]>
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 | ||
cec47e3d KU |
109 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
110 | { | |
111 | if (rq && rq->end_io_data) | |
112 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
113 | return NULL; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
116 | ||
ba61fdd1 JM |
117 | #define MINOR_ALLOCED ((void *)-1) |
118 | ||
1da177e4 LT |
119 | /* |
120 | * Bits for the md->flags field. | |
121 | */ | |
1eb787ec | 122 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 123 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 124 | #define DMF_FROZEN 2 |
fba9f90e | 125 | #define DMF_FREEING 3 |
5c6bd75d | 126 | #define DMF_DELETING 4 |
2e93ccc1 | 127 | #define DMF_NOFLUSH_SUSPENDING 5 |
8ae12666 KO |
128 | #define DMF_DEFERRED_REMOVE 6 |
129 | #define DMF_SUSPENDED_INTERNALLY 7 | |
1da177e4 | 130 | |
83d5e5b0 MP |
131 | /* |
132 | * A dummy definition to make RCU happy. | |
133 | * struct dm_table should never be dereferenced in this file. | |
134 | */ | |
135 | struct dm_table { | |
136 | int undefined__; | |
137 | }; | |
138 | ||
304f3f6a MB |
139 | /* |
140 | * Work processed by per-device workqueue. | |
141 | */ | |
1da177e4 | 142 | struct mapped_device { |
83d5e5b0 | 143 | struct srcu_struct io_barrier; |
e61290a4 | 144 | struct mutex suspend_lock; |
1da177e4 | 145 | atomic_t holders; |
5c6bd75d | 146 | atomic_t open_count; |
1da177e4 | 147 | |
2a7faeb1 MP |
148 | /* |
149 | * The current mapping. | |
150 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
151 | * dereference. | |
152 | */ | |
6fa99520 | 153 | struct dm_table __rcu *map; |
2a7faeb1 | 154 | |
86f1152b BM |
155 | struct list_head table_devices; |
156 | struct mutex table_devices_lock; | |
157 | ||
1da177e4 LT |
158 | unsigned long flags; |
159 | ||
165125e1 | 160 | struct request_queue *queue; |
a5664dad | 161 | unsigned type; |
4a0b4ddf | 162 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
163 | struct mutex type_lock; |
164 | ||
36a0456f AK |
165 | struct target_type *immutable_target_type; |
166 | ||
1da177e4 | 167 | struct gendisk *disk; |
7e51f257 | 168 | char name[16]; |
1da177e4 LT |
169 | |
170 | void *interface_ptr; | |
171 | ||
172 | /* | |
173 | * A list of ios that arrived while we were suspended. | |
174 | */ | |
316d315b | 175 | atomic_t pending[2]; |
1da177e4 | 176 | wait_queue_head_t wait; |
53d5914f | 177 | struct work_struct work; |
74859364 | 178 | struct bio_list deferred; |
022c2611 | 179 | spinlock_t deferred_lock; |
1da177e4 | 180 | |
af7e466a | 181 | /* |
29e4013d | 182 | * Processing queue (flush) |
304f3f6a MB |
183 | */ |
184 | struct workqueue_struct *wq; | |
185 | ||
1da177e4 LT |
186 | /* |
187 | * io objects are allocated from here. | |
188 | */ | |
189 | mempool_t *io_pool; | |
1ae49ea2 | 190 | mempool_t *rq_pool; |
1da177e4 | 191 | |
9faf400f SB |
192 | struct bio_set *bs; |
193 | ||
1da177e4 LT |
194 | /* |
195 | * Event handling. | |
196 | */ | |
197 | atomic_t event_nr; | |
198 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
199 | atomic_t uevent_seq; |
200 | struct list_head uevent_list; | |
201 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
202 | |
203 | /* | |
204 | * freeze/thaw support require holding onto a super block | |
205 | */ | |
206 | struct super_block *frozen_sb; | |
db8fef4f | 207 | struct block_device *bdev; |
3ac51e74 DW |
208 | |
209 | /* forced geometry settings */ | |
210 | struct hd_geometry geometry; | |
784aae73 | 211 | |
2995fa78 MP |
212 | /* kobject and completion */ |
213 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 214 | |
d87f4c14 TH |
215 | /* zero-length flush that will be cloned and submitted to targets */ |
216 | struct bio flush_bio; | |
fd2ed4d2 | 217 | |
96b26c8c MP |
218 | /* the number of internal suspends */ |
219 | unsigned internal_suspend_count; | |
220 | ||
fd2ed4d2 | 221 | struct dm_stats stats; |
2eb6e1e3 KB |
222 | |
223 | struct kthread_worker kworker; | |
224 | struct task_struct *kworker_task; | |
de3ec86d MS |
225 | |
226 | /* for request-based merge heuristic in dm_request_fn() */ | |
0ce65797 | 227 | unsigned seq_rq_merge_deadline_usecs; |
de3ec86d | 228 | int last_rq_rw; |
0ce65797 MS |
229 | sector_t last_rq_pos; |
230 | ktime_t last_rq_start_time; | |
bfebd1cd MS |
231 | |
232 | /* for blk-mq request-based DM support */ | |
233 | struct blk_mq_tag_set tag_set; | |
17e149b8 | 234 | bool use_blk_mq; |
1da177e4 LT |
235 | }; |
236 | ||
17e149b8 MS |
237 | #ifdef CONFIG_DM_MQ_DEFAULT |
238 | static bool use_blk_mq = true; | |
239 | #else | |
240 | static bool use_blk_mq = false; | |
241 | #endif | |
242 | ||
243 | bool dm_use_blk_mq(struct mapped_device *md) | |
244 | { | |
245 | return md->use_blk_mq; | |
246 | } | |
247 | ||
e6ee8c0b KU |
248 | /* |
249 | * For mempools pre-allocation at the table loading time. | |
250 | */ | |
251 | struct dm_md_mempools { | |
252 | mempool_t *io_pool; | |
1ae49ea2 | 253 | mempool_t *rq_pool; |
e6ee8c0b KU |
254 | struct bio_set *bs; |
255 | }; | |
256 | ||
86f1152b BM |
257 | struct table_device { |
258 | struct list_head list; | |
259 | atomic_t count; | |
260 | struct dm_dev dm_dev; | |
261 | }; | |
262 | ||
6cfa5857 MS |
263 | #define RESERVED_BIO_BASED_IOS 16 |
264 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 265 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 266 | static struct kmem_cache *_io_cache; |
8fbf26ad | 267 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 268 | static struct kmem_cache *_rq_cache; |
94818742 | 269 | |
e8603136 MS |
270 | /* |
271 | * Bio-based DM's mempools' reserved IOs set by the user. | |
272 | */ | |
273 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
274 | ||
f4790826 MS |
275 | /* |
276 | * Request-based DM's mempools' reserved IOs set by the user. | |
277 | */ | |
278 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
279 | ||
09c2d531 | 280 | static unsigned __dm_get_module_param(unsigned *module_param, |
f4790826 MS |
281 | unsigned def, unsigned max) |
282 | { | |
09c2d531 MS |
283 | unsigned param = ACCESS_ONCE(*module_param); |
284 | unsigned modified_param = 0; | |
f4790826 | 285 | |
09c2d531 MS |
286 | if (!param) |
287 | modified_param = def; | |
288 | else if (param > max) | |
289 | modified_param = max; | |
f4790826 | 290 | |
09c2d531 MS |
291 | if (modified_param) { |
292 | (void)cmpxchg(module_param, param, modified_param); | |
293 | param = modified_param; | |
f4790826 MS |
294 | } |
295 | ||
09c2d531 | 296 | return param; |
f4790826 MS |
297 | } |
298 | ||
e8603136 MS |
299 | unsigned dm_get_reserved_bio_based_ios(void) |
300 | { | |
09c2d531 | 301 | return __dm_get_module_param(&reserved_bio_based_ios, |
e8603136 MS |
302 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); |
303 | } | |
304 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
305 | ||
f4790826 MS |
306 | unsigned dm_get_reserved_rq_based_ios(void) |
307 | { | |
09c2d531 | 308 | return __dm_get_module_param(&reserved_rq_based_ios, |
f4790826 MS |
309 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); |
310 | } | |
311 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
312 | ||
1da177e4 LT |
313 | static int __init local_init(void) |
314 | { | |
51157b4a | 315 | int r = -ENOMEM; |
1da177e4 | 316 | |
1da177e4 | 317 | /* allocate a slab for the dm_ios */ |
028867ac | 318 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 319 | if (!_io_cache) |
51157b4a | 320 | return r; |
1da177e4 | 321 | |
8fbf26ad KU |
322 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
323 | if (!_rq_tio_cache) | |
dba14160 | 324 | goto out_free_io_cache; |
8fbf26ad | 325 | |
1ae49ea2 MS |
326 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
327 | __alignof__(struct request), 0, NULL); | |
328 | if (!_rq_cache) | |
329 | goto out_free_rq_tio_cache; | |
330 | ||
51e5b2bd | 331 | r = dm_uevent_init(); |
51157b4a | 332 | if (r) |
1ae49ea2 | 333 | goto out_free_rq_cache; |
51e5b2bd | 334 | |
acfe0ad7 MP |
335 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
336 | if (!deferred_remove_workqueue) { | |
337 | r = -ENOMEM; | |
338 | goto out_uevent_exit; | |
339 | } | |
340 | ||
1da177e4 LT |
341 | _major = major; |
342 | r = register_blkdev(_major, _name); | |
51157b4a | 343 | if (r < 0) |
acfe0ad7 | 344 | goto out_free_workqueue; |
1da177e4 LT |
345 | |
346 | if (!_major) | |
347 | _major = r; | |
348 | ||
349 | return 0; | |
51157b4a | 350 | |
acfe0ad7 MP |
351 | out_free_workqueue: |
352 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
353 | out_uevent_exit: |
354 | dm_uevent_exit(); | |
1ae49ea2 MS |
355 | out_free_rq_cache: |
356 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
357 | out_free_rq_tio_cache: |
358 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
359 | out_free_io_cache: |
360 | kmem_cache_destroy(_io_cache); | |
361 | ||
362 | return r; | |
1da177e4 LT |
363 | } |
364 | ||
365 | static void local_exit(void) | |
366 | { | |
2c140a24 | 367 | flush_scheduled_work(); |
acfe0ad7 | 368 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 369 | |
1ae49ea2 | 370 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 371 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 372 | kmem_cache_destroy(_io_cache); |
00d59405 | 373 | unregister_blkdev(_major, _name); |
51e5b2bd | 374 | dm_uevent_exit(); |
1da177e4 LT |
375 | |
376 | _major = 0; | |
377 | ||
378 | DMINFO("cleaned up"); | |
379 | } | |
380 | ||
b9249e55 | 381 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
382 | local_init, |
383 | dm_target_init, | |
384 | dm_linear_init, | |
385 | dm_stripe_init, | |
952b3557 | 386 | dm_io_init, |
945fa4d2 | 387 | dm_kcopyd_init, |
1da177e4 | 388 | dm_interface_init, |
fd2ed4d2 | 389 | dm_statistics_init, |
1da177e4 LT |
390 | }; |
391 | ||
b9249e55 | 392 | static void (*_exits[])(void) = { |
1da177e4 LT |
393 | local_exit, |
394 | dm_target_exit, | |
395 | dm_linear_exit, | |
396 | dm_stripe_exit, | |
952b3557 | 397 | dm_io_exit, |
945fa4d2 | 398 | dm_kcopyd_exit, |
1da177e4 | 399 | dm_interface_exit, |
fd2ed4d2 | 400 | dm_statistics_exit, |
1da177e4 LT |
401 | }; |
402 | ||
403 | static int __init dm_init(void) | |
404 | { | |
405 | const int count = ARRAY_SIZE(_inits); | |
406 | ||
407 | int r, i; | |
408 | ||
409 | for (i = 0; i < count; i++) { | |
410 | r = _inits[i](); | |
411 | if (r) | |
412 | goto bad; | |
413 | } | |
414 | ||
415 | return 0; | |
416 | ||
417 | bad: | |
418 | while (i--) | |
419 | _exits[i](); | |
420 | ||
421 | return r; | |
422 | } | |
423 | ||
424 | static void __exit dm_exit(void) | |
425 | { | |
426 | int i = ARRAY_SIZE(_exits); | |
427 | ||
428 | while (i--) | |
429 | _exits[i](); | |
d15b774c AK |
430 | |
431 | /* | |
432 | * Should be empty by this point. | |
433 | */ | |
d15b774c | 434 | idr_destroy(&_minor_idr); |
1da177e4 LT |
435 | } |
436 | ||
437 | /* | |
438 | * Block device functions | |
439 | */ | |
432a212c MA |
440 | int dm_deleting_md(struct mapped_device *md) |
441 | { | |
442 | return test_bit(DMF_DELETING, &md->flags); | |
443 | } | |
444 | ||
fe5f9f2c | 445 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
446 | { |
447 | struct mapped_device *md; | |
448 | ||
fba9f90e JM |
449 | spin_lock(&_minor_lock); |
450 | ||
fe5f9f2c | 451 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
452 | if (!md) |
453 | goto out; | |
454 | ||
5c6bd75d | 455 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 456 | dm_deleting_md(md)) { |
fba9f90e JM |
457 | md = NULL; |
458 | goto out; | |
459 | } | |
460 | ||
1da177e4 | 461 | dm_get(md); |
5c6bd75d | 462 | atomic_inc(&md->open_count); |
fba9f90e JM |
463 | out: |
464 | spin_unlock(&_minor_lock); | |
465 | ||
466 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
467 | } |
468 | ||
db2a144b | 469 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 470 | { |
63a4f065 | 471 | struct mapped_device *md; |
6e9624b8 | 472 | |
4a1aeb98 MB |
473 | spin_lock(&_minor_lock); |
474 | ||
63a4f065 MS |
475 | md = disk->private_data; |
476 | if (WARN_ON(!md)) | |
477 | goto out; | |
478 | ||
2c140a24 MP |
479 | if (atomic_dec_and_test(&md->open_count) && |
480 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 481 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 482 | |
1da177e4 | 483 | dm_put(md); |
63a4f065 | 484 | out: |
4a1aeb98 | 485 | spin_unlock(&_minor_lock); |
1da177e4 LT |
486 | } |
487 | ||
5c6bd75d AK |
488 | int dm_open_count(struct mapped_device *md) |
489 | { | |
490 | return atomic_read(&md->open_count); | |
491 | } | |
492 | ||
493 | /* | |
494 | * Guarantees nothing is using the device before it's deleted. | |
495 | */ | |
2c140a24 | 496 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
497 | { |
498 | int r = 0; | |
499 | ||
500 | spin_lock(&_minor_lock); | |
501 | ||
2c140a24 | 502 | if (dm_open_count(md)) { |
5c6bd75d | 503 | r = -EBUSY; |
2c140a24 MP |
504 | if (mark_deferred) |
505 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
506 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
507 | r = -EEXIST; | |
5c6bd75d AK |
508 | else |
509 | set_bit(DMF_DELETING, &md->flags); | |
510 | ||
511 | spin_unlock(&_minor_lock); | |
512 | ||
513 | return r; | |
514 | } | |
515 | ||
2c140a24 MP |
516 | int dm_cancel_deferred_remove(struct mapped_device *md) |
517 | { | |
518 | int r = 0; | |
519 | ||
520 | spin_lock(&_minor_lock); | |
521 | ||
522 | if (test_bit(DMF_DELETING, &md->flags)) | |
523 | r = -EBUSY; | |
524 | else | |
525 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
526 | ||
527 | spin_unlock(&_minor_lock); | |
528 | ||
529 | return r; | |
530 | } | |
531 | ||
532 | static void do_deferred_remove(struct work_struct *w) | |
533 | { | |
534 | dm_deferred_remove(); | |
535 | } | |
536 | ||
fd2ed4d2 MP |
537 | sector_t dm_get_size(struct mapped_device *md) |
538 | { | |
539 | return get_capacity(md->disk); | |
540 | } | |
541 | ||
9974fa2c MS |
542 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
543 | { | |
544 | return md->queue; | |
545 | } | |
546 | ||
fd2ed4d2 MP |
547 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
548 | { | |
549 | return &md->stats; | |
550 | } | |
551 | ||
3ac51e74 DW |
552 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
553 | { | |
554 | struct mapped_device *md = bdev->bd_disk->private_data; | |
555 | ||
556 | return dm_get_geometry(md, geo); | |
557 | } | |
558 | ||
e56f81e0 CH |
559 | static int dm_get_live_table_for_ioctl(struct mapped_device *md, |
560 | struct dm_target **tgt, struct block_device **bdev, | |
561 | fmode_t *mode, int *srcu_idx) | |
aa129a22 | 562 | { |
6c182cd8 | 563 | struct dm_table *map; |
e56f81e0 | 564 | int r; |
aa129a22 | 565 | |
6c182cd8 | 566 | retry: |
e56f81e0 CH |
567 | r = -ENOTTY; |
568 | map = dm_get_live_table(md, srcu_idx); | |
aa129a22 MB |
569 | if (!map || !dm_table_get_size(map)) |
570 | goto out; | |
571 | ||
572 | /* We only support devices that have a single target */ | |
573 | if (dm_table_get_num_targets(map) != 1) | |
574 | goto out; | |
575 | ||
e56f81e0 CH |
576 | *tgt = dm_table_get_target(map, 0); |
577 | ||
578 | if (!(*tgt)->type->prepare_ioctl) | |
4d341d82 | 579 | goto out; |
aa129a22 | 580 | |
4f186f8b | 581 | if (dm_suspended_md(md)) { |
aa129a22 MB |
582 | r = -EAGAIN; |
583 | goto out; | |
584 | } | |
585 | ||
e56f81e0 CH |
586 | r = (*tgt)->type->prepare_ioctl(*tgt, bdev, mode); |
587 | if (r < 0) | |
588 | goto out; | |
aa129a22 | 589 | |
e56f81e0 | 590 | return r; |
aa129a22 | 591 | |
aa129a22 | 592 | out: |
e56f81e0 | 593 | dm_put_live_table(md, *srcu_idx); |
5bbbfdf6 | 594 | if (r == -ENOTCONN && !fatal_signal_pending(current)) { |
6c182cd8 HR |
595 | msleep(10); |
596 | goto retry; | |
597 | } | |
e56f81e0 CH |
598 | return r; |
599 | } | |
600 | ||
601 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, | |
602 | unsigned int cmd, unsigned long arg) | |
603 | { | |
604 | struct mapped_device *md = bdev->bd_disk->private_data; | |
605 | struct dm_target *tgt; | |
647a20d5 | 606 | struct block_device *tgt_bdev = NULL; |
e56f81e0 CH |
607 | int srcu_idx, r; |
608 | ||
647a20d5 | 609 | r = dm_get_live_table_for_ioctl(md, &tgt, &tgt_bdev, &mode, &srcu_idx); |
e56f81e0 CH |
610 | if (r < 0) |
611 | return r; | |
6c182cd8 | 612 | |
e56f81e0 CH |
613 | if (r > 0) { |
614 | /* | |
615 | * Target determined this ioctl is being issued against | |
616 | * a logical partition of the parent bdev; so extra | |
617 | * validation is needed. | |
618 | */ | |
619 | r = scsi_verify_blk_ioctl(NULL, cmd); | |
620 | if (r) | |
621 | goto out; | |
622 | } | |
6c182cd8 | 623 | |
647a20d5 | 624 | r = __blkdev_driver_ioctl(tgt_bdev, mode, cmd, arg); |
e56f81e0 CH |
625 | out: |
626 | dm_put_live_table(md, srcu_idx); | |
aa129a22 MB |
627 | return r; |
628 | } | |
629 | ||
028867ac | 630 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
631 | { |
632 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
633 | } | |
634 | ||
028867ac | 635 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
636 | { |
637 | mempool_free(io, md->io_pool); | |
638 | } | |
639 | ||
028867ac | 640 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 641 | { |
dba14160 | 642 | bio_put(&tio->clone); |
1da177e4 LT |
643 | } |
644 | ||
08885643 KU |
645 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
646 | gfp_t gfp_mask) | |
cec47e3d | 647 | { |
5f015204 | 648 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
649 | } |
650 | ||
651 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
652 | { | |
5f015204 | 653 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
654 | } |
655 | ||
1ae49ea2 MS |
656 | static struct request *alloc_clone_request(struct mapped_device *md, |
657 | gfp_t gfp_mask) | |
658 | { | |
659 | return mempool_alloc(md->rq_pool, gfp_mask); | |
660 | } | |
661 | ||
662 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
663 | { | |
664 | mempool_free(rq, md->rq_pool); | |
665 | } | |
666 | ||
90abb8c4 KU |
667 | static int md_in_flight(struct mapped_device *md) |
668 | { | |
669 | return atomic_read(&md->pending[READ]) + | |
670 | atomic_read(&md->pending[WRITE]); | |
671 | } | |
672 | ||
3eaf840e JNN |
673 | static void start_io_acct(struct dm_io *io) |
674 | { | |
675 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 676 | struct bio *bio = io->bio; |
c9959059 | 677 | int cpu; |
fd2ed4d2 | 678 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
679 | |
680 | io->start_time = jiffies; | |
681 | ||
074a7aca TH |
682 | cpu = part_stat_lock(); |
683 | part_round_stats(cpu, &dm_disk(md)->part0); | |
684 | part_stat_unlock(); | |
1e9bb880 SL |
685 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
686 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
687 | |
688 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 689 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 690 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
691 | } |
692 | ||
d221d2e7 | 693 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
694 | { |
695 | struct mapped_device *md = io->md; | |
696 | struct bio *bio = io->bio; | |
697 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 698 | int pending; |
3eaf840e JNN |
699 | int rw = bio_data_dir(bio); |
700 | ||
18c0b223 | 701 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 702 | |
fd2ed4d2 | 703 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 704 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
705 | bio_sectors(bio), true, duration, &io->stats_aux); |
706 | ||
af7e466a MP |
707 | /* |
708 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 709 | * a flush. |
af7e466a | 710 | */ |
1e9bb880 SL |
711 | pending = atomic_dec_return(&md->pending[rw]); |
712 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 713 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 714 | |
d221d2e7 MP |
715 | /* nudge anyone waiting on suspend queue */ |
716 | if (!pending) | |
717 | wake_up(&md->wait); | |
3eaf840e JNN |
718 | } |
719 | ||
1da177e4 LT |
720 | /* |
721 | * Add the bio to the list of deferred io. | |
722 | */ | |
92c63902 | 723 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 724 | { |
05447420 | 725 | unsigned long flags; |
1da177e4 | 726 | |
05447420 | 727 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 728 | bio_list_add(&md->deferred, bio); |
05447420 | 729 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 730 | queue_work(md->wq, &md->work); |
1da177e4 LT |
731 | } |
732 | ||
733 | /* | |
734 | * Everyone (including functions in this file), should use this | |
735 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 736 | * dm_put_live_table() when finished. |
1da177e4 | 737 | */ |
83d5e5b0 | 738 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 739 | { |
83d5e5b0 MP |
740 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
741 | ||
742 | return srcu_dereference(md->map, &md->io_barrier); | |
743 | } | |
1da177e4 | 744 | |
83d5e5b0 MP |
745 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
746 | { | |
747 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
748 | } | |
749 | ||
750 | void dm_sync_table(struct mapped_device *md) | |
751 | { | |
752 | synchronize_srcu(&md->io_barrier); | |
753 | synchronize_rcu_expedited(); | |
754 | } | |
755 | ||
756 | /* | |
757 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
758 | * The caller must not block between these two functions. | |
759 | */ | |
760 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
761 | { | |
762 | rcu_read_lock(); | |
763 | return rcu_dereference(md->map); | |
764 | } | |
1da177e4 | 765 | |
83d5e5b0 MP |
766 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
767 | { | |
768 | rcu_read_unlock(); | |
1da177e4 LT |
769 | } |
770 | ||
86f1152b BM |
771 | /* |
772 | * Open a table device so we can use it as a map destination. | |
773 | */ | |
774 | static int open_table_device(struct table_device *td, dev_t dev, | |
775 | struct mapped_device *md) | |
776 | { | |
777 | static char *_claim_ptr = "I belong to device-mapper"; | |
778 | struct block_device *bdev; | |
779 | ||
780 | int r; | |
781 | ||
782 | BUG_ON(td->dm_dev.bdev); | |
783 | ||
784 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
785 | if (IS_ERR(bdev)) | |
786 | return PTR_ERR(bdev); | |
787 | ||
788 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
789 | if (r) { | |
790 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
791 | return r; | |
792 | } | |
793 | ||
794 | td->dm_dev.bdev = bdev; | |
795 | return 0; | |
796 | } | |
797 | ||
798 | /* | |
799 | * Close a table device that we've been using. | |
800 | */ | |
801 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
802 | { | |
803 | if (!td->dm_dev.bdev) | |
804 | return; | |
805 | ||
806 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
807 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
808 | td->dm_dev.bdev = NULL; | |
809 | } | |
810 | ||
811 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
812 | fmode_t mode) { | |
813 | struct table_device *td; | |
814 | ||
815 | list_for_each_entry(td, l, list) | |
816 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
817 | return td; | |
818 | ||
819 | return NULL; | |
820 | } | |
821 | ||
822 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
823 | struct dm_dev **result) { | |
824 | int r; | |
825 | struct table_device *td; | |
826 | ||
827 | mutex_lock(&md->table_devices_lock); | |
828 | td = find_table_device(&md->table_devices, dev, mode); | |
829 | if (!td) { | |
830 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
831 | if (!td) { | |
832 | mutex_unlock(&md->table_devices_lock); | |
833 | return -ENOMEM; | |
834 | } | |
835 | ||
836 | td->dm_dev.mode = mode; | |
837 | td->dm_dev.bdev = NULL; | |
838 | ||
839 | if ((r = open_table_device(td, dev, md))) { | |
840 | mutex_unlock(&md->table_devices_lock); | |
841 | kfree(td); | |
842 | return r; | |
843 | } | |
844 | ||
845 | format_dev_t(td->dm_dev.name, dev); | |
846 | ||
847 | atomic_set(&td->count, 0); | |
848 | list_add(&td->list, &md->table_devices); | |
849 | } | |
850 | atomic_inc(&td->count); | |
851 | mutex_unlock(&md->table_devices_lock); | |
852 | ||
853 | *result = &td->dm_dev; | |
854 | return 0; | |
855 | } | |
856 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
857 | ||
858 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
859 | { | |
860 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
861 | ||
862 | mutex_lock(&md->table_devices_lock); | |
863 | if (atomic_dec_and_test(&td->count)) { | |
864 | close_table_device(td, md); | |
865 | list_del(&td->list); | |
866 | kfree(td); | |
867 | } | |
868 | mutex_unlock(&md->table_devices_lock); | |
869 | } | |
870 | EXPORT_SYMBOL(dm_put_table_device); | |
871 | ||
872 | static void free_table_devices(struct list_head *devices) | |
873 | { | |
874 | struct list_head *tmp, *next; | |
875 | ||
876 | list_for_each_safe(tmp, next, devices) { | |
877 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
878 | ||
879 | DMWARN("dm_destroy: %s still exists with %d references", | |
880 | td->dm_dev.name, atomic_read(&td->count)); | |
881 | kfree(td); | |
882 | } | |
883 | } | |
884 | ||
3ac51e74 DW |
885 | /* |
886 | * Get the geometry associated with a dm device | |
887 | */ | |
888 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
889 | { | |
890 | *geo = md->geometry; | |
891 | ||
892 | return 0; | |
893 | } | |
894 | ||
895 | /* | |
896 | * Set the geometry of a device. | |
897 | */ | |
898 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
899 | { | |
900 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
901 | ||
902 | if (geo->start > sz) { | |
903 | DMWARN("Start sector is beyond the geometry limits."); | |
904 | return -EINVAL; | |
905 | } | |
906 | ||
907 | md->geometry = *geo; | |
908 | ||
909 | return 0; | |
910 | } | |
911 | ||
1da177e4 LT |
912 | /*----------------------------------------------------------------- |
913 | * CRUD START: | |
914 | * A more elegant soln is in the works that uses the queue | |
915 | * merge fn, unfortunately there are a couple of changes to | |
916 | * the block layer that I want to make for this. So in the | |
917 | * interests of getting something for people to use I give | |
918 | * you this clearly demarcated crap. | |
919 | *---------------------------------------------------------------*/ | |
920 | ||
2e93ccc1 KU |
921 | static int __noflush_suspending(struct mapped_device *md) |
922 | { | |
923 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
924 | } | |
925 | ||
1da177e4 LT |
926 | /* |
927 | * Decrements the number of outstanding ios that a bio has been | |
928 | * cloned into, completing the original io if necc. | |
929 | */ | |
858119e1 | 930 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 931 | { |
2e93ccc1 | 932 | unsigned long flags; |
b35f8caa MB |
933 | int io_error; |
934 | struct bio *bio; | |
935 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
936 | |
937 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
938 | if (unlikely(error)) { |
939 | spin_lock_irqsave(&io->endio_lock, flags); | |
940 | if (!(io->error > 0 && __noflush_suspending(md))) | |
941 | io->error = error; | |
942 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
943 | } | |
1da177e4 LT |
944 | |
945 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
946 | if (io->error == DM_ENDIO_REQUEUE) { |
947 | /* | |
948 | * Target requested pushing back the I/O. | |
2e93ccc1 | 949 | */ |
022c2611 | 950 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
951 | if (__noflush_suspending(md)) |
952 | bio_list_add_head(&md->deferred, io->bio); | |
953 | else | |
2e93ccc1 KU |
954 | /* noflush suspend was interrupted. */ |
955 | io->error = -EIO; | |
022c2611 | 956 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
957 | } |
958 | ||
b35f8caa MB |
959 | io_error = io->error; |
960 | bio = io->bio; | |
6a8736d1 TH |
961 | end_io_acct(io); |
962 | free_io(md, io); | |
963 | ||
964 | if (io_error == DM_ENDIO_REQUEUE) | |
965 | return; | |
2e93ccc1 | 966 | |
4f024f37 | 967 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 968 | /* |
6a8736d1 TH |
969 | * Preflush done for flush with data, reissue |
970 | * without REQ_FLUSH. | |
af7e466a | 971 | */ |
6a8736d1 TH |
972 | bio->bi_rw &= ~REQ_FLUSH; |
973 | queue_io(md, bio); | |
af7e466a | 974 | } else { |
b372d360 | 975 | /* done with normal IO or empty flush */ |
0a82a8d1 | 976 | trace_block_bio_complete(md->queue, bio, io_error); |
4246a0b6 CH |
977 | bio->bi_error = io_error; |
978 | bio_endio(bio); | |
b35f8caa | 979 | } |
1da177e4 LT |
980 | } |
981 | } | |
982 | ||
7eee4ae2 MS |
983 | static void disable_write_same(struct mapped_device *md) |
984 | { | |
985 | struct queue_limits *limits = dm_get_queue_limits(md); | |
986 | ||
987 | /* device doesn't really support WRITE SAME, disable it */ | |
988 | limits->max_write_same_sectors = 0; | |
989 | } | |
990 | ||
4246a0b6 | 991 | static void clone_endio(struct bio *bio) |
1da177e4 | 992 | { |
4246a0b6 | 993 | int error = bio->bi_error; |
5164bece | 994 | int r = error; |
bfc6d41c | 995 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 996 | struct dm_io *io = tio->io; |
9faf400f | 997 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
998 | dm_endio_fn endio = tio->ti->type->end_io; |
999 | ||
1da177e4 | 1000 | if (endio) { |
7de3ee57 | 1001 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
1002 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
1003 | /* | |
1004 | * error and requeue request are handled | |
1005 | * in dec_pending(). | |
1006 | */ | |
1da177e4 | 1007 | error = r; |
45cbcd79 KU |
1008 | else if (r == DM_ENDIO_INCOMPLETE) |
1009 | /* The target will handle the io */ | |
6712ecf8 | 1010 | return; |
45cbcd79 KU |
1011 | else if (r) { |
1012 | DMWARN("unimplemented target endio return value: %d", r); | |
1013 | BUG(); | |
1014 | } | |
1da177e4 LT |
1015 | } |
1016 | ||
7eee4ae2 MS |
1017 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
1018 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
1019 | disable_write_same(md); | |
1020 | ||
9faf400f | 1021 | free_tio(md, tio); |
b35f8caa | 1022 | dec_pending(io, error); |
1da177e4 LT |
1023 | } |
1024 | ||
78d8e58a MS |
1025 | /* |
1026 | * Partial completion handling for request-based dm | |
1027 | */ | |
4246a0b6 | 1028 | static void end_clone_bio(struct bio *clone) |
78d8e58a MS |
1029 | { |
1030 | struct dm_rq_clone_bio_info *info = | |
1031 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
1032 | struct dm_rq_target_io *tio = info->tio; | |
1033 | struct bio *bio = info->orig; | |
1034 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; | |
50887bd1 | 1035 | int error = clone->bi_error; |
78d8e58a MS |
1036 | |
1037 | bio_put(clone); | |
1038 | ||
1039 | if (tio->error) | |
1040 | /* | |
1041 | * An error has already been detected on the request. | |
1042 | * Once error occurred, just let clone->end_io() handle | |
1043 | * the remainder. | |
1044 | */ | |
1045 | return; | |
50887bd1 | 1046 | else if (error) { |
78d8e58a MS |
1047 | /* |
1048 | * Don't notice the error to the upper layer yet. | |
1049 | * The error handling decision is made by the target driver, | |
1050 | * when the request is completed. | |
1051 | */ | |
50887bd1 | 1052 | tio->error = error; |
78d8e58a MS |
1053 | return; |
1054 | } | |
1055 | ||
1056 | /* | |
1057 | * I/O for the bio successfully completed. | |
1058 | * Notice the data completion to the upper layer. | |
1059 | */ | |
1060 | ||
1061 | /* | |
1062 | * bios are processed from the head of the list. | |
1063 | * So the completing bio should always be rq->bio. | |
1064 | * If it's not, something wrong is happening. | |
1065 | */ | |
1066 | if (tio->orig->bio != bio) | |
1067 | DMERR("bio completion is going in the middle of the request"); | |
1068 | ||
1069 | /* | |
1070 | * Update the original request. | |
1071 | * Do not use blk_end_request() here, because it may complete | |
1072 | * the original request before the clone, and break the ordering. | |
1073 | */ | |
1074 | blk_update_request(tio->orig, 0, nr_bytes); | |
1075 | } | |
1076 | ||
bfebd1cd MS |
1077 | static struct dm_rq_target_io *tio_from_request(struct request *rq) |
1078 | { | |
1079 | return (rq->q->mq_ops ? blk_mq_rq_to_pdu(rq) : rq->special); | |
1080 | } | |
1081 | ||
e262f347 MP |
1082 | static void rq_end_stats(struct mapped_device *md, struct request *orig) |
1083 | { | |
1084 | if (unlikely(dm_stats_used(&md->stats))) { | |
1085 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
1086 | tio->duration_jiffies = jiffies - tio->duration_jiffies; | |
1087 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
1088 | tio->n_sectors, true, tio->duration_jiffies, | |
1089 | &tio->stats_aux); | |
1090 | } | |
1091 | } | |
1092 | ||
cec47e3d KU |
1093 | /* |
1094 | * Don't touch any member of the md after calling this function because | |
1095 | * the md may be freed in dm_put() at the end of this function. | |
1096 | * Or do dm_get() before calling this function and dm_put() later. | |
1097 | */ | |
466d89a6 | 1098 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1099 | { |
b4324fee | 1100 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1101 | |
1102 | /* nudge anyone waiting on suspend queue */ | |
621739b0 | 1103 | if (!md_in_flight(md)) |
cec47e3d KU |
1104 | wake_up(&md->wait); |
1105 | ||
a8c32a5c JA |
1106 | /* |
1107 | * Run this off this callpath, as drivers could invoke end_io while | |
1108 | * inside their request_fn (and holding the queue lock). Calling | |
1109 | * back into ->request_fn() could deadlock attempting to grab the | |
1110 | * queue lock again. | |
1111 | */ | |
9a0e609e | 1112 | if (run_queue) { |
bfebd1cd MS |
1113 | if (md->queue->mq_ops) |
1114 | blk_mq_run_hw_queues(md->queue, true); | |
621739b0 | 1115 | else |
9a0e609e MS |
1116 | blk_run_queue_async(md->queue); |
1117 | } | |
cec47e3d KU |
1118 | |
1119 | /* | |
1120 | * dm_put() must be at the end of this function. See the comment above | |
1121 | */ | |
1122 | dm_put(md); | |
1123 | } | |
1124 | ||
e5d8de32 | 1125 | static void free_rq_clone(struct request *clone) |
a77e28c7 KU |
1126 | { |
1127 | struct dm_rq_target_io *tio = clone->end_io_data; | |
bfebd1cd | 1128 | struct mapped_device *md = tio->md; |
a77e28c7 | 1129 | |
78d8e58a MS |
1130 | blk_rq_unprep_clone(clone); |
1131 | ||
aa6df8dd MS |
1132 | if (md->type == DM_TYPE_MQ_REQUEST_BASED) |
1133 | /* stacked on blk-mq queue(s) */ | |
e5863d9a | 1134 | tio->ti->type->release_clone_rq(clone); |
02233342 MS |
1135 | else if (!md->queue->mq_ops) |
1136 | /* request_fn queue stacked on request_fn queue(s) */ | |
bfebd1cd | 1137 | free_clone_request(md, clone); |
aa6df8dd MS |
1138 | /* |
1139 | * NOTE: for the blk-mq queue stacked on request_fn queue(s) case: | |
1140 | * no need to call free_clone_request() because we leverage blk-mq by | |
1141 | * allocating the clone at the end of the blk-mq pdu (see: clone_rq) | |
1142 | */ | |
bfebd1cd MS |
1143 | |
1144 | if (!md->queue->mq_ops) | |
1145 | free_rq_tio(tio); | |
a77e28c7 KU |
1146 | } |
1147 | ||
980691e5 KU |
1148 | /* |
1149 | * Complete the clone and the original request. | |
466d89a6 KB |
1150 | * Must be called without clone's queue lock held, |
1151 | * see end_clone_request() for more details. | |
980691e5 KU |
1152 | */ |
1153 | static void dm_end_request(struct request *clone, int error) | |
1154 | { | |
1155 | int rw = rq_data_dir(clone); | |
1156 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1157 | struct mapped_device *md = tio->md; | |
1158 | struct request *rq = tio->orig; | |
1159 | ||
29e4013d | 1160 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1161 | rq->errors = clone->errors; |
1162 | rq->resid_len = clone->resid_len; | |
1163 | ||
1164 | if (rq->sense) | |
1165 | /* | |
1166 | * We are using the sense buffer of the original | |
1167 | * request. | |
1168 | * So setting the length of the sense data is enough. | |
1169 | */ | |
1170 | rq->sense_len = clone->sense_len; | |
1171 | } | |
1172 | ||
e5d8de32 | 1173 | free_rq_clone(clone); |
e262f347 | 1174 | rq_end_stats(md, rq); |
bfebd1cd MS |
1175 | if (!rq->q->mq_ops) |
1176 | blk_end_request_all(rq, error); | |
1177 | else | |
1178 | blk_mq_end_request(rq, error); | |
29e4013d | 1179 | rq_completed(md, rw, true); |
980691e5 KU |
1180 | } |
1181 | ||
cec47e3d KU |
1182 | static void dm_unprep_request(struct request *rq) |
1183 | { | |
bfebd1cd | 1184 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1185 | struct request *clone = tio->clone; |
cec47e3d | 1186 | |
bfebd1cd MS |
1187 | if (!rq->q->mq_ops) { |
1188 | rq->special = NULL; | |
1189 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1190 | } | |
cec47e3d | 1191 | |
e5863d9a | 1192 | if (clone) |
e5d8de32 | 1193 | free_rq_clone(clone); |
cec47e3d KU |
1194 | } |
1195 | ||
1196 | /* | |
1197 | * Requeue the original request of a clone. | |
1198 | */ | |
bfebd1cd | 1199 | static void old_requeue_request(struct request *rq) |
cec47e3d | 1200 | { |
cec47e3d KU |
1201 | struct request_queue *q = rq->q; |
1202 | unsigned long flags; | |
1203 | ||
cec47e3d | 1204 | spin_lock_irqsave(q->queue_lock, flags); |
cec47e3d | 1205 | blk_requeue_request(q, rq); |
4ae9944d | 1206 | blk_run_queue_async(q); |
cec47e3d | 1207 | spin_unlock_irqrestore(q->queue_lock, flags); |
bfebd1cd MS |
1208 | } |
1209 | ||
2d76fff1 MS |
1210 | static void dm_requeue_original_request(struct mapped_device *md, |
1211 | struct request *rq) | |
bfebd1cd MS |
1212 | { |
1213 | int rw = rq_data_dir(rq); | |
1214 | ||
1215 | dm_unprep_request(rq); | |
1216 | ||
e262f347 | 1217 | rq_end_stats(md, rq); |
bfebd1cd MS |
1218 | if (!rq->q->mq_ops) |
1219 | old_requeue_request(rq); | |
1220 | else { | |
1221 | blk_mq_requeue_request(rq); | |
1222 | blk_mq_kick_requeue_list(rq->q); | |
1223 | } | |
cec47e3d | 1224 | |
466d89a6 KB |
1225 | rq_completed(md, rw, false); |
1226 | } | |
1227 | ||
bfebd1cd | 1228 | static void old_stop_queue(struct request_queue *q) |
cec47e3d KU |
1229 | { |
1230 | unsigned long flags; | |
1231 | ||
bfebd1cd MS |
1232 | if (blk_queue_stopped(q)) |
1233 | return; | |
1234 | ||
cec47e3d | 1235 | spin_lock_irqsave(q->queue_lock, flags); |
bfebd1cd | 1236 | blk_stop_queue(q); |
cec47e3d KU |
1237 | spin_unlock_irqrestore(q->queue_lock, flags); |
1238 | } | |
1239 | ||
bfebd1cd | 1240 | static void stop_queue(struct request_queue *q) |
cec47e3d | 1241 | { |
bfebd1cd MS |
1242 | if (!q->mq_ops) |
1243 | old_stop_queue(q); | |
1244 | else | |
1245 | blk_mq_stop_hw_queues(q); | |
cec47e3d KU |
1246 | } |
1247 | ||
bfebd1cd | 1248 | static void old_start_queue(struct request_queue *q) |
cec47e3d KU |
1249 | { |
1250 | unsigned long flags; | |
1251 | ||
1252 | spin_lock_irqsave(q->queue_lock, flags); | |
bfebd1cd MS |
1253 | if (blk_queue_stopped(q)) |
1254 | blk_start_queue(q); | |
cec47e3d KU |
1255 | spin_unlock_irqrestore(q->queue_lock, flags); |
1256 | } | |
1257 | ||
bfebd1cd MS |
1258 | static void start_queue(struct request_queue *q) |
1259 | { | |
1260 | if (!q->mq_ops) | |
1261 | old_start_queue(q); | |
1262 | else | |
1263 | blk_mq_start_stopped_hw_queues(q, true); | |
1264 | } | |
1265 | ||
11a68244 | 1266 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1267 | { |
11a68244 | 1268 | int r = error; |
cec47e3d | 1269 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1270 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1271 | |
ba1cbad9 MS |
1272 | if (tio->ti) { |
1273 | rq_end_io = tio->ti->type->rq_end_io; | |
1274 | ||
1275 | if (mapped && rq_end_io) | |
1276 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1277 | } | |
cec47e3d | 1278 | |
7eee4ae2 MS |
1279 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1280 | !clone->q->limits.max_write_same_sectors)) | |
1281 | disable_write_same(tio->md); | |
1282 | ||
11a68244 | 1283 | if (r <= 0) |
cec47e3d | 1284 | /* The target wants to complete the I/O */ |
11a68244 KU |
1285 | dm_end_request(clone, r); |
1286 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1287 | /* The target will handle the I/O */ |
1288 | return; | |
11a68244 | 1289 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d | 1290 | /* The target wants to requeue the I/O */ |
2d76fff1 | 1291 | dm_requeue_original_request(tio->md, tio->orig); |
cec47e3d | 1292 | else { |
11a68244 | 1293 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1294 | BUG(); |
1295 | } | |
1296 | } | |
1297 | ||
11a68244 KU |
1298 | /* |
1299 | * Request completion handler for request-based dm | |
1300 | */ | |
1301 | static void dm_softirq_done(struct request *rq) | |
1302 | { | |
1303 | bool mapped = true; | |
bfebd1cd | 1304 | struct dm_rq_target_io *tio = tio_from_request(rq); |
466d89a6 | 1305 | struct request *clone = tio->clone; |
bfebd1cd | 1306 | int rw; |
11a68244 | 1307 | |
e5863d9a | 1308 | if (!clone) { |
e262f347 | 1309 | rq_end_stats(tio->md, rq); |
bfebd1cd MS |
1310 | rw = rq_data_dir(rq); |
1311 | if (!rq->q->mq_ops) { | |
1312 | blk_end_request_all(rq, tio->error); | |
1313 | rq_completed(tio->md, rw, false); | |
1314 | free_rq_tio(tio); | |
1315 | } else { | |
1316 | blk_mq_end_request(rq, tio->error); | |
1317 | rq_completed(tio->md, rw, false); | |
1318 | } | |
e5863d9a MS |
1319 | return; |
1320 | } | |
11a68244 KU |
1321 | |
1322 | if (rq->cmd_flags & REQ_FAILED) | |
1323 | mapped = false; | |
1324 | ||
1325 | dm_done(clone, tio->error, mapped); | |
1326 | } | |
1327 | ||
cec47e3d KU |
1328 | /* |
1329 | * Complete the clone and the original request with the error status | |
1330 | * through softirq context. | |
1331 | */ | |
466d89a6 | 1332 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1333 | { |
bfebd1cd | 1334 | struct dm_rq_target_io *tio = tio_from_request(rq); |
cec47e3d KU |
1335 | |
1336 | tio->error = error; | |
cec47e3d KU |
1337 | blk_complete_request(rq); |
1338 | } | |
1339 | ||
1340 | /* | |
1341 | * Complete the not-mapped clone and the original request with the error status | |
1342 | * through softirq context. | |
1343 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1344 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1345 | */ |
466d89a6 | 1346 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1347 | { |
cec47e3d | 1348 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1349 | dm_complete_request(rq, error); |
cec47e3d | 1350 | } |
cec47e3d KU |
1351 | |
1352 | /* | |
bfebd1cd | 1353 | * Called with the clone's queue lock held (for non-blk-mq) |
cec47e3d KU |
1354 | */ |
1355 | static void end_clone_request(struct request *clone, int error) | |
1356 | { | |
466d89a6 KB |
1357 | struct dm_rq_target_io *tio = clone->end_io_data; |
1358 | ||
e5863d9a MS |
1359 | if (!clone->q->mq_ops) { |
1360 | /* | |
1361 | * For just cleaning up the information of the queue in which | |
1362 | * the clone was dispatched. | |
1363 | * The clone is *NOT* freed actually here because it is alloced | |
1364 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1365 | */ | |
1366 | __blk_put_request(clone->q, clone); | |
1367 | } | |
cec47e3d KU |
1368 | |
1369 | /* | |
1370 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1371 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1372 | * - another request may be submitted by the upper level driver |
1373 | * of the stacking during the completion | |
1374 | * - the submission which requires queue lock may be done | |
466d89a6 | 1375 | * against this clone's queue |
cec47e3d | 1376 | */ |
466d89a6 | 1377 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1378 | } |
1379 | ||
56a67df7 MS |
1380 | /* |
1381 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1382 | * target boundary. | |
1383 | */ | |
1384 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1385 | { | |
1386 | sector_t target_offset = dm_target_offset(ti, sector); | |
1387 | ||
1388 | return ti->len - target_offset; | |
1389 | } | |
1390 | ||
1391 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1392 | { |
56a67df7 | 1393 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1394 | sector_t offset, max_len; |
1da177e4 LT |
1395 | |
1396 | /* | |
542f9038 | 1397 | * Does the target need to split even further? |
1da177e4 | 1398 | */ |
542f9038 MS |
1399 | if (ti->max_io_len) { |
1400 | offset = dm_target_offset(ti, sector); | |
1401 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1402 | max_len = sector_div(offset, ti->max_io_len); | |
1403 | else | |
1404 | max_len = offset & (ti->max_io_len - 1); | |
1405 | max_len = ti->max_io_len - max_len; | |
1406 | ||
1407 | if (len > max_len) | |
1408 | len = max_len; | |
1da177e4 LT |
1409 | } |
1410 | ||
1411 | return len; | |
1412 | } | |
1413 | ||
542f9038 MS |
1414 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1415 | { | |
1416 | if (len > UINT_MAX) { | |
1417 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1418 | (unsigned long long)len, UINT_MAX); | |
1419 | ti->error = "Maximum size of target IO is too large"; | |
1420 | return -EINVAL; | |
1421 | } | |
1422 | ||
1423 | ti->max_io_len = (uint32_t) len; | |
1424 | ||
1425 | return 0; | |
1426 | } | |
1427 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1428 | ||
1dd40c3e MP |
1429 | /* |
1430 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1431 | * allowed for all bio types except REQ_FLUSH. | |
1432 | * | |
1433 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1434 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1435 | * sent in a next bio. | |
1436 | * | |
1437 | * A diagram that explains the arithmetics: | |
1438 | * +--------------------+---------------+-------+ | |
1439 | * | 1 | 2 | 3 | | |
1440 | * +--------------------+---------------+-------+ | |
1441 | * | |
1442 | * <-------------- *tio->len_ptr ---------------> | |
1443 | * <------- bi_size -------> | |
1444 | * <-- n_sectors --> | |
1445 | * | |
1446 | * Region 1 was already iterated over with bio_advance or similar function. | |
1447 | * (it may be empty if the target doesn't use bio_advance) | |
1448 | * Region 2 is the remaining bio size that the target wants to process. | |
1449 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1450 | * to make it empty) | |
1451 | * The target requires that region 3 is to be sent in the next bio. | |
1452 | * | |
1453 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1454 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1455 | * copies of the bio. | |
1456 | */ | |
1457 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1458 | { | |
1459 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1460 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1461 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1462 | BUG_ON(bi_size > *tio->len_ptr); | |
1463 | BUG_ON(n_sectors > bi_size); | |
1464 | *tio->len_ptr -= bi_size - n_sectors; | |
1465 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1466 | } | |
1467 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1468 | ||
bd2a49b8 | 1469 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1470 | { |
1471 | int r; | |
2056a782 | 1472 | sector_t sector; |
9faf400f | 1473 | struct mapped_device *md; |
dba14160 | 1474 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1475 | struct dm_target *ti = tio->ti; |
1da177e4 | 1476 | |
1da177e4 | 1477 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1478 | |
1479 | /* | |
1480 | * Map the clone. If r == 0 we don't need to do | |
1481 | * anything, the target has assumed ownership of | |
1482 | * this io. | |
1483 | */ | |
1484 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1485 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1486 | r = ti->type->map(ti, clone); |
45cbcd79 | 1487 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1488 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1489 | |
d07335e5 MS |
1490 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1491 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1492 | |
1da177e4 | 1493 | generic_make_request(clone); |
2e93ccc1 KU |
1494 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1495 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1496 | md = tio->io->md; |
1497 | dec_pending(tio->io, r); | |
9faf400f | 1498 | free_tio(md, tio); |
ab37844d | 1499 | } else if (r != DM_MAPIO_SUBMITTED) { |
45cbcd79 KU |
1500 | DMWARN("unimplemented target map return value: %d", r); |
1501 | BUG(); | |
1da177e4 LT |
1502 | } |
1503 | } | |
1504 | ||
1505 | struct clone_info { | |
1506 | struct mapped_device *md; | |
1507 | struct dm_table *map; | |
1508 | struct bio *bio; | |
1509 | struct dm_io *io; | |
1510 | sector_t sector; | |
e0d6609a | 1511 | unsigned sector_count; |
1da177e4 LT |
1512 | }; |
1513 | ||
e0d6609a | 1514 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1515 | { |
4f024f37 KO |
1516 | bio->bi_iter.bi_sector = sector; |
1517 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1518 | } |
1519 | ||
1520 | /* | |
1521 | * Creates a bio that consists of range of complete bvecs. | |
1522 | */ | |
dba14160 | 1523 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1524 | sector_t sector, unsigned len) |
1da177e4 | 1525 | { |
dba14160 | 1526 | struct bio *clone = &tio->clone; |
1da177e4 | 1527 | |
1c3b13e6 KO |
1528 | __bio_clone_fast(clone, bio); |
1529 | ||
1530 | if (bio_integrity(bio)) | |
1531 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1532 | |
1c3b13e6 KO |
1533 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1534 | clone->bi_iter.bi_size = to_bytes(len); | |
1535 | ||
1536 | if (bio_integrity(bio)) | |
1537 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1538 | } |
1539 | ||
9015df24 | 1540 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1541 | struct dm_target *ti, |
55a62eef | 1542 | unsigned target_bio_nr) |
f9ab94ce | 1543 | { |
dba14160 MP |
1544 | struct dm_target_io *tio; |
1545 | struct bio *clone; | |
1546 | ||
99778273 | 1547 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1548 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1549 | |
1550 | tio->io = ci->io; | |
1551 | tio->ti = ti; | |
55a62eef | 1552 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1553 | |
1554 | return tio; | |
1555 | } | |
1556 | ||
14fe594d AK |
1557 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1558 | struct dm_target *ti, | |
1dd40c3e | 1559 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1560 | { |
99778273 | 1561 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1562 | struct bio *clone = &tio->clone; |
9015df24 | 1563 | |
1dd40c3e MP |
1564 | tio->len_ptr = len; |
1565 | ||
99778273 | 1566 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1567 | if (len) |
1dd40c3e | 1568 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1569 | |
bd2a49b8 | 1570 | __map_bio(tio); |
f9ab94ce MP |
1571 | } |
1572 | ||
14fe594d | 1573 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1574 | unsigned num_bios, unsigned *len) |
06a426ce | 1575 | { |
55a62eef | 1576 | unsigned target_bio_nr; |
06a426ce | 1577 | |
55a62eef | 1578 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1579 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1580 | } |
1581 | ||
14fe594d | 1582 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1583 | { |
06a426ce | 1584 | unsigned target_nr = 0; |
f9ab94ce MP |
1585 | struct dm_target *ti; |
1586 | ||
b372d360 | 1587 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1588 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1589 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1590 | |
f9ab94ce MP |
1591 | return 0; |
1592 | } | |
1593 | ||
e4c93811 | 1594 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1595 | sector_t sector, unsigned *len) |
5ae89a87 | 1596 | { |
dba14160 | 1597 | struct bio *bio = ci->bio; |
5ae89a87 | 1598 | struct dm_target_io *tio; |
b0d8ed4d AK |
1599 | unsigned target_bio_nr; |
1600 | unsigned num_target_bios = 1; | |
5ae89a87 | 1601 | |
b0d8ed4d AK |
1602 | /* |
1603 | * Does the target want to receive duplicate copies of the bio? | |
1604 | */ | |
1605 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1606 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1607 | |
b0d8ed4d | 1608 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1609 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1610 | tio->len_ptr = len; |
1611 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1612 | __map_bio(tio); |
1613 | } | |
5ae89a87 MS |
1614 | } |
1615 | ||
55a62eef | 1616 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1617 | |
55a62eef | 1618 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1619 | { |
55a62eef | 1620 | return ti->num_discard_bios; |
23508a96 MS |
1621 | } |
1622 | ||
55a62eef | 1623 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1624 | { |
55a62eef | 1625 | return ti->num_write_same_bios; |
23508a96 MS |
1626 | } |
1627 | ||
1628 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1629 | ||
1630 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1631 | { | |
55a62eef | 1632 | return ti->split_discard_bios; |
23508a96 MS |
1633 | } |
1634 | ||
14fe594d AK |
1635 | static int __send_changing_extent_only(struct clone_info *ci, |
1636 | get_num_bios_fn get_num_bios, | |
1637 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1638 | { |
1639 | struct dm_target *ti; | |
e0d6609a | 1640 | unsigned len; |
55a62eef | 1641 | unsigned num_bios; |
5ae89a87 | 1642 | |
a79245b3 MS |
1643 | do { |
1644 | ti = dm_table_find_target(ci->map, ci->sector); | |
1645 | if (!dm_target_is_valid(ti)) | |
1646 | return -EIO; | |
5ae89a87 | 1647 | |
5ae89a87 | 1648 | /* |
23508a96 MS |
1649 | * Even though the device advertised support for this type of |
1650 | * request, that does not mean every target supports it, and | |
936688d7 | 1651 | * reconfiguration might also have changed that since the |
a79245b3 | 1652 | * check was performed. |
5ae89a87 | 1653 | */ |
55a62eef AK |
1654 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1655 | if (!num_bios) | |
a79245b3 | 1656 | return -EOPNOTSUPP; |
5ae89a87 | 1657 | |
23508a96 | 1658 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1659 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1660 | else |
e0d6609a | 1661 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1662 | |
1dd40c3e | 1663 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1664 | |
1665 | ci->sector += len; | |
1666 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1667 | |
1668 | return 0; | |
1669 | } | |
1670 | ||
14fe594d | 1671 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1672 | { |
14fe594d AK |
1673 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1674 | is_split_required_for_discard); | |
23508a96 MS |
1675 | } |
1676 | ||
14fe594d | 1677 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1678 | { |
14fe594d | 1679 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1680 | } |
1681 | ||
e4c93811 AK |
1682 | /* |
1683 | * Select the correct strategy for processing a non-flush bio. | |
1684 | */ | |
14fe594d | 1685 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1686 | { |
dba14160 | 1687 | struct bio *bio = ci->bio; |
512875bd | 1688 | struct dm_target *ti; |
1c3b13e6 | 1689 | unsigned len; |
1da177e4 | 1690 | |
5ae89a87 | 1691 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1692 | return __send_discard(ci); |
23508a96 | 1693 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1694 | return __send_write_same(ci); |
5ae89a87 | 1695 | |
512875bd JN |
1696 | ti = dm_table_find_target(ci->map, ci->sector); |
1697 | if (!dm_target_is_valid(ti)) | |
1698 | return -EIO; | |
1699 | ||
1c3b13e6 | 1700 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1701 | |
1dd40c3e | 1702 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1703 | |
1c3b13e6 KO |
1704 | ci->sector += len; |
1705 | ci->sector_count -= len; | |
1da177e4 | 1706 | |
1c3b13e6 | 1707 | return 0; |
1da177e4 LT |
1708 | } |
1709 | ||
1710 | /* | |
14fe594d | 1711 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1712 | */ |
83d5e5b0 MP |
1713 | static void __split_and_process_bio(struct mapped_device *md, |
1714 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1715 | { |
1716 | struct clone_info ci; | |
512875bd | 1717 | int error = 0; |
1da177e4 | 1718 | |
83d5e5b0 | 1719 | if (unlikely(!map)) { |
6a8736d1 | 1720 | bio_io_error(bio); |
f0b9a450 MP |
1721 | return; |
1722 | } | |
692d0eb9 | 1723 | |
83d5e5b0 | 1724 | ci.map = map; |
1da177e4 | 1725 | ci.md = md; |
1da177e4 LT |
1726 | ci.io = alloc_io(md); |
1727 | ci.io->error = 0; | |
1728 | atomic_set(&ci.io->io_count, 1); | |
1729 | ci.io->bio = bio; | |
1730 | ci.io->md = md; | |
f88fb981 | 1731 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1732 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1733 | |
3eaf840e | 1734 | start_io_acct(ci.io); |
bd2a49b8 | 1735 | |
b372d360 MS |
1736 | if (bio->bi_rw & REQ_FLUSH) { |
1737 | ci.bio = &ci.md->flush_bio; | |
1738 | ci.sector_count = 0; | |
14fe594d | 1739 | error = __send_empty_flush(&ci); |
b372d360 MS |
1740 | /* dec_pending submits any data associated with flush */ |
1741 | } else { | |
6a8736d1 | 1742 | ci.bio = bio; |
d87f4c14 | 1743 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1744 | while (ci.sector_count && !error) |
14fe594d | 1745 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1746 | } |
1da177e4 LT |
1747 | |
1748 | /* drop the extra reference count */ | |
512875bd | 1749 | dec_pending(ci.io, error); |
1da177e4 LT |
1750 | } |
1751 | /*----------------------------------------------------------------- | |
1752 | * CRUD END | |
1753 | *---------------------------------------------------------------*/ | |
1754 | ||
1755 | /* | |
1756 | * The request function that just remaps the bio built up by | |
1757 | * dm_merge_bvec. | |
1758 | */ | |
dece1635 | 1759 | static blk_qc_t dm_make_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1760 | { |
12f03a49 | 1761 | int rw = bio_data_dir(bio); |
1da177e4 | 1762 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1763 | int srcu_idx; |
1764 | struct dm_table *map; | |
1da177e4 | 1765 | |
83d5e5b0 | 1766 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1767 | |
18c0b223 | 1768 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1769 | |
6a8736d1 TH |
1770 | /* if we're suspended, we have to queue this io for later */ |
1771 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1772 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1773 | |
6a8736d1 TH |
1774 | if (bio_rw(bio) != READA) |
1775 | queue_io(md, bio); | |
1776 | else | |
54d9a1b4 | 1777 | bio_io_error(bio); |
dece1635 | 1778 | return BLK_QC_T_NONE; |
1da177e4 LT |
1779 | } |
1780 | ||
83d5e5b0 MP |
1781 | __split_and_process_bio(md, map, bio); |
1782 | dm_put_live_table(md, srcu_idx); | |
dece1635 | 1783 | return BLK_QC_T_NONE; |
cec47e3d KU |
1784 | } |
1785 | ||
fd2ed4d2 | 1786 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1787 | { |
1788 | return blk_queue_stackable(md->queue); | |
1789 | } | |
1790 | ||
466d89a6 | 1791 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1792 | { |
1793 | int r; | |
1794 | ||
466d89a6 KB |
1795 | if (blk_queue_io_stat(clone->q)) |
1796 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1797 | |
466d89a6 KB |
1798 | clone->start_time = jiffies; |
1799 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1800 | if (r) |
466d89a6 | 1801 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1802 | dm_complete_request(rq, r); |
1803 | } | |
cec47e3d | 1804 | |
78d8e58a MS |
1805 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1806 | void *data) | |
cec47e3d | 1807 | { |
78d8e58a MS |
1808 | struct dm_rq_target_io *tio = data; |
1809 | struct dm_rq_clone_bio_info *info = | |
1810 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
1811 | ||
1812 | info->orig = bio_orig; | |
1813 | info->tio = tio; | |
1814 | bio->bi_end_io = end_clone_bio; | |
1815 | ||
1816 | return 0; | |
1817 | } | |
1818 | ||
1819 | static int setup_clone(struct request *clone, struct request *rq, | |
1820 | struct dm_rq_target_io *tio, gfp_t gfp_mask) | |
cec47e3d | 1821 | { |
78d8e58a MS |
1822 | int r; |
1823 | ||
1824 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, | |
1825 | dm_rq_bio_constructor, tio); | |
1826 | if (r) | |
1827 | return r; | |
1828 | ||
1829 | clone->cmd = rq->cmd; | |
1830 | clone->cmd_len = rq->cmd_len; | |
1831 | clone->sense = rq->sense; | |
cec47e3d KU |
1832 | clone->end_io = end_clone_request; |
1833 | clone->end_io_data = tio; | |
78d8e58a | 1834 | |
1ae49ea2 | 1835 | tio->clone = clone; |
78d8e58a MS |
1836 | |
1837 | return 0; | |
cec47e3d KU |
1838 | } |
1839 | ||
6facdaff | 1840 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1841 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 | 1842 | { |
02233342 MS |
1843 | /* |
1844 | * Do not allocate a clone if tio->clone was already set | |
1845 | * (see: dm_mq_queue_rq). | |
1846 | */ | |
1847 | bool alloc_clone = !tio->clone; | |
1848 | struct request *clone; | |
1ae49ea2 | 1849 | |
02233342 MS |
1850 | if (alloc_clone) { |
1851 | clone = alloc_clone_request(md, gfp_mask); | |
1852 | if (!clone) | |
1853 | return NULL; | |
1854 | } else | |
1855 | clone = tio->clone; | |
1ae49ea2 MS |
1856 | |
1857 | blk_rq_init(NULL, clone); | |
78d8e58a MS |
1858 | if (setup_clone(clone, rq, tio, gfp_mask)) { |
1859 | /* -ENOMEM */ | |
1860 | if (alloc_clone) | |
1861 | free_clone_request(md, clone); | |
1862 | return NULL; | |
1863 | } | |
1ae49ea2 MS |
1864 | |
1865 | return clone; | |
1866 | } | |
1867 | ||
2eb6e1e3 KB |
1868 | static void map_tio_request(struct kthread_work *work); |
1869 | ||
bfebd1cd MS |
1870 | static void init_tio(struct dm_rq_target_io *tio, struct request *rq, |
1871 | struct mapped_device *md) | |
1872 | { | |
1873 | tio->md = md; | |
1874 | tio->ti = NULL; | |
1875 | tio->clone = NULL; | |
1876 | tio->orig = rq; | |
1877 | tio->error = 0; | |
1878 | memset(&tio->info, 0, sizeof(tio->info)); | |
02233342 MS |
1879 | if (md->kworker_task) |
1880 | init_kthread_work(&tio->work, map_tio_request); | |
bfebd1cd MS |
1881 | } |
1882 | ||
466d89a6 KB |
1883 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1884 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1885 | { |
6facdaff | 1886 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1887 | int srcu_idx; |
1888 | struct dm_table *table; | |
6facdaff KU |
1889 | |
1890 | tio = alloc_rq_tio(md, gfp_mask); | |
1891 | if (!tio) | |
1892 | return NULL; | |
1893 | ||
bfebd1cd | 1894 | init_tio(tio, rq, md); |
6facdaff | 1895 | |
e5863d9a MS |
1896 | table = dm_get_live_table(md, &srcu_idx); |
1897 | if (!dm_table_mq_request_based(table)) { | |
1898 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1899 | dm_put_live_table(md, srcu_idx); | |
1900 | free_rq_tio(tio); | |
1901 | return NULL; | |
1902 | } | |
6facdaff | 1903 | } |
e5863d9a | 1904 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1905 | |
466d89a6 | 1906 | return tio; |
6facdaff KU |
1907 | } |
1908 | ||
cec47e3d KU |
1909 | /* |
1910 | * Called with the queue lock held. | |
1911 | */ | |
1912 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1913 | { | |
1914 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1915 | struct dm_rq_target_io *tio; |
cec47e3d | 1916 | |
cec47e3d KU |
1917 | if (unlikely(rq->special)) { |
1918 | DMWARN("Already has something in rq->special."); | |
1919 | return BLKPREP_KILL; | |
1920 | } | |
1921 | ||
466d89a6 KB |
1922 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1923 | if (!tio) | |
cec47e3d | 1924 | return BLKPREP_DEFER; |
cec47e3d | 1925 | |
466d89a6 | 1926 | rq->special = tio; |
cec47e3d KU |
1927 | rq->cmd_flags |= REQ_DONTPREP; |
1928 | ||
1929 | return BLKPREP_OK; | |
1930 | } | |
1931 | ||
9eef87da KU |
1932 | /* |
1933 | * Returns: | |
e5863d9a MS |
1934 | * 0 : the request has been processed |
1935 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1936 | * < 0 : the request was completed due to failure | |
9eef87da | 1937 | */ |
bfebd1cd | 1938 | static int map_request(struct dm_rq_target_io *tio, struct request *rq, |
9eef87da | 1939 | struct mapped_device *md) |
cec47e3d | 1940 | { |
e5863d9a | 1941 | int r; |
bfebd1cd | 1942 | struct dm_target *ti = tio->ti; |
e5863d9a MS |
1943 | struct request *clone = NULL; |
1944 | ||
1945 | if (tio->clone) { | |
1946 | clone = tio->clone; | |
1947 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1948 | } else { | |
1949 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1950 | if (r < 0) { | |
1951 | /* The target wants to complete the I/O */ | |
1952 | dm_kill_unmapped_request(rq, r); | |
1953 | return r; | |
1954 | } | |
3a140755 JN |
1955 | if (r != DM_MAPIO_REMAPPED) |
1956 | return r; | |
78d8e58a MS |
1957 | if (setup_clone(clone, rq, tio, GFP_ATOMIC)) { |
1958 | /* -ENOMEM */ | |
1959 | ti->type->release_clone_rq(clone); | |
1960 | return DM_MAPIO_REQUEUE; | |
1961 | } | |
e5863d9a | 1962 | } |
cec47e3d | 1963 | |
cec47e3d KU |
1964 | switch (r) { |
1965 | case DM_MAPIO_SUBMITTED: | |
1966 | /* The target has taken the I/O to submit by itself later */ | |
1967 | break; | |
1968 | case DM_MAPIO_REMAPPED: | |
1969 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1970 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1971 | blk_rq_pos(rq)); |
1972 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1973 | break; |
1974 | case DM_MAPIO_REQUEUE: | |
1975 | /* The target wants to requeue the I/O */ | |
2d76fff1 | 1976 | dm_requeue_original_request(md, tio->orig); |
cec47e3d KU |
1977 | break; |
1978 | default: | |
1979 | if (r > 0) { | |
1980 | DMWARN("unimplemented target map return value: %d", r); | |
1981 | BUG(); | |
1982 | } | |
1983 | ||
1984 | /* The target wants to complete the I/O */ | |
466d89a6 | 1985 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1986 | return r; |
cec47e3d | 1987 | } |
9eef87da | 1988 | |
e5863d9a | 1989 | return 0; |
cec47e3d KU |
1990 | } |
1991 | ||
2eb6e1e3 | 1992 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 1993 | { |
2eb6e1e3 | 1994 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
1995 | struct request *rq = tio->orig; |
1996 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 1997 | |
bfebd1cd | 1998 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) |
2d76fff1 | 1999 | dm_requeue_original_request(md, rq); |
2eb6e1e3 KB |
2000 | } |
2001 | ||
466d89a6 | 2002 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 2003 | { |
bfebd1cd MS |
2004 | if (!orig->q->mq_ops) |
2005 | blk_start_request(orig); | |
2006 | else | |
2007 | blk_mq_start_request(orig); | |
466d89a6 | 2008 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 | 2009 | |
0ce65797 MS |
2010 | if (md->seq_rq_merge_deadline_usecs) { |
2011 | md->last_rq_pos = rq_end_sector(orig); | |
2012 | md->last_rq_rw = rq_data_dir(orig); | |
2013 | md->last_rq_start_time = ktime_get(); | |
2014 | } | |
de3ec86d | 2015 | |
e262f347 MP |
2016 | if (unlikely(dm_stats_used(&md->stats))) { |
2017 | struct dm_rq_target_io *tio = tio_from_request(orig); | |
2018 | tio->duration_jiffies = jiffies; | |
2019 | tio->n_sectors = blk_rq_sectors(orig); | |
2020 | dm_stats_account_io(&md->stats, orig->cmd_flags, blk_rq_pos(orig), | |
2021 | tio->n_sectors, false, 0, &tio->stats_aux); | |
2022 | } | |
2023 | ||
ba1cbad9 MS |
2024 | /* |
2025 | * Hold the md reference here for the in-flight I/O. | |
2026 | * We can't rely on the reference count by device opener, | |
2027 | * because the device may be closed during the request completion | |
2028 | * when all bios are completed. | |
2029 | * See the comment in rq_completed() too. | |
2030 | */ | |
2031 | dm_get(md); | |
ba1cbad9 MS |
2032 | } |
2033 | ||
0ce65797 MS |
2034 | #define MAX_SEQ_RQ_MERGE_DEADLINE_USECS 100000 |
2035 | ||
2036 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_show(struct mapped_device *md, char *buf) | |
2037 | { | |
2038 | return sprintf(buf, "%u\n", md->seq_rq_merge_deadline_usecs); | |
2039 | } | |
2040 | ||
2041 | ssize_t dm_attr_rq_based_seq_io_merge_deadline_store(struct mapped_device *md, | |
2042 | const char *buf, size_t count) | |
2043 | { | |
2044 | unsigned deadline; | |
2045 | ||
17e149b8 | 2046 | if (!dm_request_based(md) || md->use_blk_mq) |
0ce65797 MS |
2047 | return count; |
2048 | ||
2049 | if (kstrtouint(buf, 10, &deadline)) | |
2050 | return -EINVAL; | |
2051 | ||
2052 | if (deadline > MAX_SEQ_RQ_MERGE_DEADLINE_USECS) | |
2053 | deadline = MAX_SEQ_RQ_MERGE_DEADLINE_USECS; | |
2054 | ||
2055 | md->seq_rq_merge_deadline_usecs = deadline; | |
2056 | ||
2057 | return count; | |
2058 | } | |
2059 | ||
2060 | static bool dm_request_peeked_before_merge_deadline(struct mapped_device *md) | |
2061 | { | |
2062 | ktime_t kt_deadline; | |
2063 | ||
2064 | if (!md->seq_rq_merge_deadline_usecs) | |
2065 | return false; | |
2066 | ||
2067 | kt_deadline = ns_to_ktime((u64)md->seq_rq_merge_deadline_usecs * NSEC_PER_USEC); | |
2068 | kt_deadline = ktime_add_safe(md->last_rq_start_time, kt_deadline); | |
2069 | ||
2070 | return !ktime_after(ktime_get(), kt_deadline); | |
2071 | } | |
2072 | ||
cec47e3d KU |
2073 | /* |
2074 | * q->request_fn for request-based dm. | |
2075 | * Called with the queue lock held. | |
2076 | */ | |
2077 | static void dm_request_fn(struct request_queue *q) | |
2078 | { | |
2079 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
2080 | int srcu_idx; |
2081 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 2082 | struct dm_target *ti; |
466d89a6 | 2083 | struct request *rq; |
2eb6e1e3 | 2084 | struct dm_rq_target_io *tio; |
29e4013d | 2085 | sector_t pos; |
cec47e3d KU |
2086 | |
2087 | /* | |
b4324fee KU |
2088 | * For suspend, check blk_queue_stopped() and increment |
2089 | * ->pending within a single queue_lock not to increment the | |
2090 | * number of in-flight I/Os after the queue is stopped in | |
2091 | * dm_suspend(). | |
cec47e3d | 2092 | */ |
7eaceacc | 2093 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
2094 | rq = blk_peek_request(q); |
2095 | if (!rq) | |
9d1deb83 | 2096 | goto out; |
cec47e3d | 2097 | |
29e4013d TH |
2098 | /* always use block 0 to find the target for flushes for now */ |
2099 | pos = 0; | |
2100 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2101 | pos = blk_rq_pos(rq); | |
2102 | ||
2103 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
2104 | if (!dm_target_is_valid(ti)) { |
2105 | /* | |
466d89a6 | 2106 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
2107 | * before calling dm_kill_unmapped_request |
2108 | */ | |
2109 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
2110 | dm_start_request(md, rq); |
2111 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
2112 | continue; |
2113 | } | |
d0bcb878 | 2114 | |
0ce65797 MS |
2115 | if (dm_request_peeked_before_merge_deadline(md) && |
2116 | md_in_flight(md) && rq->bio && rq->bio->bi_vcnt == 1 && | |
de3ec86d MS |
2117 | md->last_rq_pos == pos && md->last_rq_rw == rq_data_dir(rq)) |
2118 | goto delay_and_out; | |
2119 | ||
cec47e3d | 2120 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 2121 | goto delay_and_out; |
cec47e3d | 2122 | |
466d89a6 | 2123 | dm_start_request(md, rq); |
9eef87da | 2124 | |
bfebd1cd | 2125 | tio = tio_from_request(rq); |
2eb6e1e3 KB |
2126 | /* Establish tio->ti before queuing work (map_tio_request) */ |
2127 | tio->ti = ti; | |
2128 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 2129 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
2130 | } |
2131 | ||
2132 | goto out; | |
2133 | ||
7eaceacc | 2134 | delay_and_out: |
d548b34b | 2135 | blk_delay_queue(q, HZ / 100); |
cec47e3d | 2136 | out: |
83d5e5b0 | 2137 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2138 | } |
2139 | ||
1da177e4 LT |
2140 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2141 | { | |
8a57dfc6 CS |
2142 | int r = bdi_bits; |
2143 | struct mapped_device *md = congested_data; | |
2144 | struct dm_table *map; | |
1da177e4 | 2145 | |
1eb787ec | 2146 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2147 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2148 | if (map) { |
cec47e3d KU |
2149 | /* |
2150 | * Request-based dm cares about only own queue for | |
2151 | * the query about congestion status of request_queue | |
2152 | */ | |
2153 | if (dm_request_based(md)) | |
4452226e | 2154 | r = md->queue->backing_dev_info.wb.state & |
cec47e3d KU |
2155 | bdi_bits; |
2156 | else | |
2157 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2158 | } |
83d5e5b0 | 2159 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2160 | } |
2161 | ||
1da177e4 LT |
2162 | return r; |
2163 | } | |
2164 | ||
2165 | /*----------------------------------------------------------------- | |
2166 | * An IDR is used to keep track of allocated minor numbers. | |
2167 | *---------------------------------------------------------------*/ | |
2b06cfff | 2168 | static void free_minor(int minor) |
1da177e4 | 2169 | { |
f32c10b0 | 2170 | spin_lock(&_minor_lock); |
1da177e4 | 2171 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2172 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | /* | |
2176 | * See if the device with a specific minor # is free. | |
2177 | */ | |
cf13ab8e | 2178 | static int specific_minor(int minor) |
1da177e4 | 2179 | { |
c9d76be6 | 2180 | int r; |
1da177e4 LT |
2181 | |
2182 | if (minor >= (1 << MINORBITS)) | |
2183 | return -EINVAL; | |
2184 | ||
c9d76be6 | 2185 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2186 | spin_lock(&_minor_lock); |
1da177e4 | 2187 | |
c9d76be6 | 2188 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2189 | |
f32c10b0 | 2190 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2191 | idr_preload_end(); |
2192 | if (r < 0) | |
2193 | return r == -ENOSPC ? -EBUSY : r; | |
2194 | return 0; | |
1da177e4 LT |
2195 | } |
2196 | ||
cf13ab8e | 2197 | static int next_free_minor(int *minor) |
1da177e4 | 2198 | { |
c9d76be6 | 2199 | int r; |
62f75c2f | 2200 | |
c9d76be6 | 2201 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2202 | spin_lock(&_minor_lock); |
1da177e4 | 2203 | |
c9d76be6 | 2204 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2205 | |
f32c10b0 | 2206 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2207 | idr_preload_end(); |
2208 | if (r < 0) | |
2209 | return r; | |
2210 | *minor = r; | |
2211 | return 0; | |
1da177e4 LT |
2212 | } |
2213 | ||
83d5cde4 | 2214 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2215 | |
53d5914f MP |
2216 | static void dm_wq_work(struct work_struct *work); |
2217 | ||
4a0b4ddf MS |
2218 | static void dm_init_md_queue(struct mapped_device *md) |
2219 | { | |
2220 | /* | |
2221 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
bfebd1cd | 2222 | * devices. The type of this dm device may not have been decided yet. |
4a0b4ddf MS |
2223 | * The type is decided at the first table loading time. |
2224 | * To prevent problematic device stacking, clear the queue flag | |
2225 | * for request stacking support until then. | |
2226 | * | |
2227 | * This queue is new, so no concurrency on the queue_flags. | |
2228 | */ | |
2229 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
ad5f498f MP |
2230 | |
2231 | /* | |
2232 | * Initialize data that will only be used by a non-blk-mq DM queue | |
2233 | * - must do so here (in alloc_dev callchain) before queue is used | |
2234 | */ | |
2235 | md->queue->queuedata = md; | |
2236 | md->queue->backing_dev_info.congested_data = md; | |
bfebd1cd | 2237 | } |
4a0b4ddf | 2238 | |
bfebd1cd MS |
2239 | static void dm_init_old_md_queue(struct mapped_device *md) |
2240 | { | |
17e149b8 | 2241 | md->use_blk_mq = false; |
bfebd1cd MS |
2242 | dm_init_md_queue(md); |
2243 | ||
2244 | /* | |
2245 | * Initialize aspects of queue that aren't relevant for blk-mq | |
2246 | */ | |
4a0b4ddf | 2247 | md->queue->backing_dev_info.congested_fn = dm_any_congested; |
4a0b4ddf | 2248 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
4a0b4ddf MS |
2249 | } |
2250 | ||
0f20972f MS |
2251 | static void cleanup_mapped_device(struct mapped_device *md) |
2252 | { | |
0f20972f MS |
2253 | if (md->wq) |
2254 | destroy_workqueue(md->wq); | |
2255 | if (md->kworker_task) | |
2256 | kthread_stop(md->kworker_task); | |
6f65985e JL |
2257 | mempool_destroy(md->io_pool); |
2258 | mempool_destroy(md->rq_pool); | |
0f20972f MS |
2259 | if (md->bs) |
2260 | bioset_free(md->bs); | |
2261 | ||
b06075a9 MP |
2262 | cleanup_srcu_struct(&md->io_barrier); |
2263 | ||
0f20972f MS |
2264 | if (md->disk) { |
2265 | spin_lock(&_minor_lock); | |
2266 | md->disk->private_data = NULL; | |
2267 | spin_unlock(&_minor_lock); | |
0f20972f MS |
2268 | del_gendisk(md->disk); |
2269 | put_disk(md->disk); | |
2270 | } | |
2271 | ||
2272 | if (md->queue) | |
2273 | blk_cleanup_queue(md->queue); | |
2274 | ||
2275 | if (md->bdev) { | |
2276 | bdput(md->bdev); | |
2277 | md->bdev = NULL; | |
2278 | } | |
2279 | } | |
2280 | ||
1da177e4 LT |
2281 | /* |
2282 | * Allocate and initialise a blank device with a given minor. | |
2283 | */ | |
2b06cfff | 2284 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2285 | { |
2286 | int r; | |
cf13ab8e | 2287 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2288 | void *old_md; |
1da177e4 LT |
2289 | |
2290 | if (!md) { | |
2291 | DMWARN("unable to allocate device, out of memory."); | |
2292 | return NULL; | |
2293 | } | |
2294 | ||
10da4f79 | 2295 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2296 | goto bad_module_get; |
10da4f79 | 2297 | |
1da177e4 | 2298 | /* get a minor number for the dev */ |
2b06cfff | 2299 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2300 | r = next_free_minor(&minor); |
2b06cfff | 2301 | else |
cf13ab8e | 2302 | r = specific_minor(minor); |
1da177e4 | 2303 | if (r < 0) |
6ed7ade8 | 2304 | goto bad_minor; |
1da177e4 | 2305 | |
83d5e5b0 MP |
2306 | r = init_srcu_struct(&md->io_barrier); |
2307 | if (r < 0) | |
2308 | goto bad_io_barrier; | |
2309 | ||
17e149b8 | 2310 | md->use_blk_mq = use_blk_mq; |
a5664dad | 2311 | md->type = DM_TYPE_NONE; |
e61290a4 | 2312 | mutex_init(&md->suspend_lock); |
a5664dad | 2313 | mutex_init(&md->type_lock); |
86f1152b | 2314 | mutex_init(&md->table_devices_lock); |
022c2611 | 2315 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2316 | atomic_set(&md->holders, 1); |
5c6bd75d | 2317 | atomic_set(&md->open_count, 0); |
1da177e4 | 2318 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2319 | atomic_set(&md->uevent_seq, 0); |
2320 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2321 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2322 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2323 | |
4a0b4ddf | 2324 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2325 | if (!md->queue) |
0f20972f | 2326 | goto bad; |
1da177e4 | 2327 | |
4a0b4ddf | 2328 | dm_init_md_queue(md); |
9faf400f | 2329 | |
1da177e4 LT |
2330 | md->disk = alloc_disk(1); |
2331 | if (!md->disk) | |
0f20972f | 2332 | goto bad; |
1da177e4 | 2333 | |
316d315b NK |
2334 | atomic_set(&md->pending[0], 0); |
2335 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2336 | init_waitqueue_head(&md->wait); |
53d5914f | 2337 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2338 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2339 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2340 | md->kworker_task = NULL; |
f0b04115 | 2341 | |
1da177e4 LT |
2342 | md->disk->major = _major; |
2343 | md->disk->first_minor = minor; | |
2344 | md->disk->fops = &dm_blk_dops; | |
2345 | md->disk->queue = md->queue; | |
2346 | md->disk->private_data = md; | |
2347 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2348 | add_disk(md->disk); | |
7e51f257 | 2349 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2350 | |
670368a8 | 2351 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a | 2352 | if (!md->wq) |
0f20972f | 2353 | goto bad; |
304f3f6a | 2354 | |
32a926da MP |
2355 | md->bdev = bdget_disk(md->disk, 0); |
2356 | if (!md->bdev) | |
0f20972f | 2357 | goto bad; |
32a926da | 2358 | |
6a8736d1 TH |
2359 | bio_init(&md->flush_bio); |
2360 | md->flush_bio.bi_bdev = md->bdev; | |
2361 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2362 | ||
fd2ed4d2 MP |
2363 | dm_stats_init(&md->stats); |
2364 | ||
ba61fdd1 | 2365 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2366 | spin_lock(&_minor_lock); |
ba61fdd1 | 2367 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2368 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2369 | |
2370 | BUG_ON(old_md != MINOR_ALLOCED); | |
2371 | ||
1da177e4 LT |
2372 | return md; |
2373 | ||
0f20972f MS |
2374 | bad: |
2375 | cleanup_mapped_device(md); | |
83d5e5b0 | 2376 | bad_io_barrier: |
1da177e4 | 2377 | free_minor(minor); |
6ed7ade8 | 2378 | bad_minor: |
10da4f79 | 2379 | module_put(THIS_MODULE); |
6ed7ade8 | 2380 | bad_module_get: |
1da177e4 LT |
2381 | kfree(md); |
2382 | return NULL; | |
2383 | } | |
2384 | ||
ae9da83f JN |
2385 | static void unlock_fs(struct mapped_device *md); |
2386 | ||
1da177e4 LT |
2387 | static void free_dev(struct mapped_device *md) |
2388 | { | |
f331c029 | 2389 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2390 | |
32a926da | 2391 | unlock_fs(md); |
2eb6e1e3 | 2392 | |
0f20972f MS |
2393 | cleanup_mapped_device(md); |
2394 | if (md->use_blk_mq) | |
2395 | blk_mq_free_tag_set(&md->tag_set); | |
63a4f065 | 2396 | |
86f1152b | 2397 | free_table_devices(&md->table_devices); |
63a4f065 | 2398 | dm_stats_cleanup(&md->stats); |
63a4f065 MS |
2399 | free_minor(minor); |
2400 | ||
10da4f79 | 2401 | module_put(THIS_MODULE); |
1da177e4 LT |
2402 | kfree(md); |
2403 | } | |
2404 | ||
e6ee8c0b KU |
2405 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2406 | { | |
c0820cf5 | 2407 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2408 | |
4e6e36c3 MS |
2409 | if (md->bs) { |
2410 | /* The md already has necessary mempools. */ | |
2411 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
16245bdc JN |
2412 | /* |
2413 | * Reload bioset because front_pad may have changed | |
2414 | * because a different table was loaded. | |
2415 | */ | |
2416 | bioset_free(md->bs); | |
2417 | md->bs = p->bs; | |
2418 | p->bs = NULL; | |
16245bdc | 2419 | } |
4e6e36c3 MS |
2420 | /* |
2421 | * There's no need to reload with request-based dm | |
2422 | * because the size of front_pad doesn't change. | |
2423 | * Note for future: If you are to reload bioset, | |
2424 | * prep-ed requests in the queue may refer | |
2425 | * to bio from the old bioset, so you must walk | |
2426 | * through the queue to unprep. | |
2427 | */ | |
2428 | goto out; | |
c0820cf5 | 2429 | } |
e6ee8c0b | 2430 | |
cbc4e3c1 MS |
2431 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
2432 | ||
e6ee8c0b KU |
2433 | md->io_pool = p->io_pool; |
2434 | p->io_pool = NULL; | |
1ae49ea2 MS |
2435 | md->rq_pool = p->rq_pool; |
2436 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2437 | md->bs = p->bs; |
2438 | p->bs = NULL; | |
4e6e36c3 | 2439 | |
e6ee8c0b | 2440 | out: |
02233342 | 2441 | /* mempool bind completed, no longer need any mempools in the table */ |
e6ee8c0b KU |
2442 | dm_table_free_md_mempools(t); |
2443 | } | |
2444 | ||
1da177e4 LT |
2445 | /* |
2446 | * Bind a table to the device. | |
2447 | */ | |
2448 | static void event_callback(void *context) | |
2449 | { | |
7a8c3d3b MA |
2450 | unsigned long flags; |
2451 | LIST_HEAD(uevents); | |
1da177e4 LT |
2452 | struct mapped_device *md = (struct mapped_device *) context; |
2453 | ||
7a8c3d3b MA |
2454 | spin_lock_irqsave(&md->uevent_lock, flags); |
2455 | list_splice_init(&md->uevent_list, &uevents); | |
2456 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2457 | ||
ed9e1982 | 2458 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2459 | |
1da177e4 LT |
2460 | atomic_inc(&md->event_nr); |
2461 | wake_up(&md->eventq); | |
2462 | } | |
2463 | ||
c217649b MS |
2464 | /* |
2465 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2466 | */ | |
4e90188b | 2467 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2468 | { |
4e90188b | 2469 | set_capacity(md->disk, size); |
1da177e4 | 2470 | |
db8fef4f | 2471 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2472 | } |
2473 | ||
042d2a9b AK |
2474 | /* |
2475 | * Returns old map, which caller must destroy. | |
2476 | */ | |
2477 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2478 | struct queue_limits *limits) | |
1da177e4 | 2479 | { |
042d2a9b | 2480 | struct dm_table *old_map; |
165125e1 | 2481 | struct request_queue *q = md->queue; |
1da177e4 LT |
2482 | sector_t size; |
2483 | ||
2484 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2485 | |
2486 | /* | |
2487 | * Wipe any geometry if the size of the table changed. | |
2488 | */ | |
fd2ed4d2 | 2489 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2490 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2491 | ||
32a926da | 2492 | __set_size(md, size); |
d5816876 | 2493 | |
2ca3310e AK |
2494 | dm_table_event_callback(t, event_callback, md); |
2495 | ||
e6ee8c0b KU |
2496 | /* |
2497 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2498 | * for request-based during suspension. So stop it to prevent | |
2499 | * I/O mapping before resume. | |
2500 | * This must be done before setting the queue restrictions, | |
2501 | * because request-based dm may be run just after the setting. | |
2502 | */ | |
bfebd1cd | 2503 | if (dm_table_request_based(t)) |
e6ee8c0b KU |
2504 | stop_queue(q); |
2505 | ||
2506 | __bind_mempools(md, t); | |
2507 | ||
a12f5d48 | 2508 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2509 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2510 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2511 | ||
754c5fc7 | 2512 | dm_table_set_restrictions(t, q, limits); |
41abc4e1 HR |
2513 | if (old_map) |
2514 | dm_sync_table(md); | |
1da177e4 | 2515 | |
042d2a9b | 2516 | return old_map; |
1da177e4 LT |
2517 | } |
2518 | ||
a7940155 AK |
2519 | /* |
2520 | * Returns unbound table for the caller to free. | |
2521 | */ | |
2522 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2523 | { |
a12f5d48 | 2524 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2525 | |
2526 | if (!map) | |
a7940155 | 2527 | return NULL; |
1da177e4 LT |
2528 | |
2529 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2530 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2531 | dm_sync_table(md); |
a7940155 AK |
2532 | |
2533 | return map; | |
1da177e4 LT |
2534 | } |
2535 | ||
2536 | /* | |
2537 | * Constructor for a new device. | |
2538 | */ | |
2b06cfff | 2539 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2540 | { |
2541 | struct mapped_device *md; | |
2542 | ||
2b06cfff | 2543 | md = alloc_dev(minor); |
1da177e4 LT |
2544 | if (!md) |
2545 | return -ENXIO; | |
2546 | ||
784aae73 MB |
2547 | dm_sysfs_init(md); |
2548 | ||
1da177e4 LT |
2549 | *result = md; |
2550 | return 0; | |
2551 | } | |
2552 | ||
a5664dad MS |
2553 | /* |
2554 | * Functions to manage md->type. | |
2555 | * All are required to hold md->type_lock. | |
2556 | */ | |
2557 | void dm_lock_md_type(struct mapped_device *md) | |
2558 | { | |
2559 | mutex_lock(&md->type_lock); | |
2560 | } | |
2561 | ||
2562 | void dm_unlock_md_type(struct mapped_device *md) | |
2563 | { | |
2564 | mutex_unlock(&md->type_lock); | |
2565 | } | |
2566 | ||
2567 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2568 | { | |
00c4fc3b | 2569 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2570 | md->type = type; |
2571 | } | |
2572 | ||
2573 | unsigned dm_get_md_type(struct mapped_device *md) | |
2574 | { | |
00c4fc3b | 2575 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2576 | return md->type; |
2577 | } | |
2578 | ||
36a0456f AK |
2579 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2580 | { | |
2581 | return md->immutable_target_type; | |
2582 | } | |
2583 | ||
f84cb8a4 MS |
2584 | /* |
2585 | * The queue_limits are only valid as long as you have a reference | |
2586 | * count on 'md'. | |
2587 | */ | |
2588 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2589 | { | |
2590 | BUG_ON(!atomic_read(&md->holders)); | |
2591 | return &md->queue->limits; | |
2592 | } | |
2593 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2594 | ||
bfebd1cd MS |
2595 | static void init_rq_based_worker_thread(struct mapped_device *md) |
2596 | { | |
2597 | /* Initialize the request-based DM worker thread */ | |
2598 | init_kthread_worker(&md->kworker); | |
2599 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2600 | "kdmwork-%s", dm_device_name(md)); | |
2601 | } | |
2602 | ||
4a0b4ddf MS |
2603 | /* |
2604 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2605 | */ | |
2606 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2607 | { | |
2608 | struct request_queue *q = NULL; | |
2609 | ||
4a0b4ddf MS |
2610 | /* Fully initialize the queue */ |
2611 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2612 | if (!q) | |
bfebd1cd | 2613 | return -EINVAL; |
4a0b4ddf | 2614 | |
0ce65797 MS |
2615 | /* disable dm_request_fn's merge heuristic by default */ |
2616 | md->seq_rq_merge_deadline_usecs = 0; | |
2617 | ||
4a0b4ddf | 2618 | md->queue = q; |
bfebd1cd | 2619 | dm_init_old_md_queue(md); |
4a0b4ddf MS |
2620 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
2621 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
4a0b4ddf | 2622 | |
bfebd1cd | 2623 | init_rq_based_worker_thread(md); |
2eb6e1e3 | 2624 | |
4a0b4ddf MS |
2625 | elv_register_queue(md->queue); |
2626 | ||
bfebd1cd MS |
2627 | return 0; |
2628 | } | |
2629 | ||
2630 | static int dm_mq_init_request(void *data, struct request *rq, | |
2631 | unsigned int hctx_idx, unsigned int request_idx, | |
2632 | unsigned int numa_node) | |
2633 | { | |
2634 | struct mapped_device *md = data; | |
2635 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2636 | ||
2637 | /* | |
2638 | * Must initialize md member of tio, otherwise it won't | |
2639 | * be available in dm_mq_queue_rq. | |
2640 | */ | |
2641 | tio->md = md; | |
2642 | ||
2643 | return 0; | |
2644 | } | |
2645 | ||
2646 | static int dm_mq_queue_rq(struct blk_mq_hw_ctx *hctx, | |
2647 | const struct blk_mq_queue_data *bd) | |
2648 | { | |
2649 | struct request *rq = bd->rq; | |
2650 | struct dm_rq_target_io *tio = blk_mq_rq_to_pdu(rq); | |
2651 | struct mapped_device *md = tio->md; | |
2652 | int srcu_idx; | |
2653 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
2654 | struct dm_target *ti; | |
2655 | sector_t pos; | |
2656 | ||
2657 | /* always use block 0 to find the target for flushes for now */ | |
2658 | pos = 0; | |
2659 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
2660 | pos = blk_rq_pos(rq); | |
2661 | ||
2662 | ti = dm_table_find_target(map, pos); | |
2663 | if (!dm_target_is_valid(ti)) { | |
2664 | dm_put_live_table(md, srcu_idx); | |
2665 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
2666 | /* | |
2667 | * Must perform setup, that rq_completed() requires, | |
2668 | * before returning BLK_MQ_RQ_QUEUE_ERROR | |
2669 | */ | |
2670 | dm_start_request(md, rq); | |
2671 | return BLK_MQ_RQ_QUEUE_ERROR; | |
2672 | } | |
2673 | dm_put_live_table(md, srcu_idx); | |
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 MS |
2688 | |
2689 | /* Clone the request if underlying devices aren't blk-mq */ | |
2690 | if (dm_table_get_type(map) == DM_TYPE_REQUEST_BASED) { | |
2691 | /* clone request is allocated at the end of the pdu */ | |
2692 | tio->clone = (void *)blk_mq_rq_to_pdu(rq) + sizeof(struct dm_rq_target_io); | |
45714fbe | 2693 | (void) clone_rq(rq, md, tio, GFP_ATOMIC); |
02233342 MS |
2694 | queue_kthread_work(&md->kworker, &tio->work); |
2695 | } else { | |
2696 | /* Direct call is fine since .queue_rq allows allocations */ | |
45714fbe MS |
2697 | if (map_request(tio, rq, md) == DM_MAPIO_REQUEUE) { |
2698 | /* Undo dm_start_request() before requeuing */ | |
e262f347 | 2699 | rq_end_stats(md, rq); |
45714fbe MS |
2700 | rq_completed(md, rq_data_dir(rq), false); |
2701 | return BLK_MQ_RQ_QUEUE_BUSY; | |
2702 | } | |
02233342 | 2703 | } |
bfebd1cd MS |
2704 | |
2705 | return BLK_MQ_RQ_QUEUE_OK; | |
2706 | } | |
2707 | ||
2708 | static struct blk_mq_ops dm_mq_ops = { | |
2709 | .queue_rq = dm_mq_queue_rq, | |
2710 | .map_queue = blk_mq_map_queue, | |
2711 | .complete = dm_softirq_done, | |
2712 | .init_request = dm_mq_init_request, | |
2713 | }; | |
2714 | ||
2715 | static int dm_init_request_based_blk_mq_queue(struct mapped_device *md) | |
2716 | { | |
02233342 | 2717 | unsigned md_type = dm_get_md_type(md); |
bfebd1cd MS |
2718 | struct request_queue *q; |
2719 | int err; | |
2720 | ||
2721 | memset(&md->tag_set, 0, sizeof(md->tag_set)); | |
2722 | md->tag_set.ops = &dm_mq_ops; | |
2723 | md->tag_set.queue_depth = BLKDEV_MAX_RQ; | |
2724 | md->tag_set.numa_node = NUMA_NO_NODE; | |
2725 | md->tag_set.flags = BLK_MQ_F_SHOULD_MERGE | BLK_MQ_F_SG_MERGE; | |
2726 | md->tag_set.nr_hw_queues = 1; | |
02233342 MS |
2727 | if (md_type == DM_TYPE_REQUEST_BASED) { |
2728 | /* make the memory for non-blk-mq clone part of the pdu */ | |
2729 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io) + sizeof(struct request); | |
2730 | } else | |
2731 | md->tag_set.cmd_size = sizeof(struct dm_rq_target_io); | |
bfebd1cd MS |
2732 | md->tag_set.driver_data = md; |
2733 | ||
2734 | err = blk_mq_alloc_tag_set(&md->tag_set); | |
2735 | if (err) | |
2736 | return err; | |
2737 | ||
2738 | q = blk_mq_init_allocated_queue(&md->tag_set, md->queue); | |
2739 | if (IS_ERR(q)) { | |
2740 | err = PTR_ERR(q); | |
2741 | goto out_tag_set; | |
2742 | } | |
2743 | md->queue = q; | |
2744 | dm_init_md_queue(md); | |
2745 | ||
2746 | /* backfill 'mq' sysfs registration normally done in blk_register_queue */ | |
2747 | blk_mq_register_disk(md->disk); | |
2748 | ||
02233342 MS |
2749 | if (md_type == DM_TYPE_REQUEST_BASED) |
2750 | init_rq_based_worker_thread(md); | |
bfebd1cd MS |
2751 | |
2752 | return 0; | |
2753 | ||
2754 | out_tag_set: | |
2755 | blk_mq_free_tag_set(&md->tag_set); | |
2756 | return err; | |
4a0b4ddf MS |
2757 | } |
2758 | ||
4e6e36c3 MS |
2759 | static unsigned filter_md_type(unsigned type, struct mapped_device *md) |
2760 | { | |
2761 | if (type == DM_TYPE_BIO_BASED) | |
2762 | return type; | |
2763 | ||
2764 | return !md->use_blk_mq ? DM_TYPE_REQUEST_BASED : DM_TYPE_MQ_REQUEST_BASED; | |
2765 | } | |
2766 | ||
4a0b4ddf MS |
2767 | /* |
2768 | * Setup the DM device's queue based on md's type | |
2769 | */ | |
2770 | int dm_setup_md_queue(struct mapped_device *md) | |
2771 | { | |
bfebd1cd | 2772 | int r; |
17e149b8 | 2773 | unsigned md_type = filter_md_type(dm_get_md_type(md), md); |
bfebd1cd MS |
2774 | |
2775 | switch (md_type) { | |
2776 | case DM_TYPE_REQUEST_BASED: | |
2777 | r = dm_init_request_based_queue(md); | |
2778 | if (r) { | |
ff36ab34 | 2779 | DMWARN("Cannot initialize queue for request-based mapped device"); |
bfebd1cd | 2780 | return r; |
ff36ab34 | 2781 | } |
bfebd1cd MS |
2782 | break; |
2783 | case DM_TYPE_MQ_REQUEST_BASED: | |
2784 | r = dm_init_request_based_blk_mq_queue(md); | |
2785 | if (r) { | |
2786 | DMWARN("Cannot initialize queue for request-based blk-mq mapped device"); | |
2787 | return r; | |
2788 | } | |
2789 | break; | |
2790 | case DM_TYPE_BIO_BASED: | |
2791 | dm_init_old_md_queue(md); | |
ff36ab34 | 2792 | blk_queue_make_request(md->queue, dm_make_request); |
dbba42d8 MP |
2793 | /* |
2794 | * DM handles splitting bios as needed. Free the bio_split bioset | |
2795 | * since it won't be used (saves 1 process per bio-based DM device). | |
2796 | */ | |
2797 | bioset_free(md->queue->bio_split); | |
2798 | md->queue->bio_split = NULL; | |
bfebd1cd | 2799 | break; |
4a0b4ddf MS |
2800 | } |
2801 | ||
2802 | return 0; | |
2803 | } | |
2804 | ||
2bec1f4a | 2805 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2806 | { |
2807 | struct mapped_device *md; | |
1da177e4 LT |
2808 | unsigned minor = MINOR(dev); |
2809 | ||
2810 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2811 | return NULL; | |
2812 | ||
f32c10b0 | 2813 | spin_lock(&_minor_lock); |
1da177e4 LT |
2814 | |
2815 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2816 | if (md) { |
2817 | if ((md == MINOR_ALLOCED || | |
2818 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2819 | dm_deleting_md(md) || | |
2820 | test_bit(DMF_FREEING, &md->flags))) { | |
2821 | md = NULL; | |
2822 | goto out; | |
2823 | } | |
2824 | dm_get(md); | |
fba9f90e | 2825 | } |
1da177e4 | 2826 | |
fba9f90e | 2827 | out: |
f32c10b0 | 2828 | spin_unlock(&_minor_lock); |
1da177e4 | 2829 | |
637842cf DT |
2830 | return md; |
2831 | } | |
3cf2e4ba | 2832 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2833 | |
9ade92a9 | 2834 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2835 | { |
9ade92a9 | 2836 | return md->interface_ptr; |
1da177e4 LT |
2837 | } |
2838 | ||
2839 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2840 | { | |
2841 | md->interface_ptr = ptr; | |
2842 | } | |
2843 | ||
2844 | void dm_get(struct mapped_device *md) | |
2845 | { | |
2846 | atomic_inc(&md->holders); | |
3f77316d | 2847 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2848 | } |
2849 | ||
09ee96b2 MP |
2850 | int dm_hold(struct mapped_device *md) |
2851 | { | |
2852 | spin_lock(&_minor_lock); | |
2853 | if (test_bit(DMF_FREEING, &md->flags)) { | |
2854 | spin_unlock(&_minor_lock); | |
2855 | return -EBUSY; | |
2856 | } | |
2857 | dm_get(md); | |
2858 | spin_unlock(&_minor_lock); | |
2859 | return 0; | |
2860 | } | |
2861 | EXPORT_SYMBOL_GPL(dm_hold); | |
2862 | ||
72d94861 AK |
2863 | const char *dm_device_name(struct mapped_device *md) |
2864 | { | |
2865 | return md->name; | |
2866 | } | |
2867 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2868 | ||
3f77316d | 2869 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2870 | { |
1134e5ae | 2871 | struct dm_table *map; |
83d5e5b0 | 2872 | int srcu_idx; |
1da177e4 | 2873 | |
3f77316d | 2874 | might_sleep(); |
fba9f90e | 2875 | |
63a4f065 | 2876 | spin_lock(&_minor_lock); |
3f77316d KU |
2877 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2878 | set_bit(DMF_FREEING, &md->flags); | |
2879 | spin_unlock(&_minor_lock); | |
2880 | ||
02233342 | 2881 | if (dm_request_based(md) && md->kworker_task) |
2eb6e1e3 KB |
2882 | flush_kthread_worker(&md->kworker); |
2883 | ||
ab7c7bb6 MP |
2884 | /* |
2885 | * Take suspend_lock so that presuspend and postsuspend methods | |
2886 | * do not race with internal suspend. | |
2887 | */ | |
2888 | mutex_lock(&md->suspend_lock); | |
2a708cff | 2889 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2890 | if (!dm_suspended_md(md)) { |
2891 | dm_table_presuspend_targets(map); | |
2892 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2893 | } |
83d5e5b0 MP |
2894 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2895 | dm_put_live_table(md, srcu_idx); | |
2a708cff | 2896 | mutex_unlock(&md->suspend_lock); |
83d5e5b0 | 2897 | |
3f77316d KU |
2898 | /* |
2899 | * Rare, but there may be I/O requests still going to complete, | |
2900 | * for example. Wait for all references to disappear. | |
2901 | * No one should increment the reference count of the mapped_device, | |
2902 | * after the mapped_device state becomes DMF_FREEING. | |
2903 | */ | |
2904 | if (wait) | |
2905 | while (atomic_read(&md->holders)) | |
2906 | msleep(1); | |
2907 | else if (atomic_read(&md->holders)) | |
2908 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2909 | dm_device_name(md), atomic_read(&md->holders)); | |
2910 | ||
2911 | dm_sysfs_exit(md); | |
3f77316d KU |
2912 | dm_table_destroy(__unbind(md)); |
2913 | free_dev(md); | |
2914 | } | |
2915 | ||
2916 | void dm_destroy(struct mapped_device *md) | |
2917 | { | |
2918 | __dm_destroy(md, true); | |
2919 | } | |
2920 | ||
2921 | void dm_destroy_immediate(struct mapped_device *md) | |
2922 | { | |
2923 | __dm_destroy(md, false); | |
2924 | } | |
2925 | ||
2926 | void dm_put(struct mapped_device *md) | |
2927 | { | |
2928 | atomic_dec(&md->holders); | |
1da177e4 | 2929 | } |
79eb885c | 2930 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2931 | |
401600df | 2932 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2933 | { |
2934 | int r = 0; | |
b44ebeb0 MP |
2935 | DECLARE_WAITQUEUE(wait, current); |
2936 | ||
b44ebeb0 | 2937 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2938 | |
2939 | while (1) { | |
401600df | 2940 | set_current_state(interruptible); |
46125c1c | 2941 | |
b4324fee | 2942 | if (!md_in_flight(md)) |
46125c1c MB |
2943 | break; |
2944 | ||
401600df MP |
2945 | if (interruptible == TASK_INTERRUPTIBLE && |
2946 | signal_pending(current)) { | |
46125c1c MB |
2947 | r = -EINTR; |
2948 | break; | |
2949 | } | |
2950 | ||
2951 | io_schedule(); | |
2952 | } | |
2953 | set_current_state(TASK_RUNNING); | |
2954 | ||
b44ebeb0 MP |
2955 | remove_wait_queue(&md->wait, &wait); |
2956 | ||
46125c1c MB |
2957 | return r; |
2958 | } | |
2959 | ||
1da177e4 LT |
2960 | /* |
2961 | * Process the deferred bios | |
2962 | */ | |
ef208587 | 2963 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2964 | { |
ef208587 MP |
2965 | struct mapped_device *md = container_of(work, struct mapped_device, |
2966 | work); | |
6d6f10df | 2967 | struct bio *c; |
83d5e5b0 MP |
2968 | int srcu_idx; |
2969 | struct dm_table *map; | |
1da177e4 | 2970 | |
83d5e5b0 | 2971 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2972 | |
3b00b203 | 2973 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2974 | spin_lock_irq(&md->deferred_lock); |
2975 | c = bio_list_pop(&md->deferred); | |
2976 | spin_unlock_irq(&md->deferred_lock); | |
2977 | ||
6a8736d1 | 2978 | if (!c) |
df12ee99 | 2979 | break; |
022c2611 | 2980 | |
e6ee8c0b KU |
2981 | if (dm_request_based(md)) |
2982 | generic_make_request(c); | |
6a8736d1 | 2983 | else |
83d5e5b0 | 2984 | __split_and_process_bio(md, map, c); |
022c2611 | 2985 | } |
73d410c0 | 2986 | |
83d5e5b0 | 2987 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2988 | } |
2989 | ||
9a1fb464 | 2990 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2991 | { |
3b00b203 | 2992 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2993 | smp_mb__after_atomic(); |
53d5914f | 2994 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2995 | } |
2996 | ||
1da177e4 | 2997 | /* |
042d2a9b | 2998 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2999 | */ |
042d2a9b | 3000 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 3001 | { |
87eb5b21 | 3002 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 3003 | struct queue_limits limits; |
042d2a9b | 3004 | int r; |
1da177e4 | 3005 | |
e61290a4 | 3006 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
3007 | |
3008 | /* device must be suspended */ | |
4f186f8b | 3009 | if (!dm_suspended_md(md)) |
93c534ae | 3010 | goto out; |
1da177e4 | 3011 | |
3ae70656 MS |
3012 | /* |
3013 | * If the new table has no data devices, retain the existing limits. | |
3014 | * This helps multipath with queue_if_no_path if all paths disappear, | |
3015 | * then new I/O is queued based on these limits, and then some paths | |
3016 | * reappear. | |
3017 | */ | |
3018 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 3019 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
3020 | if (live_map) |
3021 | limits = md->queue->limits; | |
83d5e5b0 | 3022 | dm_put_live_table_fast(md); |
3ae70656 MS |
3023 | } |
3024 | ||
87eb5b21 MC |
3025 | if (!live_map) { |
3026 | r = dm_calculate_queue_limits(table, &limits); | |
3027 | if (r) { | |
3028 | map = ERR_PTR(r); | |
3029 | goto out; | |
3030 | } | |
042d2a9b | 3031 | } |
754c5fc7 | 3032 | |
042d2a9b | 3033 | map = __bind(md, table, &limits); |
1da177e4 | 3034 | |
93c534ae | 3035 | out: |
e61290a4 | 3036 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 3037 | return map; |
1da177e4 LT |
3038 | } |
3039 | ||
3040 | /* | |
3041 | * Functions to lock and unlock any filesystem running on the | |
3042 | * device. | |
3043 | */ | |
2ca3310e | 3044 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 3045 | { |
e39e2e95 | 3046 | int r; |
1da177e4 LT |
3047 | |
3048 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 3049 | |
db8fef4f | 3050 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 3051 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 3052 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
3053 | md->frozen_sb = NULL; |
3054 | return r; | |
dfbe03f6 AK |
3055 | } |
3056 | ||
aa8d7c2f AK |
3057 | set_bit(DMF_FROZEN, &md->flags); |
3058 | ||
1da177e4 LT |
3059 | return 0; |
3060 | } | |
3061 | ||
2ca3310e | 3062 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 3063 | { |
aa8d7c2f AK |
3064 | if (!test_bit(DMF_FROZEN, &md->flags)) |
3065 | return; | |
3066 | ||
db8fef4f | 3067 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 3068 | md->frozen_sb = NULL; |
aa8d7c2f | 3069 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
3070 | } |
3071 | ||
3072 | /* | |
ffcc3936 MS |
3073 | * If __dm_suspend returns 0, the device is completely quiescent |
3074 | * now. There is no request-processing activity. All new requests | |
3075 | * are being added to md->deferred list. | |
cec47e3d | 3076 | * |
ffcc3936 | 3077 | * Caller must hold md->suspend_lock |
cec47e3d | 3078 | */ |
ffcc3936 MS |
3079 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
3080 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 3081 | { |
ffcc3936 MS |
3082 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
3083 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
3084 | int r; | |
1da177e4 | 3085 | |
2e93ccc1 KU |
3086 | /* |
3087 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
3088 | * This flag is cleared before dm_suspend returns. | |
3089 | */ | |
3090 | if (noflush) | |
3091 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
3092 | ||
d67ee213 MS |
3093 | /* |
3094 | * This gets reverted if there's an error later and the targets | |
3095 | * provide the .presuspend_undo hook. | |
3096 | */ | |
cf222b37 AK |
3097 | dm_table_presuspend_targets(map); |
3098 | ||
32a926da | 3099 | /* |
9f518b27 KU |
3100 | * Flush I/O to the device. |
3101 | * Any I/O submitted after lock_fs() may not be flushed. | |
3102 | * noflush takes precedence over do_lockfs. | |
3103 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
3104 | */ |
3105 | if (!noflush && do_lockfs) { | |
3106 | r = lock_fs(md); | |
d67ee213 MS |
3107 | if (r) { |
3108 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 3109 | return r; |
d67ee213 | 3110 | } |
aa8d7c2f | 3111 | } |
1da177e4 LT |
3112 | |
3113 | /* | |
3b00b203 MP |
3114 | * Here we must make sure that no processes are submitting requests |
3115 | * to target drivers i.e. no one may be executing | |
3116 | * __split_and_process_bio. This is called from dm_request and | |
3117 | * dm_wq_work. | |
3118 | * | |
3119 | * To get all processes out of __split_and_process_bio in dm_request, | |
3120 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
3121 | * __split_and_process_bio from dm_request and quiesce the thread |
3122 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
3123 | * flush_workqueue(md->wq). | |
1da177e4 | 3124 | */ |
1eb787ec | 3125 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
3126 | if (map) |
3127 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 3128 | |
d0bcb878 | 3129 | /* |
29e4013d TH |
3130 | * Stop md->queue before flushing md->wq in case request-based |
3131 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 3132 | */ |
2eb6e1e3 | 3133 | if (dm_request_based(md)) { |
9f518b27 | 3134 | stop_queue(md->queue); |
02233342 MS |
3135 | if (md->kworker_task) |
3136 | flush_kthread_worker(&md->kworker); | |
2eb6e1e3 | 3137 | } |
cec47e3d | 3138 | |
d0bcb878 KU |
3139 | flush_workqueue(md->wq); |
3140 | ||
1da177e4 | 3141 | /* |
3b00b203 MP |
3142 | * At this point no more requests are entering target request routines. |
3143 | * We call dm_wait_for_completion to wait for all existing requests | |
3144 | * to finish. | |
1da177e4 | 3145 | */ |
ffcc3936 | 3146 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 3147 | |
6d6f10df | 3148 | if (noflush) |
022c2611 | 3149 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
3150 | if (map) |
3151 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 3152 | |
1da177e4 | 3153 | /* were we interrupted ? */ |
46125c1c | 3154 | if (r < 0) { |
9a1fb464 | 3155 | dm_queue_flush(md); |
73d410c0 | 3156 | |
cec47e3d | 3157 | if (dm_request_based(md)) |
9f518b27 | 3158 | start_queue(md->queue); |
cec47e3d | 3159 | |
2ca3310e | 3160 | unlock_fs(md); |
d67ee213 | 3161 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 3162 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 3163 | } |
1da177e4 | 3164 | |
ffcc3936 MS |
3165 | return r; |
3166 | } | |
3167 | ||
3168 | /* | |
3169 | * We need to be able to change a mapping table under a mounted | |
3170 | * filesystem. For example we might want to move some data in | |
3171 | * the background. Before the table can be swapped with | |
3172 | * dm_bind_table, dm_suspend must be called to flush any in | |
3173 | * flight bios and ensure that any further io gets deferred. | |
3174 | */ | |
3175 | /* | |
3176 | * Suspend mechanism in request-based dm. | |
3177 | * | |
3178 | * 1. Flush all I/Os by lock_fs() if needed. | |
3179 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
3180 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
3181 | * | |
3182 | * To abort suspend, start the request_queue. | |
3183 | */ | |
3184 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
3185 | { | |
3186 | struct dm_table *map = NULL; | |
3187 | int r = 0; | |
3188 | ||
3189 | retry: | |
3190 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3191 | ||
3192 | if (dm_suspended_md(md)) { | |
3193 | r = -EINVAL; | |
3194 | goto out_unlock; | |
3195 | } | |
3196 | ||
3197 | if (dm_suspended_internally_md(md)) { | |
3198 | /* already internally suspended, wait for internal resume */ | |
3199 | mutex_unlock(&md->suspend_lock); | |
3200 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3201 | if (r) | |
3202 | return r; | |
3203 | goto retry; | |
3204 | } | |
3205 | ||
a12f5d48 | 3206 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3207 | |
3208 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
3209 | if (r) | |
3210 | goto out_unlock; | |
3b00b203 | 3211 | |
2ca3310e | 3212 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 3213 | |
4d4471cb KU |
3214 | dm_table_postsuspend_targets(map); |
3215 | ||
d287483d | 3216 | out_unlock: |
e61290a4 | 3217 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 3218 | return r; |
1da177e4 LT |
3219 | } |
3220 | ||
ffcc3936 MS |
3221 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
3222 | { | |
3223 | if (map) { | |
3224 | int r = dm_table_resume_targets(map); | |
3225 | if (r) | |
3226 | return r; | |
3227 | } | |
3228 | ||
3229 | dm_queue_flush(md); | |
3230 | ||
3231 | /* | |
3232 | * Flushing deferred I/Os must be done after targets are resumed | |
3233 | * so that mapping of targets can work correctly. | |
3234 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
3235 | */ | |
3236 | if (dm_request_based(md)) | |
3237 | start_queue(md->queue); | |
3238 | ||
3239 | unlock_fs(md); | |
3240 | ||
3241 | return 0; | |
3242 | } | |
3243 | ||
1da177e4 LT |
3244 | int dm_resume(struct mapped_device *md) |
3245 | { | |
cf222b37 | 3246 | int r = -EINVAL; |
cf222b37 | 3247 | struct dm_table *map = NULL; |
1da177e4 | 3248 | |
ffcc3936 MS |
3249 | retry: |
3250 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3251 | ||
4f186f8b | 3252 | if (!dm_suspended_md(md)) |
cf222b37 | 3253 | goto out; |
cf222b37 | 3254 | |
ffcc3936 MS |
3255 | if (dm_suspended_internally_md(md)) { |
3256 | /* already internally suspended, wait for internal resume */ | |
3257 | mutex_unlock(&md->suspend_lock); | |
3258 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3259 | if (r) | |
3260 | return r; | |
3261 | goto retry; | |
3262 | } | |
3263 | ||
a12f5d48 | 3264 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3265 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3266 | goto out; |
1da177e4 | 3267 | |
ffcc3936 | 3268 | r = __dm_resume(md, map); |
8757b776 MB |
3269 | if (r) |
3270 | goto out; | |
2ca3310e | 3271 | |
2ca3310e AK |
3272 | clear_bit(DMF_SUSPENDED, &md->flags); |
3273 | ||
cf222b37 AK |
3274 | r = 0; |
3275 | out: | |
e61290a4 | 3276 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3277 | |
cf222b37 | 3278 | return r; |
1da177e4 LT |
3279 | } |
3280 | ||
fd2ed4d2 MP |
3281 | /* |
3282 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3283 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3284 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3285 | */ |
3286 | ||
ffcc3936 | 3287 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3288 | { |
ffcc3936 MS |
3289 | struct dm_table *map = NULL; |
3290 | ||
96b26c8c | 3291 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3292 | return; /* nested internal suspend */ |
3293 | ||
3294 | if (dm_suspended_md(md)) { | |
3295 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3296 | return; /* nest suspend */ | |
3297 | } | |
3298 | ||
a12f5d48 | 3299 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3300 | |
3301 | /* | |
3302 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3303 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3304 | * would require changing .presuspend to return an error -- avoid this | |
3305 | * until there is a need for more elaborate variants of internal suspend. | |
3306 | */ | |
3307 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3308 | ||
3309 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3310 | ||
3311 | dm_table_postsuspend_targets(map); | |
3312 | } | |
3313 | ||
3314 | static void __dm_internal_resume(struct mapped_device *md) | |
3315 | { | |
96b26c8c MP |
3316 | BUG_ON(!md->internal_suspend_count); |
3317 | ||
3318 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3319 | return; /* resume from nested internal suspend */ |
3320 | ||
fd2ed4d2 | 3321 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3322 | goto done; /* resume from nested suspend */ |
3323 | ||
3324 | /* | |
3325 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3326 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3327 | */ | |
3328 | (void) __dm_resume(md, NULL); | |
3329 | ||
3330 | done: | |
3331 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3332 | smp_mb__after_atomic(); | |
3333 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3334 | } | |
3335 | ||
3336 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3337 | { | |
3338 | mutex_lock(&md->suspend_lock); | |
3339 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3340 | mutex_unlock(&md->suspend_lock); | |
3341 | } | |
3342 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3343 | ||
3344 | void dm_internal_resume(struct mapped_device *md) | |
3345 | { | |
3346 | mutex_lock(&md->suspend_lock); | |
3347 | __dm_internal_resume(md); | |
3348 | mutex_unlock(&md->suspend_lock); | |
3349 | } | |
3350 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3351 | ||
3352 | /* | |
3353 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3354 | * which prevents interaction with userspace-driven suspend. | |
3355 | */ | |
3356 | ||
3357 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3358 | { | |
3359 | mutex_lock(&md->suspend_lock); | |
3360 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3361 | return; |
3362 | ||
3363 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3364 | synchronize_srcu(&md->io_barrier); | |
3365 | flush_workqueue(md->wq); | |
3366 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3367 | } | |
b735fede | 3368 | EXPORT_SYMBOL_GPL(dm_internal_suspend_fast); |
fd2ed4d2 | 3369 | |
ffcc3936 | 3370 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3371 | { |
ffcc3936 | 3372 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3373 | goto done; |
3374 | ||
3375 | dm_queue_flush(md); | |
3376 | ||
3377 | done: | |
3378 | mutex_unlock(&md->suspend_lock); | |
3379 | } | |
b735fede | 3380 | EXPORT_SYMBOL_GPL(dm_internal_resume_fast); |
fd2ed4d2 | 3381 | |
1da177e4 LT |
3382 | /*----------------------------------------------------------------- |
3383 | * Event notification. | |
3384 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3385 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3386 | unsigned cookie) |
69267a30 | 3387 | { |
60935eb2 MB |
3388 | char udev_cookie[DM_COOKIE_LENGTH]; |
3389 | char *envp[] = { udev_cookie, NULL }; | |
3390 | ||
3391 | if (!cookie) | |
3abf85b5 | 3392 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3393 | else { |
3394 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3395 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3396 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3397 | action, envp); | |
60935eb2 | 3398 | } |
69267a30 AK |
3399 | } |
3400 | ||
7a8c3d3b MA |
3401 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3402 | { | |
3403 | return atomic_add_return(1, &md->uevent_seq); | |
3404 | } | |
3405 | ||
1da177e4 LT |
3406 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3407 | { | |
3408 | return atomic_read(&md->event_nr); | |
3409 | } | |
3410 | ||
3411 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3412 | { | |
3413 | return wait_event_interruptible(md->eventq, | |
3414 | (event_nr != atomic_read(&md->event_nr))); | |
3415 | } | |
3416 | ||
7a8c3d3b MA |
3417 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3418 | { | |
3419 | unsigned long flags; | |
3420 | ||
3421 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3422 | list_add(elist, &md->uevent_list); | |
3423 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3424 | } | |
3425 | ||
1da177e4 LT |
3426 | /* |
3427 | * The gendisk is only valid as long as you have a reference | |
3428 | * count on 'md'. | |
3429 | */ | |
3430 | struct gendisk *dm_disk(struct mapped_device *md) | |
3431 | { | |
3432 | return md->disk; | |
3433 | } | |
65ff5b7d | 3434 | EXPORT_SYMBOL_GPL(dm_disk); |
1da177e4 | 3435 | |
784aae73 MB |
3436 | struct kobject *dm_kobject(struct mapped_device *md) |
3437 | { | |
2995fa78 | 3438 | return &md->kobj_holder.kobj; |
784aae73 MB |
3439 | } |
3440 | ||
784aae73 MB |
3441 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3442 | { | |
3443 | struct mapped_device *md; | |
3444 | ||
2995fa78 | 3445 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3446 | |
4d89b7b4 | 3447 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3448 | dm_deleting_md(md)) |
4d89b7b4 MB |
3449 | return NULL; |
3450 | ||
784aae73 MB |
3451 | dm_get(md); |
3452 | return md; | |
3453 | } | |
3454 | ||
4f186f8b | 3455 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3456 | { |
3457 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3458 | } | |
3459 | ||
ffcc3936 MS |
3460 | int dm_suspended_internally_md(struct mapped_device *md) |
3461 | { | |
3462 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3463 | } | |
3464 | ||
2c140a24 MP |
3465 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3466 | { | |
3467 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3468 | } | |
3469 | ||
64dbce58 KU |
3470 | int dm_suspended(struct dm_target *ti) |
3471 | { | |
ecdb2e25 | 3472 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3473 | } |
3474 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3475 | ||
2e93ccc1 KU |
3476 | int dm_noflush_suspending(struct dm_target *ti) |
3477 | { | |
ecdb2e25 | 3478 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3479 | } |
3480 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3481 | ||
78d8e58a MS |
3482 | struct dm_md_mempools *dm_alloc_md_mempools(struct mapped_device *md, unsigned type, |
3483 | unsigned integrity, unsigned per_bio_data_size) | |
e6ee8c0b | 3484 | { |
78d8e58a MS |
3485 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3486 | struct kmem_cache *cachep = NULL; | |
3487 | unsigned int pool_size = 0; | |
5f015204 | 3488 | unsigned int front_pad; |
e6ee8c0b KU |
3489 | |
3490 | if (!pools) | |
4e6e36c3 | 3491 | return NULL; |
e6ee8c0b | 3492 | |
78d8e58a | 3493 | type = filter_md_type(type, md); |
17e149b8 | 3494 | |
78d8e58a MS |
3495 | switch (type) { |
3496 | case DM_TYPE_BIO_BASED: | |
3497 | cachep = _io_cache; | |
3498 | pool_size = dm_get_reserved_bio_based_ios(); | |
3499 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
3500 | break; | |
3501 | case DM_TYPE_REQUEST_BASED: | |
3502 | cachep = _rq_tio_cache; | |
3503 | pool_size = dm_get_reserved_rq_based_ios(); | |
3504 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); | |
3505 | if (!pools->rq_pool) | |
3506 | goto out; | |
3507 | /* fall through to setup remaining rq-based pools */ | |
3508 | case DM_TYPE_MQ_REQUEST_BASED: | |
3509 | if (!pool_size) | |
3510 | pool_size = dm_get_reserved_rq_based_ios(); | |
3511 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
3512 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3513 | WARN_ON(per_bio_data_size != 0); | |
3514 | break; | |
3515 | default: | |
3516 | BUG(); | |
3517 | } | |
3518 | ||
3519 | if (cachep) { | |
3520 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); | |
3521 | if (!pools->io_pool) | |
3522 | goto out; | |
3523 | } | |
e6ee8c0b | 3524 | |
3d8aab2d | 3525 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3526 | if (!pools->bs) |
5f015204 | 3527 | goto out; |
e6ee8c0b | 3528 | |
a91a2785 | 3529 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3530 | goto out; |
a91a2785 | 3531 | |
e6ee8c0b | 3532 | return pools; |
5f1b670d | 3533 | |
5f1b670d CH |
3534 | out: |
3535 | dm_free_md_mempools(pools); | |
78d8e58a | 3536 | |
4e6e36c3 | 3537 | return NULL; |
e6ee8c0b KU |
3538 | } |
3539 | ||
3540 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3541 | { | |
3542 | if (!pools) | |
3543 | return; | |
3544 | ||
6f65985e JL |
3545 | mempool_destroy(pools->io_pool); |
3546 | mempool_destroy(pools->rq_pool); | |
1ae49ea2 | 3547 | |
e6ee8c0b KU |
3548 | if (pools->bs) |
3549 | bioset_free(pools->bs); | |
3550 | ||
3551 | kfree(pools); | |
3552 | } | |
3553 | ||
71cdb697 CH |
3554 | static int dm_pr_register(struct block_device *bdev, u64 old_key, u64 new_key, |
3555 | u32 flags) | |
3556 | { | |
3557 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3558 | const struct pr_ops *ops; | |
3559 | struct dm_target *tgt; | |
3560 | fmode_t mode; | |
3561 | int srcu_idx, r; | |
3562 | ||
3563 | r = dm_get_live_table_for_ioctl(md, &tgt, &bdev, &mode, &srcu_idx); | |
3564 | if (r < 0) | |
3565 | return r; | |
3566 | ||
3567 | ops = bdev->bd_disk->fops->pr_ops; | |
3568 | if (ops && ops->pr_register) | |
3569 | r = ops->pr_register(bdev, old_key, new_key, flags); | |
3570 | else | |
3571 | r = -EOPNOTSUPP; | |
3572 | ||
3573 | dm_put_live_table(md, srcu_idx); | |
3574 | return r; | |
3575 | } | |
3576 | ||
3577 | static int dm_pr_reserve(struct block_device *bdev, u64 key, enum pr_type type, | |
3578 | u32 flags) | |
3579 | { | |
3580 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3581 | const struct pr_ops *ops; | |
3582 | struct dm_target *tgt; | |
3583 | fmode_t mode; | |
3584 | int srcu_idx, r; | |
3585 | ||
3586 | r = dm_get_live_table_for_ioctl(md, &tgt, &bdev, &mode, &srcu_idx); | |
3587 | if (r < 0) | |
3588 | return r; | |
3589 | ||
3590 | ops = bdev->bd_disk->fops->pr_ops; | |
3591 | if (ops && ops->pr_reserve) | |
3592 | r = ops->pr_reserve(bdev, key, type, flags); | |
3593 | else | |
3594 | r = -EOPNOTSUPP; | |
3595 | ||
3596 | dm_put_live_table(md, srcu_idx); | |
3597 | return r; | |
3598 | } | |
3599 | ||
3600 | static int dm_pr_release(struct block_device *bdev, u64 key, enum pr_type type) | |
3601 | { | |
3602 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3603 | const struct pr_ops *ops; | |
3604 | struct dm_target *tgt; | |
3605 | fmode_t mode; | |
3606 | int srcu_idx, r; | |
3607 | ||
3608 | r = dm_get_live_table_for_ioctl(md, &tgt, &bdev, &mode, &srcu_idx); | |
3609 | if (r < 0) | |
3610 | return r; | |
3611 | ||
3612 | ops = bdev->bd_disk->fops->pr_ops; | |
3613 | if (ops && ops->pr_release) | |
3614 | r = ops->pr_release(bdev, key, type); | |
3615 | else | |
3616 | r = -EOPNOTSUPP; | |
3617 | ||
3618 | dm_put_live_table(md, srcu_idx); | |
3619 | return r; | |
3620 | } | |
3621 | ||
3622 | static int dm_pr_preempt(struct block_device *bdev, u64 old_key, u64 new_key, | |
3623 | enum pr_type type, bool abort) | |
3624 | { | |
3625 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3626 | const struct pr_ops *ops; | |
3627 | struct dm_target *tgt; | |
3628 | fmode_t mode; | |
3629 | int srcu_idx, r; | |
3630 | ||
3631 | r = dm_get_live_table_for_ioctl(md, &tgt, &bdev, &mode, &srcu_idx); | |
3632 | if (r < 0) | |
3633 | return r; | |
3634 | ||
3635 | ops = bdev->bd_disk->fops->pr_ops; | |
3636 | if (ops && ops->pr_preempt) | |
3637 | r = ops->pr_preempt(bdev, old_key, new_key, type, abort); | |
3638 | else | |
3639 | r = -EOPNOTSUPP; | |
3640 | ||
3641 | dm_put_live_table(md, srcu_idx); | |
3642 | return r; | |
3643 | } | |
3644 | ||
3645 | static int dm_pr_clear(struct block_device *bdev, u64 key) | |
3646 | { | |
3647 | struct mapped_device *md = bdev->bd_disk->private_data; | |
3648 | const struct pr_ops *ops; | |
3649 | struct dm_target *tgt; | |
3650 | fmode_t mode; | |
3651 | int srcu_idx, r; | |
3652 | ||
3653 | r = dm_get_live_table_for_ioctl(md, &tgt, &bdev, &mode, &srcu_idx); | |
3654 | if (r < 0) | |
3655 | return r; | |
3656 | ||
3657 | ops = bdev->bd_disk->fops->pr_ops; | |
3658 | if (ops && ops->pr_clear) | |
3659 | r = ops->pr_clear(bdev, key); | |
3660 | else | |
3661 | r = -EOPNOTSUPP; | |
3662 | ||
3663 | dm_put_live_table(md, srcu_idx); | |
3664 | return r; | |
3665 | } | |
3666 | ||
3667 | static const struct pr_ops dm_pr_ops = { | |
3668 | .pr_register = dm_pr_register, | |
3669 | .pr_reserve = dm_pr_reserve, | |
3670 | .pr_release = dm_pr_release, | |
3671 | .pr_preempt = dm_pr_preempt, | |
3672 | .pr_clear = dm_pr_clear, | |
3673 | }; | |
3674 | ||
83d5cde4 | 3675 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3676 | .open = dm_blk_open, |
3677 | .release = dm_blk_close, | |
aa129a22 | 3678 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3679 | .getgeo = dm_blk_getgeo, |
71cdb697 | 3680 | .pr_ops = &dm_pr_ops, |
1da177e4 LT |
3681 | .owner = THIS_MODULE |
3682 | }; | |
3683 | ||
1da177e4 LT |
3684 | /* |
3685 | * module hooks | |
3686 | */ | |
3687 | module_init(dm_init); | |
3688 | module_exit(dm_exit); | |
3689 | ||
3690 | module_param(major, uint, 0); | |
3691 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3692 | |
e8603136 MS |
3693 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3694 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3695 | ||
f4790826 MS |
3696 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3697 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3698 | ||
17e149b8 MS |
3699 | module_param(use_blk_mq, bool, S_IRUGO | S_IWUSR); |
3700 | MODULE_PARM_DESC(use_blk_mq, "Use block multiqueue for request-based DM devices"); | |
3701 | ||
1da177e4 LT |
3702 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3703 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3704 | MODULE_LICENSE("GPL"); |