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> | |
17 | #include <linux/buffer_head.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
3ac51e74 | 21 | #include <linux/hdreg.h> |
3f77316d | 22 | #include <linux/delay.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); |
1da177e4 | 53 | /* |
8fbf26ad | 54 | * For bio-based dm. |
1da177e4 LT |
55 | * One of these is allocated per bio. |
56 | */ | |
57 | struct dm_io { | |
58 | struct mapped_device *md; | |
59 | int error; | |
1da177e4 | 60 | atomic_t io_count; |
6ae2fa67 | 61 | struct bio *bio; |
3eaf840e | 62 | unsigned long start_time; |
f88fb981 | 63 | spinlock_t endio_lock; |
1da177e4 LT |
64 | }; |
65 | ||
66 | /* | |
8fbf26ad | 67 | * For bio-based dm. |
1da177e4 LT |
68 | * One of these is allocated per target within a bio. Hopefully |
69 | * this will be simplified out one day. | |
70 | */ | |
028867ac | 71 | struct dm_target_io { |
1da177e4 LT |
72 | struct dm_io *io; |
73 | struct dm_target *ti; | |
74 | union map_info info; | |
75 | }; | |
76 | ||
8fbf26ad KU |
77 | /* |
78 | * For request-based dm. | |
79 | * One of these is allocated per request. | |
80 | */ | |
81 | struct dm_rq_target_io { | |
82 | struct mapped_device *md; | |
83 | struct dm_target *ti; | |
84 | struct request *orig, clone; | |
85 | int error; | |
86 | union map_info info; | |
87 | }; | |
88 | ||
89 | /* | |
90 | * For request-based dm. | |
91 | * One of these is allocated per bio. | |
92 | */ | |
93 | struct dm_rq_clone_bio_info { | |
94 | struct bio *orig; | |
cec47e3d | 95 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
96 | }; |
97 | ||
1da177e4 LT |
98 | union map_info *dm_get_mapinfo(struct bio *bio) |
99 | { | |
17b2f66f | 100 | if (bio && bio->bi_private) |
028867ac | 101 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 102 | return NULL; |
1da177e4 LT |
103 | } |
104 | ||
cec47e3d KU |
105 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
106 | { | |
107 | if (rq && rq->end_io_data) | |
108 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
109 | return NULL; | |
110 | } | |
111 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
112 | ||
ba61fdd1 JM |
113 | #define MINOR_ALLOCED ((void *)-1) |
114 | ||
1da177e4 LT |
115 | /* |
116 | * Bits for the md->flags field. | |
117 | */ | |
1eb787ec | 118 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 119 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 120 | #define DMF_FROZEN 2 |
fba9f90e | 121 | #define DMF_FREEING 3 |
5c6bd75d | 122 | #define DMF_DELETING 4 |
2e93ccc1 | 123 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 124 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 125 | |
304f3f6a MB |
126 | /* |
127 | * Work processed by per-device workqueue. | |
128 | */ | |
1da177e4 | 129 | struct mapped_device { |
2ca3310e | 130 | struct rw_semaphore io_lock; |
e61290a4 | 131 | struct mutex suspend_lock; |
1da177e4 LT |
132 | rwlock_t map_lock; |
133 | atomic_t holders; | |
5c6bd75d | 134 | atomic_t open_count; |
1da177e4 LT |
135 | |
136 | unsigned long flags; | |
137 | ||
165125e1 | 138 | struct request_queue *queue; |
a5664dad | 139 | unsigned type; |
4a0b4ddf | 140 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
141 | struct mutex type_lock; |
142 | ||
36a0456f AK |
143 | struct target_type *immutable_target_type; |
144 | ||
1da177e4 | 145 | struct gendisk *disk; |
7e51f257 | 146 | char name[16]; |
1da177e4 LT |
147 | |
148 | void *interface_ptr; | |
149 | ||
150 | /* | |
151 | * A list of ios that arrived while we were suspended. | |
152 | */ | |
316d315b | 153 | atomic_t pending[2]; |
1da177e4 | 154 | wait_queue_head_t wait; |
53d5914f | 155 | struct work_struct work; |
74859364 | 156 | struct bio_list deferred; |
022c2611 | 157 | spinlock_t deferred_lock; |
1da177e4 | 158 | |
af7e466a | 159 | /* |
29e4013d | 160 | * Processing queue (flush) |
304f3f6a MB |
161 | */ |
162 | struct workqueue_struct *wq; | |
163 | ||
1da177e4 LT |
164 | /* |
165 | * The current mapping. | |
166 | */ | |
167 | struct dm_table *map; | |
168 | ||
169 | /* | |
170 | * io objects are allocated from here. | |
171 | */ | |
172 | mempool_t *io_pool; | |
173 | mempool_t *tio_pool; | |
174 | ||
9faf400f SB |
175 | struct bio_set *bs; |
176 | ||
1da177e4 LT |
177 | /* |
178 | * Event handling. | |
179 | */ | |
180 | atomic_t event_nr; | |
181 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
182 | atomic_t uevent_seq; |
183 | struct list_head uevent_list; | |
184 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
185 | |
186 | /* | |
187 | * freeze/thaw support require holding onto a super block | |
188 | */ | |
189 | struct super_block *frozen_sb; | |
db8fef4f | 190 | struct block_device *bdev; |
3ac51e74 DW |
191 | |
192 | /* forced geometry settings */ | |
193 | struct hd_geometry geometry; | |
784aae73 | 194 | |
cec47e3d KU |
195 | /* For saving the address of __make_request for request based dm */ |
196 | make_request_fn *saved_make_request_fn; | |
197 | ||
784aae73 MB |
198 | /* sysfs handle */ |
199 | struct kobject kobj; | |
52b1fd5a | 200 | |
d87f4c14 TH |
201 | /* zero-length flush that will be cloned and submitted to targets */ |
202 | struct bio flush_bio; | |
1da177e4 LT |
203 | }; |
204 | ||
e6ee8c0b KU |
205 | /* |
206 | * For mempools pre-allocation at the table loading time. | |
207 | */ | |
208 | struct dm_md_mempools { | |
209 | mempool_t *io_pool; | |
210 | mempool_t *tio_pool; | |
211 | struct bio_set *bs; | |
212 | }; | |
213 | ||
1da177e4 | 214 | #define MIN_IOS 256 |
e18b890b CL |
215 | static struct kmem_cache *_io_cache; |
216 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
217 | static struct kmem_cache *_rq_tio_cache; |
218 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 219 | |
1da177e4 LT |
220 | static int __init local_init(void) |
221 | { | |
51157b4a | 222 | int r = -ENOMEM; |
1da177e4 | 223 | |
1da177e4 | 224 | /* allocate a slab for the dm_ios */ |
028867ac | 225 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 226 | if (!_io_cache) |
51157b4a | 227 | return r; |
1da177e4 LT |
228 | |
229 | /* allocate a slab for the target ios */ | |
028867ac | 230 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
231 | if (!_tio_cache) |
232 | goto out_free_io_cache; | |
1da177e4 | 233 | |
8fbf26ad KU |
234 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
235 | if (!_rq_tio_cache) | |
236 | goto out_free_tio_cache; | |
237 | ||
238 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
239 | if (!_rq_bio_info_cache) | |
240 | goto out_free_rq_tio_cache; | |
241 | ||
51e5b2bd | 242 | r = dm_uevent_init(); |
51157b4a | 243 | if (r) |
8fbf26ad | 244 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 245 | |
1da177e4 LT |
246 | _major = major; |
247 | r = register_blkdev(_major, _name); | |
51157b4a KU |
248 | if (r < 0) |
249 | goto out_uevent_exit; | |
1da177e4 LT |
250 | |
251 | if (!_major) | |
252 | _major = r; | |
253 | ||
254 | return 0; | |
51157b4a KU |
255 | |
256 | out_uevent_exit: | |
257 | dm_uevent_exit(); | |
8fbf26ad KU |
258 | out_free_rq_bio_info_cache: |
259 | kmem_cache_destroy(_rq_bio_info_cache); | |
260 | out_free_rq_tio_cache: | |
261 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
262 | out_free_tio_cache: |
263 | kmem_cache_destroy(_tio_cache); | |
264 | out_free_io_cache: | |
265 | kmem_cache_destroy(_io_cache); | |
266 | ||
267 | return r; | |
1da177e4 LT |
268 | } |
269 | ||
270 | static void local_exit(void) | |
271 | { | |
8fbf26ad KU |
272 | kmem_cache_destroy(_rq_bio_info_cache); |
273 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
274 | kmem_cache_destroy(_tio_cache); |
275 | kmem_cache_destroy(_io_cache); | |
00d59405 | 276 | unregister_blkdev(_major, _name); |
51e5b2bd | 277 | dm_uevent_exit(); |
1da177e4 LT |
278 | |
279 | _major = 0; | |
280 | ||
281 | DMINFO("cleaned up"); | |
282 | } | |
283 | ||
b9249e55 | 284 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
285 | local_init, |
286 | dm_target_init, | |
287 | dm_linear_init, | |
288 | dm_stripe_init, | |
952b3557 | 289 | dm_io_init, |
945fa4d2 | 290 | dm_kcopyd_init, |
1da177e4 LT |
291 | dm_interface_init, |
292 | }; | |
293 | ||
b9249e55 | 294 | static void (*_exits[])(void) = { |
1da177e4 LT |
295 | local_exit, |
296 | dm_target_exit, | |
297 | dm_linear_exit, | |
298 | dm_stripe_exit, | |
952b3557 | 299 | dm_io_exit, |
945fa4d2 | 300 | dm_kcopyd_exit, |
1da177e4 LT |
301 | dm_interface_exit, |
302 | }; | |
303 | ||
304 | static int __init dm_init(void) | |
305 | { | |
306 | const int count = ARRAY_SIZE(_inits); | |
307 | ||
308 | int r, i; | |
309 | ||
310 | for (i = 0; i < count; i++) { | |
311 | r = _inits[i](); | |
312 | if (r) | |
313 | goto bad; | |
314 | } | |
315 | ||
316 | return 0; | |
317 | ||
318 | bad: | |
319 | while (i--) | |
320 | _exits[i](); | |
321 | ||
322 | return r; | |
323 | } | |
324 | ||
325 | static void __exit dm_exit(void) | |
326 | { | |
327 | int i = ARRAY_SIZE(_exits); | |
328 | ||
329 | while (i--) | |
330 | _exits[i](); | |
d15b774c AK |
331 | |
332 | /* | |
333 | * Should be empty by this point. | |
334 | */ | |
335 | idr_remove_all(&_minor_idr); | |
336 | idr_destroy(&_minor_idr); | |
1da177e4 LT |
337 | } |
338 | ||
339 | /* | |
340 | * Block device functions | |
341 | */ | |
432a212c MA |
342 | int dm_deleting_md(struct mapped_device *md) |
343 | { | |
344 | return test_bit(DMF_DELETING, &md->flags); | |
345 | } | |
346 | ||
fe5f9f2c | 347 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
348 | { |
349 | struct mapped_device *md; | |
350 | ||
fba9f90e JM |
351 | spin_lock(&_minor_lock); |
352 | ||
fe5f9f2c | 353 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
354 | if (!md) |
355 | goto out; | |
356 | ||
5c6bd75d | 357 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 358 | dm_deleting_md(md)) { |
fba9f90e JM |
359 | md = NULL; |
360 | goto out; | |
361 | } | |
362 | ||
1da177e4 | 363 | dm_get(md); |
5c6bd75d | 364 | atomic_inc(&md->open_count); |
fba9f90e JM |
365 | |
366 | out: | |
367 | spin_unlock(&_minor_lock); | |
368 | ||
369 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
370 | } |
371 | ||
fe5f9f2c | 372 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 373 | { |
fe5f9f2c | 374 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 375 | |
4a1aeb98 MB |
376 | spin_lock(&_minor_lock); |
377 | ||
5c6bd75d | 378 | atomic_dec(&md->open_count); |
1da177e4 | 379 | dm_put(md); |
4a1aeb98 MB |
380 | |
381 | spin_unlock(&_minor_lock); | |
6e9624b8 | 382 | |
1da177e4 LT |
383 | return 0; |
384 | } | |
385 | ||
5c6bd75d AK |
386 | int dm_open_count(struct mapped_device *md) |
387 | { | |
388 | return atomic_read(&md->open_count); | |
389 | } | |
390 | ||
391 | /* | |
392 | * Guarantees nothing is using the device before it's deleted. | |
393 | */ | |
394 | int dm_lock_for_deletion(struct mapped_device *md) | |
395 | { | |
396 | int r = 0; | |
397 | ||
398 | spin_lock(&_minor_lock); | |
399 | ||
400 | if (dm_open_count(md)) | |
401 | r = -EBUSY; | |
402 | else | |
403 | set_bit(DMF_DELETING, &md->flags); | |
404 | ||
405 | spin_unlock(&_minor_lock); | |
406 | ||
407 | return r; | |
408 | } | |
409 | ||
3ac51e74 DW |
410 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
411 | { | |
412 | struct mapped_device *md = bdev->bd_disk->private_data; | |
413 | ||
414 | return dm_get_geometry(md, geo); | |
415 | } | |
416 | ||
fe5f9f2c | 417 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
418 | unsigned int cmd, unsigned long arg) |
419 | { | |
fe5f9f2c | 420 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 421 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
422 | struct dm_target *tgt; |
423 | int r = -ENOTTY; | |
424 | ||
aa129a22 MB |
425 | if (!map || !dm_table_get_size(map)) |
426 | goto out; | |
427 | ||
428 | /* We only support devices that have a single target */ | |
429 | if (dm_table_get_num_targets(map) != 1) | |
430 | goto out; | |
431 | ||
432 | tgt = dm_table_get_target(map, 0); | |
433 | ||
4f186f8b | 434 | if (dm_suspended_md(md)) { |
aa129a22 MB |
435 | r = -EAGAIN; |
436 | goto out; | |
437 | } | |
438 | ||
439 | if (tgt->type->ioctl) | |
647b3d00 | 440 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
441 | |
442 | out: | |
443 | dm_table_put(map); | |
444 | ||
aa129a22 MB |
445 | return r; |
446 | } | |
447 | ||
028867ac | 448 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
449 | { |
450 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
451 | } | |
452 | ||
028867ac | 453 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
454 | { |
455 | mempool_free(io, md->io_pool); | |
456 | } | |
457 | ||
028867ac | 458 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
459 | { |
460 | mempool_free(tio, md->tio_pool); | |
461 | } | |
462 | ||
08885643 KU |
463 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
464 | gfp_t gfp_mask) | |
cec47e3d | 465 | { |
08885643 | 466 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
467 | } |
468 | ||
469 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
470 | { | |
471 | mempool_free(tio, tio->md->tio_pool); | |
472 | } | |
473 | ||
474 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
475 | { | |
476 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
477 | } | |
478 | ||
479 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
480 | { | |
481 | mempool_free(info, info->tio->md->io_pool); | |
482 | } | |
483 | ||
90abb8c4 KU |
484 | static int md_in_flight(struct mapped_device *md) |
485 | { | |
486 | return atomic_read(&md->pending[READ]) + | |
487 | atomic_read(&md->pending[WRITE]); | |
488 | } | |
489 | ||
3eaf840e JNN |
490 | static void start_io_acct(struct dm_io *io) |
491 | { | |
492 | struct mapped_device *md = io->md; | |
c9959059 | 493 | int cpu; |
316d315b | 494 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
495 | |
496 | io->start_time = jiffies; | |
497 | ||
074a7aca TH |
498 | cpu = part_stat_lock(); |
499 | part_round_stats(cpu, &dm_disk(md)->part0); | |
500 | part_stat_unlock(); | |
1e9bb880 SL |
501 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
502 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
503 | } |
504 | ||
d221d2e7 | 505 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
506 | { |
507 | struct mapped_device *md = io->md; | |
508 | struct bio *bio = io->bio; | |
509 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 510 | int pending, cpu; |
3eaf840e JNN |
511 | int rw = bio_data_dir(bio); |
512 | ||
074a7aca TH |
513 | cpu = part_stat_lock(); |
514 | part_round_stats(cpu, &dm_disk(md)->part0); | |
515 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
516 | part_stat_unlock(); | |
3eaf840e | 517 | |
af7e466a MP |
518 | /* |
519 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 520 | * a flush. |
af7e466a | 521 | */ |
1e9bb880 SL |
522 | pending = atomic_dec_return(&md->pending[rw]); |
523 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 524 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 525 | |
d221d2e7 MP |
526 | /* nudge anyone waiting on suspend queue */ |
527 | if (!pending) | |
528 | wake_up(&md->wait); | |
3eaf840e JNN |
529 | } |
530 | ||
1da177e4 LT |
531 | /* |
532 | * Add the bio to the list of deferred io. | |
533 | */ | |
92c63902 | 534 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 535 | { |
05447420 | 536 | unsigned long flags; |
1da177e4 | 537 | |
05447420 | 538 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 539 | bio_list_add(&md->deferred, bio); |
05447420 | 540 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 541 | queue_work(md->wq, &md->work); |
1da177e4 LT |
542 | } |
543 | ||
544 | /* | |
545 | * Everyone (including functions in this file), should use this | |
546 | * function to access the md->map field, and make sure they call | |
547 | * dm_table_put() when finished. | |
548 | */ | |
7c666411 | 549 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
550 | { |
551 | struct dm_table *t; | |
523d9297 | 552 | unsigned long flags; |
1da177e4 | 553 | |
523d9297 | 554 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
555 | t = md->map; |
556 | if (t) | |
557 | dm_table_get(t); | |
523d9297 | 558 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
559 | |
560 | return t; | |
561 | } | |
562 | ||
3ac51e74 DW |
563 | /* |
564 | * Get the geometry associated with a dm device | |
565 | */ | |
566 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
567 | { | |
568 | *geo = md->geometry; | |
569 | ||
570 | return 0; | |
571 | } | |
572 | ||
573 | /* | |
574 | * Set the geometry of a device. | |
575 | */ | |
576 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
577 | { | |
578 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
579 | ||
580 | if (geo->start > sz) { | |
581 | DMWARN("Start sector is beyond the geometry limits."); | |
582 | return -EINVAL; | |
583 | } | |
584 | ||
585 | md->geometry = *geo; | |
586 | ||
587 | return 0; | |
588 | } | |
589 | ||
1da177e4 LT |
590 | /*----------------------------------------------------------------- |
591 | * CRUD START: | |
592 | * A more elegant soln is in the works that uses the queue | |
593 | * merge fn, unfortunately there are a couple of changes to | |
594 | * the block layer that I want to make for this. So in the | |
595 | * interests of getting something for people to use I give | |
596 | * you this clearly demarcated crap. | |
597 | *---------------------------------------------------------------*/ | |
598 | ||
2e93ccc1 KU |
599 | static int __noflush_suspending(struct mapped_device *md) |
600 | { | |
601 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
602 | } | |
603 | ||
1da177e4 LT |
604 | /* |
605 | * Decrements the number of outstanding ios that a bio has been | |
606 | * cloned into, completing the original io if necc. | |
607 | */ | |
858119e1 | 608 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 609 | { |
2e93ccc1 | 610 | unsigned long flags; |
b35f8caa MB |
611 | int io_error; |
612 | struct bio *bio; | |
613 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
614 | |
615 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
616 | if (unlikely(error)) { |
617 | spin_lock_irqsave(&io->endio_lock, flags); | |
618 | if (!(io->error > 0 && __noflush_suspending(md))) | |
619 | io->error = error; | |
620 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
621 | } | |
1da177e4 LT |
622 | |
623 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
624 | if (io->error == DM_ENDIO_REQUEUE) { |
625 | /* | |
626 | * Target requested pushing back the I/O. | |
2e93ccc1 | 627 | */ |
022c2611 | 628 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
629 | if (__noflush_suspending(md)) |
630 | bio_list_add_head(&md->deferred, io->bio); | |
631 | else | |
2e93ccc1 KU |
632 | /* noflush suspend was interrupted. */ |
633 | io->error = -EIO; | |
022c2611 | 634 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
635 | } |
636 | ||
b35f8caa MB |
637 | io_error = io->error; |
638 | bio = io->bio; | |
6a8736d1 TH |
639 | end_io_acct(io); |
640 | free_io(md, io); | |
641 | ||
642 | if (io_error == DM_ENDIO_REQUEUE) | |
643 | return; | |
2e93ccc1 | 644 | |
b372d360 | 645 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 646 | /* |
6a8736d1 TH |
647 | * Preflush done for flush with data, reissue |
648 | * without REQ_FLUSH. | |
af7e466a | 649 | */ |
6a8736d1 TH |
650 | bio->bi_rw &= ~REQ_FLUSH; |
651 | queue_io(md, bio); | |
af7e466a | 652 | } else { |
b372d360 | 653 | /* done with normal IO or empty flush */ |
b7908c10 | 654 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 655 | bio_endio(bio, io_error); |
b35f8caa | 656 | } |
1da177e4 LT |
657 | } |
658 | } | |
659 | ||
6712ecf8 | 660 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
661 | { |
662 | int r = 0; | |
028867ac | 663 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 664 | struct dm_io *io = tio->io; |
9faf400f | 665 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
666 | dm_endio_fn endio = tio->ti->type->end_io; |
667 | ||
1da177e4 LT |
668 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
669 | error = -EIO; | |
670 | ||
671 | if (endio) { | |
672 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
673 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
674 | /* | |
675 | * error and requeue request are handled | |
676 | * in dec_pending(). | |
677 | */ | |
1da177e4 | 678 | error = r; |
45cbcd79 KU |
679 | else if (r == DM_ENDIO_INCOMPLETE) |
680 | /* The target will handle the io */ | |
6712ecf8 | 681 | return; |
45cbcd79 KU |
682 | else if (r) { |
683 | DMWARN("unimplemented target endio return value: %d", r); | |
684 | BUG(); | |
685 | } | |
1da177e4 LT |
686 | } |
687 | ||
9faf400f SB |
688 | /* |
689 | * Store md for cleanup instead of tio which is about to get freed. | |
690 | */ | |
691 | bio->bi_private = md->bs; | |
692 | ||
9faf400f | 693 | free_tio(md, tio); |
b35f8caa MB |
694 | bio_put(bio); |
695 | dec_pending(io, error); | |
1da177e4 LT |
696 | } |
697 | ||
cec47e3d KU |
698 | /* |
699 | * Partial completion handling for request-based dm | |
700 | */ | |
701 | static void end_clone_bio(struct bio *clone, int error) | |
702 | { | |
703 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
704 | struct dm_rq_target_io *tio = info->tio; | |
705 | struct bio *bio = info->orig; | |
706 | unsigned int nr_bytes = info->orig->bi_size; | |
707 | ||
708 | bio_put(clone); | |
709 | ||
710 | if (tio->error) | |
711 | /* | |
712 | * An error has already been detected on the request. | |
713 | * Once error occurred, just let clone->end_io() handle | |
714 | * the remainder. | |
715 | */ | |
716 | return; | |
717 | else if (error) { | |
718 | /* | |
719 | * Don't notice the error to the upper layer yet. | |
720 | * The error handling decision is made by the target driver, | |
721 | * when the request is completed. | |
722 | */ | |
723 | tio->error = error; | |
724 | return; | |
725 | } | |
726 | ||
727 | /* | |
728 | * I/O for the bio successfully completed. | |
729 | * Notice the data completion to the upper layer. | |
730 | */ | |
731 | ||
732 | /* | |
733 | * bios are processed from the head of the list. | |
734 | * So the completing bio should always be rq->bio. | |
735 | * If it's not, something wrong is happening. | |
736 | */ | |
737 | if (tio->orig->bio != bio) | |
738 | DMERR("bio completion is going in the middle of the request"); | |
739 | ||
740 | /* | |
741 | * Update the original request. | |
742 | * Do not use blk_end_request() here, because it may complete | |
743 | * the original request before the clone, and break the ordering. | |
744 | */ | |
745 | blk_update_request(tio->orig, 0, nr_bytes); | |
746 | } | |
747 | ||
748 | /* | |
749 | * Don't touch any member of the md after calling this function because | |
750 | * the md may be freed in dm_put() at the end of this function. | |
751 | * Or do dm_get() before calling this function and dm_put() later. | |
752 | */ | |
b4324fee | 753 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 754 | { |
b4324fee | 755 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
756 | |
757 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 758 | if (!md_in_flight(md)) |
cec47e3d KU |
759 | wake_up(&md->wait); |
760 | ||
761 | if (run_queue) | |
b4324fee | 762 | blk_run_queue(md->queue); |
cec47e3d KU |
763 | |
764 | /* | |
765 | * dm_put() must be at the end of this function. See the comment above | |
766 | */ | |
767 | dm_put(md); | |
768 | } | |
769 | ||
a77e28c7 KU |
770 | static void free_rq_clone(struct request *clone) |
771 | { | |
772 | struct dm_rq_target_io *tio = clone->end_io_data; | |
773 | ||
774 | blk_rq_unprep_clone(clone); | |
775 | free_rq_tio(tio); | |
776 | } | |
777 | ||
980691e5 KU |
778 | /* |
779 | * Complete the clone and the original request. | |
780 | * Must be called without queue lock. | |
781 | */ | |
782 | static void dm_end_request(struct request *clone, int error) | |
783 | { | |
784 | int rw = rq_data_dir(clone); | |
785 | struct dm_rq_target_io *tio = clone->end_io_data; | |
786 | struct mapped_device *md = tio->md; | |
787 | struct request *rq = tio->orig; | |
788 | ||
29e4013d | 789 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
790 | rq->errors = clone->errors; |
791 | rq->resid_len = clone->resid_len; | |
792 | ||
793 | if (rq->sense) | |
794 | /* | |
795 | * We are using the sense buffer of the original | |
796 | * request. | |
797 | * So setting the length of the sense data is enough. | |
798 | */ | |
799 | rq->sense_len = clone->sense_len; | |
800 | } | |
801 | ||
802 | free_rq_clone(clone); | |
29e4013d TH |
803 | blk_end_request_all(rq, error); |
804 | rq_completed(md, rw, true); | |
980691e5 KU |
805 | } |
806 | ||
cec47e3d KU |
807 | static void dm_unprep_request(struct request *rq) |
808 | { | |
809 | struct request *clone = rq->special; | |
cec47e3d KU |
810 | |
811 | rq->special = NULL; | |
812 | rq->cmd_flags &= ~REQ_DONTPREP; | |
813 | ||
a77e28c7 | 814 | free_rq_clone(clone); |
cec47e3d KU |
815 | } |
816 | ||
817 | /* | |
818 | * Requeue the original request of a clone. | |
819 | */ | |
820 | void dm_requeue_unmapped_request(struct request *clone) | |
821 | { | |
b4324fee | 822 | int rw = rq_data_dir(clone); |
cec47e3d KU |
823 | struct dm_rq_target_io *tio = clone->end_io_data; |
824 | struct mapped_device *md = tio->md; | |
825 | struct request *rq = tio->orig; | |
826 | struct request_queue *q = rq->q; | |
827 | unsigned long flags; | |
828 | ||
829 | dm_unprep_request(rq); | |
830 | ||
831 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
832 | blk_requeue_request(q, rq); |
833 | spin_unlock_irqrestore(q->queue_lock, flags); | |
834 | ||
b4324fee | 835 | rq_completed(md, rw, 0); |
cec47e3d KU |
836 | } |
837 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
838 | ||
839 | static void __stop_queue(struct request_queue *q) | |
840 | { | |
841 | blk_stop_queue(q); | |
842 | } | |
843 | ||
844 | static void stop_queue(struct request_queue *q) | |
845 | { | |
846 | unsigned long flags; | |
847 | ||
848 | spin_lock_irqsave(q->queue_lock, flags); | |
849 | __stop_queue(q); | |
850 | spin_unlock_irqrestore(q->queue_lock, flags); | |
851 | } | |
852 | ||
853 | static void __start_queue(struct request_queue *q) | |
854 | { | |
855 | if (blk_queue_stopped(q)) | |
856 | blk_start_queue(q); | |
857 | } | |
858 | ||
859 | static void start_queue(struct request_queue *q) | |
860 | { | |
861 | unsigned long flags; | |
862 | ||
863 | spin_lock_irqsave(q->queue_lock, flags); | |
864 | __start_queue(q); | |
865 | spin_unlock_irqrestore(q->queue_lock, flags); | |
866 | } | |
867 | ||
11a68244 | 868 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 869 | { |
11a68244 | 870 | int r = error; |
cec47e3d KU |
871 | struct dm_rq_target_io *tio = clone->end_io_data; |
872 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
cec47e3d | 873 | |
11a68244 KU |
874 | if (mapped && rq_end_io) |
875 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
cec47e3d | 876 | |
11a68244 | 877 | if (r <= 0) |
cec47e3d | 878 | /* The target wants to complete the I/O */ |
11a68244 KU |
879 | dm_end_request(clone, r); |
880 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
881 | /* The target will handle the I/O */ |
882 | return; | |
11a68244 | 883 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
884 | /* The target wants to requeue the I/O */ |
885 | dm_requeue_unmapped_request(clone); | |
886 | else { | |
11a68244 | 887 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
888 | BUG(); |
889 | } | |
890 | } | |
891 | ||
11a68244 KU |
892 | /* |
893 | * Request completion handler for request-based dm | |
894 | */ | |
895 | static void dm_softirq_done(struct request *rq) | |
896 | { | |
897 | bool mapped = true; | |
898 | struct request *clone = rq->completion_data; | |
899 | struct dm_rq_target_io *tio = clone->end_io_data; | |
900 | ||
901 | if (rq->cmd_flags & REQ_FAILED) | |
902 | mapped = false; | |
903 | ||
904 | dm_done(clone, tio->error, mapped); | |
905 | } | |
906 | ||
cec47e3d KU |
907 | /* |
908 | * Complete the clone and the original request with the error status | |
909 | * through softirq context. | |
910 | */ | |
911 | static void dm_complete_request(struct request *clone, int error) | |
912 | { | |
913 | struct dm_rq_target_io *tio = clone->end_io_data; | |
914 | struct request *rq = tio->orig; | |
915 | ||
916 | tio->error = error; | |
917 | rq->completion_data = clone; | |
918 | blk_complete_request(rq); | |
919 | } | |
920 | ||
921 | /* | |
922 | * Complete the not-mapped clone and the original request with the error status | |
923 | * through softirq context. | |
924 | * Target's rq_end_io() function isn't called. | |
925 | * This may be used when the target's map_rq() function fails. | |
926 | */ | |
927 | void dm_kill_unmapped_request(struct request *clone, int error) | |
928 | { | |
929 | struct dm_rq_target_io *tio = clone->end_io_data; | |
930 | struct request *rq = tio->orig; | |
931 | ||
932 | rq->cmd_flags |= REQ_FAILED; | |
933 | dm_complete_request(clone, error); | |
934 | } | |
935 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
936 | ||
937 | /* | |
938 | * Called with the queue lock held | |
939 | */ | |
940 | static void end_clone_request(struct request *clone, int error) | |
941 | { | |
942 | /* | |
943 | * For just cleaning up the information of the queue in which | |
944 | * the clone was dispatched. | |
945 | * The clone is *NOT* freed actually here because it is alloced from | |
946 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
947 | */ | |
948 | __blk_put_request(clone->q, clone); | |
949 | ||
950 | /* | |
951 | * Actual request completion is done in a softirq context which doesn't | |
952 | * hold the queue lock. Otherwise, deadlock could occur because: | |
953 | * - another request may be submitted by the upper level driver | |
954 | * of the stacking during the completion | |
955 | * - the submission which requires queue lock may be done | |
956 | * against this queue | |
957 | */ | |
958 | dm_complete_request(clone, error); | |
959 | } | |
960 | ||
56a67df7 MS |
961 | /* |
962 | * Return maximum size of I/O possible at the supplied sector up to the current | |
963 | * target boundary. | |
964 | */ | |
965 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
966 | { | |
967 | sector_t target_offset = dm_target_offset(ti, sector); | |
968 | ||
969 | return ti->len - target_offset; | |
970 | } | |
971 | ||
972 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 973 | { |
56a67df7 | 974 | sector_t len = max_io_len_target_boundary(sector, ti); |
1da177e4 LT |
975 | |
976 | /* | |
977 | * Does the target need to split even further ? | |
978 | */ | |
979 | if (ti->split_io) { | |
980 | sector_t boundary; | |
56a67df7 | 981 | sector_t offset = dm_target_offset(ti, sector); |
1da177e4 LT |
982 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) |
983 | - offset; | |
984 | if (len > boundary) | |
985 | len = boundary; | |
986 | } | |
987 | ||
988 | return len; | |
989 | } | |
990 | ||
991 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 992 | struct dm_target_io *tio) |
1da177e4 LT |
993 | { |
994 | int r; | |
2056a782 | 995 | sector_t sector; |
9faf400f | 996 | struct mapped_device *md; |
1da177e4 | 997 | |
1da177e4 LT |
998 | clone->bi_end_io = clone_endio; |
999 | clone->bi_private = tio; | |
1000 | ||
1001 | /* | |
1002 | * Map the clone. If r == 0 we don't need to do | |
1003 | * anything, the target has assumed ownership of | |
1004 | * this io. | |
1005 | */ | |
1006 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1007 | sector = clone->bi_sector; |
1da177e4 | 1008 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 1009 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1010 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1011 | |
d07335e5 MS |
1012 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1013 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1014 | |
1da177e4 | 1015 | generic_make_request(clone); |
2e93ccc1 KU |
1016 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1017 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1018 | md = tio->io->md; |
1019 | dec_pending(tio->io, r); | |
1020 | /* | |
1021 | * Store bio_set for cleanup. | |
1022 | */ | |
1023 | clone->bi_private = md->bs; | |
1da177e4 | 1024 | bio_put(clone); |
9faf400f | 1025 | free_tio(md, tio); |
45cbcd79 KU |
1026 | } else if (r) { |
1027 | DMWARN("unimplemented target map return value: %d", r); | |
1028 | BUG(); | |
1da177e4 LT |
1029 | } |
1030 | } | |
1031 | ||
1032 | struct clone_info { | |
1033 | struct mapped_device *md; | |
1034 | struct dm_table *map; | |
1035 | struct bio *bio; | |
1036 | struct dm_io *io; | |
1037 | sector_t sector; | |
1038 | sector_t sector_count; | |
1039 | unsigned short idx; | |
1040 | }; | |
1041 | ||
3676347a PO |
1042 | static void dm_bio_destructor(struct bio *bio) |
1043 | { | |
9faf400f SB |
1044 | struct bio_set *bs = bio->bi_private; |
1045 | ||
1046 | bio_free(bio, bs); | |
3676347a PO |
1047 | } |
1048 | ||
1da177e4 | 1049 | /* |
d87f4c14 | 1050 | * Creates a little bio that just does part of a bvec. |
1da177e4 LT |
1051 | */ |
1052 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1053 | unsigned short idx, unsigned int offset, | |
9faf400f | 1054 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1055 | { |
1056 | struct bio *clone; | |
1057 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1058 | ||
9faf400f | 1059 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1060 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1061 | *clone->bi_io_vec = *bv; |
1062 | ||
1063 | clone->bi_sector = sector; | |
1064 | clone->bi_bdev = bio->bi_bdev; | |
d87f4c14 | 1065 | clone->bi_rw = bio->bi_rw; |
1da177e4 LT |
1066 | clone->bi_vcnt = 1; |
1067 | clone->bi_size = to_bytes(len); | |
1068 | clone->bi_io_vec->bv_offset = offset; | |
1069 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1070 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1071 | |
9c47008d | 1072 | if (bio_integrity(bio)) { |
7878cba9 | 1073 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1074 | bio_integrity_trim(clone, |
1075 | bio_sector_offset(bio, idx, offset), len); | |
1076 | } | |
1077 | ||
1da177e4 LT |
1078 | return clone; |
1079 | } | |
1080 | ||
1081 | /* | |
1082 | * Creates a bio that consists of range of complete bvecs. | |
1083 | */ | |
1084 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1085 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1086 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1087 | { |
1088 | struct bio *clone; | |
1089 | ||
9faf400f SB |
1090 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1091 | __bio_clone(clone, bio); | |
1092 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
1093 | clone->bi_sector = sector; |
1094 | clone->bi_idx = idx; | |
1095 | clone->bi_vcnt = idx + bv_count; | |
1096 | clone->bi_size = to_bytes(len); | |
1097 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1098 | ||
9c47008d | 1099 | if (bio_integrity(bio)) { |
7878cba9 | 1100 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1101 | |
1102 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1103 | bio_integrity_trim(clone, | |
1104 | bio_sector_offset(bio, idx, 0), len); | |
1105 | } | |
1106 | ||
1da177e4 LT |
1107 | return clone; |
1108 | } | |
1109 | ||
9015df24 AK |
1110 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1111 | struct dm_target *ti) | |
f9ab94ce | 1112 | { |
9015df24 | 1113 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1114 | |
1115 | tio->io = ci->io; | |
1116 | tio->ti = ti; | |
f9ab94ce | 1117 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1118 | |
1119 | return tio; | |
1120 | } | |
1121 | ||
06a426ce | 1122 | static void __issue_target_request(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1123 | unsigned request_nr, sector_t len) |
9015df24 AK |
1124 | { |
1125 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1126 | struct bio *clone; | |
1127 | ||
57cba5d3 | 1128 | tio->info.target_request_nr = request_nr; |
f9ab94ce | 1129 | |
06a426ce MS |
1130 | /* |
1131 | * Discard requests require the bio's inline iovecs be initialized. | |
1132 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1133 | * and discard, so no need for concern about wasted bvec allocations. | |
1134 | */ | |
1135 | clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs); | |
f9ab94ce MP |
1136 | __bio_clone(clone, ci->bio); |
1137 | clone->bi_destructor = dm_bio_destructor; | |
a79245b3 MS |
1138 | if (len) { |
1139 | clone->bi_sector = ci->sector; | |
1140 | clone->bi_size = to_bytes(len); | |
1141 | } | |
f9ab94ce MP |
1142 | |
1143 | __map_bio(ti, clone, tio); | |
1144 | } | |
1145 | ||
06a426ce | 1146 | static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1147 | unsigned num_requests, sector_t len) |
06a426ce MS |
1148 | { |
1149 | unsigned request_nr; | |
1150 | ||
1151 | for (request_nr = 0; request_nr < num_requests; request_nr++) | |
a79245b3 | 1152 | __issue_target_request(ci, ti, request_nr, len); |
06a426ce MS |
1153 | } |
1154 | ||
b372d360 | 1155 | static int __clone_and_map_empty_flush(struct clone_info *ci) |
f9ab94ce | 1156 | { |
06a426ce | 1157 | unsigned target_nr = 0; |
f9ab94ce MP |
1158 | struct dm_target *ti; |
1159 | ||
b372d360 | 1160 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1161 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
a79245b3 | 1162 | __issue_target_requests(ci, ti, ti->num_flush_requests, 0); |
f9ab94ce | 1163 | |
f9ab94ce MP |
1164 | return 0; |
1165 | } | |
1166 | ||
5ae89a87 MS |
1167 | /* |
1168 | * Perform all io with a single clone. | |
1169 | */ | |
1170 | static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti) | |
1171 | { | |
1172 | struct bio *clone, *bio = ci->bio; | |
1173 | struct dm_target_io *tio; | |
1174 | ||
1175 | tio = alloc_tio(ci, ti); | |
1176 | clone = clone_bio(bio, ci->sector, ci->idx, | |
1177 | bio->bi_vcnt - ci->idx, ci->sector_count, | |
1178 | ci->md->bs); | |
1179 | __map_bio(ti, clone, tio); | |
1180 | ci->sector_count = 0; | |
1181 | } | |
1182 | ||
1183 | static int __clone_and_map_discard(struct clone_info *ci) | |
1184 | { | |
1185 | struct dm_target *ti; | |
a79245b3 | 1186 | sector_t len; |
5ae89a87 | 1187 | |
a79245b3 MS |
1188 | do { |
1189 | ti = dm_table_find_target(ci->map, ci->sector); | |
1190 | if (!dm_target_is_valid(ti)) | |
1191 | return -EIO; | |
5ae89a87 | 1192 | |
5ae89a87 | 1193 | /* |
a79245b3 | 1194 | * Even though the device advertised discard support, |
936688d7 MS |
1195 | * that does not mean every target supports it, and |
1196 | * reconfiguration might also have changed that since the | |
a79245b3 | 1197 | * check was performed. |
5ae89a87 | 1198 | */ |
a79245b3 MS |
1199 | if (!ti->num_discard_requests) |
1200 | return -EOPNOTSUPP; | |
5ae89a87 | 1201 | |
a79245b3 | 1202 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
06a426ce | 1203 | |
a79245b3 MS |
1204 | __issue_target_requests(ci, ti, ti->num_discard_requests, len); |
1205 | ||
1206 | ci->sector += len; | |
1207 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1208 | |
1209 | return 0; | |
1210 | } | |
1211 | ||
512875bd | 1212 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1213 | { |
1214 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1215 | struct dm_target *ti; |
1216 | sector_t len = 0, max; | |
028867ac | 1217 | struct dm_target_io *tio; |
1da177e4 | 1218 | |
5ae89a87 MS |
1219 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
1220 | return __clone_and_map_discard(ci); | |
1221 | ||
512875bd JN |
1222 | ti = dm_table_find_target(ci->map, ci->sector); |
1223 | if (!dm_target_is_valid(ti)) | |
1224 | return -EIO; | |
1225 | ||
56a67df7 | 1226 | max = max_io_len(ci->sector, ti); |
512875bd | 1227 | |
1da177e4 LT |
1228 | if (ci->sector_count <= max) { |
1229 | /* | |
1230 | * Optimise for the simple case where we can do all of | |
1231 | * the remaining io with a single clone. | |
1232 | */ | |
5ae89a87 | 1233 | __clone_and_map_simple(ci, ti); |
1da177e4 LT |
1234 | |
1235 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1236 | /* | |
1237 | * There are some bvecs that don't span targets. | |
1238 | * Do as many of these as possible. | |
1239 | */ | |
1240 | int i; | |
1241 | sector_t remaining = max; | |
1242 | sector_t bv_len; | |
1243 | ||
1244 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1245 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1246 | ||
1247 | if (bv_len > remaining) | |
1248 | break; | |
1249 | ||
1250 | remaining -= bv_len; | |
1251 | len += bv_len; | |
1252 | } | |
1253 | ||
5ae89a87 | 1254 | tio = alloc_tio(ci, ti); |
9faf400f SB |
1255 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1256 | ci->md->bs); | |
1da177e4 LT |
1257 | __map_bio(ti, clone, tio); |
1258 | ||
1259 | ci->sector += len; | |
1260 | ci->sector_count -= len; | |
1261 | ci->idx = i; | |
1262 | ||
1263 | } else { | |
1264 | /* | |
d2044a94 | 1265 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1266 | */ |
1267 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1268 | sector_t remaining = to_sector(bv->bv_len); |
1269 | unsigned int offset = 0; | |
1da177e4 | 1270 | |
d2044a94 AK |
1271 | do { |
1272 | if (offset) { | |
1273 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1274 | if (!dm_target_is_valid(ti)) |
1275 | return -EIO; | |
1276 | ||
56a67df7 | 1277 | max = max_io_len(ci->sector, ti); |
d2044a94 AK |
1278 | } |
1279 | ||
1280 | len = min(remaining, max); | |
1281 | ||
5ae89a87 | 1282 | tio = alloc_tio(ci, ti); |
d2044a94 | 1283 | clone = split_bvec(bio, ci->sector, ci->idx, |
9faf400f SB |
1284 | bv->bv_offset + offset, len, |
1285 | ci->md->bs); | |
d2044a94 AK |
1286 | |
1287 | __map_bio(ti, clone, tio); | |
1288 | ||
1289 | ci->sector += len; | |
1290 | ci->sector_count -= len; | |
1291 | offset += to_bytes(len); | |
1292 | } while (remaining -= len); | |
1da177e4 | 1293 | |
1da177e4 LT |
1294 | ci->idx++; |
1295 | } | |
512875bd JN |
1296 | |
1297 | return 0; | |
1da177e4 LT |
1298 | } |
1299 | ||
1300 | /* | |
8a53c28d | 1301 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1302 | */ |
f0b9a450 | 1303 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1304 | { |
1305 | struct clone_info ci; | |
512875bd | 1306 | int error = 0; |
1da177e4 | 1307 | |
7c666411 | 1308 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1309 | if (unlikely(!ci.map)) { |
6a8736d1 | 1310 | bio_io_error(bio); |
f0b9a450 MP |
1311 | return; |
1312 | } | |
692d0eb9 | 1313 | |
1da177e4 | 1314 | ci.md = md; |
1da177e4 LT |
1315 | ci.io = alloc_io(md); |
1316 | ci.io->error = 0; | |
1317 | atomic_set(&ci.io->io_count, 1); | |
1318 | ci.io->bio = bio; | |
1319 | ci.io->md = md; | |
f88fb981 | 1320 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1321 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1322 | ci.idx = bio->bi_idx; |
1323 | ||
3eaf840e | 1324 | start_io_acct(ci.io); |
b372d360 MS |
1325 | if (bio->bi_rw & REQ_FLUSH) { |
1326 | ci.bio = &ci.md->flush_bio; | |
1327 | ci.sector_count = 0; | |
1328 | error = __clone_and_map_empty_flush(&ci); | |
1329 | /* dec_pending submits any data associated with flush */ | |
1330 | } else { | |
6a8736d1 | 1331 | ci.bio = bio; |
d87f4c14 | 1332 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1333 | while (ci.sector_count && !error) |
d87f4c14 | 1334 | error = __clone_and_map(&ci); |
d87f4c14 | 1335 | } |
1da177e4 LT |
1336 | |
1337 | /* drop the extra reference count */ | |
512875bd | 1338 | dec_pending(ci.io, error); |
1da177e4 LT |
1339 | dm_table_put(ci.map); |
1340 | } | |
1341 | /*----------------------------------------------------------------- | |
1342 | * CRUD END | |
1343 | *---------------------------------------------------------------*/ | |
1344 | ||
f6fccb12 MB |
1345 | static int dm_merge_bvec(struct request_queue *q, |
1346 | struct bvec_merge_data *bvm, | |
1347 | struct bio_vec *biovec) | |
1348 | { | |
1349 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1350 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1351 | struct dm_target *ti; |
1352 | sector_t max_sectors; | |
5037108a | 1353 | int max_size = 0; |
f6fccb12 MB |
1354 | |
1355 | if (unlikely(!map)) | |
5037108a | 1356 | goto out; |
f6fccb12 MB |
1357 | |
1358 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1359 | if (!dm_target_is_valid(ti)) |
1360 | goto out_table; | |
f6fccb12 MB |
1361 | |
1362 | /* | |
1363 | * Find maximum amount of I/O that won't need splitting | |
1364 | */ | |
56a67df7 | 1365 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1366 | (sector_t) BIO_MAX_SECTORS); |
1367 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1368 | if (max_size < 0) | |
1369 | max_size = 0; | |
1370 | ||
1371 | /* | |
1372 | * merge_bvec_fn() returns number of bytes | |
1373 | * it can accept at this offset | |
1374 | * max is precomputed maximal io size | |
1375 | */ | |
1376 | if (max_size && ti->type->merge) | |
1377 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1378 | /* |
1379 | * If the target doesn't support merge method and some of the devices | |
1380 | * provided their merge_bvec method (we know this by looking at | |
1381 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1382 | * entries. So always set max_size to 0, and the code below allows | |
1383 | * just one page. | |
1384 | */ | |
1385 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1386 | ||
1387 | max_size = 0; | |
f6fccb12 | 1388 | |
b01cd5ac | 1389 | out_table: |
5037108a MP |
1390 | dm_table_put(map); |
1391 | ||
1392 | out: | |
f6fccb12 MB |
1393 | /* |
1394 | * Always allow an entire first page | |
1395 | */ | |
1396 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1397 | max_size = biovec->bv_len; | |
1398 | ||
f6fccb12 MB |
1399 | return max_size; |
1400 | } | |
1401 | ||
1da177e4 LT |
1402 | /* |
1403 | * The request function that just remaps the bio built up by | |
1404 | * dm_merge_bvec. | |
1405 | */ | |
cec47e3d | 1406 | static int _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1407 | { |
12f03a49 | 1408 | int rw = bio_data_dir(bio); |
1da177e4 | 1409 | struct mapped_device *md = q->queuedata; |
c9959059 | 1410 | int cpu; |
1da177e4 | 1411 | |
2ca3310e | 1412 | down_read(&md->io_lock); |
1da177e4 | 1413 | |
074a7aca TH |
1414 | cpu = part_stat_lock(); |
1415 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1416 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1417 | part_stat_unlock(); | |
12f03a49 | 1418 | |
6a8736d1 TH |
1419 | /* if we're suspended, we have to queue this io for later */ |
1420 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1421 | up_read(&md->io_lock); |
1da177e4 | 1422 | |
6a8736d1 TH |
1423 | if (bio_rw(bio) != READA) |
1424 | queue_io(md, bio); | |
1425 | else | |
54d9a1b4 | 1426 | bio_io_error(bio); |
92c63902 | 1427 | return 0; |
1da177e4 LT |
1428 | } |
1429 | ||
f0b9a450 | 1430 | __split_and_process_bio(md, bio); |
2ca3310e | 1431 | up_read(&md->io_lock); |
f0b9a450 | 1432 | return 0; |
1da177e4 LT |
1433 | } |
1434 | ||
cec47e3d KU |
1435 | static int dm_make_request(struct request_queue *q, struct bio *bio) |
1436 | { | |
1437 | struct mapped_device *md = q->queuedata; | |
1438 | ||
cec47e3d KU |
1439 | return md->saved_make_request_fn(q, bio); /* call __make_request() */ |
1440 | } | |
1441 | ||
1442 | static int dm_request_based(struct mapped_device *md) | |
1443 | { | |
1444 | return blk_queue_stackable(md->queue); | |
1445 | } | |
1446 | ||
1447 | static int dm_request(struct request_queue *q, struct bio *bio) | |
1448 | { | |
1449 | struct mapped_device *md = q->queuedata; | |
1450 | ||
1451 | if (dm_request_based(md)) | |
1452 | return dm_make_request(q, bio); | |
1453 | ||
1454 | return _dm_request(q, bio); | |
1455 | } | |
1456 | ||
1457 | void dm_dispatch_request(struct request *rq) | |
1458 | { | |
1459 | int r; | |
1460 | ||
1461 | if (blk_queue_io_stat(rq->q)) | |
1462 | rq->cmd_flags |= REQ_IO_STAT; | |
1463 | ||
1464 | rq->start_time = jiffies; | |
1465 | r = blk_insert_cloned_request(rq->q, rq); | |
1466 | if (r) | |
1467 | dm_complete_request(rq, r); | |
1468 | } | |
1469 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1470 | ||
1471 | static void dm_rq_bio_destructor(struct bio *bio) | |
1472 | { | |
1473 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1474 | struct mapped_device *md = info->tio->md; | |
1475 | ||
1476 | free_bio_info(info); | |
1477 | bio_free(bio, md->bs); | |
1478 | } | |
1479 | ||
1480 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1481 | void *data) | |
1482 | { | |
1483 | struct dm_rq_target_io *tio = data; | |
1484 | struct mapped_device *md = tio->md; | |
1485 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1486 | ||
1487 | if (!info) | |
1488 | return -ENOMEM; | |
1489 | ||
1490 | info->orig = bio_orig; | |
1491 | info->tio = tio; | |
1492 | bio->bi_end_io = end_clone_bio; | |
1493 | bio->bi_private = info; | |
1494 | bio->bi_destructor = dm_rq_bio_destructor; | |
1495 | ||
1496 | return 0; | |
1497 | } | |
1498 | ||
1499 | static int setup_clone(struct request *clone, struct request *rq, | |
1500 | struct dm_rq_target_io *tio) | |
1501 | { | |
d0bcb878 | 1502 | int r; |
cec47e3d | 1503 | |
29e4013d TH |
1504 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1505 | dm_rq_bio_constructor, tio); | |
1506 | if (r) | |
1507 | return r; | |
cec47e3d | 1508 | |
29e4013d TH |
1509 | clone->cmd = rq->cmd; |
1510 | clone->cmd_len = rq->cmd_len; | |
1511 | clone->sense = rq->sense; | |
1512 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1513 | clone->end_io = end_clone_request; |
1514 | clone->end_io_data = tio; | |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
6facdaff KU |
1519 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1520 | gfp_t gfp_mask) | |
1521 | { | |
1522 | struct request *clone; | |
1523 | struct dm_rq_target_io *tio; | |
1524 | ||
1525 | tio = alloc_rq_tio(md, gfp_mask); | |
1526 | if (!tio) | |
1527 | return NULL; | |
1528 | ||
1529 | tio->md = md; | |
1530 | tio->ti = NULL; | |
1531 | tio->orig = rq; | |
1532 | tio->error = 0; | |
1533 | memset(&tio->info, 0, sizeof(tio->info)); | |
1534 | ||
1535 | clone = &tio->clone; | |
1536 | if (setup_clone(clone, rq, tio)) { | |
1537 | /* -ENOMEM */ | |
1538 | free_rq_tio(tio); | |
1539 | return NULL; | |
1540 | } | |
1541 | ||
1542 | return clone; | |
1543 | } | |
1544 | ||
cec47e3d KU |
1545 | /* |
1546 | * Called with the queue lock held. | |
1547 | */ | |
1548 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1549 | { | |
1550 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1551 | struct request *clone; |
1552 | ||
cec47e3d KU |
1553 | if (unlikely(rq->special)) { |
1554 | DMWARN("Already has something in rq->special."); | |
1555 | return BLKPREP_KILL; | |
1556 | } | |
1557 | ||
6facdaff KU |
1558 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1559 | if (!clone) | |
cec47e3d | 1560 | return BLKPREP_DEFER; |
cec47e3d KU |
1561 | |
1562 | rq->special = clone; | |
1563 | rq->cmd_flags |= REQ_DONTPREP; | |
1564 | ||
1565 | return BLKPREP_OK; | |
1566 | } | |
1567 | ||
9eef87da KU |
1568 | /* |
1569 | * Returns: | |
1570 | * 0 : the request has been processed (not requeued) | |
1571 | * !0 : the request has been requeued | |
1572 | */ | |
1573 | static int map_request(struct dm_target *ti, struct request *clone, | |
1574 | struct mapped_device *md) | |
cec47e3d | 1575 | { |
9eef87da | 1576 | int r, requeued = 0; |
cec47e3d KU |
1577 | struct dm_rq_target_io *tio = clone->end_io_data; |
1578 | ||
1579 | /* | |
1580 | * Hold the md reference here for the in-flight I/O. | |
1581 | * We can't rely on the reference count by device opener, | |
1582 | * because the device may be closed during the request completion | |
1583 | * when all bios are completed. | |
1584 | * See the comment in rq_completed() too. | |
1585 | */ | |
1586 | dm_get(md); | |
1587 | ||
1588 | tio->ti = ti; | |
1589 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1590 | switch (r) { | |
1591 | case DM_MAPIO_SUBMITTED: | |
1592 | /* The target has taken the I/O to submit by itself later */ | |
1593 | break; | |
1594 | case DM_MAPIO_REMAPPED: | |
1595 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1596 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1597 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1598 | dm_dispatch_request(clone); |
1599 | break; | |
1600 | case DM_MAPIO_REQUEUE: | |
1601 | /* The target wants to requeue the I/O */ | |
1602 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1603 | requeued = 1; |
cec47e3d KU |
1604 | break; |
1605 | default: | |
1606 | if (r > 0) { | |
1607 | DMWARN("unimplemented target map return value: %d", r); | |
1608 | BUG(); | |
1609 | } | |
1610 | ||
1611 | /* The target wants to complete the I/O */ | |
1612 | dm_kill_unmapped_request(clone, r); | |
1613 | break; | |
1614 | } | |
9eef87da KU |
1615 | |
1616 | return requeued; | |
cec47e3d KU |
1617 | } |
1618 | ||
1619 | /* | |
1620 | * q->request_fn for request-based dm. | |
1621 | * Called with the queue lock held. | |
1622 | */ | |
1623 | static void dm_request_fn(struct request_queue *q) | |
1624 | { | |
1625 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1626 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1627 | struct dm_target *ti; |
b4324fee | 1628 | struct request *rq, *clone; |
29e4013d | 1629 | sector_t pos; |
cec47e3d KU |
1630 | |
1631 | /* | |
b4324fee KU |
1632 | * For suspend, check blk_queue_stopped() and increment |
1633 | * ->pending within a single queue_lock not to increment the | |
1634 | * number of in-flight I/Os after the queue is stopped in | |
1635 | * dm_suspend(). | |
cec47e3d | 1636 | */ |
7eaceacc | 1637 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1638 | rq = blk_peek_request(q); |
1639 | if (!rq) | |
7eaceacc | 1640 | goto delay_and_out; |
cec47e3d | 1641 | |
29e4013d TH |
1642 | /* always use block 0 to find the target for flushes for now */ |
1643 | pos = 0; | |
1644 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1645 | pos = blk_rq_pos(rq); | |
1646 | ||
1647 | ti = dm_table_find_target(map, pos); | |
1648 | BUG_ON(!dm_target_is_valid(ti)); | |
d0bcb878 | 1649 | |
cec47e3d | 1650 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1651 | goto delay_and_out; |
cec47e3d KU |
1652 | |
1653 | blk_start_request(rq); | |
b4324fee KU |
1654 | clone = rq->special; |
1655 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1656 | ||
cec47e3d | 1657 | spin_unlock(q->queue_lock); |
9eef87da KU |
1658 | if (map_request(ti, clone, md)) |
1659 | goto requeued; | |
1660 | ||
052189a2 KU |
1661 | BUG_ON(!irqs_disabled()); |
1662 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1663 | } |
1664 | ||
1665 | goto out; | |
1666 | ||
9eef87da | 1667 | requeued: |
052189a2 KU |
1668 | BUG_ON(!irqs_disabled()); |
1669 | spin_lock(q->queue_lock); | |
9eef87da | 1670 | |
7eaceacc JA |
1671 | delay_and_out: |
1672 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1673 | out: |
1674 | dm_table_put(map); | |
1675 | ||
1676 | return; | |
1677 | } | |
1678 | ||
1679 | int dm_underlying_device_busy(struct request_queue *q) | |
1680 | { | |
1681 | return blk_lld_busy(q); | |
1682 | } | |
1683 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1684 | ||
1685 | static int dm_lld_busy(struct request_queue *q) | |
1686 | { | |
1687 | int r; | |
1688 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1689 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1690 | |
1691 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1692 | r = 1; | |
1693 | else | |
1694 | r = dm_table_any_busy_target(map); | |
1695 | ||
1696 | dm_table_put(map); | |
1697 | ||
1698 | return r; | |
1699 | } | |
1700 | ||
1da177e4 LT |
1701 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1702 | { | |
8a57dfc6 CS |
1703 | int r = bdi_bits; |
1704 | struct mapped_device *md = congested_data; | |
1705 | struct dm_table *map; | |
1da177e4 | 1706 | |
1eb787ec | 1707 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1708 | map = dm_get_live_table(md); |
8a57dfc6 | 1709 | if (map) { |
cec47e3d KU |
1710 | /* |
1711 | * Request-based dm cares about only own queue for | |
1712 | * the query about congestion status of request_queue | |
1713 | */ | |
1714 | if (dm_request_based(md)) | |
1715 | r = md->queue->backing_dev_info.state & | |
1716 | bdi_bits; | |
1717 | else | |
1718 | r = dm_table_any_congested(map, bdi_bits); | |
1719 | ||
8a57dfc6 CS |
1720 | dm_table_put(map); |
1721 | } | |
1722 | } | |
1723 | ||
1da177e4 LT |
1724 | return r; |
1725 | } | |
1726 | ||
1727 | /*----------------------------------------------------------------- | |
1728 | * An IDR is used to keep track of allocated minor numbers. | |
1729 | *---------------------------------------------------------------*/ | |
2b06cfff | 1730 | static void free_minor(int minor) |
1da177e4 | 1731 | { |
f32c10b0 | 1732 | spin_lock(&_minor_lock); |
1da177e4 | 1733 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1734 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1735 | } |
1736 | ||
1737 | /* | |
1738 | * See if the device with a specific minor # is free. | |
1739 | */ | |
cf13ab8e | 1740 | static int specific_minor(int minor) |
1da177e4 LT |
1741 | { |
1742 | int r, m; | |
1743 | ||
1744 | if (minor >= (1 << MINORBITS)) | |
1745 | return -EINVAL; | |
1746 | ||
62f75c2f JM |
1747 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1748 | if (!r) | |
1749 | return -ENOMEM; | |
1750 | ||
f32c10b0 | 1751 | spin_lock(&_minor_lock); |
1da177e4 LT |
1752 | |
1753 | if (idr_find(&_minor_idr, minor)) { | |
1754 | r = -EBUSY; | |
1755 | goto out; | |
1756 | } | |
1757 | ||
ba61fdd1 | 1758 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1759 | if (r) |
1da177e4 | 1760 | goto out; |
1da177e4 LT |
1761 | |
1762 | if (m != minor) { | |
1763 | idr_remove(&_minor_idr, m); | |
1764 | r = -EBUSY; | |
1765 | goto out; | |
1766 | } | |
1767 | ||
1768 | out: | |
f32c10b0 | 1769 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1770 | return r; |
1771 | } | |
1772 | ||
cf13ab8e | 1773 | static int next_free_minor(int *minor) |
1da177e4 | 1774 | { |
2b06cfff | 1775 | int r, m; |
1da177e4 | 1776 | |
1da177e4 | 1777 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1778 | if (!r) |
1779 | return -ENOMEM; | |
1780 | ||
f32c10b0 | 1781 | spin_lock(&_minor_lock); |
1da177e4 | 1782 | |
ba61fdd1 | 1783 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1784 | if (r) |
1da177e4 | 1785 | goto out; |
1da177e4 LT |
1786 | |
1787 | if (m >= (1 << MINORBITS)) { | |
1788 | idr_remove(&_minor_idr, m); | |
1789 | r = -ENOSPC; | |
1790 | goto out; | |
1791 | } | |
1792 | ||
1793 | *minor = m; | |
1794 | ||
1795 | out: | |
f32c10b0 | 1796 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1797 | return r; |
1798 | } | |
1799 | ||
83d5cde4 | 1800 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1801 | |
53d5914f MP |
1802 | static void dm_wq_work(struct work_struct *work); |
1803 | ||
4a0b4ddf MS |
1804 | static void dm_init_md_queue(struct mapped_device *md) |
1805 | { | |
1806 | /* | |
1807 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1808 | * devices. The type of this dm device has not been decided yet. | |
1809 | * The type is decided at the first table loading time. | |
1810 | * To prevent problematic device stacking, clear the queue flag | |
1811 | * for request stacking support until then. | |
1812 | * | |
1813 | * This queue is new, so no concurrency on the queue_flags. | |
1814 | */ | |
1815 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1816 | ||
1817 | md->queue->queuedata = md; | |
1818 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1819 | md->queue->backing_dev_info.congested_data = md; | |
1820 | blk_queue_make_request(md->queue, dm_request); | |
1821 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1822 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1823 | } | |
1824 | ||
1da177e4 LT |
1825 | /* |
1826 | * Allocate and initialise a blank device with a given minor. | |
1827 | */ | |
2b06cfff | 1828 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1829 | { |
1830 | int r; | |
cf13ab8e | 1831 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1832 | void *old_md; |
1da177e4 LT |
1833 | |
1834 | if (!md) { | |
1835 | DMWARN("unable to allocate device, out of memory."); | |
1836 | return NULL; | |
1837 | } | |
1838 | ||
10da4f79 | 1839 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1840 | goto bad_module_get; |
10da4f79 | 1841 | |
1da177e4 | 1842 | /* get a minor number for the dev */ |
2b06cfff | 1843 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1844 | r = next_free_minor(&minor); |
2b06cfff | 1845 | else |
cf13ab8e | 1846 | r = specific_minor(minor); |
1da177e4 | 1847 | if (r < 0) |
6ed7ade8 | 1848 | goto bad_minor; |
1da177e4 | 1849 | |
a5664dad | 1850 | md->type = DM_TYPE_NONE; |
2ca3310e | 1851 | init_rwsem(&md->io_lock); |
e61290a4 | 1852 | mutex_init(&md->suspend_lock); |
a5664dad | 1853 | mutex_init(&md->type_lock); |
022c2611 | 1854 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1855 | rwlock_init(&md->map_lock); |
1856 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1857 | atomic_set(&md->open_count, 0); |
1da177e4 | 1858 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1859 | atomic_set(&md->uevent_seq, 0); |
1860 | INIT_LIST_HEAD(&md->uevent_list); | |
1861 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1862 | |
4a0b4ddf | 1863 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1864 | if (!md->queue) |
6ed7ade8 | 1865 | goto bad_queue; |
1da177e4 | 1866 | |
4a0b4ddf | 1867 | dm_init_md_queue(md); |
9faf400f | 1868 | |
1da177e4 LT |
1869 | md->disk = alloc_disk(1); |
1870 | if (!md->disk) | |
6ed7ade8 | 1871 | goto bad_disk; |
1da177e4 | 1872 | |
316d315b NK |
1873 | atomic_set(&md->pending[0], 0); |
1874 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1875 | init_waitqueue_head(&md->wait); |
53d5914f | 1876 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1877 | init_waitqueue_head(&md->eventq); |
1878 | ||
1da177e4 LT |
1879 | md->disk->major = _major; |
1880 | md->disk->first_minor = minor; | |
1881 | md->disk->fops = &dm_blk_dops; | |
1882 | md->disk->queue = md->queue; | |
1883 | md->disk->private_data = md; | |
1884 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1885 | add_disk(md->disk); | |
7e51f257 | 1886 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1887 | |
9c4376de TH |
1888 | md->wq = alloc_workqueue("kdmflush", |
1889 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1890 | if (!md->wq) |
1891 | goto bad_thread; | |
1892 | ||
32a926da MP |
1893 | md->bdev = bdget_disk(md->disk, 0); |
1894 | if (!md->bdev) | |
1895 | goto bad_bdev; | |
1896 | ||
6a8736d1 TH |
1897 | bio_init(&md->flush_bio); |
1898 | md->flush_bio.bi_bdev = md->bdev; | |
1899 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1900 | ||
ba61fdd1 | 1901 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1902 | spin_lock(&_minor_lock); |
ba61fdd1 | 1903 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1904 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1905 | |
1906 | BUG_ON(old_md != MINOR_ALLOCED); | |
1907 | ||
1da177e4 LT |
1908 | return md; |
1909 | ||
32a926da MP |
1910 | bad_bdev: |
1911 | destroy_workqueue(md->wq); | |
304f3f6a | 1912 | bad_thread: |
03022c54 | 1913 | del_gendisk(md->disk); |
304f3f6a | 1914 | put_disk(md->disk); |
6ed7ade8 | 1915 | bad_disk: |
1312f40e | 1916 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1917 | bad_queue: |
1da177e4 | 1918 | free_minor(minor); |
6ed7ade8 | 1919 | bad_minor: |
10da4f79 | 1920 | module_put(THIS_MODULE); |
6ed7ade8 | 1921 | bad_module_get: |
1da177e4 LT |
1922 | kfree(md); |
1923 | return NULL; | |
1924 | } | |
1925 | ||
ae9da83f JN |
1926 | static void unlock_fs(struct mapped_device *md); |
1927 | ||
1da177e4 LT |
1928 | static void free_dev(struct mapped_device *md) |
1929 | { | |
f331c029 | 1930 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1931 | |
32a926da MP |
1932 | unlock_fs(md); |
1933 | bdput(md->bdev); | |
304f3f6a | 1934 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1935 | if (md->tio_pool) |
1936 | mempool_destroy(md->tio_pool); | |
1937 | if (md->io_pool) | |
1938 | mempool_destroy(md->io_pool); | |
1939 | if (md->bs) | |
1940 | bioset_free(md->bs); | |
9c47008d | 1941 | blk_integrity_unregister(md->disk); |
1da177e4 | 1942 | del_gendisk(md->disk); |
63d94e48 | 1943 | free_minor(minor); |
fba9f90e JM |
1944 | |
1945 | spin_lock(&_minor_lock); | |
1946 | md->disk->private_data = NULL; | |
1947 | spin_unlock(&_minor_lock); | |
1948 | ||
1da177e4 | 1949 | put_disk(md->disk); |
1312f40e | 1950 | blk_cleanup_queue(md->queue); |
10da4f79 | 1951 | module_put(THIS_MODULE); |
1da177e4 LT |
1952 | kfree(md); |
1953 | } | |
1954 | ||
e6ee8c0b KU |
1955 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1956 | { | |
1957 | struct dm_md_mempools *p; | |
1958 | ||
1959 | if (md->io_pool && md->tio_pool && md->bs) | |
1960 | /* the md already has necessary mempools */ | |
1961 | goto out; | |
1962 | ||
1963 | p = dm_table_get_md_mempools(t); | |
1964 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1965 | ||
1966 | md->io_pool = p->io_pool; | |
1967 | p->io_pool = NULL; | |
1968 | md->tio_pool = p->tio_pool; | |
1969 | p->tio_pool = NULL; | |
1970 | md->bs = p->bs; | |
1971 | p->bs = NULL; | |
1972 | ||
1973 | out: | |
1974 | /* mempool bind completed, now no need any mempools in the table */ | |
1975 | dm_table_free_md_mempools(t); | |
1976 | } | |
1977 | ||
1da177e4 LT |
1978 | /* |
1979 | * Bind a table to the device. | |
1980 | */ | |
1981 | static void event_callback(void *context) | |
1982 | { | |
7a8c3d3b MA |
1983 | unsigned long flags; |
1984 | LIST_HEAD(uevents); | |
1da177e4 LT |
1985 | struct mapped_device *md = (struct mapped_device *) context; |
1986 | ||
7a8c3d3b MA |
1987 | spin_lock_irqsave(&md->uevent_lock, flags); |
1988 | list_splice_init(&md->uevent_list, &uevents); | |
1989 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1990 | ||
ed9e1982 | 1991 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1992 | |
1da177e4 LT |
1993 | atomic_inc(&md->event_nr); |
1994 | wake_up(&md->eventq); | |
1995 | } | |
1996 | ||
c217649b MS |
1997 | /* |
1998 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1999 | */ | |
4e90188b | 2000 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2001 | { |
4e90188b | 2002 | set_capacity(md->disk, size); |
1da177e4 | 2003 | |
db8fef4f | 2004 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2005 | } |
2006 | ||
d5b9dd04 MP |
2007 | /* |
2008 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2009 | * | |
2010 | * If this function returns 0, then the device is either a non-dm | |
2011 | * device without a merge_bvec_fn, or it is a dm device that is | |
2012 | * able to split any bios it receives that are too big. | |
2013 | */ | |
2014 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2015 | { | |
2016 | struct mapped_device *dev_md; | |
2017 | ||
2018 | if (!q->merge_bvec_fn) | |
2019 | return 0; | |
2020 | ||
2021 | if (q->make_request_fn == dm_request) { | |
2022 | dev_md = q->queuedata; | |
2023 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2024 | return 0; | |
2025 | } | |
2026 | ||
2027 | return 1; | |
2028 | } | |
2029 | ||
2030 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2031 | struct dm_dev *dev, sector_t start, | |
2032 | sector_t len, void *data) | |
2033 | { | |
2034 | struct block_device *bdev = dev->bdev; | |
2035 | struct request_queue *q = bdev_get_queue(bdev); | |
2036 | ||
2037 | return dm_queue_merge_is_compulsory(q); | |
2038 | } | |
2039 | ||
2040 | /* | |
2041 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2042 | * on the properties of the underlying devices. | |
2043 | */ | |
2044 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2045 | { | |
2046 | unsigned i = 0; | |
2047 | struct dm_target *ti; | |
2048 | ||
2049 | while (i < dm_table_get_num_targets(table)) { | |
2050 | ti = dm_table_get_target(table, i++); | |
2051 | ||
2052 | if (ti->type->iterate_devices && | |
2053 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2054 | return 0; | |
2055 | } | |
2056 | ||
2057 | return 1; | |
2058 | } | |
2059 | ||
042d2a9b AK |
2060 | /* |
2061 | * Returns old map, which caller must destroy. | |
2062 | */ | |
2063 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2064 | struct queue_limits *limits) | |
1da177e4 | 2065 | { |
042d2a9b | 2066 | struct dm_table *old_map; |
165125e1 | 2067 | struct request_queue *q = md->queue; |
1da177e4 | 2068 | sector_t size; |
523d9297 | 2069 | unsigned long flags; |
d5b9dd04 | 2070 | int merge_is_optional; |
1da177e4 LT |
2071 | |
2072 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2073 | |
2074 | /* | |
2075 | * Wipe any geometry if the size of the table changed. | |
2076 | */ | |
2077 | if (size != get_capacity(md->disk)) | |
2078 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2079 | ||
32a926da | 2080 | __set_size(md, size); |
d5816876 | 2081 | |
2ca3310e AK |
2082 | dm_table_event_callback(t, event_callback, md); |
2083 | ||
e6ee8c0b KU |
2084 | /* |
2085 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2086 | * for request-based during suspension. So stop it to prevent | |
2087 | * I/O mapping before resume. | |
2088 | * This must be done before setting the queue restrictions, | |
2089 | * because request-based dm may be run just after the setting. | |
2090 | */ | |
2091 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2092 | stop_queue(q); | |
2093 | ||
2094 | __bind_mempools(md, t); | |
2095 | ||
d5b9dd04 MP |
2096 | merge_is_optional = dm_table_merge_is_optional(t); |
2097 | ||
523d9297 | 2098 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2099 | old_map = md->map; |
1da177e4 | 2100 | md->map = t; |
36a0456f AK |
2101 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2102 | ||
754c5fc7 | 2103 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2104 | if (merge_is_optional) |
2105 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2106 | else | |
2107 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2108 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2109 | |
042d2a9b | 2110 | return old_map; |
1da177e4 LT |
2111 | } |
2112 | ||
a7940155 AK |
2113 | /* |
2114 | * Returns unbound table for the caller to free. | |
2115 | */ | |
2116 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2117 | { |
2118 | struct dm_table *map = md->map; | |
523d9297 | 2119 | unsigned long flags; |
1da177e4 LT |
2120 | |
2121 | if (!map) | |
a7940155 | 2122 | return NULL; |
1da177e4 LT |
2123 | |
2124 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2125 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2126 | md->map = NULL; |
523d9297 | 2127 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2128 | |
2129 | return map; | |
1da177e4 LT |
2130 | } |
2131 | ||
2132 | /* | |
2133 | * Constructor for a new device. | |
2134 | */ | |
2b06cfff | 2135 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2136 | { |
2137 | struct mapped_device *md; | |
2138 | ||
2b06cfff | 2139 | md = alloc_dev(minor); |
1da177e4 LT |
2140 | if (!md) |
2141 | return -ENXIO; | |
2142 | ||
784aae73 MB |
2143 | dm_sysfs_init(md); |
2144 | ||
1da177e4 LT |
2145 | *result = md; |
2146 | return 0; | |
2147 | } | |
2148 | ||
a5664dad MS |
2149 | /* |
2150 | * Functions to manage md->type. | |
2151 | * All are required to hold md->type_lock. | |
2152 | */ | |
2153 | void dm_lock_md_type(struct mapped_device *md) | |
2154 | { | |
2155 | mutex_lock(&md->type_lock); | |
2156 | } | |
2157 | ||
2158 | void dm_unlock_md_type(struct mapped_device *md) | |
2159 | { | |
2160 | mutex_unlock(&md->type_lock); | |
2161 | } | |
2162 | ||
2163 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2164 | { | |
2165 | md->type = type; | |
2166 | } | |
2167 | ||
2168 | unsigned dm_get_md_type(struct mapped_device *md) | |
2169 | { | |
2170 | return md->type; | |
2171 | } | |
2172 | ||
36a0456f AK |
2173 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2174 | { | |
2175 | return md->immutable_target_type; | |
2176 | } | |
2177 | ||
4a0b4ddf MS |
2178 | /* |
2179 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2180 | */ | |
2181 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2182 | { | |
2183 | struct request_queue *q = NULL; | |
2184 | ||
2185 | if (md->queue->elevator) | |
2186 | return 1; | |
2187 | ||
2188 | /* Fully initialize the queue */ | |
2189 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2190 | if (!q) | |
2191 | return 0; | |
2192 | ||
2193 | md->queue = q; | |
2194 | md->saved_make_request_fn = md->queue->make_request_fn; | |
2195 | dm_init_md_queue(md); | |
2196 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2197 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2198 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2199 | |
2200 | elv_register_queue(md->queue); | |
2201 | ||
2202 | return 1; | |
2203 | } | |
2204 | ||
2205 | /* | |
2206 | * Setup the DM device's queue based on md's type | |
2207 | */ | |
2208 | int dm_setup_md_queue(struct mapped_device *md) | |
2209 | { | |
2210 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2211 | !dm_init_request_based_queue(md)) { | |
2212 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2213 | return -EINVAL; | |
2214 | } | |
2215 | ||
2216 | return 0; | |
2217 | } | |
2218 | ||
637842cf | 2219 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2220 | { |
2221 | struct mapped_device *md; | |
1da177e4 LT |
2222 | unsigned minor = MINOR(dev); |
2223 | ||
2224 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2225 | return NULL; | |
2226 | ||
f32c10b0 | 2227 | spin_lock(&_minor_lock); |
1da177e4 LT |
2228 | |
2229 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2230 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2231 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2232 | dm_deleting_md(md) || |
17b2f66f | 2233 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2234 | md = NULL; |
fba9f90e JM |
2235 | goto out; |
2236 | } | |
1da177e4 | 2237 | |
fba9f90e | 2238 | out: |
f32c10b0 | 2239 | spin_unlock(&_minor_lock); |
1da177e4 | 2240 | |
637842cf DT |
2241 | return md; |
2242 | } | |
2243 | ||
d229a958 DT |
2244 | struct mapped_device *dm_get_md(dev_t dev) |
2245 | { | |
2246 | struct mapped_device *md = dm_find_md(dev); | |
2247 | ||
2248 | if (md) | |
2249 | dm_get(md); | |
2250 | ||
2251 | return md; | |
2252 | } | |
2253 | ||
9ade92a9 | 2254 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2255 | { |
9ade92a9 | 2256 | return md->interface_ptr; |
1da177e4 LT |
2257 | } |
2258 | ||
2259 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2260 | { | |
2261 | md->interface_ptr = ptr; | |
2262 | } | |
2263 | ||
2264 | void dm_get(struct mapped_device *md) | |
2265 | { | |
2266 | atomic_inc(&md->holders); | |
3f77316d | 2267 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2268 | } |
2269 | ||
72d94861 AK |
2270 | const char *dm_device_name(struct mapped_device *md) |
2271 | { | |
2272 | return md->name; | |
2273 | } | |
2274 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2275 | ||
3f77316d | 2276 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2277 | { |
1134e5ae | 2278 | struct dm_table *map; |
1da177e4 | 2279 | |
3f77316d | 2280 | might_sleep(); |
fba9f90e | 2281 | |
3f77316d KU |
2282 | spin_lock(&_minor_lock); |
2283 | map = dm_get_live_table(md); | |
2284 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2285 | set_bit(DMF_FREEING, &md->flags); | |
2286 | spin_unlock(&_minor_lock); | |
2287 | ||
2288 | if (!dm_suspended_md(md)) { | |
2289 | dm_table_presuspend_targets(map); | |
2290 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2291 | } |
3f77316d KU |
2292 | |
2293 | /* | |
2294 | * Rare, but there may be I/O requests still going to complete, | |
2295 | * for example. Wait for all references to disappear. | |
2296 | * No one should increment the reference count of the mapped_device, | |
2297 | * after the mapped_device state becomes DMF_FREEING. | |
2298 | */ | |
2299 | if (wait) | |
2300 | while (atomic_read(&md->holders)) | |
2301 | msleep(1); | |
2302 | else if (atomic_read(&md->holders)) | |
2303 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2304 | dm_device_name(md), atomic_read(&md->holders)); | |
2305 | ||
2306 | dm_sysfs_exit(md); | |
2307 | dm_table_put(map); | |
2308 | dm_table_destroy(__unbind(md)); | |
2309 | free_dev(md); | |
2310 | } | |
2311 | ||
2312 | void dm_destroy(struct mapped_device *md) | |
2313 | { | |
2314 | __dm_destroy(md, true); | |
2315 | } | |
2316 | ||
2317 | void dm_destroy_immediate(struct mapped_device *md) | |
2318 | { | |
2319 | __dm_destroy(md, false); | |
2320 | } | |
2321 | ||
2322 | void dm_put(struct mapped_device *md) | |
2323 | { | |
2324 | atomic_dec(&md->holders); | |
1da177e4 | 2325 | } |
79eb885c | 2326 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2327 | |
401600df | 2328 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2329 | { |
2330 | int r = 0; | |
b44ebeb0 MP |
2331 | DECLARE_WAITQUEUE(wait, current); |
2332 | ||
b44ebeb0 | 2333 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2334 | |
2335 | while (1) { | |
401600df | 2336 | set_current_state(interruptible); |
46125c1c | 2337 | |
b4324fee | 2338 | if (!md_in_flight(md)) |
46125c1c MB |
2339 | break; |
2340 | ||
401600df MP |
2341 | if (interruptible == TASK_INTERRUPTIBLE && |
2342 | signal_pending(current)) { | |
46125c1c MB |
2343 | r = -EINTR; |
2344 | break; | |
2345 | } | |
2346 | ||
2347 | io_schedule(); | |
2348 | } | |
2349 | set_current_state(TASK_RUNNING); | |
2350 | ||
b44ebeb0 MP |
2351 | remove_wait_queue(&md->wait, &wait); |
2352 | ||
46125c1c MB |
2353 | return r; |
2354 | } | |
2355 | ||
1da177e4 LT |
2356 | /* |
2357 | * Process the deferred bios | |
2358 | */ | |
ef208587 | 2359 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2360 | { |
ef208587 MP |
2361 | struct mapped_device *md = container_of(work, struct mapped_device, |
2362 | work); | |
6d6f10df | 2363 | struct bio *c; |
1da177e4 | 2364 | |
6a8736d1 | 2365 | down_read(&md->io_lock); |
ef208587 | 2366 | |
3b00b203 | 2367 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2368 | spin_lock_irq(&md->deferred_lock); |
2369 | c = bio_list_pop(&md->deferred); | |
2370 | spin_unlock_irq(&md->deferred_lock); | |
2371 | ||
6a8736d1 | 2372 | if (!c) |
df12ee99 | 2373 | break; |
022c2611 | 2374 | |
6a8736d1 | 2375 | up_read(&md->io_lock); |
3b00b203 | 2376 | |
e6ee8c0b KU |
2377 | if (dm_request_based(md)) |
2378 | generic_make_request(c); | |
6a8736d1 TH |
2379 | else |
2380 | __split_and_process_bio(md, c); | |
3b00b203 | 2381 | |
6a8736d1 | 2382 | down_read(&md->io_lock); |
022c2611 | 2383 | } |
73d410c0 | 2384 | |
6a8736d1 | 2385 | up_read(&md->io_lock); |
1da177e4 LT |
2386 | } |
2387 | ||
9a1fb464 | 2388 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2389 | { |
3b00b203 MP |
2390 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2391 | smp_mb__after_clear_bit(); | |
53d5914f | 2392 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2393 | } |
2394 | ||
1da177e4 | 2395 | /* |
042d2a9b | 2396 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2397 | */ |
042d2a9b | 2398 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2399 | { |
042d2a9b | 2400 | struct dm_table *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2401 | struct queue_limits limits; |
042d2a9b | 2402 | int r; |
1da177e4 | 2403 | |
e61290a4 | 2404 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2405 | |
2406 | /* device must be suspended */ | |
4f186f8b | 2407 | if (!dm_suspended_md(md)) |
93c534ae | 2408 | goto out; |
1da177e4 | 2409 | |
754c5fc7 | 2410 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2411 | if (r) { |
2412 | map = ERR_PTR(r); | |
754c5fc7 | 2413 | goto out; |
042d2a9b | 2414 | } |
754c5fc7 | 2415 | |
042d2a9b | 2416 | map = __bind(md, table, &limits); |
1da177e4 | 2417 | |
93c534ae | 2418 | out: |
e61290a4 | 2419 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2420 | return map; |
1da177e4 LT |
2421 | } |
2422 | ||
2423 | /* | |
2424 | * Functions to lock and unlock any filesystem running on the | |
2425 | * device. | |
2426 | */ | |
2ca3310e | 2427 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2428 | { |
e39e2e95 | 2429 | int r; |
1da177e4 LT |
2430 | |
2431 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2432 | |
db8fef4f | 2433 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2434 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2435 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2436 | md->frozen_sb = NULL; |
2437 | return r; | |
dfbe03f6 AK |
2438 | } |
2439 | ||
aa8d7c2f AK |
2440 | set_bit(DMF_FROZEN, &md->flags); |
2441 | ||
1da177e4 LT |
2442 | return 0; |
2443 | } | |
2444 | ||
2ca3310e | 2445 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2446 | { |
aa8d7c2f AK |
2447 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2448 | return; | |
2449 | ||
db8fef4f | 2450 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2451 | md->frozen_sb = NULL; |
aa8d7c2f | 2452 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2453 | } |
2454 | ||
2455 | /* | |
2456 | * We need to be able to change a mapping table under a mounted | |
2457 | * filesystem. For example we might want to move some data in | |
2458 | * the background. Before the table can be swapped with | |
2459 | * dm_bind_table, dm_suspend must be called to flush any in | |
2460 | * flight bios and ensure that any further io gets deferred. | |
2461 | */ | |
cec47e3d KU |
2462 | /* |
2463 | * Suspend mechanism in request-based dm. | |
2464 | * | |
9f518b27 KU |
2465 | * 1. Flush all I/Os by lock_fs() if needed. |
2466 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2467 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2468 | * |
9f518b27 | 2469 | * To abort suspend, start the request_queue. |
cec47e3d | 2470 | */ |
a3d77d35 | 2471 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2472 | { |
2ca3310e | 2473 | struct dm_table *map = NULL; |
46125c1c | 2474 | int r = 0; |
a3d77d35 | 2475 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2476 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2477 | |
e61290a4 | 2478 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2479 | |
4f186f8b | 2480 | if (dm_suspended_md(md)) { |
73d410c0 | 2481 | r = -EINVAL; |
d287483d | 2482 | goto out_unlock; |
73d410c0 | 2483 | } |
1da177e4 | 2484 | |
7c666411 | 2485 | map = dm_get_live_table(md); |
1da177e4 | 2486 | |
2e93ccc1 KU |
2487 | /* |
2488 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2489 | * This flag is cleared before dm_suspend returns. | |
2490 | */ | |
2491 | if (noflush) | |
2492 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2493 | ||
cf222b37 AK |
2494 | /* This does not get reverted if there's an error later. */ |
2495 | dm_table_presuspend_targets(map); | |
2496 | ||
32a926da | 2497 | /* |
9f518b27 KU |
2498 | * Flush I/O to the device. |
2499 | * Any I/O submitted after lock_fs() may not be flushed. | |
2500 | * noflush takes precedence over do_lockfs. | |
2501 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2502 | */ |
2503 | if (!noflush && do_lockfs) { | |
2504 | r = lock_fs(md); | |
2505 | if (r) | |
f431d966 | 2506 | goto out; |
aa8d7c2f | 2507 | } |
1da177e4 LT |
2508 | |
2509 | /* | |
3b00b203 MP |
2510 | * Here we must make sure that no processes are submitting requests |
2511 | * to target drivers i.e. no one may be executing | |
2512 | * __split_and_process_bio. This is called from dm_request and | |
2513 | * dm_wq_work. | |
2514 | * | |
2515 | * To get all processes out of __split_and_process_bio in dm_request, | |
2516 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2517 | * __split_and_process_bio from dm_request and quiesce the thread |
2518 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2519 | * flush_workqueue(md->wq). | |
1da177e4 | 2520 | */ |
2ca3310e | 2521 | down_write(&md->io_lock); |
1eb787ec | 2522 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2523 | up_write(&md->io_lock); |
1da177e4 | 2524 | |
d0bcb878 | 2525 | /* |
29e4013d TH |
2526 | * Stop md->queue before flushing md->wq in case request-based |
2527 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2528 | */ |
cec47e3d | 2529 | if (dm_request_based(md)) |
9f518b27 | 2530 | stop_queue(md->queue); |
cec47e3d | 2531 | |
d0bcb878 KU |
2532 | flush_workqueue(md->wq); |
2533 | ||
1da177e4 | 2534 | /* |
3b00b203 MP |
2535 | * At this point no more requests are entering target request routines. |
2536 | * We call dm_wait_for_completion to wait for all existing requests | |
2537 | * to finish. | |
1da177e4 | 2538 | */ |
401600df | 2539 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2540 | |
2ca3310e | 2541 | down_write(&md->io_lock); |
6d6f10df | 2542 | if (noflush) |
022c2611 | 2543 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2544 | up_write(&md->io_lock); |
2e93ccc1 | 2545 | |
1da177e4 | 2546 | /* were we interrupted ? */ |
46125c1c | 2547 | if (r < 0) { |
9a1fb464 | 2548 | dm_queue_flush(md); |
73d410c0 | 2549 | |
cec47e3d | 2550 | if (dm_request_based(md)) |
9f518b27 | 2551 | start_queue(md->queue); |
cec47e3d | 2552 | |
2ca3310e | 2553 | unlock_fs(md); |
2e93ccc1 | 2554 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2555 | } |
1da177e4 | 2556 | |
3b00b203 MP |
2557 | /* |
2558 | * If dm_wait_for_completion returned 0, the device is completely | |
2559 | * quiescent now. There is no request-processing activity. All new | |
2560 | * requests are being added to md->deferred list. | |
2561 | */ | |
2562 | ||
2ca3310e | 2563 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2564 | |
4d4471cb KU |
2565 | dm_table_postsuspend_targets(map); |
2566 | ||
2ca3310e AK |
2567 | out: |
2568 | dm_table_put(map); | |
d287483d AK |
2569 | |
2570 | out_unlock: | |
e61290a4 | 2571 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2572 | return r; |
1da177e4 LT |
2573 | } |
2574 | ||
2575 | int dm_resume(struct mapped_device *md) | |
2576 | { | |
cf222b37 | 2577 | int r = -EINVAL; |
cf222b37 | 2578 | struct dm_table *map = NULL; |
1da177e4 | 2579 | |
e61290a4 | 2580 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2581 | if (!dm_suspended_md(md)) |
cf222b37 | 2582 | goto out; |
cf222b37 | 2583 | |
7c666411 | 2584 | map = dm_get_live_table(md); |
2ca3310e | 2585 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2586 | goto out; |
1da177e4 | 2587 | |
8757b776 MB |
2588 | r = dm_table_resume_targets(map); |
2589 | if (r) | |
2590 | goto out; | |
2ca3310e | 2591 | |
9a1fb464 | 2592 | dm_queue_flush(md); |
2ca3310e | 2593 | |
cec47e3d KU |
2594 | /* |
2595 | * Flushing deferred I/Os must be done after targets are resumed | |
2596 | * so that mapping of targets can work correctly. | |
2597 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2598 | */ | |
2599 | if (dm_request_based(md)) | |
2600 | start_queue(md->queue); | |
2601 | ||
2ca3310e AK |
2602 | unlock_fs(md); |
2603 | ||
2604 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2605 | ||
cf222b37 AK |
2606 | r = 0; |
2607 | out: | |
2608 | dm_table_put(map); | |
e61290a4 | 2609 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2610 | |
cf222b37 | 2611 | return r; |
1da177e4 LT |
2612 | } |
2613 | ||
2614 | /*----------------------------------------------------------------- | |
2615 | * Event notification. | |
2616 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2617 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2618 | unsigned cookie) |
69267a30 | 2619 | { |
60935eb2 MB |
2620 | char udev_cookie[DM_COOKIE_LENGTH]; |
2621 | char *envp[] = { udev_cookie, NULL }; | |
2622 | ||
2623 | if (!cookie) | |
3abf85b5 | 2624 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2625 | else { |
2626 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2627 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2628 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2629 | action, envp); | |
60935eb2 | 2630 | } |
69267a30 AK |
2631 | } |
2632 | ||
7a8c3d3b MA |
2633 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2634 | { | |
2635 | return atomic_add_return(1, &md->uevent_seq); | |
2636 | } | |
2637 | ||
1da177e4 LT |
2638 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2639 | { | |
2640 | return atomic_read(&md->event_nr); | |
2641 | } | |
2642 | ||
2643 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2644 | { | |
2645 | return wait_event_interruptible(md->eventq, | |
2646 | (event_nr != atomic_read(&md->event_nr))); | |
2647 | } | |
2648 | ||
7a8c3d3b MA |
2649 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2650 | { | |
2651 | unsigned long flags; | |
2652 | ||
2653 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2654 | list_add(elist, &md->uevent_list); | |
2655 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2656 | } | |
2657 | ||
1da177e4 LT |
2658 | /* |
2659 | * The gendisk is only valid as long as you have a reference | |
2660 | * count on 'md'. | |
2661 | */ | |
2662 | struct gendisk *dm_disk(struct mapped_device *md) | |
2663 | { | |
2664 | return md->disk; | |
2665 | } | |
2666 | ||
784aae73 MB |
2667 | struct kobject *dm_kobject(struct mapped_device *md) |
2668 | { | |
2669 | return &md->kobj; | |
2670 | } | |
2671 | ||
2672 | /* | |
2673 | * struct mapped_device should not be exported outside of dm.c | |
2674 | * so use this check to verify that kobj is part of md structure | |
2675 | */ | |
2676 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2677 | { | |
2678 | struct mapped_device *md; | |
2679 | ||
2680 | md = container_of(kobj, struct mapped_device, kobj); | |
2681 | if (&md->kobj != kobj) | |
2682 | return NULL; | |
2683 | ||
4d89b7b4 | 2684 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2685 | dm_deleting_md(md)) |
4d89b7b4 MB |
2686 | return NULL; |
2687 | ||
784aae73 MB |
2688 | dm_get(md); |
2689 | return md; | |
2690 | } | |
2691 | ||
4f186f8b | 2692 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2693 | { |
2694 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2695 | } | |
2696 | ||
64dbce58 KU |
2697 | int dm_suspended(struct dm_target *ti) |
2698 | { | |
ecdb2e25 | 2699 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2700 | } |
2701 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2702 | ||
2e93ccc1 KU |
2703 | int dm_noflush_suspending(struct dm_target *ti) |
2704 | { | |
ecdb2e25 | 2705 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2706 | } |
2707 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2708 | ||
a91a2785 | 2709 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity) |
e6ee8c0b KU |
2710 | { |
2711 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2712 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2713 | |
2714 | if (!pools) | |
2715 | return NULL; | |
2716 | ||
2717 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2718 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2719 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2720 | if (!pools->io_pool) | |
2721 | goto free_pools_and_out; | |
2722 | ||
2723 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2724 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2725 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2726 | if (!pools->tio_pool) | |
2727 | goto free_io_pool_and_out; | |
2728 | ||
a91a2785 | 2729 | pools->bs = bioset_create(pool_size, 0); |
e6ee8c0b KU |
2730 | if (!pools->bs) |
2731 | goto free_tio_pool_and_out; | |
2732 | ||
a91a2785 MP |
2733 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2734 | goto free_bioset_and_out; | |
2735 | ||
e6ee8c0b KU |
2736 | return pools; |
2737 | ||
a91a2785 MP |
2738 | free_bioset_and_out: |
2739 | bioset_free(pools->bs); | |
2740 | ||
e6ee8c0b KU |
2741 | free_tio_pool_and_out: |
2742 | mempool_destroy(pools->tio_pool); | |
2743 | ||
2744 | free_io_pool_and_out: | |
2745 | mempool_destroy(pools->io_pool); | |
2746 | ||
2747 | free_pools_and_out: | |
2748 | kfree(pools); | |
2749 | ||
2750 | return NULL; | |
2751 | } | |
2752 | ||
2753 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2754 | { | |
2755 | if (!pools) | |
2756 | return; | |
2757 | ||
2758 | if (pools->io_pool) | |
2759 | mempool_destroy(pools->io_pool); | |
2760 | ||
2761 | if (pools->tio_pool) | |
2762 | mempool_destroy(pools->tio_pool); | |
2763 | ||
2764 | if (pools->bs) | |
2765 | bioset_free(pools->bs); | |
2766 | ||
2767 | kfree(pools); | |
2768 | } | |
2769 | ||
83d5cde4 | 2770 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2771 | .open = dm_blk_open, |
2772 | .release = dm_blk_close, | |
aa129a22 | 2773 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2774 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2775 | .owner = THIS_MODULE |
2776 | }; | |
2777 | ||
2778 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2779 | ||
2780 | /* | |
2781 | * module hooks | |
2782 | */ | |
2783 | module_init(dm_init); | |
2784 | module_exit(dm_exit); | |
2785 | ||
2786 | module_param(major, uint, 0); | |
2787 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2788 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2789 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2790 | MODULE_LICENSE("GPL"); |