]>
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> |
55782138 LZ |
24 | |
25 | #include <trace/events/block.h> | |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "core" |
28 | ||
71a16736 NK |
29 | #ifdef CONFIG_PRINTK |
30 | /* | |
31 | * ratelimit state to be used in DMXXX_LIMIT(). | |
32 | */ | |
33 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
34 | DEFAULT_RATELIMIT_INTERVAL, | |
35 | DEFAULT_RATELIMIT_BURST); | |
36 | EXPORT_SYMBOL(dm_ratelimit_state); | |
37 | #endif | |
38 | ||
60935eb2 MB |
39 | /* |
40 | * Cookies are numeric values sent with CHANGE and REMOVE | |
41 | * uevents while resuming, removing or renaming the device. | |
42 | */ | |
43 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
44 | #define DM_COOKIE_LENGTH 24 | |
45 | ||
1da177e4 LT |
46 | static const char *_name = DM_NAME; |
47 | ||
48 | static unsigned int major = 0; | |
49 | static unsigned int _major = 0; | |
50 | ||
d15b774c AK |
51 | static DEFINE_IDR(_minor_idr); |
52 | ||
f32c10b0 | 53 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
54 | |
55 | static void do_deferred_remove(struct work_struct *w); | |
56 | ||
57 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
58 | ||
acfe0ad7 MP |
59 | static struct workqueue_struct *deferred_remove_workqueue; |
60 | ||
1da177e4 | 61 | /* |
8fbf26ad | 62 | * For bio-based dm. |
1da177e4 LT |
63 | * One of these is allocated per bio. |
64 | */ | |
65 | struct dm_io { | |
66 | struct mapped_device *md; | |
67 | int error; | |
1da177e4 | 68 | atomic_t io_count; |
6ae2fa67 | 69 | struct bio *bio; |
3eaf840e | 70 | unsigned long start_time; |
f88fb981 | 71 | spinlock_t endio_lock; |
fd2ed4d2 | 72 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
73 | }; |
74 | ||
8fbf26ad KU |
75 | /* |
76 | * For request-based dm. | |
77 | * One of these is allocated per request. | |
78 | */ | |
79 | struct dm_rq_target_io { | |
80 | struct mapped_device *md; | |
81 | struct dm_target *ti; | |
1ae49ea2 | 82 | struct request *orig, *clone; |
2eb6e1e3 | 83 | struct kthread_work work; |
8fbf26ad KU |
84 | int error; |
85 | union map_info info; | |
86 | }; | |
87 | ||
88 | /* | |
94818742 KO |
89 | * For request-based dm - the bio clones we allocate are embedded in these |
90 | * structs. | |
91 | * | |
92 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
93 | * the bioset is created - this means the bio has to come at the end of the | |
94 | * struct. | |
8fbf26ad KU |
95 | */ |
96 | struct dm_rq_clone_bio_info { | |
97 | struct bio *orig; | |
cec47e3d | 98 | struct dm_rq_target_io *tio; |
94818742 | 99 | struct bio clone; |
8fbf26ad KU |
100 | }; |
101 | ||
cec47e3d KU |
102 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
103 | { | |
104 | if (rq && rq->end_io_data) | |
105 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
106 | return NULL; | |
107 | } | |
108 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
109 | ||
ba61fdd1 JM |
110 | #define MINOR_ALLOCED ((void *)-1) |
111 | ||
1da177e4 LT |
112 | /* |
113 | * Bits for the md->flags field. | |
114 | */ | |
1eb787ec | 115 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 116 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 117 | #define DMF_FROZEN 2 |
fba9f90e | 118 | #define DMF_FREEING 3 |
5c6bd75d | 119 | #define DMF_DELETING 4 |
2e93ccc1 | 120 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 121 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 122 | #define DMF_DEFERRED_REMOVE 7 |
ffcc3936 | 123 | #define DMF_SUSPENDED_INTERNALLY 8 |
1da177e4 | 124 | |
83d5e5b0 MP |
125 | /* |
126 | * A dummy definition to make RCU happy. | |
127 | * struct dm_table should never be dereferenced in this file. | |
128 | */ | |
129 | struct dm_table { | |
130 | int undefined__; | |
131 | }; | |
132 | ||
304f3f6a MB |
133 | /* |
134 | * Work processed by per-device workqueue. | |
135 | */ | |
1da177e4 | 136 | struct mapped_device { |
83d5e5b0 | 137 | struct srcu_struct io_barrier; |
e61290a4 | 138 | struct mutex suspend_lock; |
1da177e4 | 139 | atomic_t holders; |
5c6bd75d | 140 | atomic_t open_count; |
1da177e4 | 141 | |
2a7faeb1 MP |
142 | /* |
143 | * The current mapping. | |
144 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
145 | * dereference. | |
146 | */ | |
6fa99520 | 147 | struct dm_table __rcu *map; |
2a7faeb1 | 148 | |
86f1152b BM |
149 | struct list_head table_devices; |
150 | struct mutex table_devices_lock; | |
151 | ||
1da177e4 LT |
152 | unsigned long flags; |
153 | ||
165125e1 | 154 | struct request_queue *queue; |
a5664dad | 155 | unsigned type; |
4a0b4ddf | 156 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
157 | struct mutex type_lock; |
158 | ||
36a0456f AK |
159 | struct target_type *immutable_target_type; |
160 | ||
1da177e4 | 161 | struct gendisk *disk; |
7e51f257 | 162 | char name[16]; |
1da177e4 LT |
163 | |
164 | void *interface_ptr; | |
165 | ||
166 | /* | |
167 | * A list of ios that arrived while we were suspended. | |
168 | */ | |
316d315b | 169 | atomic_t pending[2]; |
1da177e4 | 170 | wait_queue_head_t wait; |
53d5914f | 171 | struct work_struct work; |
74859364 | 172 | struct bio_list deferred; |
022c2611 | 173 | spinlock_t deferred_lock; |
1da177e4 | 174 | |
af7e466a | 175 | /* |
29e4013d | 176 | * Processing queue (flush) |
304f3f6a MB |
177 | */ |
178 | struct workqueue_struct *wq; | |
179 | ||
1da177e4 LT |
180 | /* |
181 | * io objects are allocated from here. | |
182 | */ | |
183 | mempool_t *io_pool; | |
1ae49ea2 | 184 | mempool_t *rq_pool; |
1da177e4 | 185 | |
9faf400f SB |
186 | struct bio_set *bs; |
187 | ||
1da177e4 LT |
188 | /* |
189 | * Event handling. | |
190 | */ | |
191 | atomic_t event_nr; | |
192 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
193 | atomic_t uevent_seq; |
194 | struct list_head uevent_list; | |
195 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
196 | |
197 | /* | |
198 | * freeze/thaw support require holding onto a super block | |
199 | */ | |
200 | struct super_block *frozen_sb; | |
db8fef4f | 201 | struct block_device *bdev; |
3ac51e74 DW |
202 | |
203 | /* forced geometry settings */ | |
204 | struct hd_geometry geometry; | |
784aae73 | 205 | |
2995fa78 MP |
206 | /* kobject and completion */ |
207 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 208 | |
d87f4c14 TH |
209 | /* zero-length flush that will be cloned and submitted to targets */ |
210 | struct bio flush_bio; | |
fd2ed4d2 | 211 | |
96b26c8c MP |
212 | /* the number of internal suspends */ |
213 | unsigned internal_suspend_count; | |
214 | ||
fd2ed4d2 | 215 | struct dm_stats stats; |
2eb6e1e3 KB |
216 | |
217 | struct kthread_worker kworker; | |
218 | struct task_struct *kworker_task; | |
1da177e4 LT |
219 | }; |
220 | ||
e6ee8c0b KU |
221 | /* |
222 | * For mempools pre-allocation at the table loading time. | |
223 | */ | |
224 | struct dm_md_mempools { | |
225 | mempool_t *io_pool; | |
1ae49ea2 | 226 | mempool_t *rq_pool; |
e6ee8c0b KU |
227 | struct bio_set *bs; |
228 | }; | |
229 | ||
86f1152b BM |
230 | struct table_device { |
231 | struct list_head list; | |
232 | atomic_t count; | |
233 | struct dm_dev dm_dev; | |
234 | }; | |
235 | ||
6cfa5857 MS |
236 | #define RESERVED_BIO_BASED_IOS 16 |
237 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 238 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 239 | static struct kmem_cache *_io_cache; |
8fbf26ad | 240 | static struct kmem_cache *_rq_tio_cache; |
1ae49ea2 | 241 | static struct kmem_cache *_rq_cache; |
94818742 | 242 | |
e8603136 MS |
243 | /* |
244 | * Bio-based DM's mempools' reserved IOs set by the user. | |
245 | */ | |
246 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
247 | ||
f4790826 MS |
248 | /* |
249 | * Request-based DM's mempools' reserved IOs set by the user. | |
250 | */ | |
251 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
252 | ||
253 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
254 | unsigned def, unsigned max) | |
255 | { | |
256 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
257 | unsigned modified_ios = 0; | |
258 | ||
259 | if (!ios) | |
260 | modified_ios = def; | |
261 | else if (ios > max) | |
262 | modified_ios = max; | |
263 | ||
264 | if (modified_ios) { | |
265 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
266 | ios = modified_ios; | |
267 | } | |
268 | ||
269 | return ios; | |
270 | } | |
271 | ||
e8603136 MS |
272 | unsigned dm_get_reserved_bio_based_ios(void) |
273 | { | |
274 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
275 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
276 | } | |
277 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
278 | ||
f4790826 MS |
279 | unsigned dm_get_reserved_rq_based_ios(void) |
280 | { | |
281 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
282 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
283 | } | |
284 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
285 | ||
1da177e4 LT |
286 | static int __init local_init(void) |
287 | { | |
51157b4a | 288 | int r = -ENOMEM; |
1da177e4 | 289 | |
1da177e4 | 290 | /* allocate a slab for the dm_ios */ |
028867ac | 291 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 292 | if (!_io_cache) |
51157b4a | 293 | return r; |
1da177e4 | 294 | |
8fbf26ad KU |
295 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
296 | if (!_rq_tio_cache) | |
dba14160 | 297 | goto out_free_io_cache; |
8fbf26ad | 298 | |
1ae49ea2 MS |
299 | _rq_cache = kmem_cache_create("dm_clone_request", sizeof(struct request), |
300 | __alignof__(struct request), 0, NULL); | |
301 | if (!_rq_cache) | |
302 | goto out_free_rq_tio_cache; | |
303 | ||
51e5b2bd | 304 | r = dm_uevent_init(); |
51157b4a | 305 | if (r) |
1ae49ea2 | 306 | goto out_free_rq_cache; |
51e5b2bd | 307 | |
acfe0ad7 MP |
308 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
309 | if (!deferred_remove_workqueue) { | |
310 | r = -ENOMEM; | |
311 | goto out_uevent_exit; | |
312 | } | |
313 | ||
1da177e4 LT |
314 | _major = major; |
315 | r = register_blkdev(_major, _name); | |
51157b4a | 316 | if (r < 0) |
acfe0ad7 | 317 | goto out_free_workqueue; |
1da177e4 LT |
318 | |
319 | if (!_major) | |
320 | _major = r; | |
321 | ||
322 | return 0; | |
51157b4a | 323 | |
acfe0ad7 MP |
324 | out_free_workqueue: |
325 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
326 | out_uevent_exit: |
327 | dm_uevent_exit(); | |
1ae49ea2 MS |
328 | out_free_rq_cache: |
329 | kmem_cache_destroy(_rq_cache); | |
8fbf26ad KU |
330 | out_free_rq_tio_cache: |
331 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
332 | out_free_io_cache: |
333 | kmem_cache_destroy(_io_cache); | |
334 | ||
335 | return r; | |
1da177e4 LT |
336 | } |
337 | ||
338 | static void local_exit(void) | |
339 | { | |
2c140a24 | 340 | flush_scheduled_work(); |
acfe0ad7 | 341 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 342 | |
1ae49ea2 | 343 | kmem_cache_destroy(_rq_cache); |
8fbf26ad | 344 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 345 | kmem_cache_destroy(_io_cache); |
00d59405 | 346 | unregister_blkdev(_major, _name); |
51e5b2bd | 347 | dm_uevent_exit(); |
1da177e4 LT |
348 | |
349 | _major = 0; | |
350 | ||
351 | DMINFO("cleaned up"); | |
352 | } | |
353 | ||
b9249e55 | 354 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
355 | local_init, |
356 | dm_target_init, | |
357 | dm_linear_init, | |
358 | dm_stripe_init, | |
952b3557 | 359 | dm_io_init, |
945fa4d2 | 360 | dm_kcopyd_init, |
1da177e4 | 361 | dm_interface_init, |
fd2ed4d2 | 362 | dm_statistics_init, |
1da177e4 LT |
363 | }; |
364 | ||
b9249e55 | 365 | static void (*_exits[])(void) = { |
1da177e4 LT |
366 | local_exit, |
367 | dm_target_exit, | |
368 | dm_linear_exit, | |
369 | dm_stripe_exit, | |
952b3557 | 370 | dm_io_exit, |
945fa4d2 | 371 | dm_kcopyd_exit, |
1da177e4 | 372 | dm_interface_exit, |
fd2ed4d2 | 373 | dm_statistics_exit, |
1da177e4 LT |
374 | }; |
375 | ||
376 | static int __init dm_init(void) | |
377 | { | |
378 | const int count = ARRAY_SIZE(_inits); | |
379 | ||
380 | int r, i; | |
381 | ||
382 | for (i = 0; i < count; i++) { | |
383 | r = _inits[i](); | |
384 | if (r) | |
385 | goto bad; | |
386 | } | |
387 | ||
388 | return 0; | |
389 | ||
390 | bad: | |
391 | while (i--) | |
392 | _exits[i](); | |
393 | ||
394 | return r; | |
395 | } | |
396 | ||
397 | static void __exit dm_exit(void) | |
398 | { | |
399 | int i = ARRAY_SIZE(_exits); | |
400 | ||
401 | while (i--) | |
402 | _exits[i](); | |
d15b774c AK |
403 | |
404 | /* | |
405 | * Should be empty by this point. | |
406 | */ | |
d15b774c | 407 | idr_destroy(&_minor_idr); |
1da177e4 LT |
408 | } |
409 | ||
410 | /* | |
411 | * Block device functions | |
412 | */ | |
432a212c MA |
413 | int dm_deleting_md(struct mapped_device *md) |
414 | { | |
415 | return test_bit(DMF_DELETING, &md->flags); | |
416 | } | |
417 | ||
fe5f9f2c | 418 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
419 | { |
420 | struct mapped_device *md; | |
421 | ||
fba9f90e JM |
422 | spin_lock(&_minor_lock); |
423 | ||
fe5f9f2c | 424 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
425 | if (!md) |
426 | goto out; | |
427 | ||
5c6bd75d | 428 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 429 | dm_deleting_md(md)) { |
fba9f90e JM |
430 | md = NULL; |
431 | goto out; | |
432 | } | |
433 | ||
1da177e4 | 434 | dm_get(md); |
5c6bd75d | 435 | atomic_inc(&md->open_count); |
fba9f90e JM |
436 | |
437 | out: | |
438 | spin_unlock(&_minor_lock); | |
439 | ||
440 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
441 | } |
442 | ||
db2a144b | 443 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 444 | { |
fe5f9f2c | 445 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 446 | |
4a1aeb98 MB |
447 | spin_lock(&_minor_lock); |
448 | ||
2c140a24 MP |
449 | if (atomic_dec_and_test(&md->open_count) && |
450 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 451 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 452 | |
1da177e4 | 453 | dm_put(md); |
4a1aeb98 MB |
454 | |
455 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
456 | } |
457 | ||
5c6bd75d AK |
458 | int dm_open_count(struct mapped_device *md) |
459 | { | |
460 | return atomic_read(&md->open_count); | |
461 | } | |
462 | ||
463 | /* | |
464 | * Guarantees nothing is using the device before it's deleted. | |
465 | */ | |
2c140a24 | 466 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
467 | { |
468 | int r = 0; | |
469 | ||
470 | spin_lock(&_minor_lock); | |
471 | ||
2c140a24 | 472 | if (dm_open_count(md)) { |
5c6bd75d | 473 | r = -EBUSY; |
2c140a24 MP |
474 | if (mark_deferred) |
475 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
476 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
477 | r = -EEXIST; | |
5c6bd75d AK |
478 | else |
479 | set_bit(DMF_DELETING, &md->flags); | |
480 | ||
481 | spin_unlock(&_minor_lock); | |
482 | ||
483 | return r; | |
484 | } | |
485 | ||
2c140a24 MP |
486 | int dm_cancel_deferred_remove(struct mapped_device *md) |
487 | { | |
488 | int r = 0; | |
489 | ||
490 | spin_lock(&_minor_lock); | |
491 | ||
492 | if (test_bit(DMF_DELETING, &md->flags)) | |
493 | r = -EBUSY; | |
494 | else | |
495 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
496 | ||
497 | spin_unlock(&_minor_lock); | |
498 | ||
499 | return r; | |
500 | } | |
501 | ||
502 | static void do_deferred_remove(struct work_struct *w) | |
503 | { | |
504 | dm_deferred_remove(); | |
505 | } | |
506 | ||
fd2ed4d2 MP |
507 | sector_t dm_get_size(struct mapped_device *md) |
508 | { | |
509 | return get_capacity(md->disk); | |
510 | } | |
511 | ||
9974fa2c MS |
512 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
513 | { | |
514 | return md->queue; | |
515 | } | |
516 | ||
fd2ed4d2 MP |
517 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
518 | { | |
519 | return &md->stats; | |
520 | } | |
521 | ||
3ac51e74 DW |
522 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
523 | { | |
524 | struct mapped_device *md = bdev->bd_disk->private_data; | |
525 | ||
526 | return dm_get_geometry(md, geo); | |
527 | } | |
528 | ||
fe5f9f2c | 529 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
530 | unsigned int cmd, unsigned long arg) |
531 | { | |
fe5f9f2c | 532 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 533 | int srcu_idx; |
6c182cd8 | 534 | struct dm_table *map; |
aa129a22 MB |
535 | struct dm_target *tgt; |
536 | int r = -ENOTTY; | |
537 | ||
6c182cd8 | 538 | retry: |
83d5e5b0 MP |
539 | map = dm_get_live_table(md, &srcu_idx); |
540 | ||
aa129a22 MB |
541 | if (!map || !dm_table_get_size(map)) |
542 | goto out; | |
543 | ||
544 | /* We only support devices that have a single target */ | |
545 | if (dm_table_get_num_targets(map) != 1) | |
546 | goto out; | |
547 | ||
548 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
549 | if (!tgt->type->ioctl) |
550 | goto out; | |
aa129a22 | 551 | |
4f186f8b | 552 | if (dm_suspended_md(md)) { |
aa129a22 MB |
553 | r = -EAGAIN; |
554 | goto out; | |
555 | } | |
556 | ||
4d341d82 | 557 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
558 | |
559 | out: | |
83d5e5b0 | 560 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 561 | |
6c182cd8 HR |
562 | if (r == -ENOTCONN) { |
563 | msleep(10); | |
564 | goto retry; | |
565 | } | |
566 | ||
aa129a22 MB |
567 | return r; |
568 | } | |
569 | ||
028867ac | 570 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
571 | { |
572 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
573 | } | |
574 | ||
028867ac | 575 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
576 | { |
577 | mempool_free(io, md->io_pool); | |
578 | } | |
579 | ||
028867ac | 580 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 581 | { |
dba14160 | 582 | bio_put(&tio->clone); |
1da177e4 LT |
583 | } |
584 | ||
08885643 KU |
585 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
586 | gfp_t gfp_mask) | |
cec47e3d | 587 | { |
5f015204 | 588 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
589 | } |
590 | ||
591 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
592 | { | |
5f015204 | 593 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
594 | } |
595 | ||
1ae49ea2 MS |
596 | static struct request *alloc_clone_request(struct mapped_device *md, |
597 | gfp_t gfp_mask) | |
598 | { | |
599 | return mempool_alloc(md->rq_pool, gfp_mask); | |
600 | } | |
601 | ||
602 | static void free_clone_request(struct mapped_device *md, struct request *rq) | |
603 | { | |
604 | mempool_free(rq, md->rq_pool); | |
605 | } | |
606 | ||
90abb8c4 KU |
607 | static int md_in_flight(struct mapped_device *md) |
608 | { | |
609 | return atomic_read(&md->pending[READ]) + | |
610 | atomic_read(&md->pending[WRITE]); | |
611 | } | |
612 | ||
3eaf840e JNN |
613 | static void start_io_acct(struct dm_io *io) |
614 | { | |
615 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 616 | struct bio *bio = io->bio; |
c9959059 | 617 | int cpu; |
fd2ed4d2 | 618 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
619 | |
620 | io->start_time = jiffies; | |
621 | ||
074a7aca TH |
622 | cpu = part_stat_lock(); |
623 | part_round_stats(cpu, &dm_disk(md)->part0); | |
624 | part_stat_unlock(); | |
1e9bb880 SL |
625 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
626 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
627 | |
628 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 629 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 630 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
631 | } |
632 | ||
d221d2e7 | 633 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
634 | { |
635 | struct mapped_device *md = io->md; | |
636 | struct bio *bio = io->bio; | |
637 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 638 | int pending; |
3eaf840e JNN |
639 | int rw = bio_data_dir(bio); |
640 | ||
18c0b223 | 641 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 642 | |
fd2ed4d2 | 643 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 644 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
645 | bio_sectors(bio), true, duration, &io->stats_aux); |
646 | ||
af7e466a MP |
647 | /* |
648 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 649 | * a flush. |
af7e466a | 650 | */ |
1e9bb880 SL |
651 | pending = atomic_dec_return(&md->pending[rw]); |
652 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 653 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 654 | |
d221d2e7 MP |
655 | /* nudge anyone waiting on suspend queue */ |
656 | if (!pending) | |
657 | wake_up(&md->wait); | |
3eaf840e JNN |
658 | } |
659 | ||
1da177e4 LT |
660 | /* |
661 | * Add the bio to the list of deferred io. | |
662 | */ | |
92c63902 | 663 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 664 | { |
05447420 | 665 | unsigned long flags; |
1da177e4 | 666 | |
05447420 | 667 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 668 | bio_list_add(&md->deferred, bio); |
05447420 | 669 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 670 | queue_work(md->wq, &md->work); |
1da177e4 LT |
671 | } |
672 | ||
673 | /* | |
674 | * Everyone (including functions in this file), should use this | |
675 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 676 | * dm_put_live_table() when finished. |
1da177e4 | 677 | */ |
83d5e5b0 | 678 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 679 | { |
83d5e5b0 MP |
680 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
681 | ||
682 | return srcu_dereference(md->map, &md->io_barrier); | |
683 | } | |
1da177e4 | 684 | |
83d5e5b0 MP |
685 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
686 | { | |
687 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
688 | } | |
689 | ||
690 | void dm_sync_table(struct mapped_device *md) | |
691 | { | |
692 | synchronize_srcu(&md->io_barrier); | |
693 | synchronize_rcu_expedited(); | |
694 | } | |
695 | ||
696 | /* | |
697 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
698 | * The caller must not block between these two functions. | |
699 | */ | |
700 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
701 | { | |
702 | rcu_read_lock(); | |
703 | return rcu_dereference(md->map); | |
704 | } | |
1da177e4 | 705 | |
83d5e5b0 MP |
706 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
707 | { | |
708 | rcu_read_unlock(); | |
1da177e4 LT |
709 | } |
710 | ||
86f1152b BM |
711 | /* |
712 | * Open a table device so we can use it as a map destination. | |
713 | */ | |
714 | static int open_table_device(struct table_device *td, dev_t dev, | |
715 | struct mapped_device *md) | |
716 | { | |
717 | static char *_claim_ptr = "I belong to device-mapper"; | |
718 | struct block_device *bdev; | |
719 | ||
720 | int r; | |
721 | ||
722 | BUG_ON(td->dm_dev.bdev); | |
723 | ||
724 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
725 | if (IS_ERR(bdev)) | |
726 | return PTR_ERR(bdev); | |
727 | ||
728 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
729 | if (r) { | |
730 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
731 | return r; | |
732 | } | |
733 | ||
734 | td->dm_dev.bdev = bdev; | |
735 | return 0; | |
736 | } | |
737 | ||
738 | /* | |
739 | * Close a table device that we've been using. | |
740 | */ | |
741 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
742 | { | |
743 | if (!td->dm_dev.bdev) | |
744 | return; | |
745 | ||
746 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
747 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
748 | td->dm_dev.bdev = NULL; | |
749 | } | |
750 | ||
751 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
752 | fmode_t mode) { | |
753 | struct table_device *td; | |
754 | ||
755 | list_for_each_entry(td, l, list) | |
756 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
757 | return td; | |
758 | ||
759 | return NULL; | |
760 | } | |
761 | ||
762 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
763 | struct dm_dev **result) { | |
764 | int r; | |
765 | struct table_device *td; | |
766 | ||
767 | mutex_lock(&md->table_devices_lock); | |
768 | td = find_table_device(&md->table_devices, dev, mode); | |
769 | if (!td) { | |
770 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
771 | if (!td) { | |
772 | mutex_unlock(&md->table_devices_lock); | |
773 | return -ENOMEM; | |
774 | } | |
775 | ||
776 | td->dm_dev.mode = mode; | |
777 | td->dm_dev.bdev = NULL; | |
778 | ||
779 | if ((r = open_table_device(td, dev, md))) { | |
780 | mutex_unlock(&md->table_devices_lock); | |
781 | kfree(td); | |
782 | return r; | |
783 | } | |
784 | ||
785 | format_dev_t(td->dm_dev.name, dev); | |
786 | ||
787 | atomic_set(&td->count, 0); | |
788 | list_add(&td->list, &md->table_devices); | |
789 | } | |
790 | atomic_inc(&td->count); | |
791 | mutex_unlock(&md->table_devices_lock); | |
792 | ||
793 | *result = &td->dm_dev; | |
794 | return 0; | |
795 | } | |
796 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
797 | ||
798 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
799 | { | |
800 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
801 | ||
802 | mutex_lock(&md->table_devices_lock); | |
803 | if (atomic_dec_and_test(&td->count)) { | |
804 | close_table_device(td, md); | |
805 | list_del(&td->list); | |
806 | kfree(td); | |
807 | } | |
808 | mutex_unlock(&md->table_devices_lock); | |
809 | } | |
810 | EXPORT_SYMBOL(dm_put_table_device); | |
811 | ||
812 | static void free_table_devices(struct list_head *devices) | |
813 | { | |
814 | struct list_head *tmp, *next; | |
815 | ||
816 | list_for_each_safe(tmp, next, devices) { | |
817 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
818 | ||
819 | DMWARN("dm_destroy: %s still exists with %d references", | |
820 | td->dm_dev.name, atomic_read(&td->count)); | |
821 | kfree(td); | |
822 | } | |
823 | } | |
824 | ||
3ac51e74 DW |
825 | /* |
826 | * Get the geometry associated with a dm device | |
827 | */ | |
828 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
829 | { | |
830 | *geo = md->geometry; | |
831 | ||
832 | return 0; | |
833 | } | |
834 | ||
835 | /* | |
836 | * Set the geometry of a device. | |
837 | */ | |
838 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
839 | { | |
840 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
841 | ||
842 | if (geo->start > sz) { | |
843 | DMWARN("Start sector is beyond the geometry limits."); | |
844 | return -EINVAL; | |
845 | } | |
846 | ||
847 | md->geometry = *geo; | |
848 | ||
849 | return 0; | |
850 | } | |
851 | ||
1da177e4 LT |
852 | /*----------------------------------------------------------------- |
853 | * CRUD START: | |
854 | * A more elegant soln is in the works that uses the queue | |
855 | * merge fn, unfortunately there are a couple of changes to | |
856 | * the block layer that I want to make for this. So in the | |
857 | * interests of getting something for people to use I give | |
858 | * you this clearly demarcated crap. | |
859 | *---------------------------------------------------------------*/ | |
860 | ||
2e93ccc1 KU |
861 | static int __noflush_suspending(struct mapped_device *md) |
862 | { | |
863 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
864 | } | |
865 | ||
1da177e4 LT |
866 | /* |
867 | * Decrements the number of outstanding ios that a bio has been | |
868 | * cloned into, completing the original io if necc. | |
869 | */ | |
858119e1 | 870 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 871 | { |
2e93ccc1 | 872 | unsigned long flags; |
b35f8caa MB |
873 | int io_error; |
874 | struct bio *bio; | |
875 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
876 | |
877 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
878 | if (unlikely(error)) { |
879 | spin_lock_irqsave(&io->endio_lock, flags); | |
880 | if (!(io->error > 0 && __noflush_suspending(md))) | |
881 | io->error = error; | |
882 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
883 | } | |
1da177e4 LT |
884 | |
885 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
886 | if (io->error == DM_ENDIO_REQUEUE) { |
887 | /* | |
888 | * Target requested pushing back the I/O. | |
2e93ccc1 | 889 | */ |
022c2611 | 890 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
891 | if (__noflush_suspending(md)) |
892 | bio_list_add_head(&md->deferred, io->bio); | |
893 | else | |
2e93ccc1 KU |
894 | /* noflush suspend was interrupted. */ |
895 | io->error = -EIO; | |
022c2611 | 896 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
897 | } |
898 | ||
b35f8caa MB |
899 | io_error = io->error; |
900 | bio = io->bio; | |
6a8736d1 TH |
901 | end_io_acct(io); |
902 | free_io(md, io); | |
903 | ||
904 | if (io_error == DM_ENDIO_REQUEUE) | |
905 | return; | |
2e93ccc1 | 906 | |
4f024f37 | 907 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 908 | /* |
6a8736d1 TH |
909 | * Preflush done for flush with data, reissue |
910 | * without REQ_FLUSH. | |
af7e466a | 911 | */ |
6a8736d1 TH |
912 | bio->bi_rw &= ~REQ_FLUSH; |
913 | queue_io(md, bio); | |
af7e466a | 914 | } else { |
b372d360 | 915 | /* done with normal IO or empty flush */ |
0a82a8d1 | 916 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 917 | bio_endio(bio, io_error); |
b35f8caa | 918 | } |
1da177e4 LT |
919 | } |
920 | } | |
921 | ||
7eee4ae2 MS |
922 | static void disable_write_same(struct mapped_device *md) |
923 | { | |
924 | struct queue_limits *limits = dm_get_queue_limits(md); | |
925 | ||
926 | /* device doesn't really support WRITE SAME, disable it */ | |
927 | limits->max_write_same_sectors = 0; | |
928 | } | |
929 | ||
6712ecf8 | 930 | static void clone_endio(struct bio *bio, int error) |
1da177e4 | 931 | { |
5164bece | 932 | int r = error; |
bfc6d41c | 933 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 934 | struct dm_io *io = tio->io; |
9faf400f | 935 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
936 | dm_endio_fn endio = tio->ti->type->end_io; |
937 | ||
1da177e4 LT |
938 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
939 | error = -EIO; | |
940 | ||
941 | if (endio) { | |
7de3ee57 | 942 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
943 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
944 | /* | |
945 | * error and requeue request are handled | |
946 | * in dec_pending(). | |
947 | */ | |
1da177e4 | 948 | error = r; |
45cbcd79 KU |
949 | else if (r == DM_ENDIO_INCOMPLETE) |
950 | /* The target will handle the io */ | |
6712ecf8 | 951 | return; |
45cbcd79 KU |
952 | else if (r) { |
953 | DMWARN("unimplemented target endio return value: %d", r); | |
954 | BUG(); | |
955 | } | |
1da177e4 LT |
956 | } |
957 | ||
7eee4ae2 MS |
958 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
959 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
960 | disable_write_same(md); | |
961 | ||
9faf400f | 962 | free_tio(md, tio); |
b35f8caa | 963 | dec_pending(io, error); |
1da177e4 LT |
964 | } |
965 | ||
cec47e3d KU |
966 | /* |
967 | * Partial completion handling for request-based dm | |
968 | */ | |
969 | static void end_clone_bio(struct bio *clone, int error) | |
970 | { | |
bfc6d41c MP |
971 | struct dm_rq_clone_bio_info *info = |
972 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
973 | struct dm_rq_target_io *tio = info->tio; |
974 | struct bio *bio = info->orig; | |
4f024f37 | 975 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
976 | |
977 | bio_put(clone); | |
978 | ||
979 | if (tio->error) | |
980 | /* | |
981 | * An error has already been detected on the request. | |
982 | * Once error occurred, just let clone->end_io() handle | |
983 | * the remainder. | |
984 | */ | |
985 | return; | |
986 | else if (error) { | |
987 | /* | |
988 | * Don't notice the error to the upper layer yet. | |
989 | * The error handling decision is made by the target driver, | |
990 | * when the request is completed. | |
991 | */ | |
992 | tio->error = error; | |
993 | return; | |
994 | } | |
995 | ||
996 | /* | |
997 | * I/O for the bio successfully completed. | |
998 | * Notice the data completion to the upper layer. | |
999 | */ | |
1000 | ||
1001 | /* | |
1002 | * bios are processed from the head of the list. | |
1003 | * So the completing bio should always be rq->bio. | |
1004 | * If it's not, something wrong is happening. | |
1005 | */ | |
1006 | if (tio->orig->bio != bio) | |
1007 | DMERR("bio completion is going in the middle of the request"); | |
1008 | ||
1009 | /* | |
1010 | * Update the original request. | |
1011 | * Do not use blk_end_request() here, because it may complete | |
1012 | * the original request before the clone, and break the ordering. | |
1013 | */ | |
1014 | blk_update_request(tio->orig, 0, nr_bytes); | |
1015 | } | |
1016 | ||
1017 | /* | |
1018 | * Don't touch any member of the md after calling this function because | |
1019 | * the md may be freed in dm_put() at the end of this function. | |
1020 | * Or do dm_get() before calling this function and dm_put() later. | |
1021 | */ | |
466d89a6 | 1022 | static void rq_completed(struct mapped_device *md, int rw, bool run_queue) |
cec47e3d | 1023 | { |
b4324fee | 1024 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
1025 | |
1026 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 1027 | if (!md_in_flight(md)) |
cec47e3d KU |
1028 | wake_up(&md->wait); |
1029 | ||
a8c32a5c JA |
1030 | /* |
1031 | * Run this off this callpath, as drivers could invoke end_io while | |
1032 | * inside their request_fn (and holding the queue lock). Calling | |
1033 | * back into ->request_fn() could deadlock attempting to grab the | |
1034 | * queue lock again. | |
1035 | */ | |
cec47e3d | 1036 | if (run_queue) |
a8c32a5c | 1037 | blk_run_queue_async(md->queue); |
cec47e3d KU |
1038 | |
1039 | /* | |
1040 | * dm_put() must be at the end of this function. See the comment above | |
1041 | */ | |
1042 | dm_put(md); | |
1043 | } | |
1044 | ||
a77e28c7 KU |
1045 | static void free_rq_clone(struct request *clone) |
1046 | { | |
1047 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1048 | ||
1049 | blk_rq_unprep_clone(clone); | |
e5863d9a MS |
1050 | if (clone->q && clone->q->mq_ops) |
1051 | tio->ti->type->release_clone_rq(clone); | |
1052 | else | |
1053 | free_clone_request(tio->md, clone); | |
a77e28c7 KU |
1054 | free_rq_tio(tio); |
1055 | } | |
1056 | ||
980691e5 KU |
1057 | /* |
1058 | * Complete the clone and the original request. | |
466d89a6 KB |
1059 | * Must be called without clone's queue lock held, |
1060 | * see end_clone_request() for more details. | |
980691e5 KU |
1061 | */ |
1062 | static void dm_end_request(struct request *clone, int error) | |
1063 | { | |
1064 | int rw = rq_data_dir(clone); | |
1065 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1066 | struct mapped_device *md = tio->md; | |
1067 | struct request *rq = tio->orig; | |
1068 | ||
29e4013d | 1069 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1070 | rq->errors = clone->errors; |
1071 | rq->resid_len = clone->resid_len; | |
1072 | ||
1073 | if (rq->sense) | |
1074 | /* | |
1075 | * We are using the sense buffer of the original | |
1076 | * request. | |
1077 | * So setting the length of the sense data is enough. | |
1078 | */ | |
1079 | rq->sense_len = clone->sense_len; | |
1080 | } | |
1081 | ||
1082 | free_rq_clone(clone); | |
29e4013d TH |
1083 | blk_end_request_all(rq, error); |
1084 | rq_completed(md, rw, true); | |
980691e5 KU |
1085 | } |
1086 | ||
cec47e3d KU |
1087 | static void dm_unprep_request(struct request *rq) |
1088 | { | |
466d89a6 KB |
1089 | struct dm_rq_target_io *tio = rq->special; |
1090 | struct request *clone = tio->clone; | |
cec47e3d KU |
1091 | |
1092 | rq->special = NULL; | |
1093 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1094 | ||
e5863d9a MS |
1095 | if (clone) |
1096 | free_rq_clone(clone); | |
cec47e3d KU |
1097 | } |
1098 | ||
1099 | /* | |
1100 | * Requeue the original request of a clone. | |
1101 | */ | |
466d89a6 KB |
1102 | static void dm_requeue_unmapped_original_request(struct mapped_device *md, |
1103 | struct request *rq) | |
cec47e3d | 1104 | { |
466d89a6 | 1105 | int rw = rq_data_dir(rq); |
cec47e3d KU |
1106 | struct request_queue *q = rq->q; |
1107 | unsigned long flags; | |
1108 | ||
1109 | dm_unprep_request(rq); | |
1110 | ||
1111 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
1112 | blk_requeue_request(q, rq); |
1113 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1114 | ||
466d89a6 KB |
1115 | rq_completed(md, rw, false); |
1116 | } | |
1117 | ||
1118 | static void dm_requeue_unmapped_request(struct request *clone) | |
1119 | { | |
1120 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1121 | ||
1122 | dm_requeue_unmapped_original_request(tio->md, tio->orig); | |
cec47e3d | 1123 | } |
cec47e3d KU |
1124 | |
1125 | static void __stop_queue(struct request_queue *q) | |
1126 | { | |
1127 | blk_stop_queue(q); | |
1128 | } | |
1129 | ||
1130 | static void stop_queue(struct request_queue *q) | |
1131 | { | |
1132 | unsigned long flags; | |
1133 | ||
1134 | spin_lock_irqsave(q->queue_lock, flags); | |
1135 | __stop_queue(q); | |
1136 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1137 | } | |
1138 | ||
1139 | static void __start_queue(struct request_queue *q) | |
1140 | { | |
1141 | if (blk_queue_stopped(q)) | |
1142 | blk_start_queue(q); | |
1143 | } | |
1144 | ||
1145 | static void start_queue(struct request_queue *q) | |
1146 | { | |
1147 | unsigned long flags; | |
1148 | ||
1149 | spin_lock_irqsave(q->queue_lock, flags); | |
1150 | __start_queue(q); | |
1151 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1152 | } | |
1153 | ||
11a68244 | 1154 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1155 | { |
11a68244 | 1156 | int r = error; |
cec47e3d | 1157 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1158 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1159 | |
ba1cbad9 MS |
1160 | if (tio->ti) { |
1161 | rq_end_io = tio->ti->type->rq_end_io; | |
1162 | ||
1163 | if (mapped && rq_end_io) | |
1164 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1165 | } | |
cec47e3d | 1166 | |
7eee4ae2 MS |
1167 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1168 | !clone->q->limits.max_write_same_sectors)) | |
1169 | disable_write_same(tio->md); | |
1170 | ||
11a68244 | 1171 | if (r <= 0) |
cec47e3d | 1172 | /* The target wants to complete the I/O */ |
11a68244 KU |
1173 | dm_end_request(clone, r); |
1174 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1175 | /* The target will handle the I/O */ |
1176 | return; | |
11a68244 | 1177 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1178 | /* The target wants to requeue the I/O */ |
1179 | dm_requeue_unmapped_request(clone); | |
1180 | else { | |
11a68244 | 1181 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1182 | BUG(); |
1183 | } | |
1184 | } | |
1185 | ||
11a68244 KU |
1186 | /* |
1187 | * Request completion handler for request-based dm | |
1188 | */ | |
1189 | static void dm_softirq_done(struct request *rq) | |
1190 | { | |
1191 | bool mapped = true; | |
466d89a6 KB |
1192 | struct dm_rq_target_io *tio = rq->special; |
1193 | struct request *clone = tio->clone; | |
11a68244 | 1194 | |
e5863d9a MS |
1195 | if (!clone) { |
1196 | blk_end_request_all(rq, tio->error); | |
1197 | rq_completed(tio->md, rq_data_dir(rq), false); | |
1198 | free_rq_tio(tio); | |
1199 | return; | |
1200 | } | |
11a68244 KU |
1201 | |
1202 | if (rq->cmd_flags & REQ_FAILED) | |
1203 | mapped = false; | |
1204 | ||
1205 | dm_done(clone, tio->error, mapped); | |
1206 | } | |
1207 | ||
cec47e3d KU |
1208 | /* |
1209 | * Complete the clone and the original request with the error status | |
1210 | * through softirq context. | |
1211 | */ | |
466d89a6 | 1212 | static void dm_complete_request(struct request *rq, int error) |
cec47e3d | 1213 | { |
466d89a6 | 1214 | struct dm_rq_target_io *tio = rq->special; |
cec47e3d KU |
1215 | |
1216 | tio->error = error; | |
cec47e3d KU |
1217 | blk_complete_request(rq); |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * Complete the not-mapped clone and the original request with the error status | |
1222 | * through softirq context. | |
1223 | * Target's rq_end_io() function isn't called. | |
e5863d9a | 1224 | * This may be used when the target's map_rq() or clone_and_map_rq() functions fail. |
cec47e3d | 1225 | */ |
466d89a6 | 1226 | static void dm_kill_unmapped_request(struct request *rq, int error) |
cec47e3d | 1227 | { |
cec47e3d | 1228 | rq->cmd_flags |= REQ_FAILED; |
466d89a6 | 1229 | dm_complete_request(rq, error); |
cec47e3d | 1230 | } |
cec47e3d KU |
1231 | |
1232 | /* | |
466d89a6 | 1233 | * Called with the clone's queue lock held |
cec47e3d KU |
1234 | */ |
1235 | static void end_clone_request(struct request *clone, int error) | |
1236 | { | |
466d89a6 KB |
1237 | struct dm_rq_target_io *tio = clone->end_io_data; |
1238 | ||
e5863d9a MS |
1239 | if (!clone->q->mq_ops) { |
1240 | /* | |
1241 | * For just cleaning up the information of the queue in which | |
1242 | * the clone was dispatched. | |
1243 | * The clone is *NOT* freed actually here because it is alloced | |
1244 | * from dm own mempool (REQ_ALLOCED isn't set). | |
1245 | */ | |
1246 | __blk_put_request(clone->q, clone); | |
1247 | } | |
cec47e3d KU |
1248 | |
1249 | /* | |
1250 | * Actual request completion is done in a softirq context which doesn't | |
466d89a6 | 1251 | * hold the clone's queue lock. Otherwise, deadlock could occur because: |
cec47e3d KU |
1252 | * - another request may be submitted by the upper level driver |
1253 | * of the stacking during the completion | |
1254 | * - the submission which requires queue lock may be done | |
466d89a6 | 1255 | * against this clone's queue |
cec47e3d | 1256 | */ |
466d89a6 | 1257 | dm_complete_request(tio->orig, error); |
cec47e3d KU |
1258 | } |
1259 | ||
56a67df7 MS |
1260 | /* |
1261 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1262 | * target boundary. | |
1263 | */ | |
1264 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1265 | { | |
1266 | sector_t target_offset = dm_target_offset(ti, sector); | |
1267 | ||
1268 | return ti->len - target_offset; | |
1269 | } | |
1270 | ||
1271 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1272 | { |
56a67df7 | 1273 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1274 | sector_t offset, max_len; |
1da177e4 LT |
1275 | |
1276 | /* | |
542f9038 | 1277 | * Does the target need to split even further? |
1da177e4 | 1278 | */ |
542f9038 MS |
1279 | if (ti->max_io_len) { |
1280 | offset = dm_target_offset(ti, sector); | |
1281 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1282 | max_len = sector_div(offset, ti->max_io_len); | |
1283 | else | |
1284 | max_len = offset & (ti->max_io_len - 1); | |
1285 | max_len = ti->max_io_len - max_len; | |
1286 | ||
1287 | if (len > max_len) | |
1288 | len = max_len; | |
1da177e4 LT |
1289 | } |
1290 | ||
1291 | return len; | |
1292 | } | |
1293 | ||
542f9038 MS |
1294 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1295 | { | |
1296 | if (len > UINT_MAX) { | |
1297 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1298 | (unsigned long long)len, UINT_MAX); | |
1299 | ti->error = "Maximum size of target IO is too large"; | |
1300 | return -EINVAL; | |
1301 | } | |
1302 | ||
1303 | ti->max_io_len = (uint32_t) len; | |
1304 | ||
1305 | return 0; | |
1306 | } | |
1307 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1308 | ||
1dd40c3e MP |
1309 | /* |
1310 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1311 | * allowed for all bio types except REQ_FLUSH. | |
1312 | * | |
1313 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1314 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1315 | * sent in a next bio. | |
1316 | * | |
1317 | * A diagram that explains the arithmetics: | |
1318 | * +--------------------+---------------+-------+ | |
1319 | * | 1 | 2 | 3 | | |
1320 | * +--------------------+---------------+-------+ | |
1321 | * | |
1322 | * <-------------- *tio->len_ptr ---------------> | |
1323 | * <------- bi_size -------> | |
1324 | * <-- n_sectors --> | |
1325 | * | |
1326 | * Region 1 was already iterated over with bio_advance or similar function. | |
1327 | * (it may be empty if the target doesn't use bio_advance) | |
1328 | * Region 2 is the remaining bio size that the target wants to process. | |
1329 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1330 | * to make it empty) | |
1331 | * The target requires that region 3 is to be sent in the next bio. | |
1332 | * | |
1333 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1334 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1335 | * copies of the bio. | |
1336 | */ | |
1337 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1338 | { | |
1339 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1340 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1341 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1342 | BUG_ON(bi_size > *tio->len_ptr); | |
1343 | BUG_ON(n_sectors > bi_size); | |
1344 | *tio->len_ptr -= bi_size - n_sectors; | |
1345 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1346 | } | |
1347 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1348 | ||
bd2a49b8 | 1349 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1350 | { |
1351 | int r; | |
2056a782 | 1352 | sector_t sector; |
9faf400f | 1353 | struct mapped_device *md; |
dba14160 | 1354 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1355 | struct dm_target *ti = tio->ti; |
1da177e4 | 1356 | |
1da177e4 | 1357 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1358 | |
1359 | /* | |
1360 | * Map the clone. If r == 0 we don't need to do | |
1361 | * anything, the target has assumed ownership of | |
1362 | * this io. | |
1363 | */ | |
1364 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1365 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1366 | r = ti->type->map(ti, clone); |
45cbcd79 | 1367 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1368 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1369 | |
d07335e5 MS |
1370 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1371 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1372 | |
1da177e4 | 1373 | generic_make_request(clone); |
2e93ccc1 KU |
1374 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1375 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1376 | md = tio->io->md; |
1377 | dec_pending(tio->io, r); | |
9faf400f | 1378 | free_tio(md, tio); |
45cbcd79 KU |
1379 | } else if (r) { |
1380 | DMWARN("unimplemented target map return value: %d", r); | |
1381 | BUG(); | |
1da177e4 LT |
1382 | } |
1383 | } | |
1384 | ||
1385 | struct clone_info { | |
1386 | struct mapped_device *md; | |
1387 | struct dm_table *map; | |
1388 | struct bio *bio; | |
1389 | struct dm_io *io; | |
1390 | sector_t sector; | |
e0d6609a | 1391 | unsigned sector_count; |
1da177e4 LT |
1392 | }; |
1393 | ||
e0d6609a | 1394 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1395 | { |
4f024f37 KO |
1396 | bio->bi_iter.bi_sector = sector; |
1397 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1398 | } |
1399 | ||
1400 | /* | |
1401 | * Creates a bio that consists of range of complete bvecs. | |
1402 | */ | |
dba14160 | 1403 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1404 | sector_t sector, unsigned len) |
1da177e4 | 1405 | { |
dba14160 | 1406 | struct bio *clone = &tio->clone; |
1da177e4 | 1407 | |
1c3b13e6 KO |
1408 | __bio_clone_fast(clone, bio); |
1409 | ||
1410 | if (bio_integrity(bio)) | |
1411 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1412 | |
1c3b13e6 KO |
1413 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1414 | clone->bi_iter.bi_size = to_bytes(len); | |
1415 | ||
1416 | if (bio_integrity(bio)) | |
1417 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1418 | } |
1419 | ||
9015df24 | 1420 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1421 | struct dm_target *ti, |
55a62eef | 1422 | unsigned target_bio_nr) |
f9ab94ce | 1423 | { |
dba14160 MP |
1424 | struct dm_target_io *tio; |
1425 | struct bio *clone; | |
1426 | ||
99778273 | 1427 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1428 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1429 | |
1430 | tio->io = ci->io; | |
1431 | tio->ti = ti; | |
55a62eef | 1432 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1433 | |
1434 | return tio; | |
1435 | } | |
1436 | ||
14fe594d AK |
1437 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1438 | struct dm_target *ti, | |
1dd40c3e | 1439 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1440 | { |
99778273 | 1441 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1442 | struct bio *clone = &tio->clone; |
9015df24 | 1443 | |
1dd40c3e MP |
1444 | tio->len_ptr = len; |
1445 | ||
99778273 | 1446 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1447 | if (len) |
1dd40c3e | 1448 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1449 | |
bd2a49b8 | 1450 | __map_bio(tio); |
f9ab94ce MP |
1451 | } |
1452 | ||
14fe594d | 1453 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1454 | unsigned num_bios, unsigned *len) |
06a426ce | 1455 | { |
55a62eef | 1456 | unsigned target_bio_nr; |
06a426ce | 1457 | |
55a62eef | 1458 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1459 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1460 | } |
1461 | ||
14fe594d | 1462 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1463 | { |
06a426ce | 1464 | unsigned target_nr = 0; |
f9ab94ce MP |
1465 | struct dm_target *ti; |
1466 | ||
b372d360 | 1467 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1468 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1469 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1470 | |
f9ab94ce MP |
1471 | return 0; |
1472 | } | |
1473 | ||
e4c93811 | 1474 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1475 | sector_t sector, unsigned *len) |
5ae89a87 | 1476 | { |
dba14160 | 1477 | struct bio *bio = ci->bio; |
5ae89a87 | 1478 | struct dm_target_io *tio; |
b0d8ed4d AK |
1479 | unsigned target_bio_nr; |
1480 | unsigned num_target_bios = 1; | |
5ae89a87 | 1481 | |
b0d8ed4d AK |
1482 | /* |
1483 | * Does the target want to receive duplicate copies of the bio? | |
1484 | */ | |
1485 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1486 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1487 | |
b0d8ed4d | 1488 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1489 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1490 | tio->len_ptr = len; |
1491 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1492 | __map_bio(tio); |
1493 | } | |
5ae89a87 MS |
1494 | } |
1495 | ||
55a62eef | 1496 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1497 | |
55a62eef | 1498 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1499 | { |
55a62eef | 1500 | return ti->num_discard_bios; |
23508a96 MS |
1501 | } |
1502 | ||
55a62eef | 1503 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1504 | { |
55a62eef | 1505 | return ti->num_write_same_bios; |
23508a96 MS |
1506 | } |
1507 | ||
1508 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1509 | ||
1510 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1511 | { | |
55a62eef | 1512 | return ti->split_discard_bios; |
23508a96 MS |
1513 | } |
1514 | ||
14fe594d AK |
1515 | static int __send_changing_extent_only(struct clone_info *ci, |
1516 | get_num_bios_fn get_num_bios, | |
1517 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1518 | { |
1519 | struct dm_target *ti; | |
e0d6609a | 1520 | unsigned len; |
55a62eef | 1521 | unsigned num_bios; |
5ae89a87 | 1522 | |
a79245b3 MS |
1523 | do { |
1524 | ti = dm_table_find_target(ci->map, ci->sector); | |
1525 | if (!dm_target_is_valid(ti)) | |
1526 | return -EIO; | |
5ae89a87 | 1527 | |
5ae89a87 | 1528 | /* |
23508a96 MS |
1529 | * Even though the device advertised support for this type of |
1530 | * request, that does not mean every target supports it, and | |
936688d7 | 1531 | * reconfiguration might also have changed that since the |
a79245b3 | 1532 | * check was performed. |
5ae89a87 | 1533 | */ |
55a62eef AK |
1534 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1535 | if (!num_bios) | |
a79245b3 | 1536 | return -EOPNOTSUPP; |
5ae89a87 | 1537 | |
23508a96 | 1538 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1539 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1540 | else |
e0d6609a | 1541 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1542 | |
1dd40c3e | 1543 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1544 | |
1545 | ci->sector += len; | |
1546 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1547 | |
1548 | return 0; | |
1549 | } | |
1550 | ||
14fe594d | 1551 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1552 | { |
14fe594d AK |
1553 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1554 | is_split_required_for_discard); | |
23508a96 MS |
1555 | } |
1556 | ||
14fe594d | 1557 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1558 | { |
14fe594d | 1559 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1560 | } |
1561 | ||
e4c93811 AK |
1562 | /* |
1563 | * Select the correct strategy for processing a non-flush bio. | |
1564 | */ | |
14fe594d | 1565 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1566 | { |
dba14160 | 1567 | struct bio *bio = ci->bio; |
512875bd | 1568 | struct dm_target *ti; |
1c3b13e6 | 1569 | unsigned len; |
1da177e4 | 1570 | |
5ae89a87 | 1571 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1572 | return __send_discard(ci); |
23508a96 | 1573 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1574 | return __send_write_same(ci); |
5ae89a87 | 1575 | |
512875bd JN |
1576 | ti = dm_table_find_target(ci->map, ci->sector); |
1577 | if (!dm_target_is_valid(ti)) | |
1578 | return -EIO; | |
1579 | ||
1c3b13e6 | 1580 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1581 | |
1dd40c3e | 1582 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1583 | |
1c3b13e6 KO |
1584 | ci->sector += len; |
1585 | ci->sector_count -= len; | |
1da177e4 | 1586 | |
1c3b13e6 | 1587 | return 0; |
1da177e4 LT |
1588 | } |
1589 | ||
1590 | /* | |
14fe594d | 1591 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1592 | */ |
83d5e5b0 MP |
1593 | static void __split_and_process_bio(struct mapped_device *md, |
1594 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1595 | { |
1596 | struct clone_info ci; | |
512875bd | 1597 | int error = 0; |
1da177e4 | 1598 | |
83d5e5b0 | 1599 | if (unlikely(!map)) { |
6a8736d1 | 1600 | bio_io_error(bio); |
f0b9a450 MP |
1601 | return; |
1602 | } | |
692d0eb9 | 1603 | |
83d5e5b0 | 1604 | ci.map = map; |
1da177e4 | 1605 | ci.md = md; |
1da177e4 LT |
1606 | ci.io = alloc_io(md); |
1607 | ci.io->error = 0; | |
1608 | atomic_set(&ci.io->io_count, 1); | |
1609 | ci.io->bio = bio; | |
1610 | ci.io->md = md; | |
f88fb981 | 1611 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1612 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1613 | |
3eaf840e | 1614 | start_io_acct(ci.io); |
bd2a49b8 | 1615 | |
b372d360 MS |
1616 | if (bio->bi_rw & REQ_FLUSH) { |
1617 | ci.bio = &ci.md->flush_bio; | |
1618 | ci.sector_count = 0; | |
14fe594d | 1619 | error = __send_empty_flush(&ci); |
b372d360 MS |
1620 | /* dec_pending submits any data associated with flush */ |
1621 | } else { | |
6a8736d1 | 1622 | ci.bio = bio; |
d87f4c14 | 1623 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1624 | while (ci.sector_count && !error) |
14fe594d | 1625 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1626 | } |
1da177e4 LT |
1627 | |
1628 | /* drop the extra reference count */ | |
512875bd | 1629 | dec_pending(ci.io, error); |
1da177e4 LT |
1630 | } |
1631 | /*----------------------------------------------------------------- | |
1632 | * CRUD END | |
1633 | *---------------------------------------------------------------*/ | |
1634 | ||
f6fccb12 MB |
1635 | static int dm_merge_bvec(struct request_queue *q, |
1636 | struct bvec_merge_data *bvm, | |
1637 | struct bio_vec *biovec) | |
1638 | { | |
1639 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1640 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1641 | struct dm_target *ti; |
1642 | sector_t max_sectors; | |
5037108a | 1643 | int max_size = 0; |
f6fccb12 MB |
1644 | |
1645 | if (unlikely(!map)) | |
5037108a | 1646 | goto out; |
f6fccb12 MB |
1647 | |
1648 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1649 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1650 | goto out; |
f6fccb12 MB |
1651 | |
1652 | /* | |
1653 | * Find maximum amount of I/O that won't need splitting | |
1654 | */ | |
56a67df7 | 1655 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1656 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1657 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1658 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1659 | max_size = 0; |
1660 | ||
1661 | /* | |
1662 | * merge_bvec_fn() returns number of bytes | |
1663 | * it can accept at this offset | |
1664 | * max is precomputed maximal io size | |
1665 | */ | |
1666 | if (max_size && ti->type->merge) | |
1667 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1668 | /* |
1669 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1670 | * provided their merge_bvec method (we know this by looking for the |
1671 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1672 | * allow bios with multiple vector entries. So always set max_size | |
1673 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1674 | */ |
1675 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1676 | max_size = 0; |
f6fccb12 | 1677 | |
5037108a | 1678 | out: |
83d5e5b0 | 1679 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1680 | /* |
1681 | * Always allow an entire first page | |
1682 | */ | |
1683 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1684 | max_size = biovec->bv_len; | |
1685 | ||
f6fccb12 MB |
1686 | return max_size; |
1687 | } | |
1688 | ||
1da177e4 LT |
1689 | /* |
1690 | * The request function that just remaps the bio built up by | |
1691 | * dm_merge_bvec. | |
1692 | */ | |
5a7bbad2 | 1693 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1694 | { |
12f03a49 | 1695 | int rw = bio_data_dir(bio); |
1da177e4 | 1696 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1697 | int srcu_idx; |
1698 | struct dm_table *map; | |
1da177e4 | 1699 | |
83d5e5b0 | 1700 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1701 | |
18c0b223 | 1702 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1703 | |
6a8736d1 TH |
1704 | /* if we're suspended, we have to queue this io for later */ |
1705 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1706 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1707 | |
6a8736d1 TH |
1708 | if (bio_rw(bio) != READA) |
1709 | queue_io(md, bio); | |
1710 | else | |
54d9a1b4 | 1711 | bio_io_error(bio); |
5a7bbad2 | 1712 | return; |
1da177e4 LT |
1713 | } |
1714 | ||
83d5e5b0 MP |
1715 | __split_and_process_bio(md, map, bio); |
1716 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1717 | return; |
cec47e3d KU |
1718 | } |
1719 | ||
fd2ed4d2 | 1720 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1721 | { |
1722 | return blk_queue_stackable(md->queue); | |
1723 | } | |
1724 | ||
5a7bbad2 | 1725 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1726 | { |
1727 | struct mapped_device *md = q->queuedata; | |
1728 | ||
1729 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1730 | blk_queue_bio(q, bio); |
1731 | else | |
1732 | _dm_request(q, bio); | |
cec47e3d KU |
1733 | } |
1734 | ||
466d89a6 | 1735 | static void dm_dispatch_clone_request(struct request *clone, struct request *rq) |
cec47e3d KU |
1736 | { |
1737 | int r; | |
1738 | ||
466d89a6 KB |
1739 | if (blk_queue_io_stat(clone->q)) |
1740 | clone->cmd_flags |= REQ_IO_STAT; | |
cec47e3d | 1741 | |
466d89a6 KB |
1742 | clone->start_time = jiffies; |
1743 | r = blk_insert_cloned_request(clone->q, clone); | |
cec47e3d | 1744 | if (r) |
466d89a6 | 1745 | /* must complete clone in terms of original request */ |
cec47e3d KU |
1746 | dm_complete_request(rq, r); |
1747 | } | |
cec47e3d | 1748 | |
cec47e3d KU |
1749 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1750 | void *data) | |
1751 | { | |
1752 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1753 | struct dm_rq_clone_bio_info *info = |
1754 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1755 | |
1756 | info->orig = bio_orig; | |
1757 | info->tio = tio; | |
1758 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1759 | |
1760 | return 0; | |
1761 | } | |
1762 | ||
1763 | static int setup_clone(struct request *clone, struct request *rq, | |
1ae49ea2 | 1764 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
cec47e3d | 1765 | { |
d0bcb878 | 1766 | int r; |
cec47e3d | 1767 | |
1ae49ea2 | 1768 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, gfp_mask, |
29e4013d TH |
1769 | dm_rq_bio_constructor, tio); |
1770 | if (r) | |
1771 | return r; | |
cec47e3d | 1772 | |
29e4013d TH |
1773 | clone->cmd = rq->cmd; |
1774 | clone->cmd_len = rq->cmd_len; | |
1775 | clone->sense = rq->sense; | |
cec47e3d KU |
1776 | clone->end_io = end_clone_request; |
1777 | clone->end_io_data = tio; | |
1778 | ||
1ae49ea2 MS |
1779 | tio->clone = clone; |
1780 | ||
cec47e3d KU |
1781 | return 0; |
1782 | } | |
1783 | ||
6facdaff | 1784 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
466d89a6 | 1785 | struct dm_rq_target_io *tio, gfp_t gfp_mask) |
1ae49ea2 MS |
1786 | { |
1787 | struct request *clone = alloc_clone_request(md, gfp_mask); | |
1788 | ||
1789 | if (!clone) | |
1790 | return NULL; | |
1791 | ||
1792 | blk_rq_init(NULL, clone); | |
1793 | if (setup_clone(clone, rq, tio, gfp_mask)) { | |
1794 | /* -ENOMEM */ | |
1795 | free_clone_request(md, clone); | |
1796 | return NULL; | |
1797 | } | |
1798 | ||
1799 | return clone; | |
1800 | } | |
1801 | ||
2eb6e1e3 KB |
1802 | static void map_tio_request(struct kthread_work *work); |
1803 | ||
466d89a6 KB |
1804 | static struct dm_rq_target_io *prep_tio(struct request *rq, |
1805 | struct mapped_device *md, gfp_t gfp_mask) | |
6facdaff | 1806 | { |
6facdaff | 1807 | struct dm_rq_target_io *tio; |
e5863d9a MS |
1808 | int srcu_idx; |
1809 | struct dm_table *table; | |
6facdaff KU |
1810 | |
1811 | tio = alloc_rq_tio(md, gfp_mask); | |
1812 | if (!tio) | |
1813 | return NULL; | |
1814 | ||
1815 | tio->md = md; | |
1816 | tio->ti = NULL; | |
1ae49ea2 | 1817 | tio->clone = NULL; |
6facdaff KU |
1818 | tio->orig = rq; |
1819 | tio->error = 0; | |
1820 | memset(&tio->info, 0, sizeof(tio->info)); | |
2eb6e1e3 | 1821 | init_kthread_work(&tio->work, map_tio_request); |
6facdaff | 1822 | |
e5863d9a MS |
1823 | table = dm_get_live_table(md, &srcu_idx); |
1824 | if (!dm_table_mq_request_based(table)) { | |
1825 | if (!clone_rq(rq, md, tio, gfp_mask)) { | |
1826 | dm_put_live_table(md, srcu_idx); | |
1827 | free_rq_tio(tio); | |
1828 | return NULL; | |
1829 | } | |
6facdaff | 1830 | } |
e5863d9a | 1831 | dm_put_live_table(md, srcu_idx); |
6facdaff | 1832 | |
466d89a6 | 1833 | return tio; |
6facdaff KU |
1834 | } |
1835 | ||
cec47e3d KU |
1836 | /* |
1837 | * Called with the queue lock held. | |
1838 | */ | |
1839 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1840 | { | |
1841 | struct mapped_device *md = q->queuedata; | |
466d89a6 | 1842 | struct dm_rq_target_io *tio; |
cec47e3d | 1843 | |
cec47e3d KU |
1844 | if (unlikely(rq->special)) { |
1845 | DMWARN("Already has something in rq->special."); | |
1846 | return BLKPREP_KILL; | |
1847 | } | |
1848 | ||
466d89a6 KB |
1849 | tio = prep_tio(rq, md, GFP_ATOMIC); |
1850 | if (!tio) | |
cec47e3d | 1851 | return BLKPREP_DEFER; |
cec47e3d | 1852 | |
466d89a6 | 1853 | rq->special = tio; |
cec47e3d KU |
1854 | rq->cmd_flags |= REQ_DONTPREP; |
1855 | ||
1856 | return BLKPREP_OK; | |
1857 | } | |
1858 | ||
9eef87da KU |
1859 | /* |
1860 | * Returns: | |
e5863d9a MS |
1861 | * 0 : the request has been processed |
1862 | * DM_MAPIO_REQUEUE : the original request needs to be requeued | |
1863 | * < 0 : the request was completed due to failure | |
9eef87da | 1864 | */ |
466d89a6 | 1865 | static int map_request(struct dm_target *ti, struct request *rq, |
9eef87da | 1866 | struct mapped_device *md) |
cec47e3d | 1867 | { |
e5863d9a | 1868 | int r; |
466d89a6 | 1869 | struct dm_rq_target_io *tio = rq->special; |
e5863d9a MS |
1870 | struct request *clone = NULL; |
1871 | ||
1872 | if (tio->clone) { | |
1873 | clone = tio->clone; | |
1874 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1875 | } else { | |
1876 | r = ti->type->clone_and_map_rq(ti, rq, &tio->info, &clone); | |
1877 | if (r < 0) { | |
1878 | /* The target wants to complete the I/O */ | |
1879 | dm_kill_unmapped_request(rq, r); | |
1880 | return r; | |
1881 | } | |
1882 | if (IS_ERR(clone)) | |
1883 | return DM_MAPIO_REQUEUE; | |
1884 | if (setup_clone(clone, rq, tio, GFP_KERNEL)) { | |
1885 | /* -ENOMEM */ | |
1886 | ti->type->release_clone_rq(clone); | |
1887 | return DM_MAPIO_REQUEUE; | |
1888 | } | |
1889 | } | |
cec47e3d | 1890 | |
cec47e3d KU |
1891 | switch (r) { |
1892 | case DM_MAPIO_SUBMITTED: | |
1893 | /* The target has taken the I/O to submit by itself later */ | |
1894 | break; | |
1895 | case DM_MAPIO_REMAPPED: | |
1896 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 | 1897 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
466d89a6 KB |
1898 | blk_rq_pos(rq)); |
1899 | dm_dispatch_clone_request(clone, rq); | |
cec47e3d KU |
1900 | break; |
1901 | case DM_MAPIO_REQUEUE: | |
1902 | /* The target wants to requeue the I/O */ | |
1903 | dm_requeue_unmapped_request(clone); | |
1904 | break; | |
1905 | default: | |
1906 | if (r > 0) { | |
1907 | DMWARN("unimplemented target map return value: %d", r); | |
1908 | BUG(); | |
1909 | } | |
1910 | ||
1911 | /* The target wants to complete the I/O */ | |
466d89a6 | 1912 | dm_kill_unmapped_request(rq, r); |
e5863d9a | 1913 | return r; |
cec47e3d | 1914 | } |
9eef87da | 1915 | |
e5863d9a | 1916 | return 0; |
cec47e3d KU |
1917 | } |
1918 | ||
2eb6e1e3 | 1919 | static void map_tio_request(struct kthread_work *work) |
ba1cbad9 | 1920 | { |
2eb6e1e3 | 1921 | struct dm_rq_target_io *tio = container_of(work, struct dm_rq_target_io, work); |
e5863d9a MS |
1922 | struct request *rq = tio->orig; |
1923 | struct mapped_device *md = tio->md; | |
ba1cbad9 | 1924 | |
e5863d9a MS |
1925 | if (map_request(tio->ti, rq, md) == DM_MAPIO_REQUEUE) |
1926 | dm_requeue_unmapped_original_request(md, rq); | |
2eb6e1e3 KB |
1927 | } |
1928 | ||
466d89a6 | 1929 | static void dm_start_request(struct mapped_device *md, struct request *orig) |
ba1cbad9 | 1930 | { |
ba1cbad9 | 1931 | blk_start_request(orig); |
466d89a6 | 1932 | atomic_inc(&md->pending[rq_data_dir(orig)]); |
ba1cbad9 MS |
1933 | |
1934 | /* | |
1935 | * Hold the md reference here for the in-flight I/O. | |
1936 | * We can't rely on the reference count by device opener, | |
1937 | * because the device may be closed during the request completion | |
1938 | * when all bios are completed. | |
1939 | * See the comment in rq_completed() too. | |
1940 | */ | |
1941 | dm_get(md); | |
ba1cbad9 MS |
1942 | } |
1943 | ||
cec47e3d KU |
1944 | /* |
1945 | * q->request_fn for request-based dm. | |
1946 | * Called with the queue lock held. | |
1947 | */ | |
1948 | static void dm_request_fn(struct request_queue *q) | |
1949 | { | |
1950 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1951 | int srcu_idx; |
1952 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1953 | struct dm_target *ti; |
466d89a6 | 1954 | struct request *rq; |
2eb6e1e3 | 1955 | struct dm_rq_target_io *tio; |
29e4013d | 1956 | sector_t pos; |
cec47e3d KU |
1957 | |
1958 | /* | |
b4324fee KU |
1959 | * For suspend, check blk_queue_stopped() and increment |
1960 | * ->pending within a single queue_lock not to increment the | |
1961 | * number of in-flight I/Os after the queue is stopped in | |
1962 | * dm_suspend(). | |
cec47e3d | 1963 | */ |
7eaceacc | 1964 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1965 | rq = blk_peek_request(q); |
1966 | if (!rq) | |
7eaceacc | 1967 | goto delay_and_out; |
cec47e3d | 1968 | |
29e4013d TH |
1969 | /* always use block 0 to find the target for flushes for now */ |
1970 | pos = 0; | |
1971 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1972 | pos = blk_rq_pos(rq); | |
1973 | ||
1974 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1975 | if (!dm_target_is_valid(ti)) { |
1976 | /* | |
466d89a6 | 1977 | * Must perform setup, that rq_completed() requires, |
ba1cbad9 MS |
1978 | * before calling dm_kill_unmapped_request |
1979 | */ | |
1980 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
466d89a6 KB |
1981 | dm_start_request(md, rq); |
1982 | dm_kill_unmapped_request(rq, -EIO); | |
ba1cbad9 MS |
1983 | continue; |
1984 | } | |
d0bcb878 | 1985 | |
cec47e3d | 1986 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1987 | goto delay_and_out; |
cec47e3d | 1988 | |
466d89a6 | 1989 | dm_start_request(md, rq); |
9eef87da | 1990 | |
2eb6e1e3 KB |
1991 | tio = rq->special; |
1992 | /* Establish tio->ti before queuing work (map_tio_request) */ | |
1993 | tio->ti = ti; | |
1994 | queue_kthread_work(&md->kworker, &tio->work); | |
052189a2 | 1995 | BUG_ON(!irqs_disabled()); |
cec47e3d KU |
1996 | } |
1997 | ||
1998 | goto out; | |
1999 | ||
7eaceacc JA |
2000 | delay_and_out: |
2001 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 2002 | out: |
83d5e5b0 | 2003 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
2004 | } |
2005 | ||
2006 | int dm_underlying_device_busy(struct request_queue *q) | |
2007 | { | |
2008 | return blk_lld_busy(q); | |
2009 | } | |
2010 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
2011 | ||
2012 | static int dm_lld_busy(struct request_queue *q) | |
2013 | { | |
2014 | int r; | |
2015 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 2016 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
2017 | |
2018 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
2019 | r = 1; | |
2020 | else | |
2021 | r = dm_table_any_busy_target(map); | |
2022 | ||
83d5e5b0 | 2023 | dm_put_live_table_fast(md); |
cec47e3d KU |
2024 | |
2025 | return r; | |
2026 | } | |
2027 | ||
1da177e4 LT |
2028 | static int dm_any_congested(void *congested_data, int bdi_bits) |
2029 | { | |
8a57dfc6 CS |
2030 | int r = bdi_bits; |
2031 | struct mapped_device *md = congested_data; | |
2032 | struct dm_table *map; | |
1da177e4 | 2033 | |
1eb787ec | 2034 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 2035 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 2036 | if (map) { |
cec47e3d KU |
2037 | /* |
2038 | * Request-based dm cares about only own queue for | |
2039 | * the query about congestion status of request_queue | |
2040 | */ | |
2041 | if (dm_request_based(md)) | |
2042 | r = md->queue->backing_dev_info.state & | |
2043 | bdi_bits; | |
2044 | else | |
2045 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 2046 | } |
83d5e5b0 | 2047 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
2048 | } |
2049 | ||
1da177e4 LT |
2050 | return r; |
2051 | } | |
2052 | ||
2053 | /*----------------------------------------------------------------- | |
2054 | * An IDR is used to keep track of allocated minor numbers. | |
2055 | *---------------------------------------------------------------*/ | |
2b06cfff | 2056 | static void free_minor(int minor) |
1da177e4 | 2057 | { |
f32c10b0 | 2058 | spin_lock(&_minor_lock); |
1da177e4 | 2059 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 2060 | spin_unlock(&_minor_lock); |
1da177e4 LT |
2061 | } |
2062 | ||
2063 | /* | |
2064 | * See if the device with a specific minor # is free. | |
2065 | */ | |
cf13ab8e | 2066 | static int specific_minor(int minor) |
1da177e4 | 2067 | { |
c9d76be6 | 2068 | int r; |
1da177e4 LT |
2069 | |
2070 | if (minor >= (1 << MINORBITS)) | |
2071 | return -EINVAL; | |
2072 | ||
c9d76be6 | 2073 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2074 | spin_lock(&_minor_lock); |
1da177e4 | 2075 | |
c9d76be6 | 2076 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 2077 | |
f32c10b0 | 2078 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2079 | idr_preload_end(); |
2080 | if (r < 0) | |
2081 | return r == -ENOSPC ? -EBUSY : r; | |
2082 | return 0; | |
1da177e4 LT |
2083 | } |
2084 | ||
cf13ab8e | 2085 | static int next_free_minor(int *minor) |
1da177e4 | 2086 | { |
c9d76be6 | 2087 | int r; |
62f75c2f | 2088 | |
c9d76be6 | 2089 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2090 | spin_lock(&_minor_lock); |
1da177e4 | 2091 | |
c9d76be6 | 2092 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2093 | |
f32c10b0 | 2094 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2095 | idr_preload_end(); |
2096 | if (r < 0) | |
2097 | return r; | |
2098 | *minor = r; | |
2099 | return 0; | |
1da177e4 LT |
2100 | } |
2101 | ||
83d5cde4 | 2102 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2103 | |
53d5914f MP |
2104 | static void dm_wq_work(struct work_struct *work); |
2105 | ||
4a0b4ddf MS |
2106 | static void dm_init_md_queue(struct mapped_device *md) |
2107 | { | |
2108 | /* | |
2109 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
2110 | * devices. The type of this dm device has not been decided yet. | |
2111 | * The type is decided at the first table loading time. | |
2112 | * To prevent problematic device stacking, clear the queue flag | |
2113 | * for request stacking support until then. | |
2114 | * | |
2115 | * This queue is new, so no concurrency on the queue_flags. | |
2116 | */ | |
2117 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
2118 | ||
2119 | md->queue->queuedata = md; | |
2120 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2121 | md->queue->backing_dev_info.congested_data = md; | |
2122 | blk_queue_make_request(md->queue, dm_request); | |
2123 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
2124 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
2125 | } | |
2126 | ||
1da177e4 LT |
2127 | /* |
2128 | * Allocate and initialise a blank device with a given minor. | |
2129 | */ | |
2b06cfff | 2130 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2131 | { |
2132 | int r; | |
cf13ab8e | 2133 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2134 | void *old_md; |
1da177e4 LT |
2135 | |
2136 | if (!md) { | |
2137 | DMWARN("unable to allocate device, out of memory."); | |
2138 | return NULL; | |
2139 | } | |
2140 | ||
10da4f79 | 2141 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2142 | goto bad_module_get; |
10da4f79 | 2143 | |
1da177e4 | 2144 | /* get a minor number for the dev */ |
2b06cfff | 2145 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2146 | r = next_free_minor(&minor); |
2b06cfff | 2147 | else |
cf13ab8e | 2148 | r = specific_minor(minor); |
1da177e4 | 2149 | if (r < 0) |
6ed7ade8 | 2150 | goto bad_minor; |
1da177e4 | 2151 | |
83d5e5b0 MP |
2152 | r = init_srcu_struct(&md->io_barrier); |
2153 | if (r < 0) | |
2154 | goto bad_io_barrier; | |
2155 | ||
a5664dad | 2156 | md->type = DM_TYPE_NONE; |
e61290a4 | 2157 | mutex_init(&md->suspend_lock); |
a5664dad | 2158 | mutex_init(&md->type_lock); |
86f1152b | 2159 | mutex_init(&md->table_devices_lock); |
022c2611 | 2160 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2161 | atomic_set(&md->holders, 1); |
5c6bd75d | 2162 | atomic_set(&md->open_count, 0); |
1da177e4 | 2163 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2164 | atomic_set(&md->uevent_seq, 0); |
2165 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2166 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2167 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2168 | |
4a0b4ddf | 2169 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2170 | if (!md->queue) |
6ed7ade8 | 2171 | goto bad_queue; |
1da177e4 | 2172 | |
4a0b4ddf | 2173 | dm_init_md_queue(md); |
9faf400f | 2174 | |
1da177e4 LT |
2175 | md->disk = alloc_disk(1); |
2176 | if (!md->disk) | |
6ed7ade8 | 2177 | goto bad_disk; |
1da177e4 | 2178 | |
316d315b NK |
2179 | atomic_set(&md->pending[0], 0); |
2180 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2181 | init_waitqueue_head(&md->wait); |
53d5914f | 2182 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2183 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2184 | init_completion(&md->kobj_holder.completion); |
2eb6e1e3 | 2185 | md->kworker_task = NULL; |
f0b04115 | 2186 | |
1da177e4 LT |
2187 | md->disk->major = _major; |
2188 | md->disk->first_minor = minor; | |
2189 | md->disk->fops = &dm_blk_dops; | |
2190 | md->disk->queue = md->queue; | |
2191 | md->disk->private_data = md; | |
2192 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2193 | add_disk(md->disk); | |
7e51f257 | 2194 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2195 | |
670368a8 | 2196 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2197 | if (!md->wq) |
2198 | goto bad_thread; | |
2199 | ||
32a926da MP |
2200 | md->bdev = bdget_disk(md->disk, 0); |
2201 | if (!md->bdev) | |
2202 | goto bad_bdev; | |
2203 | ||
6a8736d1 TH |
2204 | bio_init(&md->flush_bio); |
2205 | md->flush_bio.bi_bdev = md->bdev; | |
2206 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2207 | ||
fd2ed4d2 MP |
2208 | dm_stats_init(&md->stats); |
2209 | ||
ba61fdd1 | 2210 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2211 | spin_lock(&_minor_lock); |
ba61fdd1 | 2212 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2213 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2214 | |
2215 | BUG_ON(old_md != MINOR_ALLOCED); | |
2216 | ||
1da177e4 LT |
2217 | return md; |
2218 | ||
32a926da MP |
2219 | bad_bdev: |
2220 | destroy_workqueue(md->wq); | |
304f3f6a | 2221 | bad_thread: |
03022c54 | 2222 | del_gendisk(md->disk); |
304f3f6a | 2223 | put_disk(md->disk); |
6ed7ade8 | 2224 | bad_disk: |
1312f40e | 2225 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2226 | bad_queue: |
83d5e5b0 MP |
2227 | cleanup_srcu_struct(&md->io_barrier); |
2228 | bad_io_barrier: | |
1da177e4 | 2229 | free_minor(minor); |
6ed7ade8 | 2230 | bad_minor: |
10da4f79 | 2231 | module_put(THIS_MODULE); |
6ed7ade8 | 2232 | bad_module_get: |
1da177e4 LT |
2233 | kfree(md); |
2234 | return NULL; | |
2235 | } | |
2236 | ||
ae9da83f JN |
2237 | static void unlock_fs(struct mapped_device *md); |
2238 | ||
1da177e4 LT |
2239 | static void free_dev(struct mapped_device *md) |
2240 | { | |
f331c029 | 2241 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2242 | |
32a926da MP |
2243 | unlock_fs(md); |
2244 | bdput(md->bdev); | |
304f3f6a | 2245 | destroy_workqueue(md->wq); |
2eb6e1e3 KB |
2246 | |
2247 | if (md->kworker_task) | |
2248 | kthread_stop(md->kworker_task); | |
e6ee8c0b KU |
2249 | if (md->io_pool) |
2250 | mempool_destroy(md->io_pool); | |
1ae49ea2 MS |
2251 | if (md->rq_pool) |
2252 | mempool_destroy(md->rq_pool); | |
e6ee8c0b KU |
2253 | if (md->bs) |
2254 | bioset_free(md->bs); | |
9c47008d | 2255 | blk_integrity_unregister(md->disk); |
1da177e4 | 2256 | del_gendisk(md->disk); |
83d5e5b0 | 2257 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2258 | free_table_devices(&md->table_devices); |
63d94e48 | 2259 | free_minor(minor); |
fba9f90e JM |
2260 | |
2261 | spin_lock(&_minor_lock); | |
2262 | md->disk->private_data = NULL; | |
2263 | spin_unlock(&_minor_lock); | |
2264 | ||
1da177e4 | 2265 | put_disk(md->disk); |
1312f40e | 2266 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2267 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2268 | module_put(THIS_MODULE); |
1da177e4 LT |
2269 | kfree(md); |
2270 | } | |
2271 | ||
e6ee8c0b KU |
2272 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2273 | { | |
c0820cf5 | 2274 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2275 | |
5f015204 | 2276 | if (md->io_pool && md->bs) { |
16245bdc JN |
2277 | /* The md already has necessary mempools. */ |
2278 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2279 | /* | |
2280 | * Reload bioset because front_pad may have changed | |
2281 | * because a different table was loaded. | |
2282 | */ | |
2283 | bioset_free(md->bs); | |
2284 | md->bs = p->bs; | |
2285 | p->bs = NULL; | |
16245bdc | 2286 | } |
466d89a6 KB |
2287 | /* |
2288 | * There's no need to reload with request-based dm | |
2289 | * because the size of front_pad doesn't change. | |
2290 | * Note for future: If you are to reload bioset, | |
2291 | * prep-ed requests in the queue may refer | |
2292 | * to bio from the old bioset, so you must walk | |
2293 | * through the queue to unprep. | |
2294 | */ | |
e6ee8c0b | 2295 | goto out; |
c0820cf5 | 2296 | } |
e6ee8c0b | 2297 | |
1ae49ea2 | 2298 | BUG_ON(!p || md->io_pool || md->rq_pool || md->bs); |
e6ee8c0b KU |
2299 | |
2300 | md->io_pool = p->io_pool; | |
2301 | p->io_pool = NULL; | |
1ae49ea2 MS |
2302 | md->rq_pool = p->rq_pool; |
2303 | p->rq_pool = NULL; | |
e6ee8c0b KU |
2304 | md->bs = p->bs; |
2305 | p->bs = NULL; | |
2306 | ||
2307 | out: | |
2308 | /* mempool bind completed, now no need any mempools in the table */ | |
2309 | dm_table_free_md_mempools(t); | |
2310 | } | |
2311 | ||
1da177e4 LT |
2312 | /* |
2313 | * Bind a table to the device. | |
2314 | */ | |
2315 | static void event_callback(void *context) | |
2316 | { | |
7a8c3d3b MA |
2317 | unsigned long flags; |
2318 | LIST_HEAD(uevents); | |
1da177e4 LT |
2319 | struct mapped_device *md = (struct mapped_device *) context; |
2320 | ||
7a8c3d3b MA |
2321 | spin_lock_irqsave(&md->uevent_lock, flags); |
2322 | list_splice_init(&md->uevent_list, &uevents); | |
2323 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2324 | ||
ed9e1982 | 2325 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2326 | |
1da177e4 LT |
2327 | atomic_inc(&md->event_nr); |
2328 | wake_up(&md->eventq); | |
2329 | } | |
2330 | ||
c217649b MS |
2331 | /* |
2332 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2333 | */ | |
4e90188b | 2334 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2335 | { |
4e90188b | 2336 | set_capacity(md->disk, size); |
1da177e4 | 2337 | |
db8fef4f | 2338 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2339 | } |
2340 | ||
d5b9dd04 MP |
2341 | /* |
2342 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2343 | * | |
2344 | * If this function returns 0, then the device is either a non-dm | |
2345 | * device without a merge_bvec_fn, or it is a dm device that is | |
2346 | * able to split any bios it receives that are too big. | |
2347 | */ | |
2348 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2349 | { | |
2350 | struct mapped_device *dev_md; | |
2351 | ||
2352 | if (!q->merge_bvec_fn) | |
2353 | return 0; | |
2354 | ||
2355 | if (q->make_request_fn == dm_request) { | |
2356 | dev_md = q->queuedata; | |
2357 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2358 | return 0; | |
2359 | } | |
2360 | ||
2361 | return 1; | |
2362 | } | |
2363 | ||
2364 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2365 | struct dm_dev *dev, sector_t start, | |
2366 | sector_t len, void *data) | |
2367 | { | |
2368 | struct block_device *bdev = dev->bdev; | |
2369 | struct request_queue *q = bdev_get_queue(bdev); | |
2370 | ||
2371 | return dm_queue_merge_is_compulsory(q); | |
2372 | } | |
2373 | ||
2374 | /* | |
2375 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2376 | * on the properties of the underlying devices. | |
2377 | */ | |
2378 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2379 | { | |
2380 | unsigned i = 0; | |
2381 | struct dm_target *ti; | |
2382 | ||
2383 | while (i < dm_table_get_num_targets(table)) { | |
2384 | ti = dm_table_get_target(table, i++); | |
2385 | ||
2386 | if (ti->type->iterate_devices && | |
2387 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2388 | return 0; | |
2389 | } | |
2390 | ||
2391 | return 1; | |
2392 | } | |
2393 | ||
042d2a9b AK |
2394 | /* |
2395 | * Returns old map, which caller must destroy. | |
2396 | */ | |
2397 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2398 | struct queue_limits *limits) | |
1da177e4 | 2399 | { |
042d2a9b | 2400 | struct dm_table *old_map; |
165125e1 | 2401 | struct request_queue *q = md->queue; |
1da177e4 | 2402 | sector_t size; |
d5b9dd04 | 2403 | int merge_is_optional; |
1da177e4 LT |
2404 | |
2405 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2406 | |
2407 | /* | |
2408 | * Wipe any geometry if the size of the table changed. | |
2409 | */ | |
fd2ed4d2 | 2410 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2411 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2412 | ||
32a926da | 2413 | __set_size(md, size); |
d5816876 | 2414 | |
2ca3310e AK |
2415 | dm_table_event_callback(t, event_callback, md); |
2416 | ||
e6ee8c0b KU |
2417 | /* |
2418 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2419 | * for request-based during suspension. So stop it to prevent | |
2420 | * I/O mapping before resume. | |
2421 | * This must be done before setting the queue restrictions, | |
2422 | * because request-based dm may be run just after the setting. | |
2423 | */ | |
2424 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2425 | stop_queue(q); | |
2426 | ||
2427 | __bind_mempools(md, t); | |
2428 | ||
d5b9dd04 MP |
2429 | merge_is_optional = dm_table_merge_is_optional(t); |
2430 | ||
a12f5d48 | 2431 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2432 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2433 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2434 | ||
754c5fc7 | 2435 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2436 | if (merge_is_optional) |
2437 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2438 | else | |
2439 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2440 | if (old_map) |
2441 | dm_sync_table(md); | |
1da177e4 | 2442 | |
042d2a9b | 2443 | return old_map; |
1da177e4 LT |
2444 | } |
2445 | ||
a7940155 AK |
2446 | /* |
2447 | * Returns unbound table for the caller to free. | |
2448 | */ | |
2449 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2450 | { |
a12f5d48 | 2451 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2452 | |
2453 | if (!map) | |
a7940155 | 2454 | return NULL; |
1da177e4 LT |
2455 | |
2456 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2457 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2458 | dm_sync_table(md); |
a7940155 AK |
2459 | |
2460 | return map; | |
1da177e4 LT |
2461 | } |
2462 | ||
2463 | /* | |
2464 | * Constructor for a new device. | |
2465 | */ | |
2b06cfff | 2466 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2467 | { |
2468 | struct mapped_device *md; | |
2469 | ||
2b06cfff | 2470 | md = alloc_dev(minor); |
1da177e4 LT |
2471 | if (!md) |
2472 | return -ENXIO; | |
2473 | ||
784aae73 MB |
2474 | dm_sysfs_init(md); |
2475 | ||
1da177e4 LT |
2476 | *result = md; |
2477 | return 0; | |
2478 | } | |
2479 | ||
a5664dad MS |
2480 | /* |
2481 | * Functions to manage md->type. | |
2482 | * All are required to hold md->type_lock. | |
2483 | */ | |
2484 | void dm_lock_md_type(struct mapped_device *md) | |
2485 | { | |
2486 | mutex_lock(&md->type_lock); | |
2487 | } | |
2488 | ||
2489 | void dm_unlock_md_type(struct mapped_device *md) | |
2490 | { | |
2491 | mutex_unlock(&md->type_lock); | |
2492 | } | |
2493 | ||
2494 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2495 | { | |
00c4fc3b | 2496 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2497 | md->type = type; |
2498 | } | |
2499 | ||
2500 | unsigned dm_get_md_type(struct mapped_device *md) | |
2501 | { | |
00c4fc3b | 2502 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2503 | return md->type; |
2504 | } | |
2505 | ||
e5863d9a MS |
2506 | static bool dm_md_type_request_based(struct mapped_device *md) |
2507 | { | |
2508 | unsigned table_type = dm_get_md_type(md); | |
2509 | ||
2510 | return (table_type == DM_TYPE_REQUEST_BASED || | |
2511 | table_type == DM_TYPE_MQ_REQUEST_BASED); | |
2512 | } | |
2513 | ||
36a0456f AK |
2514 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2515 | { | |
2516 | return md->immutable_target_type; | |
2517 | } | |
2518 | ||
f84cb8a4 MS |
2519 | /* |
2520 | * The queue_limits are only valid as long as you have a reference | |
2521 | * count on 'md'. | |
2522 | */ | |
2523 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2524 | { | |
2525 | BUG_ON(!atomic_read(&md->holders)); | |
2526 | return &md->queue->limits; | |
2527 | } | |
2528 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2529 | ||
4a0b4ddf MS |
2530 | /* |
2531 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2532 | */ | |
2533 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2534 | { | |
2535 | struct request_queue *q = NULL; | |
2536 | ||
2537 | if (md->queue->elevator) | |
2538 | return 1; | |
2539 | ||
2540 | /* Fully initialize the queue */ | |
2541 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2542 | if (!q) | |
2543 | return 0; | |
2544 | ||
2545 | md->queue = q; | |
4a0b4ddf MS |
2546 | dm_init_md_queue(md); |
2547 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2548 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2549 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf | 2550 | |
2eb6e1e3 KB |
2551 | /* Also initialize the request-based DM worker thread */ |
2552 | init_kthread_worker(&md->kworker); | |
2553 | md->kworker_task = kthread_run(kthread_worker_fn, &md->kworker, | |
2554 | "kdmwork-%s", dm_device_name(md)); | |
2555 | ||
4a0b4ddf MS |
2556 | elv_register_queue(md->queue); |
2557 | ||
2558 | return 1; | |
2559 | } | |
2560 | ||
2561 | /* | |
2562 | * Setup the DM device's queue based on md's type | |
2563 | */ | |
2564 | int dm_setup_md_queue(struct mapped_device *md) | |
2565 | { | |
e5863d9a | 2566 | if (dm_md_type_request_based(md) && !dm_init_request_based_queue(md)) { |
4a0b4ddf MS |
2567 | DMWARN("Cannot initialize queue for request-based mapped device"); |
2568 | return -EINVAL; | |
2569 | } | |
2570 | ||
2571 | return 0; | |
2572 | } | |
2573 | ||
2bec1f4a | 2574 | struct mapped_device *dm_get_md(dev_t dev) |
1da177e4 LT |
2575 | { |
2576 | struct mapped_device *md; | |
1da177e4 LT |
2577 | unsigned minor = MINOR(dev); |
2578 | ||
2579 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2580 | return NULL; | |
2581 | ||
f32c10b0 | 2582 | spin_lock(&_minor_lock); |
1da177e4 LT |
2583 | |
2584 | md = idr_find(&_minor_idr, minor); | |
2bec1f4a MP |
2585 | if (md) { |
2586 | if ((md == MINOR_ALLOCED || | |
2587 | (MINOR(disk_devt(dm_disk(md))) != minor) || | |
2588 | dm_deleting_md(md) || | |
2589 | test_bit(DMF_FREEING, &md->flags))) { | |
2590 | md = NULL; | |
2591 | goto out; | |
2592 | } | |
2593 | dm_get(md); | |
fba9f90e | 2594 | } |
1da177e4 | 2595 | |
fba9f90e | 2596 | out: |
f32c10b0 | 2597 | spin_unlock(&_minor_lock); |
1da177e4 | 2598 | |
637842cf DT |
2599 | return md; |
2600 | } | |
3cf2e4ba | 2601 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2602 | |
9ade92a9 | 2603 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2604 | { |
9ade92a9 | 2605 | return md->interface_ptr; |
1da177e4 LT |
2606 | } |
2607 | ||
2608 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2609 | { | |
2610 | md->interface_ptr = ptr; | |
2611 | } | |
2612 | ||
2613 | void dm_get(struct mapped_device *md) | |
2614 | { | |
2615 | atomic_inc(&md->holders); | |
3f77316d | 2616 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2617 | } |
2618 | ||
72d94861 AK |
2619 | const char *dm_device_name(struct mapped_device *md) |
2620 | { | |
2621 | return md->name; | |
2622 | } | |
2623 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2624 | ||
3f77316d | 2625 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2626 | { |
1134e5ae | 2627 | struct dm_table *map; |
83d5e5b0 | 2628 | int srcu_idx; |
1da177e4 | 2629 | |
3f77316d | 2630 | might_sleep(); |
fba9f90e | 2631 | |
3f77316d | 2632 | spin_lock(&_minor_lock); |
83d5e5b0 | 2633 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2634 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2635 | set_bit(DMF_FREEING, &md->flags); | |
2636 | spin_unlock(&_minor_lock); | |
2637 | ||
2eb6e1e3 KB |
2638 | if (dm_request_based(md)) |
2639 | flush_kthread_worker(&md->kworker); | |
2640 | ||
3f77316d KU |
2641 | if (!dm_suspended_md(md)) { |
2642 | dm_table_presuspend_targets(map); | |
2643 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2644 | } |
3f77316d | 2645 | |
83d5e5b0 MP |
2646 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2647 | dm_put_live_table(md, srcu_idx); | |
2648 | ||
3f77316d KU |
2649 | /* |
2650 | * Rare, but there may be I/O requests still going to complete, | |
2651 | * for example. Wait for all references to disappear. | |
2652 | * No one should increment the reference count of the mapped_device, | |
2653 | * after the mapped_device state becomes DMF_FREEING. | |
2654 | */ | |
2655 | if (wait) | |
2656 | while (atomic_read(&md->holders)) | |
2657 | msleep(1); | |
2658 | else if (atomic_read(&md->holders)) | |
2659 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2660 | dm_device_name(md), atomic_read(&md->holders)); | |
2661 | ||
2662 | dm_sysfs_exit(md); | |
3f77316d KU |
2663 | dm_table_destroy(__unbind(md)); |
2664 | free_dev(md); | |
2665 | } | |
2666 | ||
2667 | void dm_destroy(struct mapped_device *md) | |
2668 | { | |
2669 | __dm_destroy(md, true); | |
2670 | } | |
2671 | ||
2672 | void dm_destroy_immediate(struct mapped_device *md) | |
2673 | { | |
2674 | __dm_destroy(md, false); | |
2675 | } | |
2676 | ||
2677 | void dm_put(struct mapped_device *md) | |
2678 | { | |
2679 | atomic_dec(&md->holders); | |
1da177e4 | 2680 | } |
79eb885c | 2681 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2682 | |
401600df | 2683 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2684 | { |
2685 | int r = 0; | |
b44ebeb0 MP |
2686 | DECLARE_WAITQUEUE(wait, current); |
2687 | ||
b44ebeb0 | 2688 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2689 | |
2690 | while (1) { | |
401600df | 2691 | set_current_state(interruptible); |
46125c1c | 2692 | |
b4324fee | 2693 | if (!md_in_flight(md)) |
46125c1c MB |
2694 | break; |
2695 | ||
401600df MP |
2696 | if (interruptible == TASK_INTERRUPTIBLE && |
2697 | signal_pending(current)) { | |
46125c1c MB |
2698 | r = -EINTR; |
2699 | break; | |
2700 | } | |
2701 | ||
2702 | io_schedule(); | |
2703 | } | |
2704 | set_current_state(TASK_RUNNING); | |
2705 | ||
b44ebeb0 MP |
2706 | remove_wait_queue(&md->wait, &wait); |
2707 | ||
46125c1c MB |
2708 | return r; |
2709 | } | |
2710 | ||
1da177e4 LT |
2711 | /* |
2712 | * Process the deferred bios | |
2713 | */ | |
ef208587 | 2714 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2715 | { |
ef208587 MP |
2716 | struct mapped_device *md = container_of(work, struct mapped_device, |
2717 | work); | |
6d6f10df | 2718 | struct bio *c; |
83d5e5b0 MP |
2719 | int srcu_idx; |
2720 | struct dm_table *map; | |
1da177e4 | 2721 | |
83d5e5b0 | 2722 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2723 | |
3b00b203 | 2724 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2725 | spin_lock_irq(&md->deferred_lock); |
2726 | c = bio_list_pop(&md->deferred); | |
2727 | spin_unlock_irq(&md->deferred_lock); | |
2728 | ||
6a8736d1 | 2729 | if (!c) |
df12ee99 | 2730 | break; |
022c2611 | 2731 | |
e6ee8c0b KU |
2732 | if (dm_request_based(md)) |
2733 | generic_make_request(c); | |
6a8736d1 | 2734 | else |
83d5e5b0 | 2735 | __split_and_process_bio(md, map, c); |
022c2611 | 2736 | } |
73d410c0 | 2737 | |
83d5e5b0 | 2738 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2739 | } |
2740 | ||
9a1fb464 | 2741 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2742 | { |
3b00b203 | 2743 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2744 | smp_mb__after_atomic(); |
53d5914f | 2745 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2746 | } |
2747 | ||
1da177e4 | 2748 | /* |
042d2a9b | 2749 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2750 | */ |
042d2a9b | 2751 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2752 | { |
87eb5b21 | 2753 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2754 | struct queue_limits limits; |
042d2a9b | 2755 | int r; |
1da177e4 | 2756 | |
e61290a4 | 2757 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2758 | |
2759 | /* device must be suspended */ | |
4f186f8b | 2760 | if (!dm_suspended_md(md)) |
93c534ae | 2761 | goto out; |
1da177e4 | 2762 | |
3ae70656 MS |
2763 | /* |
2764 | * If the new table has no data devices, retain the existing limits. | |
2765 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2766 | * then new I/O is queued based on these limits, and then some paths | |
2767 | * reappear. | |
2768 | */ | |
2769 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2770 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2771 | if (live_map) |
2772 | limits = md->queue->limits; | |
83d5e5b0 | 2773 | dm_put_live_table_fast(md); |
3ae70656 MS |
2774 | } |
2775 | ||
87eb5b21 MC |
2776 | if (!live_map) { |
2777 | r = dm_calculate_queue_limits(table, &limits); | |
2778 | if (r) { | |
2779 | map = ERR_PTR(r); | |
2780 | goto out; | |
2781 | } | |
042d2a9b | 2782 | } |
754c5fc7 | 2783 | |
042d2a9b | 2784 | map = __bind(md, table, &limits); |
1da177e4 | 2785 | |
93c534ae | 2786 | out: |
e61290a4 | 2787 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2788 | return map; |
1da177e4 LT |
2789 | } |
2790 | ||
2791 | /* | |
2792 | * Functions to lock and unlock any filesystem running on the | |
2793 | * device. | |
2794 | */ | |
2ca3310e | 2795 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2796 | { |
e39e2e95 | 2797 | int r; |
1da177e4 LT |
2798 | |
2799 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2800 | |
db8fef4f | 2801 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2802 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2803 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2804 | md->frozen_sb = NULL; |
2805 | return r; | |
dfbe03f6 AK |
2806 | } |
2807 | ||
aa8d7c2f AK |
2808 | set_bit(DMF_FROZEN, &md->flags); |
2809 | ||
1da177e4 LT |
2810 | return 0; |
2811 | } | |
2812 | ||
2ca3310e | 2813 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2814 | { |
aa8d7c2f AK |
2815 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2816 | return; | |
2817 | ||
db8fef4f | 2818 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2819 | md->frozen_sb = NULL; |
aa8d7c2f | 2820 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2821 | } |
2822 | ||
2823 | /* | |
ffcc3936 MS |
2824 | * If __dm_suspend returns 0, the device is completely quiescent |
2825 | * now. There is no request-processing activity. All new requests | |
2826 | * are being added to md->deferred list. | |
cec47e3d | 2827 | * |
ffcc3936 | 2828 | * Caller must hold md->suspend_lock |
cec47e3d | 2829 | */ |
ffcc3936 MS |
2830 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
2831 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 2832 | { |
ffcc3936 MS |
2833 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2834 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2835 | int r; | |
1da177e4 | 2836 | |
2e93ccc1 KU |
2837 | /* |
2838 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2839 | * This flag is cleared before dm_suspend returns. | |
2840 | */ | |
2841 | if (noflush) | |
2842 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2843 | ||
d67ee213 MS |
2844 | /* |
2845 | * This gets reverted if there's an error later and the targets | |
2846 | * provide the .presuspend_undo hook. | |
2847 | */ | |
cf222b37 AK |
2848 | dm_table_presuspend_targets(map); |
2849 | ||
32a926da | 2850 | /* |
9f518b27 KU |
2851 | * Flush I/O to the device. |
2852 | * Any I/O submitted after lock_fs() may not be flushed. | |
2853 | * noflush takes precedence over do_lockfs. | |
2854 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2855 | */ |
2856 | if (!noflush && do_lockfs) { | |
2857 | r = lock_fs(md); | |
d67ee213 MS |
2858 | if (r) { |
2859 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2860 | return r; |
d67ee213 | 2861 | } |
aa8d7c2f | 2862 | } |
1da177e4 LT |
2863 | |
2864 | /* | |
3b00b203 MP |
2865 | * Here we must make sure that no processes are submitting requests |
2866 | * to target drivers i.e. no one may be executing | |
2867 | * __split_and_process_bio. This is called from dm_request and | |
2868 | * dm_wq_work. | |
2869 | * | |
2870 | * To get all processes out of __split_and_process_bio in dm_request, | |
2871 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2872 | * __split_and_process_bio from dm_request and quiesce the thread |
2873 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2874 | * flush_workqueue(md->wq). | |
1da177e4 | 2875 | */ |
1eb787ec | 2876 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2877 | if (map) |
2878 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2879 | |
d0bcb878 | 2880 | /* |
29e4013d TH |
2881 | * Stop md->queue before flushing md->wq in case request-based |
2882 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2883 | */ |
2eb6e1e3 | 2884 | if (dm_request_based(md)) { |
9f518b27 | 2885 | stop_queue(md->queue); |
2eb6e1e3 KB |
2886 | flush_kthread_worker(&md->kworker); |
2887 | } | |
cec47e3d | 2888 | |
d0bcb878 KU |
2889 | flush_workqueue(md->wq); |
2890 | ||
1da177e4 | 2891 | /* |
3b00b203 MP |
2892 | * At this point no more requests are entering target request routines. |
2893 | * We call dm_wait_for_completion to wait for all existing requests | |
2894 | * to finish. | |
1da177e4 | 2895 | */ |
ffcc3936 | 2896 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 2897 | |
6d6f10df | 2898 | if (noflush) |
022c2611 | 2899 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2900 | if (map) |
2901 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2902 | |
1da177e4 | 2903 | /* were we interrupted ? */ |
46125c1c | 2904 | if (r < 0) { |
9a1fb464 | 2905 | dm_queue_flush(md); |
73d410c0 | 2906 | |
cec47e3d | 2907 | if (dm_request_based(md)) |
9f518b27 | 2908 | start_queue(md->queue); |
cec47e3d | 2909 | |
2ca3310e | 2910 | unlock_fs(md); |
d67ee213 | 2911 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2912 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2913 | } |
1da177e4 | 2914 | |
ffcc3936 MS |
2915 | return r; |
2916 | } | |
2917 | ||
2918 | /* | |
2919 | * We need to be able to change a mapping table under a mounted | |
2920 | * filesystem. For example we might want to move some data in | |
2921 | * the background. Before the table can be swapped with | |
2922 | * dm_bind_table, dm_suspend must be called to flush any in | |
2923 | * flight bios and ensure that any further io gets deferred. | |
2924 | */ | |
2925 | /* | |
2926 | * Suspend mechanism in request-based dm. | |
2927 | * | |
2928 | * 1. Flush all I/Os by lock_fs() if needed. | |
2929 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2930 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2931 | * | |
2932 | * To abort suspend, start the request_queue. | |
2933 | */ | |
2934 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2935 | { | |
2936 | struct dm_table *map = NULL; | |
2937 | int r = 0; | |
2938 | ||
2939 | retry: | |
2940 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2941 | ||
2942 | if (dm_suspended_md(md)) { | |
2943 | r = -EINVAL; | |
2944 | goto out_unlock; | |
2945 | } | |
2946 | ||
2947 | if (dm_suspended_internally_md(md)) { | |
2948 | /* already internally suspended, wait for internal resume */ | |
2949 | mutex_unlock(&md->suspend_lock); | |
2950 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2951 | if (r) | |
2952 | return r; | |
2953 | goto retry; | |
2954 | } | |
2955 | ||
a12f5d48 | 2956 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2957 | |
2958 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
2959 | if (r) | |
2960 | goto out_unlock; | |
3b00b203 | 2961 | |
2ca3310e | 2962 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2963 | |
4d4471cb KU |
2964 | dm_table_postsuspend_targets(map); |
2965 | ||
d287483d | 2966 | out_unlock: |
e61290a4 | 2967 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2968 | return r; |
1da177e4 LT |
2969 | } |
2970 | ||
ffcc3936 MS |
2971 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2972 | { | |
2973 | if (map) { | |
2974 | int r = dm_table_resume_targets(map); | |
2975 | if (r) | |
2976 | return r; | |
2977 | } | |
2978 | ||
2979 | dm_queue_flush(md); | |
2980 | ||
2981 | /* | |
2982 | * Flushing deferred I/Os must be done after targets are resumed | |
2983 | * so that mapping of targets can work correctly. | |
2984 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2985 | */ | |
2986 | if (dm_request_based(md)) | |
2987 | start_queue(md->queue); | |
2988 | ||
2989 | unlock_fs(md); | |
2990 | ||
2991 | return 0; | |
2992 | } | |
2993 | ||
1da177e4 LT |
2994 | int dm_resume(struct mapped_device *md) |
2995 | { | |
cf222b37 | 2996 | int r = -EINVAL; |
cf222b37 | 2997 | struct dm_table *map = NULL; |
1da177e4 | 2998 | |
ffcc3936 MS |
2999 | retry: |
3000 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
3001 | ||
4f186f8b | 3002 | if (!dm_suspended_md(md)) |
cf222b37 | 3003 | goto out; |
cf222b37 | 3004 | |
ffcc3936 MS |
3005 | if (dm_suspended_internally_md(md)) { |
3006 | /* already internally suspended, wait for internal resume */ | |
3007 | mutex_unlock(&md->suspend_lock); | |
3008 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
3009 | if (r) | |
3010 | return r; | |
3011 | goto retry; | |
3012 | } | |
3013 | ||
a12f5d48 | 3014 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 3015 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 3016 | goto out; |
1da177e4 | 3017 | |
ffcc3936 | 3018 | r = __dm_resume(md, map); |
8757b776 MB |
3019 | if (r) |
3020 | goto out; | |
2ca3310e | 3021 | |
2ca3310e AK |
3022 | clear_bit(DMF_SUSPENDED, &md->flags); |
3023 | ||
cf222b37 AK |
3024 | r = 0; |
3025 | out: | |
e61290a4 | 3026 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 3027 | |
cf222b37 | 3028 | return r; |
1da177e4 LT |
3029 | } |
3030 | ||
fd2ed4d2 MP |
3031 | /* |
3032 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
3033 | * until all bios finish and prevents issuing new bios to the target drivers. | |
3034 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
3035 | */ |
3036 | ||
ffcc3936 | 3037 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 3038 | { |
ffcc3936 MS |
3039 | struct dm_table *map = NULL; |
3040 | ||
96b26c8c | 3041 | if (md->internal_suspend_count++) |
ffcc3936 MS |
3042 | return; /* nested internal suspend */ |
3043 | ||
3044 | if (dm_suspended_md(md)) { | |
3045 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3046 | return; /* nest suspend */ | |
3047 | } | |
3048 | ||
a12f5d48 | 3049 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
3050 | |
3051 | /* | |
3052 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
3053 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
3054 | * would require changing .presuspend to return an error -- avoid this | |
3055 | * until there is a need for more elaborate variants of internal suspend. | |
3056 | */ | |
3057 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
3058 | ||
3059 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3060 | ||
3061 | dm_table_postsuspend_targets(map); | |
3062 | } | |
3063 | ||
3064 | static void __dm_internal_resume(struct mapped_device *md) | |
3065 | { | |
96b26c8c MP |
3066 | BUG_ON(!md->internal_suspend_count); |
3067 | ||
3068 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
3069 | return; /* resume from nested internal suspend */ |
3070 | ||
fd2ed4d2 | 3071 | if (dm_suspended_md(md)) |
ffcc3936 MS |
3072 | goto done; /* resume from nested suspend */ |
3073 | ||
3074 | /* | |
3075 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
3076 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
3077 | */ | |
3078 | (void) __dm_resume(md, NULL); | |
3079 | ||
3080 | done: | |
3081 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3082 | smp_mb__after_atomic(); | |
3083 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
3084 | } | |
3085 | ||
3086 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
3087 | { | |
3088 | mutex_lock(&md->suspend_lock); | |
3089 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
3090 | mutex_unlock(&md->suspend_lock); | |
3091 | } | |
3092 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
3093 | ||
3094 | void dm_internal_resume(struct mapped_device *md) | |
3095 | { | |
3096 | mutex_lock(&md->suspend_lock); | |
3097 | __dm_internal_resume(md); | |
3098 | mutex_unlock(&md->suspend_lock); | |
3099 | } | |
3100 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
3101 | ||
3102 | /* | |
3103 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
3104 | * which prevents interaction with userspace-driven suspend. | |
3105 | */ | |
3106 | ||
3107 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3108 | { | |
3109 | mutex_lock(&md->suspend_lock); | |
3110 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3111 | return; |
3112 | ||
3113 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3114 | synchronize_srcu(&md->io_barrier); | |
3115 | flush_workqueue(md->wq); | |
3116 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3117 | } | |
3118 | ||
ffcc3936 | 3119 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3120 | { |
ffcc3936 | 3121 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3122 | goto done; |
3123 | ||
3124 | dm_queue_flush(md); | |
3125 | ||
3126 | done: | |
3127 | mutex_unlock(&md->suspend_lock); | |
3128 | } | |
3129 | ||
1da177e4 LT |
3130 | /*----------------------------------------------------------------- |
3131 | * Event notification. | |
3132 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3133 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3134 | unsigned cookie) |
69267a30 | 3135 | { |
60935eb2 MB |
3136 | char udev_cookie[DM_COOKIE_LENGTH]; |
3137 | char *envp[] = { udev_cookie, NULL }; | |
3138 | ||
3139 | if (!cookie) | |
3abf85b5 | 3140 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3141 | else { |
3142 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3143 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3144 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3145 | action, envp); | |
60935eb2 | 3146 | } |
69267a30 AK |
3147 | } |
3148 | ||
7a8c3d3b MA |
3149 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3150 | { | |
3151 | return atomic_add_return(1, &md->uevent_seq); | |
3152 | } | |
3153 | ||
1da177e4 LT |
3154 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3155 | { | |
3156 | return atomic_read(&md->event_nr); | |
3157 | } | |
3158 | ||
3159 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3160 | { | |
3161 | return wait_event_interruptible(md->eventq, | |
3162 | (event_nr != atomic_read(&md->event_nr))); | |
3163 | } | |
3164 | ||
7a8c3d3b MA |
3165 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3166 | { | |
3167 | unsigned long flags; | |
3168 | ||
3169 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3170 | list_add(elist, &md->uevent_list); | |
3171 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3172 | } | |
3173 | ||
1da177e4 LT |
3174 | /* |
3175 | * The gendisk is only valid as long as you have a reference | |
3176 | * count on 'md'. | |
3177 | */ | |
3178 | struct gendisk *dm_disk(struct mapped_device *md) | |
3179 | { | |
3180 | return md->disk; | |
3181 | } | |
3182 | ||
784aae73 MB |
3183 | struct kobject *dm_kobject(struct mapped_device *md) |
3184 | { | |
2995fa78 | 3185 | return &md->kobj_holder.kobj; |
784aae73 MB |
3186 | } |
3187 | ||
784aae73 MB |
3188 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3189 | { | |
3190 | struct mapped_device *md; | |
3191 | ||
2995fa78 | 3192 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3193 | |
4d89b7b4 | 3194 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3195 | dm_deleting_md(md)) |
4d89b7b4 MB |
3196 | return NULL; |
3197 | ||
784aae73 MB |
3198 | dm_get(md); |
3199 | return md; | |
3200 | } | |
3201 | ||
4f186f8b | 3202 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3203 | { |
3204 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3205 | } | |
3206 | ||
ffcc3936 MS |
3207 | int dm_suspended_internally_md(struct mapped_device *md) |
3208 | { | |
3209 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3210 | } | |
3211 | ||
2c140a24 MP |
3212 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3213 | { | |
3214 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3215 | } | |
3216 | ||
64dbce58 KU |
3217 | int dm_suspended(struct dm_target *ti) |
3218 | { | |
ecdb2e25 | 3219 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3220 | } |
3221 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3222 | ||
2e93ccc1 KU |
3223 | int dm_noflush_suspending(struct dm_target *ti) |
3224 | { | |
ecdb2e25 | 3225 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3226 | } |
3227 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3228 | ||
c0820cf5 | 3229 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 3230 | { |
5f015204 JN |
3231 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3232 | struct kmem_cache *cachep; | |
e5863d9a | 3233 | unsigned int pool_size = 0; |
5f015204 | 3234 | unsigned int front_pad; |
e6ee8c0b KU |
3235 | |
3236 | if (!pools) | |
3237 | return NULL; | |
3238 | ||
e5863d9a MS |
3239 | switch (type) { |
3240 | case DM_TYPE_BIO_BASED: | |
5f015204 | 3241 | cachep = _io_cache; |
e8603136 | 3242 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 | 3243 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
e5863d9a MS |
3244 | break; |
3245 | case DM_TYPE_REQUEST_BASED: | |
f4790826 | 3246 | pool_size = dm_get_reserved_rq_based_ios(); |
1ae49ea2 MS |
3247 | pools->rq_pool = mempool_create_slab_pool(pool_size, _rq_cache); |
3248 | if (!pools->rq_pool) | |
3249 | goto out; | |
e5863d9a MS |
3250 | /* fall through to setup remaining rq-based pools */ |
3251 | case DM_TYPE_MQ_REQUEST_BASED: | |
3252 | cachep = _rq_tio_cache; | |
3253 | if (!pool_size) | |
3254 | pool_size = dm_get_reserved_rq_based_ios(); | |
5f015204 JN |
3255 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3256 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3257 | WARN_ON(per_bio_data_size != 0); | |
e5863d9a MS |
3258 | break; |
3259 | default: | |
5f015204 | 3260 | goto out; |
e5863d9a | 3261 | } |
e6ee8c0b | 3262 | |
6cfa5857 | 3263 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
3264 | if (!pools->io_pool) |
3265 | goto out; | |
e6ee8c0b | 3266 | |
3d8aab2d | 3267 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3268 | if (!pools->bs) |
5f015204 | 3269 | goto out; |
e6ee8c0b | 3270 | |
a91a2785 | 3271 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3272 | goto out; |
a91a2785 | 3273 | |
e6ee8c0b KU |
3274 | return pools; |
3275 | ||
5f015204 JN |
3276 | out: |
3277 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3278 | |
3279 | return NULL; | |
3280 | } | |
3281 | ||
3282 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3283 | { | |
3284 | if (!pools) | |
3285 | return; | |
3286 | ||
3287 | if (pools->io_pool) | |
3288 | mempool_destroy(pools->io_pool); | |
3289 | ||
1ae49ea2 MS |
3290 | if (pools->rq_pool) |
3291 | mempool_destroy(pools->rq_pool); | |
3292 | ||
e6ee8c0b KU |
3293 | if (pools->bs) |
3294 | bioset_free(pools->bs); | |
3295 | ||
3296 | kfree(pools); | |
3297 | } | |
3298 | ||
83d5cde4 | 3299 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3300 | .open = dm_blk_open, |
3301 | .release = dm_blk_close, | |
aa129a22 | 3302 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3303 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3304 | .owner = THIS_MODULE |
3305 | }; | |
3306 | ||
1da177e4 LT |
3307 | /* |
3308 | * module hooks | |
3309 | */ | |
3310 | module_init(dm_init); | |
3311 | module_exit(dm_exit); | |
3312 | ||
3313 | module_param(major, uint, 0); | |
3314 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3315 | |
e8603136 MS |
3316 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3317 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3318 | ||
f4790826 MS |
3319 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3320 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3321 | ||
1da177e4 LT |
3322 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3323 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3324 | MODULE_LICENSE("GPL"); |