d | 2780 | r = dm_mq_init_request_queue(md); |
bfebd1cd | 2781 | if (r) { |
eca7ee6d | 2782 | DMERR("Cannot initialize queue for request-based dm-mq mapped device"); |
bfebd1cd MS |
2783 | return r; |
2784 | } | |
2785 | break; | |
2786 | case DM_TYPE_BIO_BASED: | |
eca7ee6d | 2787 | dm_init_normal_md_queue(md); |
ff36ab34 | 2788 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
2789 | /* |
2790 | * DM handles splitting bios as needed. Free the bio_split bioset | |
2791 | * since it won't be used (saves 1 process per bio-based DM device). | |
2792 | */ | |
2793 | bioset_free(md->queue->bio_split); | |
2794 | md->queue->bio_split = NULL; | |
bfebd1cd | 2795 | break; |
4a0b4ddf MS |
2796 | } |
2797 | ||
2798 | return 0; | |
2799 | } | |
2800 | ||
2bec1f4a | 2801 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2802 | { |
2803 | struct mapped_device *md; | |
1da177e4 LT |
2804 | unsigned minor = MINOR(dev); |
2805 | ||
2806 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2807 | return NULL; | |
2808 | ||
f32c10b0 | 2809 | spin_lock(&_minor_lock); |
1da177e4 LT |
2810 | |
2811 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2812 | if (md) { |
2813 | if ((md == MINOR_ALLOCED || | |
2814 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2815 | dm_deleting_md(md) || | |
2816 | test_bit(DMF_FREEING, &md->flags))) { | |
2817 | md = NULL; | |
2818 | goto out; | |
2819 | } | |
2820 | dm_get(md); | |
fba9f90e | 2821 | } |
1da177e4 | 2822 | |
fba9f90e | 2823 | out: |
f32c10b0 | 2824 | spin_unlock(&_minor_lock); |
1da177e4 | 2825 | |
637842cf DT |
2826 | return md; |
2827 | } | |
3cf2e4ba | 2828 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2829 | |
9ade92a9 | 2830 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2831 | { |
9ade92a9 | 2832 | return md->interface_ptr; |
1da177e4 LT |
2833 | } |
2834 | ||
2835 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2836 | { | |
2837 | md->interface_ptr = ptr; | |
2838 | } | |
2839 | ||
2840 | void dm_get(struct mapped_device *md) | |
2841 | { | |
2842 | atomic_inc(&md->holders); | |
3f77316d | 2843 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2844 | } |
2845 | ||
09ee96b2 MP |
2846 | int dm_hold(struct mapped_device *md) |
2847 | { | |
2848 | spin_lock(&_minor_lock); | |
2849 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2850 | spin_unlock(&_minor_lock); | |
2851 | return -EBUSY; | |
2852 | } | |
2853 | dm_get(md); | |
2854 | spin_unlock(&_minor_lock); | |
2855 | return 0; | |
2856 | } | |
2857 | EXPORT_SYMBOL_GPL(dm_hold); | |
2858 | ||
72d94861 AK |
2859 | const char *dm_device_name(struct mapped_device *md) |
2860 | { | |
2861 | return md->name; | |
2862 | } | |
2863 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2864 | ||
3f77316d | 2865 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2866 | { |
1134e5ae | 2867 | struct dm_table *map; |
83d5e5b0 | 2868 | int srcu_idx; |
1da177e4 | 2869 | |
3f77316d | 2870 | might_sleep(); |
fba9f90e | 2871 | |
63a4f065 | 2872 | spin_lock(&_minor_lock); |
3f77316d KU |
2873 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2874 | set_bit(DMF_FREEING, &md->flags); | |
2875 | spin_unlock(&_minor_lock); | |
2876 | ||
02233342 | 2877 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2878 | flush_kthread_worker(&md->kworker); |
2879 | ||
ab7c7bb6 MP |
2880 | /* |
2881 | * Take suspend_lock so that presuspend and postsuspend methods | |
2882 | * do not race with internal suspend. | |
2883 | */ | |
2884 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2885 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2886 | if (!dm_suspended_md(md)) { |
2887 | dm_table_presuspend_targets(map); | |
2888 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2889 | } |
83d5e5b0 MP |
2890 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2891 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2892 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2893 | |
3f77316d KU |
2894 | /* |
2895 | * Rare, but there may be I/O requests still going to complete, | |
2896 | * for example. Wait for all references to disappear. | |
2897 | * No one should increment the reference count of the mapped_device, | |
2898 | * after the mapped_device state becomes DMF_FREEING. | |
2899 | */ | |
2900 | if (wait) | |
2901 | while (atomic_read(&md->holders)) | |
2902 | msleep(1); | |
2903 | else if (atomic_read(&md->holders)) | |
2904 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2905 | dm_device_name(md), atomic_read(&md->holders)); | |
2906 | ||
2907 | dm_sysfs_exit(md); | |
3f77316d KU |
2908 | dm_table_destroy(__unbind(md)); |
2909 | free_dev(md); | |
2910 | } | |
2911 | ||
2912 | void dm_destroy(struct mapped_device *md) | |
2913 | { | |
2914 | __dm_destroy(md, true); | |
2915 | } | |
2916 | ||
2917 | void dm_destroy_immediate(struct mapped_device *md) | |
2918 | { | |
2919 | __dm_destroy(md, false); | |
2920 | } | |
2921 | ||
2922 | void dm_put(struct mapped_device *md) | |
2923 | { | |
2924 | atomic_dec(&md->holders); | |
1da177e4 | 2925 | } |
79eb885c | 2926 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2927 | |
401600df | 2928 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2929 | { |
2930 | int r = 0; | |
b44ebeb0 MP |
2931 | DECLARE_WAITQUEUE(wait, current); |
2932 | ||
b44ebeb0 | 2933 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2934 | |
2935 | while (1) { | |
401600df | 2936 | set_current_state(interruptible); |
46125c1c | 2937 | |
b4324fee | 2938 | if (!md_in_flight(md)) |
46125c1c MB |
2939 | break; |
2940 | ||
401600df MP |
2941 | if (interruptible == TASK_INTERRUPTIBLE && |
2942 | signal_pending(current)) { | |
46125c1c MB |
2943 | r = -EINTR; |
2944 | break; | |
2945 | } | |
2946 | ||
2947 | io_schedule(); | |
2948 | } | |
2949 | set_current_state(TASK_RUNNING); | |
2950 | ||
b44ebeb0 MP |
2951 | remove_wait_queue(&md->wait, &wait); |
2952 | ||
46125c1c MB |
2953 | return r; |
2954 | } | |
2955 | ||
1da177e4 LT |
2956 | /* |
2957 | * Process the deferred bios | |
2958 | */ | |
ef208587 | 2959 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2960 | { |
ef208587 MP |
2961 | struct mapped_device *md = container_of(work, struct mapped_device, |
2962 | work); | |
6d6f10df | 2963 | struct bio *c; |
83d5e5b0 MP |
2964 | int srcu_idx; |
2965 | struct dm_table *map; | |
1da177e4 | 2966 | |
83d5e5b0 | 2967 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2968 | |
3b00b203 | 2969 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2970 | spin_lock_irq(&md->deferred_lock); |
2971 | c = bio_list_pop(&md->deferred); | |
2972 | spin_unlock_irq(&md->deferred_lock); | |
2973 | ||
6a8736d1 | 2974 | if (!c) |
df12ee99 | 2975 | break; |
022c2611 | 2976 | |
e6ee8c0b KU |
2977 | if (dm_request_based(md)) |
2978 | generic_make_request(c); | |
6a8736d1 | 2979 | else |
83d5e5b0 | 2980 | __split_and_process_bio(md, map, c); |
022c2611 | 2981 | } |
73d410c0 | 2982 | |
83d5e5b0 | 2983 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2984 | } |
2985 | ||
9a1fb464 | 2986 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2987 | { |
3b00b203 | 2988 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2989 | smp_mb__after_atomic(); |
53d5914f | 2990 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2991 | } |
2992 | ||
1da177e4 | 2993 | /* |
042d2a9b | 2994 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2995 | */ |
042d2a9b | 2996 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2997 | { |
87eb5b21 | 2998 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2999 | struct queue_limits limits; |
042d2a9b | 3000 | int r; |
1da177e4 | 3001 | |
e61290a4 | 3002 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3003 | |
3004 | /* device must be suspended */ | |
4f186f8b | 3005 | if (!dm_suspended_md(md)) |
93c534ae | 3006 | goto out; |
1da177e4 | 3007 | |
3ae70656 MS |
3008 | /* |
3009 | * If the new table has no data devices, retain the existing limits. | |
3010 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3011 | * then new I/O is queued based on these limits, and then some paths | |
3012 | * reappear. | |
3013 | */ | |
3014 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3015 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3016 | if (live_map) |
3017 | limits = md->queue->limits; | |
83d5e5b0 | 3018 | dm_put_live_table_fast(md); |
3ae70656 MS |
3019 | } |
3020 | ||
87eb5b21 MC |
3021 | if (!live_map) { |
3022 | r = dm_calculate_queue_limits(table, &limits); | |
3023 | if (r) { | |
3024 | map = ERR_PTR(r); | |
3025 | goto out; | |
3026 | } | |
042d2a9b | 3027 | } |
754c5fc7 | 3028 | |
042d2a9b | 3029 | map = __bind(md, table, &limits); |
1da177e4 | 3030 | |
93c534ae | 3031 | out: |
e61290a4 | 3032 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3033 | return map; |
1da177e4 LT |
3034 | } |
3035 | ||
3036 | /* | |
3037 | * Functions to lock and unlock any filesystem running on the | |
3038 | * device. | |
3039 | */ | |
2ca3310e | 3040 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3041 | { |
e39e2e95 | 3042 | int r; |
1da177e4 LT |
3043 | |
3044 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3045 | |
db8fef4f | 3046 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3047 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3048 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3049 | md->frozen_sb = NULL; |
3050 | return r; | |
dfbe03f6 AK |
3051 | } |
3052 | ||
aa8d7c2f AK |
3053 | set_bit(DMF_FROZEN, &md->flags); |
3054 | ||
1da177e4 LT |
3055 | return 0; |
3056 | } | |
3057 | ||
2ca3310e | 3058 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3059 | { |
aa8d7c2f AK |
3060 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3061 | return; | |
3062 | ||
db8fef4f | 3063 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3064 | md->frozen_sb = NULL; |
aa8d7c2f | 3065 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3066 | } |
3067 | ||
3068 | /* | |
ffcc3936 MS |
3069 | * If __dm_suspend returns 0, the device is completely quiescent |
3070 | * now. There is no request-processing activity. All new requests | |
3071 | * are being added to md->deferred list. | |
cec47e3d | 3072 | * |
ffcc3936 | 3073 | * Caller must hold md->suspend_lock |
cec47e3d | 3074 | */ |
ffcc3936 MS |
3075 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3076 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3077 | { |
ffcc3936 MS |
3078 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3079 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3080 | int r; | |
1da177e4 | 3081 | |
2e93ccc1 KU |
3082 | /* |
3083 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3084 | * This flag is cleared before dm_suspend returns. | |
3085 | */ | |
3086 | if (noflush) | |
3087 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3088 | ||
d67ee213 MS |
3089 | /* |
3090 | * This gets reverted if there's an error later and the targets | |
3091 | * provide the .presuspend_undo hook. | |
3092 | */ | |
cf222b37 AK |
3093 | dm_table_presuspend_targets(map); |
3094 | ||
32a926da | 3095 | /* |
9f518b27 KU |
3096 | * Flush I/O to the device. |
3097 | * Any I/O submitted after lock_fs() may not be flushed. | |
3098 | * noflush takes precedence over do_lockfs. | |
3099 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3100 | */ |
3101 | if (!noflush && do_lockfs) { | |
3102 | r = lock_fs(md); | |
d67ee213 MS |
3103 | if (r) { |
3104 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3105 | return r; |
d67ee213 | 3106 | } |
aa8d7c2f | 3107 | } |
1da177e4 LT |
3108 | |
3109 | /* | |
3b00b203 MP |
3110 | * Here we must make sure that no processes are submitting requests |
3111 | * to target drivers i.e. no one may be executing | |
3112 | * __split_and_process_bio. This is called from dm_request and | |
3113 | * dm_wq_work. | |
3114 | * | |
3115 | * To get all processes out of __split_and_process_bio in dm_request, | |
3116 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3117 | * __split_and_process_bio from dm_request and quiesce the thread |
3118 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3119 | * flush_workqueue(md->wq). | |
1da177e4 | 3120 | */ |
1eb787ec | 3121 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3122 | if (map) |
3123 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3124 | |
d0bcb878 | 3125 | /* |
29e4013d TH |
3126 | * Stop md->queue before flushing md->wq in case request-based |
3127 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3128 | */ |
2eb6e1e3 | 3129 | if (dm_request_based(md)) { |
eca7ee6d | 3130 | dm_stop_queue(md->queue); |
02233342 MS |
3131 | if (md->kworker_task) |
3132 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3133 | } |
cec47e3d | 3134 | |
d0bcb878 KU |
3135 | flush_workqueue(md->wq); |
3136 | ||
1da177e4 | 3137 | /* |
3b00b203 MP |
3138 | * At this point no more requests are entering target request routines. |
3139 | * We call dm_wait_for_completion to wait for all existing requests | |
3140 | * to finish. | |
1da177e4 | 3141 | */ |
ffcc3936 | 3142 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3143 | |
6d6f10df | 3144 | if (noflush) |
022c2611 | 3145 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3146 | if (map) |
3147 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3148 | |
1da177e4 | 3149 | /* were we interrupted ? */ |
46125c1c | 3150 | if (r < 0) { |
9a1fb464 | 3151 | dm_queue_flush(md); |
73d410c0 | 3152 | |
cec47e3d | 3153 | if (dm_request_based(md)) |
eca7ee6d | 3154 | dm_start_queue(md->queue); |
cec47e3d | 3155 | |
2ca3310e | 3156 | unlock_fs(md); |
d67ee213 | 3157 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3158 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3159 | } |
1da177e4 | 3160 | |
ffcc3936 MS |
3161 | return r; |
3162 | } | |
3163 | ||
3164 | /* | |
3165 | * We need to be able to change a mapping table under a mounted | |
3166 | * filesystem. For example we might want to move some data in | |
3167 | * the background. Before the table can be swapped with | |
3168 | * dm_bind_table, dm_suspend must be called to flush any in | |
3169 | * flight bios and ensure that any further io gets deferred. | |
3170 | */ | |
3171 | /* | |
3172 | * Suspend mechanism in request-based dm. | |
3173 | * | |
3174 | * 1. Flush all I/Os by lock_fs() if needed. | |
3175 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3176 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3177 | * | |
3178 | * To abort suspend, start the request_queue. | |
3179 | */ | |
3180 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3181 | { | |
3182 | struct dm_table *map = NULL; | |
3183 | int r = 0; | |
3184 | ||
3185 | retry: | |
3186 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3187 | ||
3188 | if (dm_suspended_md(md)) { | |
3189 | r = -EINVAL; | |
3190 | goto out_unlock; | |
3191 | } | |
3192 | ||
3193 | if (dm_suspended_internally_md(md)) { | |
3194 | /* already internally suspended, wait for internal resume */ | |
3195 | mutex_unlock(&md->suspend_lock); | |
3196 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3197 | if (r) | |
3198 | return r; | |
3199 | goto retry; | |
3200 | } | |
3201 | ||
a12f5d48 | 3202 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3203 | |
3204 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3205 | if (r) | |
3206 | goto out_unlock; | |
3b00b203 | 3207 | |
2ca3310e | 3208 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3209 | |
4d4471cb KU |
3210 | dm_table_postsuspend_targets(map); |
3211 | ||
d287483d | 3212 | out_unlock: |
e61290a4 | 3213 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3214 | return r; |
1da177e4 LT |
3215 | } |
3216 | ||
ffcc3936 MS |
3217 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3218 | { | |
3219 | if (map) { | |
3220 | int r = dm_table_resume_targets(map); | |
3221 | if (r) | |
3222 | return r; | |
3223 | } | |
3224 | ||
3225 | dm_queue_flush(md); | |
3226 | ||
3227 | /* | |
3228 | * Flushing deferred I/Os must be done after targets are resumed | |
3229 | * so that mapping of targets can work correctly. | |
3230 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3231 | */ | |
3232 | if (dm_request_based(md)) | |
eca7ee6d | 3233 | dm_start_queue(md->queue); |
ffcc3936 MS |
3234 | |
3235 | unlock_fs(md); | |
3236 | ||
3237 | return 0; | |
3238 | } | |
3239 | ||
1da177e4 LT |
3240 | int dm_resume(struct mapped_device *md) |
3241 | { | |
cf222b37 | 3242 | int r = -EINVAL; |
cf222b37 | 3243 | struct dm_table *map = NULL; |
1da177e4 | 3244 | |
ffcc3936 MS |
3245 | retry: |
3246 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3247 | ||
4f186f8b | 3248 | if (!dm_suspended_md(md)) |
cf222b37 | 3249 | goto out; |
cf222b37 | 3250 | |
ffcc3936 MS |
3251 | if (dm_suspended_internally_md(md)) { |
3252 | /* already internally suspended, wait for internal resume */ | |
3253 | mutex_unlock(&md->suspend_lock); | |
3254 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3255 | if (r) | |
3256 | return r; | |
3257 | goto retry; | |
3258 | } | |
3259 | ||
a12f5d48 | 3260 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3261 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3262 | goto out; |
1da177e4 | 3263 | |
ffcc3936 | 3264 | r = __dm_resume(md, map); |
8757b776 MB |
3265 | if (r) |
3266 | goto out; | |
2ca3310e | 3267 | |
2ca3310e AK |
3268 | clear_bit(DMF_SUSPENDED, &md->flags); |
3269 | ||
cf222b37 AK |
3270 | r = 0; |
3271 | out: | |
e61290a4 | 3272 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3273 | |
cf222b37 | 3274 | return r; |
1da177e4 LT |
3275 | } |
3276 | ||
fd2ed4d2 MP |
3277 | /* |
3278 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3279 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3280 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3281 | */ |
3282 | ||
ffcc3936 | 3283 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3284 | { |
ffcc3936 MS |
3285 | struct dm_table *map = NULL; |
3286 | ||
96b26c8c | 3287 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3288 | return; /* nested internal suspend */ |
3289 | ||
3290 | if (dm_suspended_md(md)) { | |
3291 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3292 | return; /* nest suspend */ | |
3293 | } | |
3294 | ||
a12f5d48 | 3295 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3296 | |
3297 | /* | |
3298 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3299 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3300 | * would require changing .presuspend to return an error -- avoid this | |
3301 | * until there is a need for more elaborate variants of internal suspend. | |
3302 | */ | |
3303 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3304 | ||
3305 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3306 | ||
3307 | dm_table_postsuspend_targets(map); | |
3308 | } | |
3309 | ||
3310 | static void __dm_internal_resume(struct mapped_device *md) | |
3311 | { | |
96b26c8c MP |
3312 | BUG_ON(!md->internal_suspend_count); |
3313 | ||
3314 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3315 | return; /* resume from nested internal suspend */ |
3316 | ||
fd2ed4d2 | 3317 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3318 | goto done; /* resume from nested suspend */ |
3319 | ||
3320 | /* | |
3321 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3322 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3323 | */ | |
3324 | (void) __dm_resume(md, NULL); | |
3325 | ||
3326 | done: | |
3327 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3328 | smp_mb__after_atomic(); | |
3329 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3330 | } | |
3331 | ||
3332 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3333 | { | |
3334 | mutex_lock(&md->suspend_lock); | |
3335 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3336 | mutex_unlock(&md->suspend_lock); | |
3337 | } | |
3338 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3339 | ||
3340 | void dm_internal_resume(struct mapped_device *md) | |
3341 | { | |
3342 | mutex_lock(&md->suspend_lock); | |
3343 | __dm_internal_resume(md); | |
3344 | mutex_unlock(&md->suspend_lock); | |
3345 | } | |
3346 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3347 | ||
3348 | /* | |
3349 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3350 | * which prevents interaction with userspace-driven suspend. | |
3351 | */ | |
3352 | ||
3353 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3354 | { | |
3355 | mutex_lock(&md->suspend_lock); | |
3356 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3357 | return; |
3358 | ||
3359 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3360 | synchronize_srcu(&md->io_barrier); | |
3361 | flush_workqueue(md->wq); | |
3362 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3363 | } | |
b735fede | 3364 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3365 | |
ffcc3936 | 3366 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3367 | { |
ffcc3936 | 3368 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3369 | goto done; |
3370 | ||
3371 | dm_queue_flush(md); | |
3372 | ||
3373 | done: | |
3374 | mutex_unlock(&md->suspend_lock); | |
3375 | } | |
b735fede | 3376 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3377 | |
1da177e4 LT |
3378 | /*----------------------------------------------------------------- |
3379 | * Event notification. | |
3380 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3381 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3382 | unsigned cookie) |
69267a30 | 3383 | { |
60935eb2 MB |
3384 | char udev_cookie[DM_COOKIE_LENGTH]; |
3385 | char *envp[] = { udev_cookie, NULL }; | |
3386 | ||
3387 | if (!cookie) | |
3abf85b5 | 3388 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3389 | else { |
3390 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3391 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3392 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3393 | action, envp); | |
60935eb2 | 3394 | } |
69267a30 AK |
3395 | } |
3396 | ||
7a8c3d3b MA |
3397 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3398 | { | |
3399 | return atomic_add_return(1, &md->uevent_seq); | |
3400 | } | |
3401 | ||
1da177e4 LT |
3402 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3403 | { | |
3404 | return atomic_read(&md->event_nr); | |
3405 | } | |
3406 | ||
3407 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3408 | { | |
3409 | return wait_event_interruptible(md->eventq, | |
3410 | (event_nr != atomic_read(&md->event_nr))); | |
3411 | } | |
3412 | ||
7a8c3d3b MA |
3413 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3414 | { | |
3415 | unsigned long flags; | |
3416 | ||
3417 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3418 | list_add(elist, &md->uevent_list); | |
3419 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3420 | } | |
3421 | ||
1da177e4 LT |
3422 | /* |
3423 | * The gendisk is only valid as long as you have a reference | |
3424 | * count on 'md'. | |
3425 | */ | |
3426 | struct gendisk *dm_disk(struct mapped_device *md) | |
3427 | { | |
3428 | return md->disk; | |
3429 | } | |
65ff5b7d | 3430 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3431 | |
784aae73 MB |
3432 | struct kobject *dm_kobject(struct mapped_device *md) |
3433 | { | |
2995fa78 | 3434 | return &md->kobj_holder.kobj; |
784aae73 MB |
3435 | } |
3436 | ||
784aae73 MB |
3437 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3438 | { | |
3439 | struct mapped_device *md; | |
3440 | ||
2995fa78 | 3441 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3442 | |
4d89b7b4 | 3443 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3444 | dm_deleting_md(md)) |
4d89b7b4 MB |
3445 | return NULL; |
3446 | ||
784aae73 MB |
3447 | dm_get(md); |
3448 | return md; | |
3449 | } | |
3450 | ||
4f186f8b | 3451 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3452 | { |
3453 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3454 | } | |
3455 | ||
ffcc3936 MS |
3456 | int dm_suspended_internally_md(struct mapped_device *md) |
3457 | { | |
3458 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3459 | } | |
3460 | ||
2c140a24 MP |
3461 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3462 | { | |
3463 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3464 | } | |
3465 | ||
64dbce58 KU |
3466 | int dm_suspended(struct dm_target *ti) |
3467 | { | |
ecdb2e25 | 3468 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3469 | } |
3470 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3471 | ||
2e93ccc1 KU |
3472 | int dm_noflush_suspending(struct dm_target *ti) |
3473 | { | |
ecdb2e25 | 3474 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3475 | } |
3476 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3477 | ||
78d8e58a MS |
3478 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3479 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3480 | { |
78d8e58a MS |
3481 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3482 | struct kmem_cache *cachep = NULL; | |
3483 | unsigned int pool_size = 0; | |
5f015204 | 3484 | unsigned int front_pad; |
e6ee8c0b KU |
3485 | |
3486 | if (!pools) | |
4e6e36c3 | 3487 | return NULL; |
e6ee8c0b | 3488 | |
78d8e58a | 3489 | type = filter_md_type(type, md); |
17e149b8 | 3490 | |
78d8e58a MS |
3491 | switch (type) { |
3492 | case DM_TYPE_BIO_BASED: | |
3493 | cachep = _io_cache; | |
3494 | pool_size = dm_get_reserved_bio_based_ios(); | |
3495 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
3496 | break; | |
3497 | case DM_TYPE_REQUEST_BASED: | |
3498 | cachep = _rq_tio_cache; | |
3499 | pool_size = dm_get_reserved_rq_based_ios(); | |
3500 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3501 | if (!pools->rq_pool) | |
3502 | goto out; | |
3503 | /* fall through to setup remaining rq-based pools */ | |
3504 | case DM_TYPE_MQ_REQUEST_BASED: | |
3505 | if (!pool_size) | |
3506 | pool_size = dm_get_reserved_rq_based_ios(); | |
3507 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
3508 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3509 | WARN_ON(per_bio_data_size != 0); | |
3510 | break; | |
3511 | default: | |
3512 | BUG(); | |
3513 | } | |
3514 | ||
3515 | if (cachep) { | |
3516 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3517 | if (!pools->io_pool) | |
3518 | goto out; | |
3519 | } | |
e6ee8c0b | 3520 | |
3d8aab2d | 3521 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3522 | if (!pools->bs) |
5f015204 | 3523 | goto out; |
e6ee8c0b | 3524 | |
a91a2785 | 3525 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3526 | goto out; |
a91a2785 | 3527 | |
e6ee8c0b | 3528 | return pools; |
5f1b670d | 3529 | |
5f1b670d CH |
3530 | out: |
3531 | dm_free_md_mempools(pools); | |
78d8e58a | 3532 | |
4e6e36c3 | 3533 | return NULL; |
e6ee8c0b KU |
3534 | } |
3535 | ||
3536 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3537 | { | |
3538 | if (!pools) | |
3539 | return; | |
3540 | ||
6f65985e JL |
3541 | mempool_destroy(pools->io_pool); |
3542 | mempool_destroy(pools->rq_pool); | |
1ae49ea2 | 3543 | |
e6ee8c0b KU |
3544 | if (pools->bs) |
3545 | bioset_free(pools->bs); | |
3546 | ||
3547 | kfree(pools); | |
3548 | } | |
3549 | ||
71cdb697 | 3550 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, |
956a4025 | 3551 | u32 flags) |
71cdb697 CH |
3552 | { |
3553 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3554 | const struct pr_ops *ops; | |
71cdb697 | 3555 | fmode_t mode; |
956a4025 | 3556 | int r; |
71cdb697 | 3557 | |
956a4025 | 3558 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3559 | if (r < 0) |
3560 | return r; | |
3561 | ||
3562 | ops = bdev->bd_disk->fops->pr_ops; | |
3563 | if (ops && ops->pr_register) | |
3564 | r = ops->pr_register(bdev, old_key, new_key, flags); | |
3565 | else | |
3566 | r = -EOPNOTSUPP; | |
3567 | ||
956a4025 | 3568 | bdput(bdev); |
71cdb697 CH |
3569 | return r; |
3570 | } | |
3571 | ||
3572 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
956a4025 | 3573 | u32 flags) |
71cdb697 CH |
3574 | { |
3575 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3576 | const struct pr_ops *ops; | |
71cdb697 | 3577 | fmode_t mode; |
956a4025 | 3578 | int r; |
71cdb697 | 3579 | |
956a4025 | 3580 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3581 | if (r < 0) |
3582 | return r; | |
3583 | ||
3584 | ops = bdev->bd_disk->fops->pr_ops; | |
3585 | if (ops && ops->pr_reserve) | |
3586 | r = ops->pr_reserve(bdev, key, type, flags); | |
3587 | else | |
3588 | r = -EOPNOTSUPP; | |
3589 | ||
956a4025 | 3590 | bdput(bdev); |
71cdb697 CH |
3591 | return r; |
3592 | } | |
3593 | ||
3594 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3595 | { | |
3596 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3597 | const struct pr_ops *ops; | |
71cdb697 | 3598 | fmode_t mode; |
956a4025 | 3599 | int r; |
71cdb697 | 3600 | |
956a4025 | 3601 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3602 | if (r < 0) |
3603 | return r; | |
3604 | ||
3605 | ops = bdev->bd_disk->fops->pr_ops; | |
3606 | if (ops && ops->pr_release) | |
3607 | r = ops->pr_release(bdev, key, type); | |
3608 | else | |
3609 | r = -EOPNOTSUPP; | |
3610 | ||
956a4025 | 3611 | bdput(bdev); |
71cdb697 CH |
3612 | return r; |
3613 | } | |
3614 | ||
3615 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
956a4025 | 3616 | enum pr_type type, bool abort) |
71cdb697 CH |
3617 | { |
3618 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3619 | const struct pr_ops *ops; | |
71cdb697 | 3620 | fmode_t mode; |
956a4025 | 3621 | int r; |
71cdb697 | 3622 | |
956a4025 | 3623 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3624 | if (r < 0) |
3625 | return r; | |
3626 | ||
3627 | ops = bdev->bd_disk->fops->pr_ops; | |
3628 | if (ops && ops->pr_preempt) | |
3629 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3630 | else | |
3631 | r = -EOPNOTSUPP; | |
3632 | ||
956a4025 | 3633 | bdput(bdev); |
71cdb697 CH |
3634 | return r; |
3635 | } | |
3636 | ||
3637 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3638 | { | |
3639 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3640 | const struct pr_ops *ops; | |
71cdb697 | 3641 | fmode_t mode; |
956a4025 | 3642 | int r; |
71cdb697 | 3643 | |
956a4025 | 3644 | r = dm_grab_bdev_for_ioctl(md, &bdev, &mode); |
71cdb697 CH |
3645 | if (r < 0) |
3646 | return r; | |
3647 | ||
3648 | ops = bdev->bd_disk->fops->pr_ops; | |
3649 | if (ops && ops->pr_clear) | |
3650 | r = ops->pr_clear(bdev, key); | |
3651 | else | |
3652 | r = -EOPNOTSUPP; | |
3653 | ||
956a4025 | 3654 | bdput(bdev); |
71cdb697 CH |
3655 | return r; |
3656 | } | |
3657 | ||
3658 | static const struct pr_ops dm_pr_ops = { | |
3659 | .pr_register = dm_pr_register, | |
3660 | .pr_reserve = dm_pr_reserve, | |
3661 | .pr_release = dm_pr_release, | |
3662 | .pr_preempt = dm_pr_preempt, | |
3663 | .pr_clear = dm_pr_clear, | |
3664 | }; | |
3665 | ||
83d5cde4 | 3666 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3667 | .open = dm_blk_open, |
3668 | .release = dm_blk_close, | |
aa129a22 | 3669 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3670 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3671 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3672 | .owner = THIS_MODULE |
3673 | }; | |
3674 | ||
1da177e4 LT |
3675 | /* |
3676 | * module hooks | |
3677 | */ | |
3678 | module_init(dm_init); | |
3679 | module_exit(dm_exit); | |
3680 | ||
3681 | module_param(major, uint, 0); | |
3682 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3683 | |
e8603136 MS |
3684 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3685 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3686 | ||
f4790826 MS |
3687 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3688 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3689 | ||
17e149b8 MS |
3690 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3691 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3692 | ||
faad87df MS |
3693 | module_param(dm_mq_nr_hw_queues, uint, S_IRUGO | S_IWUSR); |
3694 | MODULE_PARM_DESC(dm_mq_nr_hw_queues, "Number of hardware queues for request-based dm-mq devices"); | |
3695 | ||
3696 | module_param(dm_mq_queue_depth, uint, S_IRUGO | S_IWUSR); | |
3697 | MODULE_PARM_DESC(dm_mq_queue_depth, "Queue depth for request-based dm-mq devices"); | |
3698 | ||
1da177e4 LT |
3699 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3700 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3701 | MODULE_LICENSE("GPL"); |