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