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> |
55782138 LZ |
23 | |
24 | #include <trace/events/block.h> | |
1da177e4 | 25 | |
72d94861 AK |
26 | #define DM_MSG_PREFIX "core" |
27 | ||
71a16736 NK |
28 | #ifdef CONFIG_PRINTK |
29 | /* | |
30 | * ratelimit state to be used in DMXXX_LIMIT(). | |
31 | */ | |
32 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
33 | DEFAULT_RATELIMIT_INTERVAL, | |
34 | DEFAULT_RATELIMIT_BURST); | |
35 | EXPORT_SYMBOL(dm_ratelimit_state); | |
36 | #endif | |
37 | ||
60935eb2 MB |
38 | /* |
39 | * Cookies are numeric values sent with CHANGE and REMOVE | |
40 | * uevents while resuming, removing or renaming the device. | |
41 | */ | |
42 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
43 | #define DM_COOKIE_LENGTH 24 | |
44 | ||
1da177e4 LT |
45 | static const char *_name = DM_NAME; |
46 | ||
47 | static unsigned int major = 0; | |
48 | static unsigned int _major = 0; | |
49 | ||
d15b774c AK |
50 | static DEFINE_IDR(_minor_idr); |
51 | ||
f32c10b0 | 52 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
53 | |
54 | static void do_deferred_remove(struct work_struct *w); | |
55 | ||
56 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
57 | ||
acfe0ad7 MP |
58 | static struct workqueue_struct *deferred_remove_workqueue; |
59 | ||
1da177e4 | 60 | /* |
8fbf26ad | 61 | * For bio-based dm. |
1da177e4 LT |
62 | * One of these is allocated per bio. |
63 | */ | |
64 | struct dm_io { | |
65 | struct mapped_device *md; | |
66 | int error; | |
1da177e4 | 67 | atomic_t io_count; |
6ae2fa67 | 68 | struct bio *bio; |
3eaf840e | 69 | unsigned long start_time; |
f88fb981 | 70 | spinlock_t endio_lock; |
fd2ed4d2 | 71 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
72 | }; |
73 | ||
8fbf26ad KU |
74 | /* |
75 | * For request-based dm. | |
76 | * One of these is allocated per request. | |
77 | */ | |
78 | struct dm_rq_target_io { | |
79 | struct mapped_device *md; | |
80 | struct dm_target *ti; | |
81 | struct request *orig, clone; | |
82 | int error; | |
83 | union map_info info; | |
84 | }; | |
85 | ||
86 | /* | |
94818742 KO |
87 | * For request-based dm - the bio clones we allocate are embedded in these |
88 | * structs. | |
89 | * | |
90 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
91 | * the bioset is created - this means the bio has to come at the end of the | |
92 | * struct. | |
8fbf26ad KU |
93 | */ |
94 | struct dm_rq_clone_bio_info { | |
95 | struct bio *orig; | |
cec47e3d | 96 | struct dm_rq_target_io *tio; |
94818742 | 97 | struct bio clone; |
8fbf26ad KU |
98 | }; |
99 | ||
cec47e3d KU |
100 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
101 | { | |
102 | if (rq && rq->end_io_data) | |
103 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
104 | return NULL; | |
105 | } | |
106 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
107 | ||
ba61fdd1 JM |
108 | #define MINOR_ALLOCED ((void *)-1) |
109 | ||
1da177e4 LT |
110 | /* |
111 | * Bits for the md->flags field. | |
112 | */ | |
1eb787ec | 113 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 114 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 115 | #define DMF_FROZEN 2 |
fba9f90e | 116 | #define DMF_FREEING 3 |
5c6bd75d | 117 | #define DMF_DELETING 4 |
2e93ccc1 | 118 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 119 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 120 | #define DMF_DEFERRED_REMOVE 7 |
ffcc3936 | 121 | #define DMF_SUSPENDED_INTERNALLY 8 |
1da177e4 | 122 | |
83d5e5b0 MP |
123 | /* |
124 | * A dummy definition to make RCU happy. | |
125 | * struct dm_table should never be dereferenced in this file. | |
126 | */ | |
127 | struct dm_table { | |
128 | int undefined__; | |
129 | }; | |
130 | ||
304f3f6a MB |
131 | /* |
132 | * Work processed by per-device workqueue. | |
133 | */ | |
1da177e4 | 134 | struct mapped_device { |
83d5e5b0 | 135 | struct srcu_struct io_barrier; |
e61290a4 | 136 | struct mutex suspend_lock; |
1da177e4 | 137 | atomic_t holders; |
5c6bd75d | 138 | atomic_t open_count; |
1da177e4 | 139 | |
2a7faeb1 MP |
140 | /* |
141 | * The current mapping. | |
142 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
143 | * dereference. | |
144 | */ | |
6fa99520 | 145 | struct dm_table __rcu *map; |
2a7faeb1 | 146 | |
86f1152b BM |
147 | struct list_head table_devices; |
148 | struct mutex table_devices_lock; | |
149 | ||
1da177e4 LT |
150 | unsigned long flags; |
151 | ||
165125e1 | 152 | struct request_queue *queue; |
a5664dad | 153 | unsigned type; |
4a0b4ddf | 154 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
155 | struct mutex type_lock; |
156 | ||
36a0456f AK |
157 | struct target_type *immutable_target_type; |
158 | ||
1da177e4 | 159 | struct gendisk *disk; |
7e51f257 | 160 | char name[16]; |
1da177e4 LT |
161 | |
162 | void *interface_ptr; | |
163 | ||
164 | /* | |
165 | * A list of ios that arrived while we were suspended. | |
166 | */ | |
316d315b | 167 | atomic_t pending[2]; |
1da177e4 | 168 | wait_queue_head_t wait; |
53d5914f | 169 | struct work_struct work; |
74859364 | 170 | struct bio_list deferred; |
022c2611 | 171 | spinlock_t deferred_lock; |
1da177e4 | 172 | |
af7e466a | 173 | /* |
29e4013d | 174 | * Processing queue (flush) |
304f3f6a MB |
175 | */ |
176 | struct workqueue_struct *wq; | |
177 | ||
1da177e4 LT |
178 | /* |
179 | * io objects are allocated from here. | |
180 | */ | |
181 | mempool_t *io_pool; | |
1da177e4 | 182 | |
9faf400f SB |
183 | struct bio_set *bs; |
184 | ||
1da177e4 LT |
185 | /* |
186 | * Event handling. | |
187 | */ | |
188 | atomic_t event_nr; | |
189 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
190 | atomic_t uevent_seq; |
191 | struct list_head uevent_list; | |
192 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
193 | |
194 | /* | |
195 | * freeze/thaw support require holding onto a super block | |
196 | */ | |
197 | struct super_block *frozen_sb; | |
db8fef4f | 198 | struct block_device *bdev; |
3ac51e74 DW |
199 | |
200 | /* forced geometry settings */ | |
201 | struct hd_geometry geometry; | |
784aae73 | 202 | |
2995fa78 MP |
203 | /* kobject and completion */ |
204 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 205 | |
d87f4c14 TH |
206 | /* zero-length flush that will be cloned and submitted to targets */ |
207 | struct bio flush_bio; | |
fd2ed4d2 | 208 | |
96b26c8c MP |
209 | /* the number of internal suspends */ |
210 | unsigned internal_suspend_count; | |
211 | ||
fd2ed4d2 | 212 | struct dm_stats stats; |
1da177e4 LT |
213 | }; |
214 | ||
e6ee8c0b KU |
215 | /* |
216 | * For mempools pre-allocation at the table loading time. | |
217 | */ | |
218 | struct dm_md_mempools { | |
219 | mempool_t *io_pool; | |
e6ee8c0b KU |
220 | struct bio_set *bs; |
221 | }; | |
222 | ||
86f1152b BM |
223 | struct table_device { |
224 | struct list_head list; | |
225 | atomic_t count; | |
226 | struct dm_dev dm_dev; | |
227 | }; | |
228 | ||
6cfa5857 MS |
229 | #define RESERVED_BIO_BASED_IOS 16 |
230 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 231 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 232 | static struct kmem_cache *_io_cache; |
8fbf26ad | 233 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 234 | |
e8603136 MS |
235 | /* |
236 | * Bio-based DM's mempools' reserved IOs set by the user. | |
237 | */ | |
238 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
239 | ||
f4790826 MS |
240 | /* |
241 | * Request-based DM's mempools' reserved IOs set by the user. | |
242 | */ | |
243 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
244 | ||
245 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
246 | unsigned def, unsigned max) | |
247 | { | |
248 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
249 | unsigned modified_ios = 0; | |
250 | ||
251 | if (!ios) | |
252 | modified_ios = def; | |
253 | else if (ios > max) | |
254 | modified_ios = max; | |
255 | ||
256 | if (modified_ios) { | |
257 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
258 | ios = modified_ios; | |
259 | } | |
260 | ||
261 | return ios; | |
262 | } | |
263 | ||
e8603136 MS |
264 | unsigned dm_get_reserved_bio_based_ios(void) |
265 | { | |
266 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
267 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
268 | } | |
269 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
270 | ||
f4790826 MS |
271 | unsigned dm_get_reserved_rq_based_ios(void) |
272 | { | |
273 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
274 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
275 | } | |
276 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
277 | ||
1da177e4 LT |
278 | static int __init local_init(void) |
279 | { | |
51157b4a | 280 | int r = -ENOMEM; |
1da177e4 | 281 | |
1da177e4 | 282 | /* allocate a slab for the dm_ios */ |
028867ac | 283 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 284 | if (!_io_cache) |
51157b4a | 285 | return r; |
1da177e4 | 286 | |
8fbf26ad KU |
287 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
288 | if (!_rq_tio_cache) | |
dba14160 | 289 | goto out_free_io_cache; |
8fbf26ad | 290 | |
51e5b2bd | 291 | r = dm_uevent_init(); |
51157b4a | 292 | if (r) |
23e5083b | 293 | goto out_free_rq_tio_cache; |
51e5b2bd | 294 | |
acfe0ad7 MP |
295 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
296 | if (!deferred_remove_workqueue) { | |
297 | r = -ENOMEM; | |
298 | goto out_uevent_exit; | |
299 | } | |
300 | ||
1da177e4 LT |
301 | _major = major; |
302 | r = register_blkdev(_major, _name); | |
51157b4a | 303 | if (r < 0) |
acfe0ad7 | 304 | goto out_free_workqueue; |
1da177e4 LT |
305 | |
306 | if (!_major) | |
307 | _major = r; | |
308 | ||
309 | return 0; | |
51157b4a | 310 | |
acfe0ad7 MP |
311 | out_free_workqueue: |
312 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
313 | out_uevent_exit: |
314 | dm_uevent_exit(); | |
8fbf26ad KU |
315 | out_free_rq_tio_cache: |
316 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
317 | out_free_io_cache: |
318 | kmem_cache_destroy(_io_cache); | |
319 | ||
320 | return r; | |
1da177e4 LT |
321 | } |
322 | ||
323 | static void local_exit(void) | |
324 | { | |
2c140a24 | 325 | flush_scheduled_work(); |
acfe0ad7 | 326 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 327 | |
8fbf26ad | 328 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 329 | kmem_cache_destroy(_io_cache); |
00d59405 | 330 | unregister_blkdev(_major, _name); |
51e5b2bd | 331 | dm_uevent_exit(); |
1da177e4 LT |
332 | |
333 | _major = 0; | |
334 | ||
335 | DMINFO("cleaned up"); | |
336 | } | |
337 | ||
b9249e55 | 338 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
339 | local_init, |
340 | dm_target_init, | |
341 | dm_linear_init, | |
342 | dm_stripe_init, | |
952b3557 | 343 | dm_io_init, |
945fa4d2 | 344 | dm_kcopyd_init, |
1da177e4 | 345 | dm_interface_init, |
fd2ed4d2 | 346 | dm_statistics_init, |
1da177e4 LT |
347 | }; |
348 | ||
b9249e55 | 349 | static void (*_exits[])(void) = { |
1da177e4 LT |
350 | local_exit, |
351 | dm_target_exit, | |
352 | dm_linear_exit, | |
353 | dm_stripe_exit, | |
952b3557 | 354 | dm_io_exit, |
945fa4d2 | 355 | dm_kcopyd_exit, |
1da177e4 | 356 | dm_interface_exit, |
fd2ed4d2 | 357 | dm_statistics_exit, |
1da177e4 LT |
358 | }; |
359 | ||
360 | static int __init dm_init(void) | |
361 | { | |
362 | const int count = ARRAY_SIZE(_inits); | |
363 | ||
364 | int r, i; | |
365 | ||
366 | for (i = 0; i < count; i++) { | |
367 | r = _inits[i](); | |
368 | if (r) | |
369 | goto bad; | |
370 | } | |
371 | ||
372 | return 0; | |
373 | ||
374 | bad: | |
375 | while (i--) | |
376 | _exits[i](); | |
377 | ||
378 | return r; | |
379 | } | |
380 | ||
381 | static void __exit dm_exit(void) | |
382 | { | |
383 | int i = ARRAY_SIZE(_exits); | |
384 | ||
385 | while (i--) | |
386 | _exits[i](); | |
d15b774c AK |
387 | |
388 | /* | |
389 | * Should be empty by this point. | |
390 | */ | |
d15b774c | 391 | idr_destroy(&_minor_idr); |
1da177e4 LT |
392 | } |
393 | ||
394 | /* | |
395 | * Block device functions | |
396 | */ | |
432a212c MA |
397 | int dm_deleting_md(struct mapped_device *md) |
398 | { | |
399 | return test_bit(DMF_DELETING, &md->flags); | |
400 | } | |
401 | ||
fe5f9f2c | 402 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
403 | { |
404 | struct mapped_device *md; | |
405 | ||
fba9f90e JM |
406 | spin_lock(&_minor_lock); |
407 | ||
fe5f9f2c | 408 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
409 | if (!md) |
410 | goto out; | |
411 | ||
5c6bd75d | 412 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 413 | dm_deleting_md(md)) { |
fba9f90e JM |
414 | md = NULL; |
415 | goto out; | |
416 | } | |
417 | ||
1da177e4 | 418 | dm_get(md); |
5c6bd75d | 419 | atomic_inc(&md->open_count); |
fba9f90e JM |
420 | |
421 | out: | |
422 | spin_unlock(&_minor_lock); | |
423 | ||
424 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
425 | } |
426 | ||
db2a144b | 427 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 428 | { |
fe5f9f2c | 429 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 430 | |
4a1aeb98 MB |
431 | spin_lock(&_minor_lock); |
432 | ||
2c140a24 MP |
433 | if (atomic_dec_and_test(&md->open_count) && |
434 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 435 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 436 | |
1da177e4 | 437 | dm_put(md); |
4a1aeb98 MB |
438 | |
439 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
440 | } |
441 | ||
5c6bd75d AK |
442 | int dm_open_count(struct mapped_device *md) |
443 | { | |
444 | return atomic_read(&md->open_count); | |
445 | } | |
446 | ||
447 | /* | |
448 | * Guarantees nothing is using the device before it's deleted. | |
449 | */ | |
2c140a24 | 450 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
451 | { |
452 | int r = 0; | |
453 | ||
454 | spin_lock(&_minor_lock); | |
455 | ||
2c140a24 | 456 | if (dm_open_count(md)) { |
5c6bd75d | 457 | r = -EBUSY; |
2c140a24 MP |
458 | if (mark_deferred) |
459 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
460 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
461 | r = -EEXIST; | |
5c6bd75d AK |
462 | else |
463 | set_bit(DMF_DELETING, &md->flags); | |
464 | ||
465 | spin_unlock(&_minor_lock); | |
466 | ||
467 | return r; | |
468 | } | |
469 | ||
2c140a24 MP |
470 | int dm_cancel_deferred_remove(struct mapped_device *md) |
471 | { | |
472 | int r = 0; | |
473 | ||
474 | spin_lock(&_minor_lock); | |
475 | ||
476 | if (test_bit(DMF_DELETING, &md->flags)) | |
477 | r = -EBUSY; | |
478 | else | |
479 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
480 | ||
481 | spin_unlock(&_minor_lock); | |
482 | ||
483 | return r; | |
484 | } | |
485 | ||
486 | static void do_deferred_remove(struct work_struct *w) | |
487 | { | |
488 | dm_deferred_remove(); | |
489 | } | |
490 | ||
fd2ed4d2 MP |
491 | sector_t dm_get_size(struct mapped_device *md) |
492 | { | |
493 | return get_capacity(md->disk); | |
494 | } | |
495 | ||
9974fa2c MS |
496 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
497 | { | |
498 | return md->queue; | |
499 | } | |
500 | ||
fd2ed4d2 MP |
501 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
502 | { | |
503 | return &md->stats; | |
504 | } | |
505 | ||
3ac51e74 DW |
506 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
507 | { | |
508 | struct mapped_device *md = bdev->bd_disk->private_data; | |
509 | ||
510 | return dm_get_geometry(md, geo); | |
511 | } | |
512 | ||
fe5f9f2c | 513 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
514 | unsigned int cmd, unsigned long arg) |
515 | { | |
fe5f9f2c | 516 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 517 | int srcu_idx; |
6c182cd8 | 518 | struct dm_table *map; |
aa129a22 MB |
519 | struct dm_target *tgt; |
520 | int r = -ENOTTY; | |
521 | ||
6c182cd8 | 522 | retry: |
83d5e5b0 MP |
523 | map = dm_get_live_table(md, &srcu_idx); |
524 | ||
aa129a22 MB |
525 | if (!map || !dm_table_get_size(map)) |
526 | goto out; | |
527 | ||
528 | /* We only support devices that have a single target */ | |
529 | if (dm_table_get_num_targets(map) != 1) | |
530 | goto out; | |
531 | ||
532 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
533 | if (!tgt->type->ioctl) |
534 | goto out; | |
aa129a22 | 535 | |
4f186f8b | 536 | if (dm_suspended_md(md)) { |
aa129a22 MB |
537 | r = -EAGAIN; |
538 | goto out; | |
539 | } | |
540 | ||
4d341d82 | 541 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
542 | |
543 | out: | |
83d5e5b0 | 544 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 545 | |
6c182cd8 HR |
546 | if (r == -ENOTCONN) { |
547 | msleep(10); | |
548 | goto retry; | |
549 | } | |
550 | ||
aa129a22 MB |
551 | return r; |
552 | } | |
553 | ||
028867ac | 554 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
555 | { |
556 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
557 | } | |
558 | ||
028867ac | 559 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
560 | { |
561 | mempool_free(io, md->io_pool); | |
562 | } | |
563 | ||
028867ac | 564 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 565 | { |
dba14160 | 566 | bio_put(&tio->clone); |
1da177e4 LT |
567 | } |
568 | ||
08885643 KU |
569 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
570 | gfp_t gfp_mask) | |
cec47e3d | 571 | { |
5f015204 | 572 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
573 | } |
574 | ||
575 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
576 | { | |
5f015204 | 577 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
578 | } |
579 | ||
90abb8c4 KU |
580 | static int md_in_flight(struct mapped_device *md) |
581 | { | |
582 | return atomic_read(&md->pending[READ]) + | |
583 | atomic_read(&md->pending[WRITE]); | |
584 | } | |
585 | ||
3eaf840e JNN |
586 | static void start_io_acct(struct dm_io *io) |
587 | { | |
588 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 589 | struct bio *bio = io->bio; |
c9959059 | 590 | int cpu; |
fd2ed4d2 | 591 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
592 | |
593 | io->start_time = jiffies; | |
594 | ||
074a7aca TH |
595 | cpu = part_stat_lock(); |
596 | part_round_stats(cpu, &dm_disk(md)->part0); | |
597 | part_stat_unlock(); | |
1e9bb880 SL |
598 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
599 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
600 | |
601 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 602 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 603 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
604 | } |
605 | ||
d221d2e7 | 606 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
607 | { |
608 | struct mapped_device *md = io->md; | |
609 | struct bio *bio = io->bio; | |
610 | unsigned long duration = jiffies - io->start_time; | |
18c0b223 | 611 | int pending; |
3eaf840e JNN |
612 | int rw = bio_data_dir(bio); |
613 | ||
18c0b223 | 614 | generic_end_io_acct(rw, &dm_disk(md)->part0, io->start_time); |
3eaf840e | 615 | |
fd2ed4d2 | 616 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 617 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
618 | bio_sectors(bio), true, duration, &io->stats_aux); |
619 | ||
af7e466a MP |
620 | /* |
621 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 622 | * a flush. |
af7e466a | 623 | */ |
1e9bb880 SL |
624 | pending = atomic_dec_return(&md->pending[rw]); |
625 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 626 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 627 | |
d221d2e7 MP |
628 | /* nudge anyone waiting on suspend queue */ |
629 | if (!pending) | |
630 | wake_up(&md->wait); | |
3eaf840e JNN |
631 | } |
632 | ||
1da177e4 LT |
633 | /* |
634 | * Add the bio to the list of deferred io. | |
635 | */ | |
92c63902 | 636 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 637 | { |
05447420 | 638 | unsigned long flags; |
1da177e4 | 639 | |
05447420 | 640 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 641 | bio_list_add(&md->deferred, bio); |
05447420 | 642 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 643 | queue_work(md->wq, &md->work); |
1da177e4 LT |
644 | } |
645 | ||
646 | /* | |
647 | * Everyone (including functions in this file), should use this | |
648 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 649 | * dm_put_live_table() when finished. |
1da177e4 | 650 | */ |
83d5e5b0 | 651 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 652 | { |
83d5e5b0 MP |
653 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
654 | ||
655 | return srcu_dereference(md->map, &md->io_barrier); | |
656 | } | |
1da177e4 | 657 | |
83d5e5b0 MP |
658 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
659 | { | |
660 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
661 | } | |
662 | ||
663 | void dm_sync_table(struct mapped_device *md) | |
664 | { | |
665 | synchronize_srcu(&md->io_barrier); | |
666 | synchronize_rcu_expedited(); | |
667 | } | |
668 | ||
669 | /* | |
670 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
671 | * The caller must not block between these two functions. | |
672 | */ | |
673 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
674 | { | |
675 | rcu_read_lock(); | |
676 | return rcu_dereference(md->map); | |
677 | } | |
1da177e4 | 678 | |
83d5e5b0 MP |
679 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
680 | { | |
681 | rcu_read_unlock(); | |
1da177e4 LT |
682 | } |
683 | ||
86f1152b BM |
684 | /* |
685 | * Open a table device so we can use it as a map destination. | |
686 | */ | |
687 | static int open_table_device(struct table_device *td, dev_t dev, | |
688 | struct mapped_device *md) | |
689 | { | |
690 | static char *_claim_ptr = "I belong to device-mapper"; | |
691 | struct block_device *bdev; | |
692 | ||
693 | int r; | |
694 | ||
695 | BUG_ON(td->dm_dev.bdev); | |
696 | ||
697 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
698 | if (IS_ERR(bdev)) | |
699 | return PTR_ERR(bdev); | |
700 | ||
701 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
702 | if (r) { | |
703 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
704 | return r; | |
705 | } | |
706 | ||
707 | td->dm_dev.bdev = bdev; | |
708 | return 0; | |
709 | } | |
710 | ||
711 | /* | |
712 | * Close a table device that we've been using. | |
713 | */ | |
714 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
715 | { | |
716 | if (!td->dm_dev.bdev) | |
717 | return; | |
718 | ||
719 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
720 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
721 | td->dm_dev.bdev = NULL; | |
722 | } | |
723 | ||
724 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
725 | fmode_t mode) { | |
726 | struct table_device *td; | |
727 | ||
728 | list_for_each_entry(td, l, list) | |
729 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
730 | return td; | |
731 | ||
732 | return NULL; | |
733 | } | |
734 | ||
735 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
736 | struct dm_dev **result) { | |
737 | int r; | |
738 | struct table_device *td; | |
739 | ||
740 | mutex_lock(&md->table_devices_lock); | |
741 | td = find_table_device(&md->table_devices, dev, mode); | |
742 | if (!td) { | |
743 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
744 | if (!td) { | |
745 | mutex_unlock(&md->table_devices_lock); | |
746 | return -ENOMEM; | |
747 | } | |
748 | ||
749 | td->dm_dev.mode = mode; | |
750 | td->dm_dev.bdev = NULL; | |
751 | ||
752 | if ((r = open_table_device(td, dev, md))) { | |
753 | mutex_unlock(&md->table_devices_lock); | |
754 | kfree(td); | |
755 | return r; | |
756 | } | |
757 | ||
758 | format_dev_t(td->dm_dev.name, dev); | |
759 | ||
760 | atomic_set(&td->count, 0); | |
761 | list_add(&td->list, &md->table_devices); | |
762 | } | |
763 | atomic_inc(&td->count); | |
764 | mutex_unlock(&md->table_devices_lock); | |
765 | ||
766 | *result = &td->dm_dev; | |
767 | return 0; | |
768 | } | |
769 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
770 | ||
771 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
772 | { | |
773 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
774 | ||
775 | mutex_lock(&md->table_devices_lock); | |
776 | if (atomic_dec_and_test(&td->count)) { | |
777 | close_table_device(td, md); | |
778 | list_del(&td->list); | |
779 | kfree(td); | |
780 | } | |
781 | mutex_unlock(&md->table_devices_lock); | |
782 | } | |
783 | EXPORT_SYMBOL(dm_put_table_device); | |
784 | ||
785 | static void free_table_devices(struct list_head *devices) | |
786 | { | |
787 | struct list_head *tmp, *next; | |
788 | ||
789 | list_for_each_safe(tmp, next, devices) { | |
790 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
791 | ||
792 | DMWARN("dm_destroy: %s still exists with %d references", | |
793 | td->dm_dev.name, atomic_read(&td->count)); | |
794 | kfree(td); | |
795 | } | |
796 | } | |
797 | ||
3ac51e74 DW |
798 | /* |
799 | * Get the geometry associated with a dm device | |
800 | */ | |
801 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
802 | { | |
803 | *geo = md->geometry; | |
804 | ||
805 | return 0; | |
806 | } | |
807 | ||
808 | /* | |
809 | * Set the geometry of a device. | |
810 | */ | |
811 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
812 | { | |
813 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
814 | ||
815 | if (geo->start > sz) { | |
816 | DMWARN("Start sector is beyond the geometry limits."); | |
817 | return -EINVAL; | |
818 | } | |
819 | ||
820 | md->geometry = *geo; | |
821 | ||
822 | return 0; | |
823 | } | |
824 | ||
1da177e4 LT |
825 | /*----------------------------------------------------------------- |
826 | * CRUD START: | |
827 | * A more elegant soln is in the works that uses the queue | |
828 | * merge fn, unfortunately there are a couple of changes to | |
829 | * the block layer that I want to make for this. So in the | |
830 | * interests of getting something for people to use I give | |
831 | * you this clearly demarcated crap. | |
832 | *---------------------------------------------------------------*/ | |
833 | ||
2e93ccc1 KU |
834 | static int __noflush_suspending(struct mapped_device *md) |
835 | { | |
836 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
837 | } | |
838 | ||
1da177e4 LT |
839 | /* |
840 | * Decrements the number of outstanding ios that a bio has been | |
841 | * cloned into, completing the original io if necc. | |
842 | */ | |
858119e1 | 843 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 844 | { |
2e93ccc1 | 845 | unsigned long flags; |
b35f8caa MB |
846 | int io_error; |
847 | struct bio *bio; | |
848 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
849 | |
850 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
851 | if (unlikely(error)) { |
852 | spin_lock_irqsave(&io->endio_lock, flags); | |
853 | if (!(io->error > 0 && __noflush_suspending(md))) | |
854 | io->error = error; | |
855 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
856 | } | |
1da177e4 LT |
857 | |
858 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
859 | if (io->error == DM_ENDIO_REQUEUE) { |
860 | /* | |
861 | * Target requested pushing back the I/O. | |
2e93ccc1 | 862 | */ |
022c2611 | 863 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
864 | if (__noflush_suspending(md)) |
865 | bio_list_add_head(&md->deferred, io->bio); | |
866 | else | |
2e93ccc1 KU |
867 | /* noflush suspend was interrupted. */ |
868 | io->error = -EIO; | |
022c2611 | 869 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
870 | } |
871 | ||
b35f8caa MB |
872 | io_error = io->error; |
873 | bio = io->bio; | |
6a8736d1 TH |
874 | end_io_acct(io); |
875 | free_io(md, io); | |
876 | ||
877 | if (io_error == DM_ENDIO_REQUEUE) | |
878 | return; | |
2e93ccc1 | 879 | |
4f024f37 | 880 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 881 | /* |
6a8736d1 TH |
882 | * Preflush done for flush with data, reissue |
883 | * without REQ_FLUSH. | |
af7e466a | 884 | */ |
6a8736d1 TH |
885 | bio->bi_rw &= ~REQ_FLUSH; |
886 | queue_io(md, bio); | |
af7e466a | 887 | } else { |
b372d360 | 888 | /* done with normal IO or empty flush */ |
0a82a8d1 | 889 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 890 | bio_endio(bio, io_error); |
b35f8caa | 891 | } |
1da177e4 LT |
892 | } |
893 | } | |
894 | ||
7eee4ae2 MS |
895 | static void disable_write_same(struct mapped_device *md) |
896 | { | |
897 | struct queue_limits *limits = dm_get_queue_limits(md); | |
898 | ||
899 | /* device doesn't really support WRITE SAME, disable it */ | |
900 | limits->max_write_same_sectors = 0; | |
901 | } | |
902 | ||
6712ecf8 | 903 | static void clone_endio(struct bio *bio, int error) |
1da177e4 | 904 | { |
5164bece | 905 | int r = error; |
bfc6d41c | 906 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 907 | struct dm_io *io = tio->io; |
9faf400f | 908 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
909 | dm_endio_fn endio = tio->ti->type->end_io; |
910 | ||
1da177e4 LT |
911 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
912 | error = -EIO; | |
913 | ||
914 | if (endio) { | |
7de3ee57 | 915 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
916 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
917 | /* | |
918 | * error and requeue request are handled | |
919 | * in dec_pending(). | |
920 | */ | |
1da177e4 | 921 | error = r; |
45cbcd79 KU |
922 | else if (r == DM_ENDIO_INCOMPLETE) |
923 | /* The target will handle the io */ | |
6712ecf8 | 924 | return; |
45cbcd79 KU |
925 | else if (r) { |
926 | DMWARN("unimplemented target endio return value: %d", r); | |
927 | BUG(); | |
928 | } | |
1da177e4 LT |
929 | } |
930 | ||
7eee4ae2 MS |
931 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
932 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
933 | disable_write_same(md); | |
934 | ||
9faf400f | 935 | free_tio(md, tio); |
b35f8caa | 936 | dec_pending(io, error); |
1da177e4 LT |
937 | } |
938 | ||
cec47e3d KU |
939 | /* |
940 | * Partial completion handling for request-based dm | |
941 | */ | |
942 | static void end_clone_bio(struct bio *clone, int error) | |
943 | { | |
bfc6d41c MP |
944 | struct dm_rq_clone_bio_info *info = |
945 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
946 | struct dm_rq_target_io *tio = info->tio; |
947 | struct bio *bio = info->orig; | |
4f024f37 | 948 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
949 | |
950 | bio_put(clone); | |
951 | ||
952 | if (tio->error) | |
953 | /* | |
954 | * An error has already been detected on the request. | |
955 | * Once error occurred, just let clone->end_io() handle | |
956 | * the remainder. | |
957 | */ | |
958 | return; | |
959 | else if (error) { | |
960 | /* | |
961 | * Don't notice the error to the upper layer yet. | |
962 | * The error handling decision is made by the target driver, | |
963 | * when the request is completed. | |
964 | */ | |
965 | tio->error = error; | |
966 | return; | |
967 | } | |
968 | ||
969 | /* | |
970 | * I/O for the bio successfully completed. | |
971 | * Notice the data completion to the upper layer. | |
972 | */ | |
973 | ||
974 | /* | |
975 | * bios are processed from the head of the list. | |
976 | * So the completing bio should always be rq->bio. | |
977 | * If it's not, something wrong is happening. | |
978 | */ | |
979 | if (tio->orig->bio != bio) | |
980 | DMERR("bio completion is going in the middle of the request"); | |
981 | ||
982 | /* | |
983 | * Update the original request. | |
984 | * Do not use blk_end_request() here, because it may complete | |
985 | * the original request before the clone, and break the ordering. | |
986 | */ | |
987 | blk_update_request(tio->orig, 0, nr_bytes); | |
988 | } | |
989 | ||
990 | /* | |
991 | * Don't touch any member of the md after calling this function because | |
992 | * the md may be freed in dm_put() at the end of this function. | |
993 | * Or do dm_get() before calling this function and dm_put() later. | |
994 | */ | |
b4324fee | 995 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 996 | { |
b4324fee | 997 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
998 | |
999 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 1000 | if (!md_in_flight(md)) |
cec47e3d KU |
1001 | wake_up(&md->wait); |
1002 | ||
a8c32a5c JA |
1003 | /* |
1004 | * Run this off this callpath, as drivers could invoke end_io while | |
1005 | * inside their request_fn (and holding the queue lock). Calling | |
1006 | * back into ->request_fn() could deadlock attempting to grab the | |
1007 | * queue lock again. | |
1008 | */ | |
cec47e3d | 1009 | if (run_queue) |
a8c32a5c | 1010 | blk_run_queue_async(md->queue); |
cec47e3d KU |
1011 | |
1012 | /* | |
1013 | * dm_put() must be at the end of this function. See the comment above | |
1014 | */ | |
1015 | dm_put(md); | |
1016 | } | |
1017 | ||
a77e28c7 KU |
1018 | static void free_rq_clone(struct request *clone) |
1019 | { | |
1020 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1021 | ||
1022 | blk_rq_unprep_clone(clone); | |
1023 | free_rq_tio(tio); | |
1024 | } | |
1025 | ||
980691e5 KU |
1026 | /* |
1027 | * Complete the clone and the original request. | |
1028 | * Must be called without queue lock. | |
1029 | */ | |
1030 | static void dm_end_request(struct request *clone, int error) | |
1031 | { | |
1032 | int rw = rq_data_dir(clone); | |
1033 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1034 | struct mapped_device *md = tio->md; | |
1035 | struct request *rq = tio->orig; | |
1036 | ||
29e4013d | 1037 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1038 | rq->errors = clone->errors; |
1039 | rq->resid_len = clone->resid_len; | |
1040 | ||
1041 | if (rq->sense) | |
1042 | /* | |
1043 | * We are using the sense buffer of the original | |
1044 | * request. | |
1045 | * So setting the length of the sense data is enough. | |
1046 | */ | |
1047 | rq->sense_len = clone->sense_len; | |
1048 | } | |
1049 | ||
1050 | free_rq_clone(clone); | |
29e4013d TH |
1051 | blk_end_request_all(rq, error); |
1052 | rq_completed(md, rw, true); | |
980691e5 KU |
1053 | } |
1054 | ||
cec47e3d KU |
1055 | static void dm_unprep_request(struct request *rq) |
1056 | { | |
1057 | struct request *clone = rq->special; | |
cec47e3d KU |
1058 | |
1059 | rq->special = NULL; | |
1060 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1061 | ||
a77e28c7 | 1062 | free_rq_clone(clone); |
cec47e3d KU |
1063 | } |
1064 | ||
1065 | /* | |
1066 | * Requeue the original request of a clone. | |
1067 | */ | |
1068 | void dm_requeue_unmapped_request(struct request *clone) | |
1069 | { | |
b4324fee | 1070 | int rw = rq_data_dir(clone); |
cec47e3d KU |
1071 | struct dm_rq_target_io *tio = clone->end_io_data; |
1072 | struct mapped_device *md = tio->md; | |
1073 | struct request *rq = tio->orig; | |
1074 | struct request_queue *q = rq->q; | |
1075 | unsigned long flags; | |
1076 | ||
1077 | dm_unprep_request(rq); | |
1078 | ||
1079 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
1080 | blk_requeue_request(q, rq); |
1081 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1082 | ||
b4324fee | 1083 | rq_completed(md, rw, 0); |
cec47e3d KU |
1084 | } |
1085 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
1086 | ||
1087 | static void __stop_queue(struct request_queue *q) | |
1088 | { | |
1089 | blk_stop_queue(q); | |
1090 | } | |
1091 | ||
1092 | static void stop_queue(struct request_queue *q) | |
1093 | { | |
1094 | unsigned long flags; | |
1095 | ||
1096 | spin_lock_irqsave(q->queue_lock, flags); | |
1097 | __stop_queue(q); | |
1098 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1099 | } | |
1100 | ||
1101 | static void __start_queue(struct request_queue *q) | |
1102 | { | |
1103 | if (blk_queue_stopped(q)) | |
1104 | blk_start_queue(q); | |
1105 | } | |
1106 | ||
1107 | static void start_queue(struct request_queue *q) | |
1108 | { | |
1109 | unsigned long flags; | |
1110 | ||
1111 | spin_lock_irqsave(q->queue_lock, flags); | |
1112 | __start_queue(q); | |
1113 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1114 | } | |
1115 | ||
11a68244 | 1116 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1117 | { |
11a68244 | 1118 | int r = error; |
cec47e3d | 1119 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1120 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1121 | |
ba1cbad9 MS |
1122 | if (tio->ti) { |
1123 | rq_end_io = tio->ti->type->rq_end_io; | |
1124 | ||
1125 | if (mapped && rq_end_io) | |
1126 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1127 | } | |
cec47e3d | 1128 | |
7eee4ae2 MS |
1129 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1130 | !clone->q->limits.max_write_same_sectors)) | |
1131 | disable_write_same(tio->md); | |
1132 | ||
11a68244 | 1133 | if (r <= 0) |
cec47e3d | 1134 | /* The target wants to complete the I/O */ |
11a68244 KU |
1135 | dm_end_request(clone, r); |
1136 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1137 | /* The target will handle the I/O */ |
1138 | return; | |
11a68244 | 1139 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1140 | /* The target wants to requeue the I/O */ |
1141 | dm_requeue_unmapped_request(clone); | |
1142 | else { | |
11a68244 | 1143 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1144 | BUG(); |
1145 | } | |
1146 | } | |
1147 | ||
11a68244 KU |
1148 | /* |
1149 | * Request completion handler for request-based dm | |
1150 | */ | |
1151 | static void dm_softirq_done(struct request *rq) | |
1152 | { | |
1153 | bool mapped = true; | |
1154 | struct request *clone = rq->completion_data; | |
1155 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1156 | ||
1157 | if (rq->cmd_flags & REQ_FAILED) | |
1158 | mapped = false; | |
1159 | ||
1160 | dm_done(clone, tio->error, mapped); | |
1161 | } | |
1162 | ||
cec47e3d KU |
1163 | /* |
1164 | * Complete the clone and the original request with the error status | |
1165 | * through softirq context. | |
1166 | */ | |
1167 | static void dm_complete_request(struct request *clone, int error) | |
1168 | { | |
1169 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1170 | struct request *rq = tio->orig; | |
1171 | ||
1172 | tio->error = error; | |
1173 | rq->completion_data = clone; | |
1174 | blk_complete_request(rq); | |
1175 | } | |
1176 | ||
1177 | /* | |
1178 | * Complete the not-mapped clone and the original request with the error status | |
1179 | * through softirq context. | |
1180 | * Target's rq_end_io() function isn't called. | |
1181 | * This may be used when the target's map_rq() function fails. | |
1182 | */ | |
1183 | void dm_kill_unmapped_request(struct request *clone, int error) | |
1184 | { | |
1185 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1186 | struct request *rq = tio->orig; | |
1187 | ||
1188 | rq->cmd_flags |= REQ_FAILED; | |
1189 | dm_complete_request(clone, error); | |
1190 | } | |
1191 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1192 | ||
1193 | /* | |
1194 | * Called with the queue lock held | |
1195 | */ | |
1196 | static void end_clone_request(struct request *clone, int error) | |
1197 | { | |
1198 | /* | |
1199 | * For just cleaning up the information of the queue in which | |
1200 | * the clone was dispatched. | |
1201 | * The clone is *NOT* freed actually here because it is alloced from | |
1202 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1203 | */ | |
1204 | __blk_put_request(clone->q, clone); | |
1205 | ||
1206 | /* | |
1207 | * Actual request completion is done in a softirq context which doesn't | |
1208 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1209 | * - another request may be submitted by the upper level driver | |
1210 | * of the stacking during the completion | |
1211 | * - the submission which requires queue lock may be done | |
1212 | * against this queue | |
1213 | */ | |
1214 | dm_complete_request(clone, error); | |
1215 | } | |
1216 | ||
56a67df7 MS |
1217 | /* |
1218 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1219 | * target boundary. | |
1220 | */ | |
1221 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1222 | { | |
1223 | sector_t target_offset = dm_target_offset(ti, sector); | |
1224 | ||
1225 | return ti->len - target_offset; | |
1226 | } | |
1227 | ||
1228 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1229 | { |
56a67df7 | 1230 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1231 | sector_t offset, max_len; |
1da177e4 LT |
1232 | |
1233 | /* | |
542f9038 | 1234 | * Does the target need to split even further? |
1da177e4 | 1235 | */ |
542f9038 MS |
1236 | if (ti->max_io_len) { |
1237 | offset = dm_target_offset(ti, sector); | |
1238 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1239 | max_len = sector_div(offset, ti->max_io_len); | |
1240 | else | |
1241 | max_len = offset & (ti->max_io_len - 1); | |
1242 | max_len = ti->max_io_len - max_len; | |
1243 | ||
1244 | if (len > max_len) | |
1245 | len = max_len; | |
1da177e4 LT |
1246 | } |
1247 | ||
1248 | return len; | |
1249 | } | |
1250 | ||
542f9038 MS |
1251 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1252 | { | |
1253 | if (len > UINT_MAX) { | |
1254 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1255 | (unsigned long long)len, UINT_MAX); | |
1256 | ti->error = "Maximum size of target IO is too large"; | |
1257 | return -EINVAL; | |
1258 | } | |
1259 | ||
1260 | ti->max_io_len = (uint32_t) len; | |
1261 | ||
1262 | return 0; | |
1263 | } | |
1264 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1265 | ||
1dd40c3e MP |
1266 | /* |
1267 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1268 | * allowed for all bio types except REQ_FLUSH. | |
1269 | * | |
1270 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1271 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1272 | * sent in a next bio. | |
1273 | * | |
1274 | * A diagram that explains the arithmetics: | |
1275 | * +--------------------+---------------+-------+ | |
1276 | * | 1 | 2 | 3 | | |
1277 | * +--------------------+---------------+-------+ | |
1278 | * | |
1279 | * <-------------- *tio->len_ptr ---------------> | |
1280 | * <------- bi_size -------> | |
1281 | * <-- n_sectors --> | |
1282 | * | |
1283 | * Region 1 was already iterated over with bio_advance or similar function. | |
1284 | * (it may be empty if the target doesn't use bio_advance) | |
1285 | * Region 2 is the remaining bio size that the target wants to process. | |
1286 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1287 | * to make it empty) | |
1288 | * The target requires that region 3 is to be sent in the next bio. | |
1289 | * | |
1290 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1291 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1292 | * copies of the bio. | |
1293 | */ | |
1294 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1295 | { | |
1296 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1297 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1298 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1299 | BUG_ON(bi_size > *tio->len_ptr); | |
1300 | BUG_ON(n_sectors > bi_size); | |
1301 | *tio->len_ptr -= bi_size - n_sectors; | |
1302 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1303 | } | |
1304 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1305 | ||
bd2a49b8 | 1306 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1307 | { |
1308 | int r; | |
2056a782 | 1309 | sector_t sector; |
9faf400f | 1310 | struct mapped_device *md; |
dba14160 | 1311 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1312 | struct dm_target *ti = tio->ti; |
1da177e4 | 1313 | |
1da177e4 | 1314 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1315 | |
1316 | /* | |
1317 | * Map the clone. If r == 0 we don't need to do | |
1318 | * anything, the target has assumed ownership of | |
1319 | * this io. | |
1320 | */ | |
1321 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1322 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1323 | r = ti->type->map(ti, clone); |
45cbcd79 | 1324 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1325 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1326 | |
d07335e5 MS |
1327 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1328 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1329 | |
1da177e4 | 1330 | generic_make_request(clone); |
2e93ccc1 KU |
1331 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1332 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1333 | md = tio->io->md; |
1334 | dec_pending(tio->io, r); | |
9faf400f | 1335 | free_tio(md, tio); |
45cbcd79 KU |
1336 | } else if (r) { |
1337 | DMWARN("unimplemented target map return value: %d", r); | |
1338 | BUG(); | |
1da177e4 LT |
1339 | } |
1340 | } | |
1341 | ||
1342 | struct clone_info { | |
1343 | struct mapped_device *md; | |
1344 | struct dm_table *map; | |
1345 | struct bio *bio; | |
1346 | struct dm_io *io; | |
1347 | sector_t sector; | |
e0d6609a | 1348 | unsigned sector_count; |
1da177e4 LT |
1349 | }; |
1350 | ||
e0d6609a | 1351 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1352 | { |
4f024f37 KO |
1353 | bio->bi_iter.bi_sector = sector; |
1354 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1355 | } |
1356 | ||
1357 | /* | |
1358 | * Creates a bio that consists of range of complete bvecs. | |
1359 | */ | |
dba14160 | 1360 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1361 | sector_t sector, unsigned len) |
1da177e4 | 1362 | { |
dba14160 | 1363 | struct bio *clone = &tio->clone; |
1da177e4 | 1364 | |
1c3b13e6 KO |
1365 | __bio_clone_fast(clone, bio); |
1366 | ||
1367 | if (bio_integrity(bio)) | |
1368 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1369 | |
1c3b13e6 KO |
1370 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1371 | clone->bi_iter.bi_size = to_bytes(len); | |
1372 | ||
1373 | if (bio_integrity(bio)) | |
1374 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1375 | } |
1376 | ||
9015df24 | 1377 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1378 | struct dm_target *ti, |
55a62eef | 1379 | unsigned target_bio_nr) |
f9ab94ce | 1380 | { |
dba14160 MP |
1381 | struct dm_target_io *tio; |
1382 | struct bio *clone; | |
1383 | ||
99778273 | 1384 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1385 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1386 | |
1387 | tio->io = ci->io; | |
1388 | tio->ti = ti; | |
55a62eef | 1389 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1390 | |
1391 | return tio; | |
1392 | } | |
1393 | ||
14fe594d AK |
1394 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1395 | struct dm_target *ti, | |
1dd40c3e | 1396 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1397 | { |
99778273 | 1398 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1399 | struct bio *clone = &tio->clone; |
9015df24 | 1400 | |
1dd40c3e MP |
1401 | tio->len_ptr = len; |
1402 | ||
99778273 | 1403 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1404 | if (len) |
1dd40c3e | 1405 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1406 | |
bd2a49b8 | 1407 | __map_bio(tio); |
f9ab94ce MP |
1408 | } |
1409 | ||
14fe594d | 1410 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1411 | unsigned num_bios, unsigned *len) |
06a426ce | 1412 | { |
55a62eef | 1413 | unsigned target_bio_nr; |
06a426ce | 1414 | |
55a62eef | 1415 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1416 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1417 | } |
1418 | ||
14fe594d | 1419 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1420 | { |
06a426ce | 1421 | unsigned target_nr = 0; |
f9ab94ce MP |
1422 | struct dm_target *ti; |
1423 | ||
b372d360 | 1424 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1425 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1426 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1427 | |
f9ab94ce MP |
1428 | return 0; |
1429 | } | |
1430 | ||
e4c93811 | 1431 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1432 | sector_t sector, unsigned *len) |
5ae89a87 | 1433 | { |
dba14160 | 1434 | struct bio *bio = ci->bio; |
5ae89a87 | 1435 | struct dm_target_io *tio; |
b0d8ed4d AK |
1436 | unsigned target_bio_nr; |
1437 | unsigned num_target_bios = 1; | |
5ae89a87 | 1438 | |
b0d8ed4d AK |
1439 | /* |
1440 | * Does the target want to receive duplicate copies of the bio? | |
1441 | */ | |
1442 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1443 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1444 | |
b0d8ed4d | 1445 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1446 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1447 | tio->len_ptr = len; |
1448 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1449 | __map_bio(tio); |
1450 | } | |
5ae89a87 MS |
1451 | } |
1452 | ||
55a62eef | 1453 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1454 | |
55a62eef | 1455 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1456 | { |
55a62eef | 1457 | return ti->num_discard_bios; |
23508a96 MS |
1458 | } |
1459 | ||
55a62eef | 1460 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1461 | { |
55a62eef | 1462 | return ti->num_write_same_bios; |
23508a96 MS |
1463 | } |
1464 | ||
1465 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1466 | ||
1467 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1468 | { | |
55a62eef | 1469 | return ti->split_discard_bios; |
23508a96 MS |
1470 | } |
1471 | ||
14fe594d AK |
1472 | static int __send_changing_extent_only(struct clone_info *ci, |
1473 | get_num_bios_fn get_num_bios, | |
1474 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1475 | { |
1476 | struct dm_target *ti; | |
e0d6609a | 1477 | unsigned len; |
55a62eef | 1478 | unsigned num_bios; |
5ae89a87 | 1479 | |
a79245b3 MS |
1480 | do { |
1481 | ti = dm_table_find_target(ci->map, ci->sector); | |
1482 | if (!dm_target_is_valid(ti)) | |
1483 | return -EIO; | |
5ae89a87 | 1484 | |
5ae89a87 | 1485 | /* |
23508a96 MS |
1486 | * Even though the device advertised support for this type of |
1487 | * request, that does not mean every target supports it, and | |
936688d7 | 1488 | * reconfiguration might also have changed that since the |
a79245b3 | 1489 | * check was performed. |
5ae89a87 | 1490 | */ |
55a62eef AK |
1491 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1492 | if (!num_bios) | |
a79245b3 | 1493 | return -EOPNOTSUPP; |
5ae89a87 | 1494 | |
23508a96 | 1495 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1496 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1497 | else |
e0d6609a | 1498 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1499 | |
1dd40c3e | 1500 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1501 | |
1502 | ci->sector += len; | |
1503 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1504 | |
1505 | return 0; | |
1506 | } | |
1507 | ||
14fe594d | 1508 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1509 | { |
14fe594d AK |
1510 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1511 | is_split_required_for_discard); | |
23508a96 MS |
1512 | } |
1513 | ||
14fe594d | 1514 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1515 | { |
14fe594d | 1516 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1517 | } |
1518 | ||
e4c93811 AK |
1519 | /* |
1520 | * Select the correct strategy for processing a non-flush bio. | |
1521 | */ | |
14fe594d | 1522 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1523 | { |
dba14160 | 1524 | struct bio *bio = ci->bio; |
512875bd | 1525 | struct dm_target *ti; |
1c3b13e6 | 1526 | unsigned len; |
1da177e4 | 1527 | |
5ae89a87 | 1528 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1529 | return __send_discard(ci); |
23508a96 | 1530 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1531 | return __send_write_same(ci); |
5ae89a87 | 1532 | |
512875bd JN |
1533 | ti = dm_table_find_target(ci->map, ci->sector); |
1534 | if (!dm_target_is_valid(ti)) | |
1535 | return -EIO; | |
1536 | ||
1c3b13e6 | 1537 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1538 | |
1dd40c3e | 1539 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1540 | |
1c3b13e6 KO |
1541 | ci->sector += len; |
1542 | ci->sector_count -= len; | |
1da177e4 | 1543 | |
1c3b13e6 | 1544 | return 0; |
1da177e4 LT |
1545 | } |
1546 | ||
1547 | /* | |
14fe594d | 1548 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1549 | */ |
83d5e5b0 MP |
1550 | static void __split_and_process_bio(struct mapped_device *md, |
1551 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1552 | { |
1553 | struct clone_info ci; | |
512875bd | 1554 | int error = 0; |
1da177e4 | 1555 | |
83d5e5b0 | 1556 | if (unlikely(!map)) { |
6a8736d1 | 1557 | bio_io_error(bio); |
f0b9a450 MP |
1558 | return; |
1559 | } | |
692d0eb9 | 1560 | |
83d5e5b0 | 1561 | ci.map = map; |
1da177e4 | 1562 | ci.md = md; |
1da177e4 LT |
1563 | ci.io = alloc_io(md); |
1564 | ci.io->error = 0; | |
1565 | atomic_set(&ci.io->io_count, 1); | |
1566 | ci.io->bio = bio; | |
1567 | ci.io->md = md; | |
f88fb981 | 1568 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1569 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1570 | |
3eaf840e | 1571 | start_io_acct(ci.io); |
bd2a49b8 | 1572 | |
b372d360 MS |
1573 | if (bio->bi_rw & REQ_FLUSH) { |
1574 | ci.bio = &ci.md->flush_bio; | |
1575 | ci.sector_count = 0; | |
14fe594d | 1576 | error = __send_empty_flush(&ci); |
b372d360 MS |
1577 | /* dec_pending submits any data associated with flush */ |
1578 | } else { | |
6a8736d1 | 1579 | ci.bio = bio; |
d87f4c14 | 1580 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1581 | while (ci.sector_count && !error) |
14fe594d | 1582 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1583 | } |
1da177e4 LT |
1584 | |
1585 | /* drop the extra reference count */ | |
512875bd | 1586 | dec_pending(ci.io, error); |
1da177e4 LT |
1587 | } |
1588 | /*----------------------------------------------------------------- | |
1589 | * CRUD END | |
1590 | *---------------------------------------------------------------*/ | |
1591 | ||
f6fccb12 MB |
1592 | static int dm_merge_bvec(struct request_queue *q, |
1593 | struct bvec_merge_data *bvm, | |
1594 | struct bio_vec *biovec) | |
1595 | { | |
1596 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1597 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1598 | struct dm_target *ti; |
1599 | sector_t max_sectors; | |
5037108a | 1600 | int max_size = 0; |
f6fccb12 MB |
1601 | |
1602 | if (unlikely(!map)) | |
5037108a | 1603 | goto out; |
f6fccb12 MB |
1604 | |
1605 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1606 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1607 | goto out; |
f6fccb12 MB |
1608 | |
1609 | /* | |
1610 | * Find maximum amount of I/O that won't need splitting | |
1611 | */ | |
56a67df7 | 1612 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1613 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1614 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1615 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1616 | max_size = 0; |
1617 | ||
1618 | /* | |
1619 | * merge_bvec_fn() returns number of bytes | |
1620 | * it can accept at this offset | |
1621 | * max is precomputed maximal io size | |
1622 | */ | |
1623 | if (max_size && ti->type->merge) | |
1624 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1625 | /* |
1626 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1627 | * provided their merge_bvec method (we know this by looking for the |
1628 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1629 | * allow bios with multiple vector entries. So always set max_size | |
1630 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1631 | */ |
1632 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1633 | max_size = 0; |
f6fccb12 | 1634 | |
5037108a | 1635 | out: |
83d5e5b0 | 1636 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1637 | /* |
1638 | * Always allow an entire first page | |
1639 | */ | |
1640 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1641 | max_size = biovec->bv_len; | |
1642 | ||
f6fccb12 MB |
1643 | return max_size; |
1644 | } | |
1645 | ||
1da177e4 LT |
1646 | /* |
1647 | * The request function that just remaps the bio built up by | |
1648 | * dm_merge_bvec. | |
1649 | */ | |
5a7bbad2 | 1650 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1651 | { |
12f03a49 | 1652 | int rw = bio_data_dir(bio); |
1da177e4 | 1653 | struct mapped_device *md = q->queuedata; |
83d5e5b0 MP |
1654 | int srcu_idx; |
1655 | struct dm_table *map; | |
1da177e4 | 1656 | |
83d5e5b0 | 1657 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1658 | |
18c0b223 | 1659 | generic_start_io_acct(rw, bio_sectors(bio), &dm_disk(md)->part0); |
12f03a49 | 1660 | |
6a8736d1 TH |
1661 | /* if we're suspended, we have to queue this io for later */ |
1662 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1663 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1664 | |
6a8736d1 TH |
1665 | if (bio_rw(bio) != READA) |
1666 | queue_io(md, bio); | |
1667 | else | |
54d9a1b4 | 1668 | bio_io_error(bio); |
5a7bbad2 | 1669 | return; |
1da177e4 LT |
1670 | } |
1671 | ||
83d5e5b0 MP |
1672 | __split_and_process_bio(md, map, bio); |
1673 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1674 | return; |
cec47e3d KU |
1675 | } |
1676 | ||
fd2ed4d2 | 1677 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1678 | { |
1679 | return blk_queue_stackable(md->queue); | |
1680 | } | |
1681 | ||
5a7bbad2 | 1682 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1683 | { |
1684 | struct mapped_device *md = q->queuedata; | |
1685 | ||
1686 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1687 | blk_queue_bio(q, bio); |
1688 | else | |
1689 | _dm_request(q, bio); | |
cec47e3d KU |
1690 | } |
1691 | ||
1692 | void dm_dispatch_request(struct request *rq) | |
1693 | { | |
1694 | int r; | |
1695 | ||
1696 | if (blk_queue_io_stat(rq->q)) | |
1697 | rq->cmd_flags |= REQ_IO_STAT; | |
1698 | ||
1699 | rq->start_time = jiffies; | |
1700 | r = blk_insert_cloned_request(rq->q, rq); | |
1701 | if (r) | |
1702 | dm_complete_request(rq, r); | |
1703 | } | |
1704 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1705 | ||
cec47e3d KU |
1706 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1707 | void *data) | |
1708 | { | |
1709 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1710 | struct dm_rq_clone_bio_info *info = |
1711 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1712 | |
1713 | info->orig = bio_orig; | |
1714 | info->tio = tio; | |
1715 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1716 | |
1717 | return 0; | |
1718 | } | |
1719 | ||
1720 | static int setup_clone(struct request *clone, struct request *rq, | |
1721 | struct dm_rq_target_io *tio) | |
1722 | { | |
d0bcb878 | 1723 | int r; |
cec47e3d | 1724 | |
29e4013d TH |
1725 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1726 | dm_rq_bio_constructor, tio); | |
1727 | if (r) | |
1728 | return r; | |
cec47e3d | 1729 | |
29e4013d TH |
1730 | clone->cmd = rq->cmd; |
1731 | clone->cmd_len = rq->cmd_len; | |
1732 | clone->sense = rq->sense; | |
cec47e3d KU |
1733 | clone->end_io = end_clone_request; |
1734 | clone->end_io_data = tio; | |
1735 | ||
1736 | return 0; | |
1737 | } | |
1738 | ||
6facdaff KU |
1739 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1740 | gfp_t gfp_mask) | |
1741 | { | |
1742 | struct request *clone; | |
1743 | struct dm_rq_target_io *tio; | |
1744 | ||
1745 | tio = alloc_rq_tio(md, gfp_mask); | |
1746 | if (!tio) | |
1747 | return NULL; | |
1748 | ||
1749 | tio->md = md; | |
1750 | tio->ti = NULL; | |
1751 | tio->orig = rq; | |
1752 | tio->error = 0; | |
1753 | memset(&tio->info, 0, sizeof(tio->info)); | |
1754 | ||
1755 | clone = &tio->clone; | |
1756 | if (setup_clone(clone, rq, tio)) { | |
1757 | /* -ENOMEM */ | |
1758 | free_rq_tio(tio); | |
1759 | return NULL; | |
1760 | } | |
1761 | ||
1762 | return clone; | |
1763 | } | |
1764 | ||
cec47e3d KU |
1765 | /* |
1766 | * Called with the queue lock held. | |
1767 | */ | |
1768 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1769 | { | |
1770 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1771 | struct request *clone; |
1772 | ||
cec47e3d KU |
1773 | if (unlikely(rq->special)) { |
1774 | DMWARN("Already has something in rq->special."); | |
1775 | return BLKPREP_KILL; | |
1776 | } | |
1777 | ||
6facdaff KU |
1778 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1779 | if (!clone) | |
cec47e3d | 1780 | return BLKPREP_DEFER; |
cec47e3d KU |
1781 | |
1782 | rq->special = clone; | |
1783 | rq->cmd_flags |= REQ_DONTPREP; | |
1784 | ||
1785 | return BLKPREP_OK; | |
1786 | } | |
1787 | ||
9eef87da KU |
1788 | /* |
1789 | * Returns: | |
1790 | * 0 : the request has been processed (not requeued) | |
1791 | * !0 : the request has been requeued | |
1792 | */ | |
1793 | static int map_request(struct dm_target *ti, struct request *clone, | |
1794 | struct mapped_device *md) | |
cec47e3d | 1795 | { |
9eef87da | 1796 | int r, requeued = 0; |
cec47e3d KU |
1797 | struct dm_rq_target_io *tio = clone->end_io_data; |
1798 | ||
cec47e3d KU |
1799 | tio->ti = ti; |
1800 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1801 | switch (r) { | |
1802 | case DM_MAPIO_SUBMITTED: | |
1803 | /* The target has taken the I/O to submit by itself later */ | |
1804 | break; | |
1805 | case DM_MAPIO_REMAPPED: | |
1806 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1807 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1808 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1809 | dm_dispatch_request(clone); |
1810 | break; | |
1811 | case DM_MAPIO_REQUEUE: | |
1812 | /* The target wants to requeue the I/O */ | |
1813 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1814 | requeued = 1; |
cec47e3d KU |
1815 | break; |
1816 | default: | |
1817 | if (r > 0) { | |
1818 | DMWARN("unimplemented target map return value: %d", r); | |
1819 | BUG(); | |
1820 | } | |
1821 | ||
1822 | /* The target wants to complete the I/O */ | |
1823 | dm_kill_unmapped_request(clone, r); | |
1824 | break; | |
1825 | } | |
9eef87da KU |
1826 | |
1827 | return requeued; | |
cec47e3d KU |
1828 | } |
1829 | ||
ba1cbad9 MS |
1830 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1831 | { | |
1832 | struct request *clone; | |
1833 | ||
1834 | blk_start_request(orig); | |
1835 | clone = orig->special; | |
1836 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1837 | ||
1838 | /* | |
1839 | * Hold the md reference here for the in-flight I/O. | |
1840 | * We can't rely on the reference count by device opener, | |
1841 | * because the device may be closed during the request completion | |
1842 | * when all bios are completed. | |
1843 | * See the comment in rq_completed() too. | |
1844 | */ | |
1845 | dm_get(md); | |
1846 | ||
1847 | return clone; | |
1848 | } | |
1849 | ||
cec47e3d KU |
1850 | /* |
1851 | * q->request_fn for request-based dm. | |
1852 | * Called with the queue lock held. | |
1853 | */ | |
1854 | static void dm_request_fn(struct request_queue *q) | |
1855 | { | |
1856 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1857 | int srcu_idx; |
1858 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1859 | struct dm_target *ti; |
b4324fee | 1860 | struct request *rq, *clone; |
29e4013d | 1861 | sector_t pos; |
cec47e3d KU |
1862 | |
1863 | /* | |
b4324fee KU |
1864 | * For suspend, check blk_queue_stopped() and increment |
1865 | * ->pending within a single queue_lock not to increment the | |
1866 | * number of in-flight I/Os after the queue is stopped in | |
1867 | * dm_suspend(). | |
cec47e3d | 1868 | */ |
7eaceacc | 1869 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1870 | rq = blk_peek_request(q); |
1871 | if (!rq) | |
7eaceacc | 1872 | goto delay_and_out; |
cec47e3d | 1873 | |
29e4013d TH |
1874 | /* always use block 0 to find the target for flushes for now */ |
1875 | pos = 0; | |
1876 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1877 | pos = blk_rq_pos(rq); | |
1878 | ||
1879 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1880 | if (!dm_target_is_valid(ti)) { |
1881 | /* | |
1882 | * Must perform setup, that dm_done() requires, | |
1883 | * before calling dm_kill_unmapped_request | |
1884 | */ | |
1885 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1886 | clone = dm_start_request(md, rq); | |
1887 | dm_kill_unmapped_request(clone, -EIO); | |
1888 | continue; | |
1889 | } | |
d0bcb878 | 1890 | |
cec47e3d | 1891 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1892 | goto delay_and_out; |
cec47e3d | 1893 | |
ba1cbad9 | 1894 | clone = dm_start_request(md, rq); |
b4324fee | 1895 | |
cec47e3d | 1896 | spin_unlock(q->queue_lock); |
9eef87da KU |
1897 | if (map_request(ti, clone, md)) |
1898 | goto requeued; | |
1899 | ||
052189a2 KU |
1900 | BUG_ON(!irqs_disabled()); |
1901 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1902 | } |
1903 | ||
1904 | goto out; | |
1905 | ||
9eef87da | 1906 | requeued: |
052189a2 KU |
1907 | BUG_ON(!irqs_disabled()); |
1908 | spin_lock(q->queue_lock); | |
9eef87da | 1909 | |
7eaceacc JA |
1910 | delay_and_out: |
1911 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1912 | out: |
83d5e5b0 | 1913 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1914 | } |
1915 | ||
1916 | int dm_underlying_device_busy(struct request_queue *q) | |
1917 | { | |
1918 | return blk_lld_busy(q); | |
1919 | } | |
1920 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1921 | ||
1922 | static int dm_lld_busy(struct request_queue *q) | |
1923 | { | |
1924 | int r; | |
1925 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1926 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1927 | |
1928 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1929 | r = 1; | |
1930 | else | |
1931 | r = dm_table_any_busy_target(map); | |
1932 | ||
83d5e5b0 | 1933 | dm_put_live_table_fast(md); |
cec47e3d KU |
1934 | |
1935 | return r; | |
1936 | } | |
1937 | ||
1da177e4 LT |
1938 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1939 | { | |
8a57dfc6 CS |
1940 | int r = bdi_bits; |
1941 | struct mapped_device *md = congested_data; | |
1942 | struct dm_table *map; | |
1da177e4 | 1943 | |
1eb787ec | 1944 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1945 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1946 | if (map) { |
cec47e3d KU |
1947 | /* |
1948 | * Request-based dm cares about only own queue for | |
1949 | * the query about congestion status of request_queue | |
1950 | */ | |
1951 | if (dm_request_based(md)) | |
1952 | r = md->queue->backing_dev_info.state & | |
1953 | bdi_bits; | |
1954 | else | |
1955 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1956 | } |
83d5e5b0 | 1957 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1958 | } |
1959 | ||
1da177e4 LT |
1960 | return r; |
1961 | } | |
1962 | ||
1963 | /*----------------------------------------------------------------- | |
1964 | * An IDR is used to keep track of allocated minor numbers. | |
1965 | *---------------------------------------------------------------*/ | |
2b06cfff | 1966 | static void free_minor(int minor) |
1da177e4 | 1967 | { |
f32c10b0 | 1968 | spin_lock(&_minor_lock); |
1da177e4 | 1969 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1970 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * See if the device with a specific minor # is free. | |
1975 | */ | |
cf13ab8e | 1976 | static int specific_minor(int minor) |
1da177e4 | 1977 | { |
c9d76be6 | 1978 | int r; |
1da177e4 LT |
1979 | |
1980 | if (minor >= (1 << MINORBITS)) | |
1981 | return -EINVAL; | |
1982 | ||
c9d76be6 | 1983 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1984 | spin_lock(&_minor_lock); |
1da177e4 | 1985 | |
c9d76be6 | 1986 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1987 | |
f32c10b0 | 1988 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1989 | idr_preload_end(); |
1990 | if (r < 0) | |
1991 | return r == -ENOSPC ? -EBUSY : r; | |
1992 | return 0; | |
1da177e4 LT |
1993 | } |
1994 | ||
cf13ab8e | 1995 | static int next_free_minor(int *minor) |
1da177e4 | 1996 | { |
c9d76be6 | 1997 | int r; |
62f75c2f | 1998 | |
c9d76be6 | 1999 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2000 | spin_lock(&_minor_lock); |
1da177e4 | 2001 | |
c9d76be6 | 2002 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2003 | |
f32c10b0 | 2004 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2005 | idr_preload_end(); |
2006 | if (r < 0) | |
2007 | return r; | |
2008 | *minor = r; | |
2009 | return 0; | |
1da177e4 LT |
2010 | } |
2011 | ||
83d5cde4 | 2012 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2013 | |
53d5914f MP |
2014 | static void dm_wq_work(struct work_struct *work); |
2015 | ||
4a0b4ddf MS |
2016 | static void dm_init_md_queue(struct mapped_device *md) |
2017 | { | |
2018 | /* | |
2019 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
2020 | * devices. The type of this dm device has not been decided yet. | |
2021 | * The type is decided at the first table loading time. | |
2022 | * To prevent problematic device stacking, clear the queue flag | |
2023 | * for request stacking support until then. | |
2024 | * | |
2025 | * This queue is new, so no concurrency on the queue_flags. | |
2026 | */ | |
2027 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
2028 | ||
2029 | md->queue->queuedata = md; | |
2030 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2031 | md->queue->backing_dev_info.congested_data = md; | |
2032 | blk_queue_make_request(md->queue, dm_request); | |
2033 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
2034 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
2035 | } | |
2036 | ||
1da177e4 LT |
2037 | /* |
2038 | * Allocate and initialise a blank device with a given minor. | |
2039 | */ | |
2b06cfff | 2040 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2041 | { |
2042 | int r; | |
cf13ab8e | 2043 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2044 | void *old_md; |
1da177e4 LT |
2045 | |
2046 | if (!md) { | |
2047 | DMWARN("unable to allocate device, out of memory."); | |
2048 | return NULL; | |
2049 | } | |
2050 | ||
10da4f79 | 2051 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2052 | goto bad_module_get; |
10da4f79 | 2053 | |
1da177e4 | 2054 | /* get a minor number for the dev */ |
2b06cfff | 2055 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2056 | r = next_free_minor(&minor); |
2b06cfff | 2057 | else |
cf13ab8e | 2058 | r = specific_minor(minor); |
1da177e4 | 2059 | if (r < 0) |
6ed7ade8 | 2060 | goto bad_minor; |
1da177e4 | 2061 | |
83d5e5b0 MP |
2062 | r = init_srcu_struct(&md->io_barrier); |
2063 | if (r < 0) | |
2064 | goto bad_io_barrier; | |
2065 | ||
a5664dad | 2066 | md->type = DM_TYPE_NONE; |
e61290a4 | 2067 | mutex_init(&md->suspend_lock); |
a5664dad | 2068 | mutex_init(&md->type_lock); |
86f1152b | 2069 | mutex_init(&md->table_devices_lock); |
022c2611 | 2070 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2071 | atomic_set(&md->holders, 1); |
5c6bd75d | 2072 | atomic_set(&md->open_count, 0); |
1da177e4 | 2073 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2074 | atomic_set(&md->uevent_seq, 0); |
2075 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2076 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2077 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2078 | |
4a0b4ddf | 2079 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2080 | if (!md->queue) |
6ed7ade8 | 2081 | goto bad_queue; |
1da177e4 | 2082 | |
4a0b4ddf | 2083 | dm_init_md_queue(md); |
9faf400f | 2084 | |
1da177e4 LT |
2085 | md->disk = alloc_disk(1); |
2086 | if (!md->disk) | |
6ed7ade8 | 2087 | goto bad_disk; |
1da177e4 | 2088 | |
316d315b NK |
2089 | atomic_set(&md->pending[0], 0); |
2090 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2091 | init_waitqueue_head(&md->wait); |
53d5914f | 2092 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2093 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2094 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 2095 | |
1da177e4 LT |
2096 | md->disk->major = _major; |
2097 | md->disk->first_minor = minor; | |
2098 | md->disk->fops = &dm_blk_dops; | |
2099 | md->disk->queue = md->queue; | |
2100 | md->disk->private_data = md; | |
2101 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2102 | add_disk(md->disk); | |
7e51f257 | 2103 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2104 | |
670368a8 | 2105 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2106 | if (!md->wq) |
2107 | goto bad_thread; | |
2108 | ||
32a926da MP |
2109 | md->bdev = bdget_disk(md->disk, 0); |
2110 | if (!md->bdev) | |
2111 | goto bad_bdev; | |
2112 | ||
6a8736d1 TH |
2113 | bio_init(&md->flush_bio); |
2114 | md->flush_bio.bi_bdev = md->bdev; | |
2115 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2116 | ||
fd2ed4d2 MP |
2117 | dm_stats_init(&md->stats); |
2118 | ||
ba61fdd1 | 2119 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2120 | spin_lock(&_minor_lock); |
ba61fdd1 | 2121 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2122 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2123 | |
2124 | BUG_ON(old_md != MINOR_ALLOCED); | |
2125 | ||
1da177e4 LT |
2126 | return md; |
2127 | ||
32a926da MP |
2128 | bad_bdev: |
2129 | destroy_workqueue(md->wq); | |
304f3f6a | 2130 | bad_thread: |
03022c54 | 2131 | del_gendisk(md->disk); |
304f3f6a | 2132 | put_disk(md->disk); |
6ed7ade8 | 2133 | bad_disk: |
1312f40e | 2134 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2135 | bad_queue: |
83d5e5b0 MP |
2136 | cleanup_srcu_struct(&md->io_barrier); |
2137 | bad_io_barrier: | |
1da177e4 | 2138 | free_minor(minor); |
6ed7ade8 | 2139 | bad_minor: |
10da4f79 | 2140 | module_put(THIS_MODULE); |
6ed7ade8 | 2141 | bad_module_get: |
1da177e4 LT |
2142 | kfree(md); |
2143 | return NULL; | |
2144 | } | |
2145 | ||
ae9da83f JN |
2146 | static void unlock_fs(struct mapped_device *md); |
2147 | ||
1da177e4 LT |
2148 | static void free_dev(struct mapped_device *md) |
2149 | { | |
f331c029 | 2150 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2151 | |
32a926da MP |
2152 | unlock_fs(md); |
2153 | bdput(md->bdev); | |
304f3f6a | 2154 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
2155 | if (md->io_pool) |
2156 | mempool_destroy(md->io_pool); | |
2157 | if (md->bs) | |
2158 | bioset_free(md->bs); | |
9c47008d | 2159 | blk_integrity_unregister(md->disk); |
1da177e4 | 2160 | del_gendisk(md->disk); |
83d5e5b0 | 2161 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2162 | free_table_devices(&md->table_devices); |
63d94e48 | 2163 | free_minor(minor); |
fba9f90e JM |
2164 | |
2165 | spin_lock(&_minor_lock); | |
2166 | md->disk->private_data = NULL; | |
2167 | spin_unlock(&_minor_lock); | |
2168 | ||
1da177e4 | 2169 | put_disk(md->disk); |
1312f40e | 2170 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2171 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2172 | module_put(THIS_MODULE); |
1da177e4 LT |
2173 | kfree(md); |
2174 | } | |
2175 | ||
e6ee8c0b KU |
2176 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2177 | { | |
c0820cf5 | 2178 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2179 | |
5f015204 | 2180 | if (md->io_pool && md->bs) { |
16245bdc JN |
2181 | /* The md already has necessary mempools. */ |
2182 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2183 | /* | |
2184 | * Reload bioset because front_pad may have changed | |
2185 | * because a different table was loaded. | |
2186 | */ | |
2187 | bioset_free(md->bs); | |
2188 | md->bs = p->bs; | |
2189 | p->bs = NULL; | |
2190 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2191 | /* |
2192 | * There's no need to reload with request-based dm | |
2193 | * because the size of front_pad doesn't change. | |
2194 | * Note for future: If you are to reload bioset, | |
2195 | * prep-ed requests in the queue may refer | |
2196 | * to bio from the old bioset, so you must walk | |
2197 | * through the queue to unprep. | |
2198 | */ | |
2199 | } | |
e6ee8c0b | 2200 | goto out; |
c0820cf5 | 2201 | } |
e6ee8c0b | 2202 | |
5f015204 | 2203 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2204 | |
2205 | md->io_pool = p->io_pool; | |
2206 | p->io_pool = NULL; | |
e6ee8c0b KU |
2207 | md->bs = p->bs; |
2208 | p->bs = NULL; | |
2209 | ||
2210 | out: | |
2211 | /* mempool bind completed, now no need any mempools in the table */ | |
2212 | dm_table_free_md_mempools(t); | |
2213 | } | |
2214 | ||
1da177e4 LT |
2215 | /* |
2216 | * Bind a table to the device. | |
2217 | */ | |
2218 | static void event_callback(void *context) | |
2219 | { | |
7a8c3d3b MA |
2220 | unsigned long flags; |
2221 | LIST_HEAD(uevents); | |
1da177e4 LT |
2222 | struct mapped_device *md = (struct mapped_device *) context; |
2223 | ||
7a8c3d3b MA |
2224 | spin_lock_irqsave(&md->uevent_lock, flags); |
2225 | list_splice_init(&md->uevent_list, &uevents); | |
2226 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2227 | ||
ed9e1982 | 2228 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2229 | |
1da177e4 LT |
2230 | atomic_inc(&md->event_nr); |
2231 | wake_up(&md->eventq); | |
2232 | } | |
2233 | ||
c217649b MS |
2234 | /* |
2235 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2236 | */ | |
4e90188b | 2237 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2238 | { |
4e90188b | 2239 | set_capacity(md->disk, size); |
1da177e4 | 2240 | |
db8fef4f | 2241 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2242 | } |
2243 | ||
d5b9dd04 MP |
2244 | /* |
2245 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2246 | * | |
2247 | * If this function returns 0, then the device is either a non-dm | |
2248 | * device without a merge_bvec_fn, or it is a dm device that is | |
2249 | * able to split any bios it receives that are too big. | |
2250 | */ | |
2251 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2252 | { | |
2253 | struct mapped_device *dev_md; | |
2254 | ||
2255 | if (!q->merge_bvec_fn) | |
2256 | return 0; | |
2257 | ||
2258 | if (q->make_request_fn == dm_request) { | |
2259 | dev_md = q->queuedata; | |
2260 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2261 | return 0; | |
2262 | } | |
2263 | ||
2264 | return 1; | |
2265 | } | |
2266 | ||
2267 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2268 | struct dm_dev *dev, sector_t start, | |
2269 | sector_t len, void *data) | |
2270 | { | |
2271 | struct block_device *bdev = dev->bdev; | |
2272 | struct request_queue *q = bdev_get_queue(bdev); | |
2273 | ||
2274 | return dm_queue_merge_is_compulsory(q); | |
2275 | } | |
2276 | ||
2277 | /* | |
2278 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2279 | * on the properties of the underlying devices. | |
2280 | */ | |
2281 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2282 | { | |
2283 | unsigned i = 0; | |
2284 | struct dm_target *ti; | |
2285 | ||
2286 | while (i < dm_table_get_num_targets(table)) { | |
2287 | ti = dm_table_get_target(table, i++); | |
2288 | ||
2289 | if (ti->type->iterate_devices && | |
2290 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2291 | return 0; | |
2292 | } | |
2293 | ||
2294 | return 1; | |
2295 | } | |
2296 | ||
042d2a9b AK |
2297 | /* |
2298 | * Returns old map, which caller must destroy. | |
2299 | */ | |
2300 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2301 | struct queue_limits *limits) | |
1da177e4 | 2302 | { |
042d2a9b | 2303 | struct dm_table *old_map; |
165125e1 | 2304 | struct request_queue *q = md->queue; |
1da177e4 | 2305 | sector_t size; |
d5b9dd04 | 2306 | int merge_is_optional; |
1da177e4 LT |
2307 | |
2308 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2309 | |
2310 | /* | |
2311 | * Wipe any geometry if the size of the table changed. | |
2312 | */ | |
fd2ed4d2 | 2313 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2314 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2315 | ||
32a926da | 2316 | __set_size(md, size); |
d5816876 | 2317 | |
2ca3310e AK |
2318 | dm_table_event_callback(t, event_callback, md); |
2319 | ||
e6ee8c0b KU |
2320 | /* |
2321 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2322 | * for request-based during suspension. So stop it to prevent | |
2323 | * I/O mapping before resume. | |
2324 | * This must be done before setting the queue restrictions, | |
2325 | * because request-based dm may be run just after the setting. | |
2326 | */ | |
2327 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2328 | stop_queue(q); | |
2329 | ||
2330 | __bind_mempools(md, t); | |
2331 | ||
d5b9dd04 MP |
2332 | merge_is_optional = dm_table_merge_is_optional(t); |
2333 | ||
a12f5d48 | 2334 | old_map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
83d5e5b0 | 2335 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2336 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2337 | ||
754c5fc7 | 2338 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2339 | if (merge_is_optional) |
2340 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2341 | else | |
2342 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2343 | if (old_map) |
2344 | dm_sync_table(md); | |
1da177e4 | 2345 | |
042d2a9b | 2346 | return old_map; |
1da177e4 LT |
2347 | } |
2348 | ||
a7940155 AK |
2349 | /* |
2350 | * Returns unbound table for the caller to free. | |
2351 | */ | |
2352 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2353 | { |
a12f5d48 | 2354 | struct dm_table *map = rcu_dereference_protected(md->map, 1); |
1da177e4 LT |
2355 | |
2356 | if (!map) | |
a7940155 | 2357 | return NULL; |
1da177e4 LT |
2358 | |
2359 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2360 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2361 | dm_sync_table(md); |
a7940155 AK |
2362 | |
2363 | return map; | |
1da177e4 LT |
2364 | } |
2365 | ||
2366 | /* | |
2367 | * Constructor for a new device. | |
2368 | */ | |
2b06cfff | 2369 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2370 | { |
2371 | struct mapped_device *md; | |
2372 | ||
2b06cfff | 2373 | md = alloc_dev(minor); |
1da177e4 LT |
2374 | if (!md) |
2375 | return -ENXIO; | |
2376 | ||
784aae73 MB |
2377 | dm_sysfs_init(md); |
2378 | ||
1da177e4 LT |
2379 | *result = md; |
2380 | return 0; | |
2381 | } | |
2382 | ||
a5664dad MS |
2383 | /* |
2384 | * Functions to manage md->type. | |
2385 | * All are required to hold md->type_lock. | |
2386 | */ | |
2387 | void dm_lock_md_type(struct mapped_device *md) | |
2388 | { | |
2389 | mutex_lock(&md->type_lock); | |
2390 | } | |
2391 | ||
2392 | void dm_unlock_md_type(struct mapped_device *md) | |
2393 | { | |
2394 | mutex_unlock(&md->type_lock); | |
2395 | } | |
2396 | ||
2397 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2398 | { | |
00c4fc3b | 2399 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2400 | md->type = type; |
2401 | } | |
2402 | ||
2403 | unsigned dm_get_md_type(struct mapped_device *md) | |
2404 | { | |
00c4fc3b | 2405 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2406 | return md->type; |
2407 | } | |
2408 | ||
36a0456f AK |
2409 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2410 | { | |
2411 | return md->immutable_target_type; | |
2412 | } | |
2413 | ||
f84cb8a4 MS |
2414 | /* |
2415 | * The queue_limits are only valid as long as you have a reference | |
2416 | * count on 'md'. | |
2417 | */ | |
2418 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2419 | { | |
2420 | BUG_ON(!atomic_read(&md->holders)); | |
2421 | return &md->queue->limits; | |
2422 | } | |
2423 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2424 | ||
4a0b4ddf MS |
2425 | /* |
2426 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2427 | */ | |
2428 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2429 | { | |
2430 | struct request_queue *q = NULL; | |
2431 | ||
2432 | if (md->queue->elevator) | |
2433 | return 1; | |
2434 | ||
2435 | /* Fully initialize the queue */ | |
2436 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2437 | if (!q) | |
2438 | return 0; | |
2439 | ||
2440 | md->queue = q; | |
4a0b4ddf MS |
2441 | dm_init_md_queue(md); |
2442 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2443 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2444 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2445 | |
2446 | elv_register_queue(md->queue); | |
2447 | ||
2448 | return 1; | |
2449 | } | |
2450 | ||
2451 | /* | |
2452 | * Setup the DM device's queue based on md's type | |
2453 | */ | |
2454 | int dm_setup_md_queue(struct mapped_device *md) | |
2455 | { | |
2456 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2457 | !dm_init_request_based_queue(md)) { | |
2458 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2459 | return -EINVAL; | |
2460 | } | |
2461 | ||
2462 | return 0; | |
2463 | } | |
2464 | ||
637842cf | 2465 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2466 | { |
2467 | struct mapped_device *md; | |
1da177e4 LT |
2468 | unsigned minor = MINOR(dev); |
2469 | ||
2470 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2471 | return NULL; | |
2472 | ||
f32c10b0 | 2473 | spin_lock(&_minor_lock); |
1da177e4 LT |
2474 | |
2475 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2476 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2477 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2478 | dm_deleting_md(md) || |
17b2f66f | 2479 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2480 | md = NULL; |
fba9f90e JM |
2481 | goto out; |
2482 | } | |
1da177e4 | 2483 | |
fba9f90e | 2484 | out: |
f32c10b0 | 2485 | spin_unlock(&_minor_lock); |
1da177e4 | 2486 | |
637842cf DT |
2487 | return md; |
2488 | } | |
2489 | ||
d229a958 DT |
2490 | struct mapped_device *dm_get_md(dev_t dev) |
2491 | { | |
2492 | struct mapped_device *md = dm_find_md(dev); | |
2493 | ||
2494 | if (md) | |
2495 | dm_get(md); | |
2496 | ||
2497 | return md; | |
2498 | } | |
3cf2e4ba | 2499 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2500 | |
9ade92a9 | 2501 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2502 | { |
9ade92a9 | 2503 | return md->interface_ptr; |
1da177e4 LT |
2504 | } |
2505 | ||
2506 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2507 | { | |
2508 | md->interface_ptr = ptr; | |
2509 | } | |
2510 | ||
2511 | void dm_get(struct mapped_device *md) | |
2512 | { | |
2513 | atomic_inc(&md->holders); | |
3f77316d | 2514 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2515 | } |
2516 | ||
72d94861 AK |
2517 | const char *dm_device_name(struct mapped_device *md) |
2518 | { | |
2519 | return md->name; | |
2520 | } | |
2521 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2522 | ||
3f77316d | 2523 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2524 | { |
1134e5ae | 2525 | struct dm_table *map; |
83d5e5b0 | 2526 | int srcu_idx; |
1da177e4 | 2527 | |
3f77316d | 2528 | might_sleep(); |
fba9f90e | 2529 | |
3f77316d | 2530 | spin_lock(&_minor_lock); |
83d5e5b0 | 2531 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2532 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2533 | set_bit(DMF_FREEING, &md->flags); | |
2534 | spin_unlock(&_minor_lock); | |
2535 | ||
2536 | if (!dm_suspended_md(md)) { | |
2537 | dm_table_presuspend_targets(map); | |
2538 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2539 | } |
3f77316d | 2540 | |
83d5e5b0 MP |
2541 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2542 | dm_put_live_table(md, srcu_idx); | |
2543 | ||
3f77316d KU |
2544 | /* |
2545 | * Rare, but there may be I/O requests still going to complete, | |
2546 | * for example. Wait for all references to disappear. | |
2547 | * No one should increment the reference count of the mapped_device, | |
2548 | * after the mapped_device state becomes DMF_FREEING. | |
2549 | */ | |
2550 | if (wait) | |
2551 | while (atomic_read(&md->holders)) | |
2552 | msleep(1); | |
2553 | else if (atomic_read(&md->holders)) | |
2554 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2555 | dm_device_name(md), atomic_read(&md->holders)); | |
2556 | ||
2557 | dm_sysfs_exit(md); | |
3f77316d KU |
2558 | dm_table_destroy(__unbind(md)); |
2559 | free_dev(md); | |
2560 | } | |
2561 | ||
2562 | void dm_destroy(struct mapped_device *md) | |
2563 | { | |
2564 | __dm_destroy(md, true); | |
2565 | } | |
2566 | ||
2567 | void dm_destroy_immediate(struct mapped_device *md) | |
2568 | { | |
2569 | __dm_destroy(md, false); | |
2570 | } | |
2571 | ||
2572 | void dm_put(struct mapped_device *md) | |
2573 | { | |
2574 | atomic_dec(&md->holders); | |
1da177e4 | 2575 | } |
79eb885c | 2576 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2577 | |
401600df | 2578 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2579 | { |
2580 | int r = 0; | |
b44ebeb0 MP |
2581 | DECLARE_WAITQUEUE(wait, current); |
2582 | ||
b44ebeb0 | 2583 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2584 | |
2585 | while (1) { | |
401600df | 2586 | set_current_state(interruptible); |
46125c1c | 2587 | |
b4324fee | 2588 | if (!md_in_flight(md)) |
46125c1c MB |
2589 | break; |
2590 | ||
401600df MP |
2591 | if (interruptible == TASK_INTERRUPTIBLE && |
2592 | signal_pending(current)) { | |
46125c1c MB |
2593 | r = -EINTR; |
2594 | break; | |
2595 | } | |
2596 | ||
2597 | io_schedule(); | |
2598 | } | |
2599 | set_current_state(TASK_RUNNING); | |
2600 | ||
b44ebeb0 MP |
2601 | remove_wait_queue(&md->wait, &wait); |
2602 | ||
46125c1c MB |
2603 | return r; |
2604 | } | |
2605 | ||
1da177e4 LT |
2606 | /* |
2607 | * Process the deferred bios | |
2608 | */ | |
ef208587 | 2609 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2610 | { |
ef208587 MP |
2611 | struct mapped_device *md = container_of(work, struct mapped_device, |
2612 | work); | |
6d6f10df | 2613 | struct bio *c; |
83d5e5b0 MP |
2614 | int srcu_idx; |
2615 | struct dm_table *map; | |
1da177e4 | 2616 | |
83d5e5b0 | 2617 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2618 | |
3b00b203 | 2619 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2620 | spin_lock_irq(&md->deferred_lock); |
2621 | c = bio_list_pop(&md->deferred); | |
2622 | spin_unlock_irq(&md->deferred_lock); | |
2623 | ||
6a8736d1 | 2624 | if (!c) |
df12ee99 | 2625 | break; |
022c2611 | 2626 | |
e6ee8c0b KU |
2627 | if (dm_request_based(md)) |
2628 | generic_make_request(c); | |
6a8736d1 | 2629 | else |
83d5e5b0 | 2630 | __split_and_process_bio(md, map, c); |
022c2611 | 2631 | } |
73d410c0 | 2632 | |
83d5e5b0 | 2633 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2634 | } |
2635 | ||
9a1fb464 | 2636 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2637 | { |
3b00b203 | 2638 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2639 | smp_mb__after_atomic(); |
53d5914f | 2640 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2641 | } |
2642 | ||
1da177e4 | 2643 | /* |
042d2a9b | 2644 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2645 | */ |
042d2a9b | 2646 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2647 | { |
87eb5b21 | 2648 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2649 | struct queue_limits limits; |
042d2a9b | 2650 | int r; |
1da177e4 | 2651 | |
e61290a4 | 2652 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2653 | |
2654 | /* device must be suspended */ | |
4f186f8b | 2655 | if (!dm_suspended_md(md)) |
93c534ae | 2656 | goto out; |
1da177e4 | 2657 | |
3ae70656 MS |
2658 | /* |
2659 | * If the new table has no data devices, retain the existing limits. | |
2660 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2661 | * then new I/O is queued based on these limits, and then some paths | |
2662 | * reappear. | |
2663 | */ | |
2664 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2665 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2666 | if (live_map) |
2667 | limits = md->queue->limits; | |
83d5e5b0 | 2668 | dm_put_live_table_fast(md); |
3ae70656 MS |
2669 | } |
2670 | ||
87eb5b21 MC |
2671 | if (!live_map) { |
2672 | r = dm_calculate_queue_limits(table, &limits); | |
2673 | if (r) { | |
2674 | map = ERR_PTR(r); | |
2675 | goto out; | |
2676 | } | |
042d2a9b | 2677 | } |
754c5fc7 | 2678 | |
042d2a9b | 2679 | map = __bind(md, table, &limits); |
1da177e4 | 2680 | |
93c534ae | 2681 | out: |
e61290a4 | 2682 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2683 | return map; |
1da177e4 LT |
2684 | } |
2685 | ||
2686 | /* | |
2687 | * Functions to lock and unlock any filesystem running on the | |
2688 | * device. | |
2689 | */ | |
2ca3310e | 2690 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2691 | { |
e39e2e95 | 2692 | int r; |
1da177e4 LT |
2693 | |
2694 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2695 | |
db8fef4f | 2696 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2697 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2698 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2699 | md->frozen_sb = NULL; |
2700 | return r; | |
dfbe03f6 AK |
2701 | } |
2702 | ||
aa8d7c2f AK |
2703 | set_bit(DMF_FROZEN, &md->flags); |
2704 | ||
1da177e4 LT |
2705 | return 0; |
2706 | } | |
2707 | ||
2ca3310e | 2708 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2709 | { |
aa8d7c2f AK |
2710 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2711 | return; | |
2712 | ||
db8fef4f | 2713 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2714 | md->frozen_sb = NULL; |
aa8d7c2f | 2715 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2716 | } |
2717 | ||
2718 | /* | |
ffcc3936 MS |
2719 | * If __dm_suspend returns 0, the device is completely quiescent |
2720 | * now. There is no request-processing activity. All new requests | |
2721 | * are being added to md->deferred list. | |
cec47e3d | 2722 | * |
ffcc3936 | 2723 | * Caller must hold md->suspend_lock |
cec47e3d | 2724 | */ |
ffcc3936 MS |
2725 | static int __dm_suspend(struct mapped_device *md, struct dm_table *map, |
2726 | unsigned suspend_flags, int interruptible) | |
1da177e4 | 2727 | { |
ffcc3936 MS |
2728 | bool do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG; |
2729 | bool noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG; | |
2730 | int r; | |
1da177e4 | 2731 | |
2e93ccc1 KU |
2732 | /* |
2733 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2734 | * This flag is cleared before dm_suspend returns. | |
2735 | */ | |
2736 | if (noflush) | |
2737 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2738 | ||
d67ee213 MS |
2739 | /* |
2740 | * This gets reverted if there's an error later and the targets | |
2741 | * provide the .presuspend_undo hook. | |
2742 | */ | |
cf222b37 AK |
2743 | dm_table_presuspend_targets(map); |
2744 | ||
32a926da | 2745 | /* |
9f518b27 KU |
2746 | * Flush I/O to the device. |
2747 | * Any I/O submitted after lock_fs() may not be flushed. | |
2748 | * noflush takes precedence over do_lockfs. | |
2749 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2750 | */ |
2751 | if (!noflush && do_lockfs) { | |
2752 | r = lock_fs(md); | |
d67ee213 MS |
2753 | if (r) { |
2754 | dm_table_presuspend_undo_targets(map); | |
ffcc3936 | 2755 | return r; |
d67ee213 | 2756 | } |
aa8d7c2f | 2757 | } |
1da177e4 LT |
2758 | |
2759 | /* | |
3b00b203 MP |
2760 | * Here we must make sure that no processes are submitting requests |
2761 | * to target drivers i.e. no one may be executing | |
2762 | * __split_and_process_bio. This is called from dm_request and | |
2763 | * dm_wq_work. | |
2764 | * | |
2765 | * To get all processes out of __split_and_process_bio in dm_request, | |
2766 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2767 | * __split_and_process_bio from dm_request and quiesce the thread |
2768 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2769 | * flush_workqueue(md->wq). | |
1da177e4 | 2770 | */ |
1eb787ec | 2771 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2772 | if (map) |
2773 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2774 | |
d0bcb878 | 2775 | /* |
29e4013d TH |
2776 | * Stop md->queue before flushing md->wq in case request-based |
2777 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2778 | */ |
cec47e3d | 2779 | if (dm_request_based(md)) |
9f518b27 | 2780 | stop_queue(md->queue); |
cec47e3d | 2781 | |
d0bcb878 KU |
2782 | flush_workqueue(md->wq); |
2783 | ||
1da177e4 | 2784 | /* |
3b00b203 MP |
2785 | * At this point no more requests are entering target request routines. |
2786 | * We call dm_wait_for_completion to wait for all existing requests | |
2787 | * to finish. | |
1da177e4 | 2788 | */ |
ffcc3936 | 2789 | r = dm_wait_for_completion(md, interruptible); |
1da177e4 | 2790 | |
6d6f10df | 2791 | if (noflush) |
022c2611 | 2792 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2793 | if (map) |
2794 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2795 | |
1da177e4 | 2796 | /* were we interrupted ? */ |
46125c1c | 2797 | if (r < 0) { |
9a1fb464 | 2798 | dm_queue_flush(md); |
73d410c0 | 2799 | |
cec47e3d | 2800 | if (dm_request_based(md)) |
9f518b27 | 2801 | start_queue(md->queue); |
cec47e3d | 2802 | |
2ca3310e | 2803 | unlock_fs(md); |
d67ee213 | 2804 | dm_table_presuspend_undo_targets(map); |
ffcc3936 | 2805 | /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2806 | } |
1da177e4 | 2807 | |
ffcc3936 MS |
2808 | return r; |
2809 | } | |
2810 | ||
2811 | /* | |
2812 | * We need to be able to change a mapping table under a mounted | |
2813 | * filesystem. For example we might want to move some data in | |
2814 | * the background. Before the table can be swapped with | |
2815 | * dm_bind_table, dm_suspend must be called to flush any in | |
2816 | * flight bios and ensure that any further io gets deferred. | |
2817 | */ | |
2818 | /* | |
2819 | * Suspend mechanism in request-based dm. | |
2820 | * | |
2821 | * 1. Flush all I/Os by lock_fs() if needed. | |
2822 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2823 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
2824 | * | |
2825 | * To abort suspend, start the request_queue. | |
2826 | */ | |
2827 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) | |
2828 | { | |
2829 | struct dm_table *map = NULL; | |
2830 | int r = 0; | |
2831 | ||
2832 | retry: | |
2833 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2834 | ||
2835 | if (dm_suspended_md(md)) { | |
2836 | r = -EINVAL; | |
2837 | goto out_unlock; | |
2838 | } | |
2839 | ||
2840 | if (dm_suspended_internally_md(md)) { | |
2841 | /* already internally suspended, wait for internal resume */ | |
2842 | mutex_unlock(&md->suspend_lock); | |
2843 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2844 | if (r) | |
2845 | return r; | |
2846 | goto retry; | |
2847 | } | |
2848 | ||
a12f5d48 | 2849 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2850 | |
2851 | r = __dm_suspend(md, map, suspend_flags, TASK_INTERRUPTIBLE); | |
2852 | if (r) | |
2853 | goto out_unlock; | |
3b00b203 | 2854 | |
2ca3310e | 2855 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2856 | |
4d4471cb KU |
2857 | dm_table_postsuspend_targets(map); |
2858 | ||
d287483d | 2859 | out_unlock: |
e61290a4 | 2860 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2861 | return r; |
1da177e4 LT |
2862 | } |
2863 | ||
ffcc3936 MS |
2864 | static int __dm_resume(struct mapped_device *md, struct dm_table *map) |
2865 | { | |
2866 | if (map) { | |
2867 | int r = dm_table_resume_targets(map); | |
2868 | if (r) | |
2869 | return r; | |
2870 | } | |
2871 | ||
2872 | dm_queue_flush(md); | |
2873 | ||
2874 | /* | |
2875 | * Flushing deferred I/Os must be done after targets are resumed | |
2876 | * so that mapping of targets can work correctly. | |
2877 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2878 | */ | |
2879 | if (dm_request_based(md)) | |
2880 | start_queue(md->queue); | |
2881 | ||
2882 | unlock_fs(md); | |
2883 | ||
2884 | return 0; | |
2885 | } | |
2886 | ||
1da177e4 LT |
2887 | int dm_resume(struct mapped_device *md) |
2888 | { | |
cf222b37 | 2889 | int r = -EINVAL; |
cf222b37 | 2890 | struct dm_table *map = NULL; |
1da177e4 | 2891 | |
ffcc3936 MS |
2892 | retry: |
2893 | mutex_lock_nested(&md->suspend_lock, SINGLE_DEPTH_NESTING); | |
2894 | ||
4f186f8b | 2895 | if (!dm_suspended_md(md)) |
cf222b37 | 2896 | goto out; |
cf222b37 | 2897 | |
ffcc3936 MS |
2898 | if (dm_suspended_internally_md(md)) { |
2899 | /* already internally suspended, wait for internal resume */ | |
2900 | mutex_unlock(&md->suspend_lock); | |
2901 | r = wait_on_bit(&md->flags, DMF_SUSPENDED_INTERNALLY, TASK_INTERRUPTIBLE); | |
2902 | if (r) | |
2903 | return r; | |
2904 | goto retry; | |
2905 | } | |
2906 | ||
a12f5d48 | 2907 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
2ca3310e | 2908 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2909 | goto out; |
1da177e4 | 2910 | |
ffcc3936 | 2911 | r = __dm_resume(md, map); |
8757b776 MB |
2912 | if (r) |
2913 | goto out; | |
2ca3310e | 2914 | |
2ca3310e AK |
2915 | clear_bit(DMF_SUSPENDED, &md->flags); |
2916 | ||
cf222b37 AK |
2917 | r = 0; |
2918 | out: | |
e61290a4 | 2919 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2920 | |
cf222b37 | 2921 | return r; |
1da177e4 LT |
2922 | } |
2923 | ||
fd2ed4d2 MP |
2924 | /* |
2925 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2926 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2927 | * It may be used only from the kernel. | |
fd2ed4d2 MP |
2928 | */ |
2929 | ||
ffcc3936 | 2930 | static void __dm_internal_suspend(struct mapped_device *md, unsigned suspend_flags) |
fd2ed4d2 | 2931 | { |
ffcc3936 MS |
2932 | struct dm_table *map = NULL; |
2933 | ||
96b26c8c | 2934 | if (md->internal_suspend_count++) |
ffcc3936 MS |
2935 | return; /* nested internal suspend */ |
2936 | ||
2937 | if (dm_suspended_md(md)) { | |
2938 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2939 | return; /* nest suspend */ | |
2940 | } | |
2941 | ||
a12f5d48 | 2942 | map = rcu_dereference_protected(md->map, lockdep_is_held(&md->suspend_lock)); |
ffcc3936 MS |
2943 | |
2944 | /* | |
2945 | * Using TASK_UNINTERRUPTIBLE because only NOFLUSH internal suspend is | |
2946 | * supported. Properly supporting a TASK_INTERRUPTIBLE internal suspend | |
2947 | * would require changing .presuspend to return an error -- avoid this | |
2948 | * until there is a need for more elaborate variants of internal suspend. | |
2949 | */ | |
2950 | (void) __dm_suspend(md, map, suspend_flags, TASK_UNINTERRUPTIBLE); | |
2951 | ||
2952 | set_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2953 | ||
2954 | dm_table_postsuspend_targets(map); | |
2955 | } | |
2956 | ||
2957 | static void __dm_internal_resume(struct mapped_device *md) | |
2958 | { | |
96b26c8c MP |
2959 | BUG_ON(!md->internal_suspend_count); |
2960 | ||
2961 | if (--md->internal_suspend_count) | |
ffcc3936 MS |
2962 | return; /* resume from nested internal suspend */ |
2963 | ||
fd2ed4d2 | 2964 | if (dm_suspended_md(md)) |
ffcc3936 MS |
2965 | goto done; /* resume from nested suspend */ |
2966 | ||
2967 | /* | |
2968 | * NOTE: existing callers don't need to call dm_table_resume_targets | |
2969 | * (which may fail -- so best to avoid it for now by passing NULL map) | |
2970 | */ | |
2971 | (void) __dm_resume(md, NULL); | |
2972 | ||
2973 | done: | |
2974 | clear_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
2975 | smp_mb__after_atomic(); | |
2976 | wake_up_bit(&md->flags, DMF_SUSPENDED_INTERNALLY); | |
2977 | } | |
2978 | ||
2979 | void dm_internal_suspend_noflush(struct mapped_device *md) | |
2980 | { | |
2981 | mutex_lock(&md->suspend_lock); | |
2982 | __dm_internal_suspend(md, DM_SUSPEND_NOFLUSH_FLAG); | |
2983 | mutex_unlock(&md->suspend_lock); | |
2984 | } | |
2985 | EXPORT_SYMBOL_GPL(dm_internal_suspend_noflush); | |
2986 | ||
2987 | void dm_internal_resume(struct mapped_device *md) | |
2988 | { | |
2989 | mutex_lock(&md->suspend_lock); | |
2990 | __dm_internal_resume(md); | |
2991 | mutex_unlock(&md->suspend_lock); | |
2992 | } | |
2993 | EXPORT_SYMBOL_GPL(dm_internal_resume); | |
2994 | ||
2995 | /* | |
2996 | * Fast variants of internal suspend/resume hold md->suspend_lock, | |
2997 | * which prevents interaction with userspace-driven suspend. | |
2998 | */ | |
2999 | ||
3000 | void dm_internal_suspend_fast(struct mapped_device *md) | |
3001 | { | |
3002 | mutex_lock(&md->suspend_lock); | |
3003 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) | |
fd2ed4d2 MP |
3004 | return; |
3005 | ||
3006 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
3007 | synchronize_srcu(&md->io_barrier); | |
3008 | flush_workqueue(md->wq); | |
3009 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
3010 | } | |
3011 | ||
ffcc3936 | 3012 | void dm_internal_resume_fast(struct mapped_device *md) |
fd2ed4d2 | 3013 | { |
ffcc3936 | 3014 | if (dm_suspended_md(md) || dm_suspended_internally_md(md)) |
fd2ed4d2 MP |
3015 | goto done; |
3016 | ||
3017 | dm_queue_flush(md); | |
3018 | ||
3019 | done: | |
3020 | mutex_unlock(&md->suspend_lock); | |
3021 | } | |
3022 | ||
1da177e4 LT |
3023 | /*----------------------------------------------------------------- |
3024 | * Event notification. | |
3025 | *---------------------------------------------------------------*/ | |
3abf85b5 | 3026 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 3027 | unsigned cookie) |
69267a30 | 3028 | { |
60935eb2 MB |
3029 | char udev_cookie[DM_COOKIE_LENGTH]; |
3030 | char *envp[] = { udev_cookie, NULL }; | |
3031 | ||
3032 | if (!cookie) | |
3abf85b5 | 3033 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
3034 | else { |
3035 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
3036 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
3037 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
3038 | action, envp); | |
60935eb2 | 3039 | } |
69267a30 AK |
3040 | } |
3041 | ||
7a8c3d3b MA |
3042 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
3043 | { | |
3044 | return atomic_add_return(1, &md->uevent_seq); | |
3045 | } | |
3046 | ||
1da177e4 LT |
3047 | uint32_t dm_get_event_nr(struct mapped_device *md) |
3048 | { | |
3049 | return atomic_read(&md->event_nr); | |
3050 | } | |
3051 | ||
3052 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
3053 | { | |
3054 | return wait_event_interruptible(md->eventq, | |
3055 | (event_nr != atomic_read(&md->event_nr))); | |
3056 | } | |
3057 | ||
7a8c3d3b MA |
3058 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
3059 | { | |
3060 | unsigned long flags; | |
3061 | ||
3062 | spin_lock_irqsave(&md->uevent_lock, flags); | |
3063 | list_add(elist, &md->uevent_list); | |
3064 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
3065 | } | |
3066 | ||
1da177e4 LT |
3067 | /* |
3068 | * The gendisk is only valid as long as you have a reference | |
3069 | * count on 'md'. | |
3070 | */ | |
3071 | struct gendisk *dm_disk(struct mapped_device *md) | |
3072 | { | |
3073 | return md->disk; | |
3074 | } | |
3075 | ||
784aae73 MB |
3076 | struct kobject *dm_kobject(struct mapped_device *md) |
3077 | { | |
2995fa78 | 3078 | return &md->kobj_holder.kobj; |
784aae73 MB |
3079 | } |
3080 | ||
784aae73 MB |
3081 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
3082 | { | |
3083 | struct mapped_device *md; | |
3084 | ||
2995fa78 | 3085 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 3086 | |
4d89b7b4 | 3087 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 3088 | dm_deleting_md(md)) |
4d89b7b4 MB |
3089 | return NULL; |
3090 | ||
784aae73 MB |
3091 | dm_get(md); |
3092 | return md; | |
3093 | } | |
3094 | ||
4f186f8b | 3095 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
3096 | { |
3097 | return test_bit(DMF_SUSPENDED, &md->flags); | |
3098 | } | |
3099 | ||
ffcc3936 MS |
3100 | int dm_suspended_internally_md(struct mapped_device *md) |
3101 | { | |
3102 | return test_bit(DMF_SUSPENDED_INTERNALLY, &md->flags); | |
3103 | } | |
3104 | ||
2c140a24 MP |
3105 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
3106 | { | |
3107 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
3108 | } | |
3109 | ||
64dbce58 KU |
3110 | int dm_suspended(struct dm_target *ti) |
3111 | { | |
ecdb2e25 | 3112 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
3113 | } |
3114 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3115 | ||
2e93ccc1 KU |
3116 | int dm_noflush_suspending(struct dm_target *ti) |
3117 | { | |
ecdb2e25 | 3118 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3119 | } |
3120 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3121 | ||
c0820cf5 | 3122 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 3123 | { |
5f015204 JN |
3124 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3125 | struct kmem_cache *cachep; | |
3126 | unsigned int pool_size; | |
3127 | unsigned int front_pad; | |
e6ee8c0b KU |
3128 | |
3129 | if (!pools) | |
3130 | return NULL; | |
3131 | ||
23e5083b | 3132 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 | 3133 | cachep = _io_cache; |
e8603136 | 3134 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 JN |
3135 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
3136 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
3137 | cachep = _rq_tio_cache; | |
f4790826 | 3138 | pool_size = dm_get_reserved_rq_based_ios(); |
5f015204 JN |
3139 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3140 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3141 | WARN_ON(per_bio_data_size != 0); | |
3142 | } else | |
3143 | goto out; | |
e6ee8c0b | 3144 | |
6cfa5857 | 3145 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
3146 | if (!pools->io_pool) |
3147 | goto out; | |
e6ee8c0b | 3148 | |
3d8aab2d | 3149 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3150 | if (!pools->bs) |
5f015204 | 3151 | goto out; |
e6ee8c0b | 3152 | |
a91a2785 | 3153 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3154 | goto out; |
a91a2785 | 3155 | |
e6ee8c0b KU |
3156 | return pools; |
3157 | ||
5f015204 JN |
3158 | out: |
3159 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3160 | |
3161 | return NULL; | |
3162 | } | |
3163 | ||
3164 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3165 | { | |
3166 | if (!pools) | |
3167 | return; | |
3168 | ||
3169 | if (pools->io_pool) | |
3170 | mempool_destroy(pools->io_pool); | |
3171 | ||
e6ee8c0b KU |
3172 | if (pools->bs) |
3173 | bioset_free(pools->bs); | |
3174 | ||
3175 | kfree(pools); | |
3176 | } | |
3177 | ||
83d5cde4 | 3178 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3179 | .open = dm_blk_open, |
3180 | .release = dm_blk_close, | |
aa129a22 | 3181 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3182 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3183 | .owner = THIS_MODULE |
3184 | }; | |
3185 | ||
1da177e4 LT |
3186 | /* |
3187 | * module hooks | |
3188 | */ | |
3189 | module_init(dm_init); | |
3190 | module_exit(dm_exit); | |
3191 | ||
3192 | module_param(major, uint, 0); | |
3193 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3194 | |
e8603136 MS |
3195 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3196 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3197 | ||
f4790826 MS |
3198 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3199 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3200 | ||
1da177e4 LT |
3201 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3202 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3203 | MODULE_LICENSE("GPL"); |