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 | ||
60935eb2 MB |
28 | /* |
29 | * Cookies are numeric values sent with CHANGE and REMOVE | |
30 | * uevents while resuming, removing or renaming the device. | |
31 | */ | |
32 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
33 | #define DM_COOKIE_LENGTH 24 | |
34 | ||
1da177e4 LT |
35 | static const char *_name = DM_NAME; |
36 | ||
37 | static unsigned int major = 0; | |
38 | static unsigned int _major = 0; | |
39 | ||
d15b774c AK |
40 | static DEFINE_IDR(_minor_idr); |
41 | ||
f32c10b0 | 42 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 43 | /* |
8fbf26ad | 44 | * For bio-based dm. |
1da177e4 LT |
45 | * One of these is allocated per bio. |
46 | */ | |
47 | struct dm_io { | |
48 | struct mapped_device *md; | |
49 | int error; | |
1da177e4 | 50 | atomic_t io_count; |
6ae2fa67 | 51 | struct bio *bio; |
3eaf840e | 52 | unsigned long start_time; |
f88fb981 | 53 | spinlock_t endio_lock; |
1da177e4 LT |
54 | }; |
55 | ||
56 | /* | |
8fbf26ad | 57 | * For bio-based dm. |
1da177e4 LT |
58 | * One of these is allocated per target within a bio. Hopefully |
59 | * this will be simplified out one day. | |
60 | */ | |
028867ac | 61 | struct dm_target_io { |
1da177e4 LT |
62 | struct dm_io *io; |
63 | struct dm_target *ti; | |
64 | union map_info info; | |
65 | }; | |
66 | ||
8fbf26ad KU |
67 | /* |
68 | * For request-based dm. | |
69 | * One of these is allocated per request. | |
70 | */ | |
71 | struct dm_rq_target_io { | |
72 | struct mapped_device *md; | |
73 | struct dm_target *ti; | |
74 | struct request *orig, clone; | |
75 | int error; | |
76 | union map_info info; | |
77 | }; | |
78 | ||
79 | /* | |
80 | * For request-based dm. | |
81 | * One of these is allocated per bio. | |
82 | */ | |
83 | struct dm_rq_clone_bio_info { | |
84 | struct bio *orig; | |
cec47e3d | 85 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
86 | }; |
87 | ||
1da177e4 LT |
88 | union map_info *dm_get_mapinfo(struct bio *bio) |
89 | { | |
17b2f66f | 90 | if (bio && bio->bi_private) |
028867ac | 91 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 92 | return NULL; |
1da177e4 LT |
93 | } |
94 | ||
cec47e3d KU |
95 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
96 | { | |
97 | if (rq && rq->end_io_data) | |
98 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
99 | return NULL; | |
100 | } | |
101 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
102 | ||
ba61fdd1 JM |
103 | #define MINOR_ALLOCED ((void *)-1) |
104 | ||
1da177e4 LT |
105 | /* |
106 | * Bits for the md->flags field. | |
107 | */ | |
1eb787ec | 108 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 109 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 110 | #define DMF_FROZEN 2 |
fba9f90e | 111 | #define DMF_FREEING 3 |
5c6bd75d | 112 | #define DMF_DELETING 4 |
2e93ccc1 | 113 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 114 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 115 | |
304f3f6a MB |
116 | /* |
117 | * Work processed by per-device workqueue. | |
118 | */ | |
1da177e4 | 119 | struct mapped_device { |
2ca3310e | 120 | struct rw_semaphore io_lock; |
e61290a4 | 121 | struct mutex suspend_lock; |
1da177e4 LT |
122 | rwlock_t map_lock; |
123 | atomic_t holders; | |
5c6bd75d | 124 | atomic_t open_count; |
1da177e4 LT |
125 | |
126 | unsigned long flags; | |
127 | ||
165125e1 | 128 | struct request_queue *queue; |
a5664dad | 129 | unsigned type; |
4a0b4ddf | 130 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
131 | struct mutex type_lock; |
132 | ||
1da177e4 | 133 | struct gendisk *disk; |
7e51f257 | 134 | char name[16]; |
1da177e4 LT |
135 | |
136 | void *interface_ptr; | |
137 | ||
138 | /* | |
139 | * A list of ios that arrived while we were suspended. | |
140 | */ | |
316d315b | 141 | atomic_t pending[2]; |
1da177e4 | 142 | wait_queue_head_t wait; |
53d5914f | 143 | struct work_struct work; |
74859364 | 144 | struct bio_list deferred; |
022c2611 | 145 | spinlock_t deferred_lock; |
1da177e4 | 146 | |
af7e466a | 147 | /* |
29e4013d | 148 | * Processing queue (flush) |
304f3f6a MB |
149 | */ |
150 | struct workqueue_struct *wq; | |
151 | ||
1da177e4 LT |
152 | /* |
153 | * The current mapping. | |
154 | */ | |
155 | struct dm_table *map; | |
156 | ||
157 | /* | |
158 | * io objects are allocated from here. | |
159 | */ | |
160 | mempool_t *io_pool; | |
161 | mempool_t *tio_pool; | |
162 | ||
9faf400f SB |
163 | struct bio_set *bs; |
164 | ||
1da177e4 LT |
165 | /* |
166 | * Event handling. | |
167 | */ | |
168 | atomic_t event_nr; | |
169 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
170 | atomic_t uevent_seq; |
171 | struct list_head uevent_list; | |
172 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
173 | |
174 | /* | |
175 | * freeze/thaw support require holding onto a super block | |
176 | */ | |
177 | struct super_block *frozen_sb; | |
db8fef4f | 178 | struct block_device *bdev; |
3ac51e74 DW |
179 | |
180 | /* forced geometry settings */ | |
181 | struct hd_geometry geometry; | |
784aae73 MB |
182 | |
183 | /* sysfs handle */ | |
184 | struct kobject kobj; | |
52b1fd5a | 185 | |
d87f4c14 TH |
186 | /* zero-length flush that will be cloned and submitted to targets */ |
187 | struct bio flush_bio; | |
1da177e4 LT |
188 | }; |
189 | ||
e6ee8c0b KU |
190 | /* |
191 | * For mempools pre-allocation at the table loading time. | |
192 | */ | |
193 | struct dm_md_mempools { | |
194 | mempool_t *io_pool; | |
195 | mempool_t *tio_pool; | |
196 | struct bio_set *bs; | |
197 | }; | |
198 | ||
1da177e4 | 199 | #define MIN_IOS 256 |
e18b890b CL |
200 | static struct kmem_cache *_io_cache; |
201 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
202 | static struct kmem_cache *_rq_tio_cache; |
203 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 204 | |
1da177e4 LT |
205 | static int __init local_init(void) |
206 | { | |
51157b4a | 207 | int r = -ENOMEM; |
1da177e4 | 208 | |
1da177e4 | 209 | /* allocate a slab for the dm_ios */ |
028867ac | 210 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 211 | if (!_io_cache) |
51157b4a | 212 | return r; |
1da177e4 LT |
213 | |
214 | /* allocate a slab for the target ios */ | |
028867ac | 215 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
216 | if (!_tio_cache) |
217 | goto out_free_io_cache; | |
1da177e4 | 218 | |
8fbf26ad KU |
219 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
220 | if (!_rq_tio_cache) | |
221 | goto out_free_tio_cache; | |
222 | ||
223 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
224 | if (!_rq_bio_info_cache) | |
225 | goto out_free_rq_tio_cache; | |
226 | ||
51e5b2bd | 227 | r = dm_uevent_init(); |
51157b4a | 228 | if (r) |
8fbf26ad | 229 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 230 | |
1da177e4 LT |
231 | _major = major; |
232 | r = register_blkdev(_major, _name); | |
51157b4a KU |
233 | if (r < 0) |
234 | goto out_uevent_exit; | |
1da177e4 LT |
235 | |
236 | if (!_major) | |
237 | _major = r; | |
238 | ||
239 | return 0; | |
51157b4a KU |
240 | |
241 | out_uevent_exit: | |
242 | dm_uevent_exit(); | |
8fbf26ad KU |
243 | out_free_rq_bio_info_cache: |
244 | kmem_cache_destroy(_rq_bio_info_cache); | |
245 | out_free_rq_tio_cache: | |
246 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
247 | out_free_tio_cache: |
248 | kmem_cache_destroy(_tio_cache); | |
249 | out_free_io_cache: | |
250 | kmem_cache_destroy(_io_cache); | |
251 | ||
252 | return r; | |
1da177e4 LT |
253 | } |
254 | ||
255 | static void local_exit(void) | |
256 | { | |
8fbf26ad KU |
257 | kmem_cache_destroy(_rq_bio_info_cache); |
258 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
259 | kmem_cache_destroy(_tio_cache); |
260 | kmem_cache_destroy(_io_cache); | |
00d59405 | 261 | unregister_blkdev(_major, _name); |
51e5b2bd | 262 | dm_uevent_exit(); |
1da177e4 LT |
263 | |
264 | _major = 0; | |
265 | ||
266 | DMINFO("cleaned up"); | |
267 | } | |
268 | ||
b9249e55 | 269 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
270 | local_init, |
271 | dm_target_init, | |
272 | dm_linear_init, | |
273 | dm_stripe_init, | |
952b3557 | 274 | dm_io_init, |
945fa4d2 | 275 | dm_kcopyd_init, |
1da177e4 LT |
276 | dm_interface_init, |
277 | }; | |
278 | ||
b9249e55 | 279 | static void (*_exits[])(void) = { |
1da177e4 LT |
280 | local_exit, |
281 | dm_target_exit, | |
282 | dm_linear_exit, | |
283 | dm_stripe_exit, | |
952b3557 | 284 | dm_io_exit, |
945fa4d2 | 285 | dm_kcopyd_exit, |
1da177e4 LT |
286 | dm_interface_exit, |
287 | }; | |
288 | ||
289 | static int __init dm_init(void) | |
290 | { | |
291 | const int count = ARRAY_SIZE(_inits); | |
292 | ||
293 | int r, i; | |
294 | ||
295 | for (i = 0; i < count; i++) { | |
296 | r = _inits[i](); | |
297 | if (r) | |
298 | goto bad; | |
299 | } | |
300 | ||
301 | return 0; | |
302 | ||
303 | bad: | |
304 | while (i--) | |
305 | _exits[i](); | |
306 | ||
307 | return r; | |
308 | } | |
309 | ||
310 | static void __exit dm_exit(void) | |
311 | { | |
312 | int i = ARRAY_SIZE(_exits); | |
313 | ||
314 | while (i--) | |
315 | _exits[i](); | |
d15b774c AK |
316 | |
317 | /* | |
318 | * Should be empty by this point. | |
319 | */ | |
320 | idr_remove_all(&_minor_idr); | |
321 | idr_destroy(&_minor_idr); | |
1da177e4 LT |
322 | } |
323 | ||
324 | /* | |
325 | * Block device functions | |
326 | */ | |
432a212c MA |
327 | int dm_deleting_md(struct mapped_device *md) |
328 | { | |
329 | return test_bit(DMF_DELETING, &md->flags); | |
330 | } | |
331 | ||
fe5f9f2c | 332 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
333 | { |
334 | struct mapped_device *md; | |
335 | ||
fba9f90e JM |
336 | spin_lock(&_minor_lock); |
337 | ||
fe5f9f2c | 338 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
339 | if (!md) |
340 | goto out; | |
341 | ||
5c6bd75d | 342 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 343 | dm_deleting_md(md)) { |
fba9f90e JM |
344 | md = NULL; |
345 | goto out; | |
346 | } | |
347 | ||
1da177e4 | 348 | dm_get(md); |
5c6bd75d | 349 | atomic_inc(&md->open_count); |
fba9f90e JM |
350 | |
351 | out: | |
352 | spin_unlock(&_minor_lock); | |
353 | ||
354 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
355 | } |
356 | ||
fe5f9f2c | 357 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 358 | { |
fe5f9f2c | 359 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 360 | |
4a1aeb98 MB |
361 | spin_lock(&_minor_lock); |
362 | ||
5c6bd75d | 363 | atomic_dec(&md->open_count); |
1da177e4 | 364 | dm_put(md); |
4a1aeb98 MB |
365 | |
366 | spin_unlock(&_minor_lock); | |
6e9624b8 | 367 | |
1da177e4 LT |
368 | return 0; |
369 | } | |
370 | ||
5c6bd75d AK |
371 | int dm_open_count(struct mapped_device *md) |
372 | { | |
373 | return atomic_read(&md->open_count); | |
374 | } | |
375 | ||
376 | /* | |
377 | * Guarantees nothing is using the device before it's deleted. | |
378 | */ | |
379 | int dm_lock_for_deletion(struct mapped_device *md) | |
380 | { | |
381 | int r = 0; | |
382 | ||
383 | spin_lock(&_minor_lock); | |
384 | ||
385 | if (dm_open_count(md)) | |
386 | r = -EBUSY; | |
387 | else | |
388 | set_bit(DMF_DELETING, &md->flags); | |
389 | ||
390 | spin_unlock(&_minor_lock); | |
391 | ||
392 | return r; | |
393 | } | |
394 | ||
3ac51e74 DW |
395 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
396 | { | |
397 | struct mapped_device *md = bdev->bd_disk->private_data; | |
398 | ||
399 | return dm_get_geometry(md, geo); | |
400 | } | |
401 | ||
fe5f9f2c | 402 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
403 | unsigned int cmd, unsigned long arg) |
404 | { | |
fe5f9f2c | 405 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 406 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
407 | struct dm_target *tgt; |
408 | int r = -ENOTTY; | |
409 | ||
aa129a22 MB |
410 | if (!map || !dm_table_get_size(map)) |
411 | goto out; | |
412 | ||
413 | /* We only support devices that have a single target */ | |
414 | if (dm_table_get_num_targets(map) != 1) | |
415 | goto out; | |
416 | ||
417 | tgt = dm_table_get_target(map, 0); | |
418 | ||
4f186f8b | 419 | if (dm_suspended_md(md)) { |
aa129a22 MB |
420 | r = -EAGAIN; |
421 | goto out; | |
422 | } | |
423 | ||
424 | if (tgt->type->ioctl) | |
647b3d00 | 425 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
426 | |
427 | out: | |
428 | dm_table_put(map); | |
429 | ||
aa129a22 MB |
430 | return r; |
431 | } | |
432 | ||
028867ac | 433 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
434 | { |
435 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
436 | } | |
437 | ||
028867ac | 438 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
439 | { |
440 | mempool_free(io, md->io_pool); | |
441 | } | |
442 | ||
028867ac | 443 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
444 | { |
445 | mempool_free(tio, md->tio_pool); | |
446 | } | |
447 | ||
08885643 KU |
448 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
449 | gfp_t gfp_mask) | |
cec47e3d | 450 | { |
08885643 | 451 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
452 | } |
453 | ||
454 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
455 | { | |
456 | mempool_free(tio, tio->md->tio_pool); | |
457 | } | |
458 | ||
459 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
460 | { | |
461 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
462 | } | |
463 | ||
464 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
465 | { | |
466 | mempool_free(info, info->tio->md->io_pool); | |
467 | } | |
468 | ||
90abb8c4 KU |
469 | static int md_in_flight(struct mapped_device *md) |
470 | { | |
471 | return atomic_read(&md->pending[READ]) + | |
472 | atomic_read(&md->pending[WRITE]); | |
473 | } | |
474 | ||
3eaf840e JNN |
475 | static void start_io_acct(struct dm_io *io) |
476 | { | |
477 | struct mapped_device *md = io->md; | |
c9959059 | 478 | int cpu; |
316d315b | 479 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
480 | |
481 | io->start_time = jiffies; | |
482 | ||
074a7aca TH |
483 | cpu = part_stat_lock(); |
484 | part_round_stats(cpu, &dm_disk(md)->part0); | |
485 | part_stat_unlock(); | |
1e9bb880 SL |
486 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
487 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
488 | } |
489 | ||
d221d2e7 | 490 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
491 | { |
492 | struct mapped_device *md = io->md; | |
493 | struct bio *bio = io->bio; | |
494 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 495 | int pending, cpu; |
3eaf840e JNN |
496 | int rw = bio_data_dir(bio); |
497 | ||
074a7aca TH |
498 | cpu = part_stat_lock(); |
499 | part_round_stats(cpu, &dm_disk(md)->part0); | |
500 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
501 | part_stat_unlock(); | |
3eaf840e | 502 | |
af7e466a MP |
503 | /* |
504 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 505 | * a flush. |
af7e466a | 506 | */ |
1e9bb880 SL |
507 | pending = atomic_dec_return(&md->pending[rw]); |
508 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 509 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 510 | |
d221d2e7 MP |
511 | /* nudge anyone waiting on suspend queue */ |
512 | if (!pending) | |
513 | wake_up(&md->wait); | |
3eaf840e JNN |
514 | } |
515 | ||
1da177e4 LT |
516 | /* |
517 | * Add the bio to the list of deferred io. | |
518 | */ | |
92c63902 | 519 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 520 | { |
05447420 | 521 | unsigned long flags; |
1da177e4 | 522 | |
05447420 | 523 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 524 | bio_list_add(&md->deferred, bio); |
05447420 | 525 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 526 | queue_work(md->wq, &md->work); |
1da177e4 LT |
527 | } |
528 | ||
529 | /* | |
530 | * Everyone (including functions in this file), should use this | |
531 | * function to access the md->map field, and make sure they call | |
532 | * dm_table_put() when finished. | |
533 | */ | |
7c666411 | 534 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
535 | { |
536 | struct dm_table *t; | |
523d9297 | 537 | unsigned long flags; |
1da177e4 | 538 | |
523d9297 | 539 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
540 | t = md->map; |
541 | if (t) | |
542 | dm_table_get(t); | |
523d9297 | 543 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
544 | |
545 | return t; | |
546 | } | |
547 | ||
3ac51e74 DW |
548 | /* |
549 | * Get the geometry associated with a dm device | |
550 | */ | |
551 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
552 | { | |
553 | *geo = md->geometry; | |
554 | ||
555 | return 0; | |
556 | } | |
557 | ||
558 | /* | |
559 | * Set the geometry of a device. | |
560 | */ | |
561 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
562 | { | |
563 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
564 | ||
565 | if (geo->start > sz) { | |
566 | DMWARN("Start sector is beyond the geometry limits."); | |
567 | return -EINVAL; | |
568 | } | |
569 | ||
570 | md->geometry = *geo; | |
571 | ||
572 | return 0; | |
573 | } | |
574 | ||
1da177e4 LT |
575 | /*----------------------------------------------------------------- |
576 | * CRUD START: | |
577 | * A more elegant soln is in the works that uses the queue | |
578 | * merge fn, unfortunately there are a couple of changes to | |
579 | * the block layer that I want to make for this. So in the | |
580 | * interests of getting something for people to use I give | |
581 | * you this clearly demarcated crap. | |
582 | *---------------------------------------------------------------*/ | |
583 | ||
2e93ccc1 KU |
584 | static int __noflush_suspending(struct mapped_device *md) |
585 | { | |
586 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
587 | } | |
588 | ||
1da177e4 LT |
589 | /* |
590 | * Decrements the number of outstanding ios that a bio has been | |
591 | * cloned into, completing the original io if necc. | |
592 | */ | |
858119e1 | 593 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 594 | { |
2e93ccc1 | 595 | unsigned long flags; |
b35f8caa MB |
596 | int io_error; |
597 | struct bio *bio; | |
598 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
599 | |
600 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
601 | if (unlikely(error)) { |
602 | spin_lock_irqsave(&io->endio_lock, flags); | |
603 | if (!(io->error > 0 && __noflush_suspending(md))) | |
604 | io->error = error; | |
605 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
606 | } | |
1da177e4 LT |
607 | |
608 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
609 | if (io->error == DM_ENDIO_REQUEUE) { |
610 | /* | |
611 | * Target requested pushing back the I/O. | |
2e93ccc1 | 612 | */ |
022c2611 | 613 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
614 | if (__noflush_suspending(md)) |
615 | bio_list_add_head(&md->deferred, io->bio); | |
616 | else | |
2e93ccc1 KU |
617 | /* noflush suspend was interrupted. */ |
618 | io->error = -EIO; | |
022c2611 | 619 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
620 | } |
621 | ||
b35f8caa MB |
622 | io_error = io->error; |
623 | bio = io->bio; | |
6a8736d1 TH |
624 | end_io_acct(io); |
625 | free_io(md, io); | |
626 | ||
627 | if (io_error == DM_ENDIO_REQUEUE) | |
628 | return; | |
2e93ccc1 | 629 | |
b372d360 | 630 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 631 | /* |
6a8736d1 TH |
632 | * Preflush done for flush with data, reissue |
633 | * without REQ_FLUSH. | |
af7e466a | 634 | */ |
6a8736d1 TH |
635 | bio->bi_rw &= ~REQ_FLUSH; |
636 | queue_io(md, bio); | |
af7e466a | 637 | } else { |
b372d360 | 638 | /* done with normal IO or empty flush */ |
b7908c10 | 639 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 640 | bio_endio(bio, io_error); |
b35f8caa | 641 | } |
1da177e4 LT |
642 | } |
643 | } | |
644 | ||
6712ecf8 | 645 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
646 | { |
647 | int r = 0; | |
028867ac | 648 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 649 | struct dm_io *io = tio->io; |
9faf400f | 650 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
651 | dm_endio_fn endio = tio->ti->type->end_io; |
652 | ||
1da177e4 LT |
653 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
654 | error = -EIO; | |
655 | ||
656 | if (endio) { | |
657 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
658 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
659 | /* | |
660 | * error and requeue request are handled | |
661 | * in dec_pending(). | |
662 | */ | |
1da177e4 | 663 | error = r; |
45cbcd79 KU |
664 | else if (r == DM_ENDIO_INCOMPLETE) |
665 | /* The target will handle the io */ | |
6712ecf8 | 666 | return; |
45cbcd79 KU |
667 | else if (r) { |
668 | DMWARN("unimplemented target endio return value: %d", r); | |
669 | BUG(); | |
670 | } | |
1da177e4 LT |
671 | } |
672 | ||
9faf400f SB |
673 | /* |
674 | * Store md for cleanup instead of tio which is about to get freed. | |
675 | */ | |
676 | bio->bi_private = md->bs; | |
677 | ||
9faf400f | 678 | free_tio(md, tio); |
b35f8caa MB |
679 | bio_put(bio); |
680 | dec_pending(io, error); | |
1da177e4 LT |
681 | } |
682 | ||
cec47e3d KU |
683 | /* |
684 | * Partial completion handling for request-based dm | |
685 | */ | |
686 | static void end_clone_bio(struct bio *clone, int error) | |
687 | { | |
688 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
689 | struct dm_rq_target_io *tio = info->tio; | |
690 | struct bio *bio = info->orig; | |
691 | unsigned int nr_bytes = info->orig->bi_size; | |
692 | ||
693 | bio_put(clone); | |
694 | ||
695 | if (tio->error) | |
696 | /* | |
697 | * An error has already been detected on the request. | |
698 | * Once error occurred, just let clone->end_io() handle | |
699 | * the remainder. | |
700 | */ | |
701 | return; | |
702 | else if (error) { | |
703 | /* | |
704 | * Don't notice the error to the upper layer yet. | |
705 | * The error handling decision is made by the target driver, | |
706 | * when the request is completed. | |
707 | */ | |
708 | tio->error = error; | |
709 | return; | |
710 | } | |
711 | ||
712 | /* | |
713 | * I/O for the bio successfully completed. | |
714 | * Notice the data completion to the upper layer. | |
715 | */ | |
716 | ||
717 | /* | |
718 | * bios are processed from the head of the list. | |
719 | * So the completing bio should always be rq->bio. | |
720 | * If it's not, something wrong is happening. | |
721 | */ | |
722 | if (tio->orig->bio != bio) | |
723 | DMERR("bio completion is going in the middle of the request"); | |
724 | ||
725 | /* | |
726 | * Update the original request. | |
727 | * Do not use blk_end_request() here, because it may complete | |
728 | * the original request before the clone, and break the ordering. | |
729 | */ | |
730 | blk_update_request(tio->orig, 0, nr_bytes); | |
731 | } | |
732 | ||
733 | /* | |
734 | * Don't touch any member of the md after calling this function because | |
735 | * the md may be freed in dm_put() at the end of this function. | |
736 | * Or do dm_get() before calling this function and dm_put() later. | |
737 | */ | |
b4324fee | 738 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 739 | { |
b4324fee | 740 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
741 | |
742 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 743 | if (!md_in_flight(md)) |
cec47e3d KU |
744 | wake_up(&md->wait); |
745 | ||
746 | if (run_queue) | |
b4324fee | 747 | blk_run_queue(md->queue); |
cec47e3d KU |
748 | |
749 | /* | |
750 | * dm_put() must be at the end of this function. See the comment above | |
751 | */ | |
752 | dm_put(md); | |
753 | } | |
754 | ||
a77e28c7 KU |
755 | static void free_rq_clone(struct request *clone) |
756 | { | |
757 | struct dm_rq_target_io *tio = clone->end_io_data; | |
758 | ||
759 | blk_rq_unprep_clone(clone); | |
760 | free_rq_tio(tio); | |
761 | } | |
762 | ||
980691e5 KU |
763 | /* |
764 | * Complete the clone and the original request. | |
765 | * Must be called without queue lock. | |
766 | */ | |
767 | static void dm_end_request(struct request *clone, int error) | |
768 | { | |
769 | int rw = rq_data_dir(clone); | |
770 | struct dm_rq_target_io *tio = clone->end_io_data; | |
771 | struct mapped_device *md = tio->md; | |
772 | struct request *rq = tio->orig; | |
773 | ||
29e4013d | 774 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
775 | rq->errors = clone->errors; |
776 | rq->resid_len = clone->resid_len; | |
777 | ||
778 | if (rq->sense) | |
779 | /* | |
780 | * We are using the sense buffer of the original | |
781 | * request. | |
782 | * So setting the length of the sense data is enough. | |
783 | */ | |
784 | rq->sense_len = clone->sense_len; | |
785 | } | |
786 | ||
787 | free_rq_clone(clone); | |
29e4013d TH |
788 | blk_end_request_all(rq, error); |
789 | rq_completed(md, rw, true); | |
980691e5 KU |
790 | } |
791 | ||
cec47e3d KU |
792 | static void dm_unprep_request(struct request *rq) |
793 | { | |
794 | struct request *clone = rq->special; | |
cec47e3d KU |
795 | |
796 | rq->special = NULL; | |
797 | rq->cmd_flags &= ~REQ_DONTPREP; | |
798 | ||
a77e28c7 | 799 | free_rq_clone(clone); |
cec47e3d KU |
800 | } |
801 | ||
802 | /* | |
803 | * Requeue the original request of a clone. | |
804 | */ | |
805 | void dm_requeue_unmapped_request(struct request *clone) | |
806 | { | |
b4324fee | 807 | int rw = rq_data_dir(clone); |
cec47e3d KU |
808 | struct dm_rq_target_io *tio = clone->end_io_data; |
809 | struct mapped_device *md = tio->md; | |
810 | struct request *rq = tio->orig; | |
811 | struct request_queue *q = rq->q; | |
812 | unsigned long flags; | |
813 | ||
814 | dm_unprep_request(rq); | |
815 | ||
816 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
817 | blk_requeue_request(q, rq); |
818 | spin_unlock_irqrestore(q->queue_lock, flags); | |
819 | ||
b4324fee | 820 | rq_completed(md, rw, 0); |
cec47e3d KU |
821 | } |
822 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
823 | ||
824 | static void __stop_queue(struct request_queue *q) | |
825 | { | |
826 | blk_stop_queue(q); | |
827 | } | |
828 | ||
829 | static void stop_queue(struct request_queue *q) | |
830 | { | |
831 | unsigned long flags; | |
832 | ||
833 | spin_lock_irqsave(q->queue_lock, flags); | |
834 | __stop_queue(q); | |
835 | spin_unlock_irqrestore(q->queue_lock, flags); | |
836 | } | |
837 | ||
838 | static void __start_queue(struct request_queue *q) | |
839 | { | |
840 | if (blk_queue_stopped(q)) | |
841 | blk_start_queue(q); | |
842 | } | |
843 | ||
844 | static void start_queue(struct request_queue *q) | |
845 | { | |
846 | unsigned long flags; | |
847 | ||
848 | spin_lock_irqsave(q->queue_lock, flags); | |
849 | __start_queue(q); | |
850 | spin_unlock_irqrestore(q->queue_lock, flags); | |
851 | } | |
852 | ||
11a68244 | 853 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 854 | { |
11a68244 | 855 | int r = error; |
cec47e3d KU |
856 | struct dm_rq_target_io *tio = clone->end_io_data; |
857 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
cec47e3d | 858 | |
11a68244 KU |
859 | if (mapped && rq_end_io) |
860 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
cec47e3d | 861 | |
11a68244 | 862 | if (r <= 0) |
cec47e3d | 863 | /* The target wants to complete the I/O */ |
11a68244 KU |
864 | dm_end_request(clone, r); |
865 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
866 | /* The target will handle the I/O */ |
867 | return; | |
11a68244 | 868 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
869 | /* The target wants to requeue the I/O */ |
870 | dm_requeue_unmapped_request(clone); | |
871 | else { | |
11a68244 | 872 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
873 | BUG(); |
874 | } | |
875 | } | |
876 | ||
11a68244 KU |
877 | /* |
878 | * Request completion handler for request-based dm | |
879 | */ | |
880 | static void dm_softirq_done(struct request *rq) | |
881 | { | |
882 | bool mapped = true; | |
883 | struct request *clone = rq->completion_data; | |
884 | struct dm_rq_target_io *tio = clone->end_io_data; | |
885 | ||
886 | if (rq->cmd_flags & REQ_FAILED) | |
887 | mapped = false; | |
888 | ||
889 | dm_done(clone, tio->error, mapped); | |
890 | } | |
891 | ||
cec47e3d KU |
892 | /* |
893 | * Complete the clone and the original request with the error status | |
894 | * through softirq context. | |
895 | */ | |
896 | static void dm_complete_request(struct request *clone, int error) | |
897 | { | |
898 | struct dm_rq_target_io *tio = clone->end_io_data; | |
899 | struct request *rq = tio->orig; | |
900 | ||
901 | tio->error = error; | |
902 | rq->completion_data = clone; | |
903 | blk_complete_request(rq); | |
904 | } | |
905 | ||
906 | /* | |
907 | * Complete the not-mapped clone and the original request with the error status | |
908 | * through softirq context. | |
909 | * Target's rq_end_io() function isn't called. | |
910 | * This may be used when the target's map_rq() function fails. | |
911 | */ | |
912 | void dm_kill_unmapped_request(struct request *clone, int error) | |
913 | { | |
914 | struct dm_rq_target_io *tio = clone->end_io_data; | |
915 | struct request *rq = tio->orig; | |
916 | ||
917 | rq->cmd_flags |= REQ_FAILED; | |
918 | dm_complete_request(clone, error); | |
919 | } | |
920 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
921 | ||
922 | /* | |
923 | * Called with the queue lock held | |
924 | */ | |
925 | static void end_clone_request(struct request *clone, int error) | |
926 | { | |
927 | /* | |
928 | * For just cleaning up the information of the queue in which | |
929 | * the clone was dispatched. | |
930 | * The clone is *NOT* freed actually here because it is alloced from | |
931 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
932 | */ | |
933 | __blk_put_request(clone->q, clone); | |
934 | ||
935 | /* | |
936 | * Actual request completion is done in a softirq context which doesn't | |
937 | * hold the queue lock. Otherwise, deadlock could occur because: | |
938 | * - another request may be submitted by the upper level driver | |
939 | * of the stacking during the completion | |
940 | * - the submission which requires queue lock may be done | |
941 | * against this queue | |
942 | */ | |
943 | dm_complete_request(clone, error); | |
944 | } | |
945 | ||
56a67df7 MS |
946 | /* |
947 | * Return maximum size of I/O possible at the supplied sector up to the current | |
948 | * target boundary. | |
949 | */ | |
950 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
951 | { | |
952 | sector_t target_offset = dm_target_offset(ti, sector); | |
953 | ||
954 | return ti->len - target_offset; | |
955 | } | |
956 | ||
957 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 958 | { |
56a67df7 | 959 | sector_t len = max_io_len_target_boundary(sector, ti); |
1da177e4 LT |
960 | |
961 | /* | |
962 | * Does the target need to split even further ? | |
963 | */ | |
964 | if (ti->split_io) { | |
965 | sector_t boundary; | |
56a67df7 | 966 | sector_t offset = dm_target_offset(ti, sector); |
1da177e4 LT |
967 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) |
968 | - offset; | |
969 | if (len > boundary) | |
970 | len = boundary; | |
971 | } | |
972 | ||
973 | return len; | |
974 | } | |
975 | ||
976 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 977 | struct dm_target_io *tio) |
1da177e4 LT |
978 | { |
979 | int r; | |
2056a782 | 980 | sector_t sector; |
9faf400f | 981 | struct mapped_device *md; |
1da177e4 | 982 | |
1da177e4 LT |
983 | clone->bi_end_io = clone_endio; |
984 | clone->bi_private = tio; | |
985 | ||
986 | /* | |
987 | * Map the clone. If r == 0 we don't need to do | |
988 | * anything, the target has assumed ownership of | |
989 | * this io. | |
990 | */ | |
991 | atomic_inc(&tio->io->io_count); | |
2056a782 | 992 | sector = clone->bi_sector; |
1da177e4 | 993 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 994 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 995 | /* the bio has been remapped so dispatch it */ |
2056a782 | 996 | |
d07335e5 MS |
997 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
998 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 999 | |
1da177e4 | 1000 | generic_make_request(clone); |
2e93ccc1 KU |
1001 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1002 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1003 | md = tio->io->md; |
1004 | dec_pending(tio->io, r); | |
1005 | /* | |
1006 | * Store bio_set for cleanup. | |
1007 | */ | |
1008 | clone->bi_private = md->bs; | |
1da177e4 | 1009 | bio_put(clone); |
9faf400f | 1010 | free_tio(md, tio); |
45cbcd79 KU |
1011 | } else if (r) { |
1012 | DMWARN("unimplemented target map return value: %d", r); | |
1013 | BUG(); | |
1da177e4 LT |
1014 | } |
1015 | } | |
1016 | ||
1017 | struct clone_info { | |
1018 | struct mapped_device *md; | |
1019 | struct dm_table *map; | |
1020 | struct bio *bio; | |
1021 | struct dm_io *io; | |
1022 | sector_t sector; | |
1023 | sector_t sector_count; | |
1024 | unsigned short idx; | |
1025 | }; | |
1026 | ||
3676347a PO |
1027 | static void dm_bio_destructor(struct bio *bio) |
1028 | { | |
9faf400f SB |
1029 | struct bio_set *bs = bio->bi_private; |
1030 | ||
1031 | bio_free(bio, bs); | |
3676347a PO |
1032 | } |
1033 | ||
1da177e4 | 1034 | /* |
d87f4c14 | 1035 | * Creates a little bio that just does part of a bvec. |
1da177e4 LT |
1036 | */ |
1037 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1038 | unsigned short idx, unsigned int offset, | |
9faf400f | 1039 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1040 | { |
1041 | struct bio *clone; | |
1042 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1043 | ||
9faf400f | 1044 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1045 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1046 | *clone->bi_io_vec = *bv; |
1047 | ||
1048 | clone->bi_sector = sector; | |
1049 | clone->bi_bdev = bio->bi_bdev; | |
d87f4c14 | 1050 | clone->bi_rw = bio->bi_rw; |
1da177e4 LT |
1051 | clone->bi_vcnt = 1; |
1052 | clone->bi_size = to_bytes(len); | |
1053 | clone->bi_io_vec->bv_offset = offset; | |
1054 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1055 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1056 | |
9c47008d | 1057 | if (bio_integrity(bio)) { |
7878cba9 | 1058 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1059 | bio_integrity_trim(clone, |
1060 | bio_sector_offset(bio, idx, offset), len); | |
1061 | } | |
1062 | ||
1da177e4 LT |
1063 | return clone; |
1064 | } | |
1065 | ||
1066 | /* | |
1067 | * Creates a bio that consists of range of complete bvecs. | |
1068 | */ | |
1069 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1070 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1071 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1072 | { |
1073 | struct bio *clone; | |
1074 | ||
9faf400f SB |
1075 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1076 | __bio_clone(clone, bio); | |
1077 | clone->bi_destructor = dm_bio_destructor; | |
1da177e4 LT |
1078 | clone->bi_sector = sector; |
1079 | clone->bi_idx = idx; | |
1080 | clone->bi_vcnt = idx + bv_count; | |
1081 | clone->bi_size = to_bytes(len); | |
1082 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1083 | ||
9c47008d | 1084 | if (bio_integrity(bio)) { |
7878cba9 | 1085 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1086 | |
1087 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1088 | bio_integrity_trim(clone, | |
1089 | bio_sector_offset(bio, idx, 0), len); | |
1090 | } | |
1091 | ||
1da177e4 LT |
1092 | return clone; |
1093 | } | |
1094 | ||
9015df24 AK |
1095 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1096 | struct dm_target *ti) | |
f9ab94ce | 1097 | { |
9015df24 | 1098 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1099 | |
1100 | tio->io = ci->io; | |
1101 | tio->ti = ti; | |
f9ab94ce | 1102 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1103 | |
1104 | return tio; | |
1105 | } | |
1106 | ||
06a426ce | 1107 | static void __issue_target_request(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1108 | unsigned request_nr, sector_t len) |
9015df24 AK |
1109 | { |
1110 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1111 | struct bio *clone; | |
1112 | ||
57cba5d3 | 1113 | tio->info.target_request_nr = request_nr; |
f9ab94ce | 1114 | |
06a426ce MS |
1115 | /* |
1116 | * Discard requests require the bio's inline iovecs be initialized. | |
1117 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1118 | * and discard, so no need for concern about wasted bvec allocations. | |
1119 | */ | |
1120 | clone = bio_alloc_bioset(GFP_NOIO, ci->bio->bi_max_vecs, ci->md->bs); | |
f9ab94ce MP |
1121 | __bio_clone(clone, ci->bio); |
1122 | clone->bi_destructor = dm_bio_destructor; | |
a79245b3 MS |
1123 | if (len) { |
1124 | clone->bi_sector = ci->sector; | |
1125 | clone->bi_size = to_bytes(len); | |
1126 | } | |
f9ab94ce MP |
1127 | |
1128 | __map_bio(ti, clone, tio); | |
1129 | } | |
1130 | ||
06a426ce | 1131 | static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1132 | unsigned num_requests, sector_t len) |
06a426ce MS |
1133 | { |
1134 | unsigned request_nr; | |
1135 | ||
1136 | for (request_nr = 0; request_nr < num_requests; request_nr++) | |
a79245b3 | 1137 | __issue_target_request(ci, ti, request_nr, len); |
06a426ce MS |
1138 | } |
1139 | ||
b372d360 | 1140 | static int __clone_and_map_empty_flush(struct clone_info *ci) |
f9ab94ce | 1141 | { |
06a426ce | 1142 | unsigned target_nr = 0; |
f9ab94ce MP |
1143 | struct dm_target *ti; |
1144 | ||
b372d360 | 1145 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1146 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
a79245b3 | 1147 | __issue_target_requests(ci, ti, ti->num_flush_requests, 0); |
f9ab94ce | 1148 | |
f9ab94ce MP |
1149 | return 0; |
1150 | } | |
1151 | ||
5ae89a87 MS |
1152 | /* |
1153 | * Perform all io with a single clone. | |
1154 | */ | |
1155 | static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti) | |
1156 | { | |
1157 | struct bio *clone, *bio = ci->bio; | |
1158 | struct dm_target_io *tio; | |
1159 | ||
1160 | tio = alloc_tio(ci, ti); | |
1161 | clone = clone_bio(bio, ci->sector, ci->idx, | |
1162 | bio->bi_vcnt - ci->idx, ci->sector_count, | |
1163 | ci->md->bs); | |
1164 | __map_bio(ti, clone, tio); | |
1165 | ci->sector_count = 0; | |
1166 | } | |
1167 | ||
1168 | static int __clone_and_map_discard(struct clone_info *ci) | |
1169 | { | |
1170 | struct dm_target *ti; | |
a79245b3 | 1171 | sector_t len; |
5ae89a87 | 1172 | |
a79245b3 MS |
1173 | do { |
1174 | ti = dm_table_find_target(ci->map, ci->sector); | |
1175 | if (!dm_target_is_valid(ti)) | |
1176 | return -EIO; | |
5ae89a87 | 1177 | |
5ae89a87 | 1178 | /* |
a79245b3 | 1179 | * Even though the device advertised discard support, |
936688d7 MS |
1180 | * that does not mean every target supports it, and |
1181 | * reconfiguration might also have changed that since the | |
a79245b3 | 1182 | * check was performed. |
5ae89a87 | 1183 | */ |
a79245b3 MS |
1184 | if (!ti->num_discard_requests) |
1185 | return -EOPNOTSUPP; | |
5ae89a87 | 1186 | |
a79245b3 | 1187 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
06a426ce | 1188 | |
a79245b3 MS |
1189 | __issue_target_requests(ci, ti, ti->num_discard_requests, len); |
1190 | ||
1191 | ci->sector += len; | |
1192 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1193 | |
1194 | return 0; | |
1195 | } | |
1196 | ||
512875bd | 1197 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1198 | { |
1199 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1200 | struct dm_target *ti; |
1201 | sector_t len = 0, max; | |
028867ac | 1202 | struct dm_target_io *tio; |
1da177e4 | 1203 | |
5ae89a87 MS |
1204 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
1205 | return __clone_and_map_discard(ci); | |
1206 | ||
512875bd JN |
1207 | ti = dm_table_find_target(ci->map, ci->sector); |
1208 | if (!dm_target_is_valid(ti)) | |
1209 | return -EIO; | |
1210 | ||
56a67df7 | 1211 | max = max_io_len(ci->sector, ti); |
512875bd | 1212 | |
1da177e4 LT |
1213 | if (ci->sector_count <= max) { |
1214 | /* | |
1215 | * Optimise for the simple case where we can do all of | |
1216 | * the remaining io with a single clone. | |
1217 | */ | |
5ae89a87 | 1218 | __clone_and_map_simple(ci, ti); |
1da177e4 LT |
1219 | |
1220 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1221 | /* | |
1222 | * There are some bvecs that don't span targets. | |
1223 | * Do as many of these as possible. | |
1224 | */ | |
1225 | int i; | |
1226 | sector_t remaining = max; | |
1227 | sector_t bv_len; | |
1228 | ||
1229 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1230 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1231 | ||
1232 | if (bv_len > remaining) | |
1233 | break; | |
1234 | ||
1235 | remaining -= bv_len; | |
1236 | len += bv_len; | |
1237 | } | |
1238 | ||
5ae89a87 | 1239 | tio = alloc_tio(ci, ti); |
9faf400f SB |
1240 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1241 | ci->md->bs); | |
1da177e4 LT |
1242 | __map_bio(ti, clone, tio); |
1243 | ||
1244 | ci->sector += len; | |
1245 | ci->sector_count -= len; | |
1246 | ci->idx = i; | |
1247 | ||
1248 | } else { | |
1249 | /* | |
d2044a94 | 1250 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1251 | */ |
1252 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1253 | sector_t remaining = to_sector(bv->bv_len); |
1254 | unsigned int offset = 0; | |
1da177e4 | 1255 | |
d2044a94 AK |
1256 | do { |
1257 | if (offset) { | |
1258 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1259 | if (!dm_target_is_valid(ti)) |
1260 | return -EIO; | |
1261 | ||
56a67df7 | 1262 | max = max_io_len(ci->sector, ti); |
d2044a94 AK |
1263 | } |
1264 | ||
1265 | len = min(remaining, max); | |
1266 | ||
5ae89a87 | 1267 | tio = alloc_tio(ci, ti); |
d2044a94 | 1268 | clone = split_bvec(bio, ci->sector, ci->idx, |
9faf400f SB |
1269 | bv->bv_offset + offset, len, |
1270 | ci->md->bs); | |
d2044a94 AK |
1271 | |
1272 | __map_bio(ti, clone, tio); | |
1273 | ||
1274 | ci->sector += len; | |
1275 | ci->sector_count -= len; | |
1276 | offset += to_bytes(len); | |
1277 | } while (remaining -= len); | |
1da177e4 | 1278 | |
1da177e4 LT |
1279 | ci->idx++; |
1280 | } | |
512875bd JN |
1281 | |
1282 | return 0; | |
1da177e4 LT |
1283 | } |
1284 | ||
1285 | /* | |
8a53c28d | 1286 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1287 | */ |
f0b9a450 | 1288 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1289 | { |
1290 | struct clone_info ci; | |
512875bd | 1291 | int error = 0; |
1da177e4 | 1292 | |
7c666411 | 1293 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1294 | if (unlikely(!ci.map)) { |
6a8736d1 | 1295 | bio_io_error(bio); |
f0b9a450 MP |
1296 | return; |
1297 | } | |
692d0eb9 | 1298 | |
1da177e4 | 1299 | ci.md = md; |
1da177e4 LT |
1300 | ci.io = alloc_io(md); |
1301 | ci.io->error = 0; | |
1302 | atomic_set(&ci.io->io_count, 1); | |
1303 | ci.io->bio = bio; | |
1304 | ci.io->md = md; | |
f88fb981 | 1305 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1306 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1307 | ci.idx = bio->bi_idx; |
1308 | ||
3eaf840e | 1309 | start_io_acct(ci.io); |
b372d360 MS |
1310 | if (bio->bi_rw & REQ_FLUSH) { |
1311 | ci.bio = &ci.md->flush_bio; | |
1312 | ci.sector_count = 0; | |
1313 | error = __clone_and_map_empty_flush(&ci); | |
1314 | /* dec_pending submits any data associated with flush */ | |
1315 | } else { | |
6a8736d1 | 1316 | ci.bio = bio; |
d87f4c14 | 1317 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1318 | while (ci.sector_count && !error) |
d87f4c14 | 1319 | error = __clone_and_map(&ci); |
d87f4c14 | 1320 | } |
1da177e4 LT |
1321 | |
1322 | /* drop the extra reference count */ | |
512875bd | 1323 | dec_pending(ci.io, error); |
1da177e4 LT |
1324 | dm_table_put(ci.map); |
1325 | } | |
1326 | /*----------------------------------------------------------------- | |
1327 | * CRUD END | |
1328 | *---------------------------------------------------------------*/ | |
1329 | ||
f6fccb12 MB |
1330 | static int dm_merge_bvec(struct request_queue *q, |
1331 | struct bvec_merge_data *bvm, | |
1332 | struct bio_vec *biovec) | |
1333 | { | |
1334 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1335 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1336 | struct dm_target *ti; |
1337 | sector_t max_sectors; | |
5037108a | 1338 | int max_size = 0; |
f6fccb12 MB |
1339 | |
1340 | if (unlikely(!map)) | |
5037108a | 1341 | goto out; |
f6fccb12 MB |
1342 | |
1343 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1344 | if (!dm_target_is_valid(ti)) |
1345 | goto out_table; | |
f6fccb12 MB |
1346 | |
1347 | /* | |
1348 | * Find maximum amount of I/O that won't need splitting | |
1349 | */ | |
56a67df7 | 1350 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1351 | (sector_t) BIO_MAX_SECTORS); |
1352 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1353 | if (max_size < 0) | |
1354 | max_size = 0; | |
1355 | ||
1356 | /* | |
1357 | * merge_bvec_fn() returns number of bytes | |
1358 | * it can accept at this offset | |
1359 | * max is precomputed maximal io size | |
1360 | */ | |
1361 | if (max_size && ti->type->merge) | |
1362 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1363 | /* |
1364 | * If the target doesn't support merge method and some of the devices | |
1365 | * provided their merge_bvec method (we know this by looking at | |
1366 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1367 | * entries. So always set max_size to 0, and the code below allows | |
1368 | * just one page. | |
1369 | */ | |
1370 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1371 | ||
1372 | max_size = 0; | |
f6fccb12 | 1373 | |
b01cd5ac | 1374 | out_table: |
5037108a MP |
1375 | dm_table_put(map); |
1376 | ||
1377 | out: | |
f6fccb12 MB |
1378 | /* |
1379 | * Always allow an entire first page | |
1380 | */ | |
1381 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1382 | max_size = biovec->bv_len; | |
1383 | ||
f6fccb12 MB |
1384 | return max_size; |
1385 | } | |
1386 | ||
1da177e4 LT |
1387 | /* |
1388 | * The request function that just remaps the bio built up by | |
1389 | * dm_merge_bvec. | |
1390 | */ | |
5a7bbad2 | 1391 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1392 | { |
12f03a49 | 1393 | int rw = bio_data_dir(bio); |
1da177e4 | 1394 | struct mapped_device *md = q->queuedata; |
c9959059 | 1395 | int cpu; |
1da177e4 | 1396 | |
2ca3310e | 1397 | down_read(&md->io_lock); |
1da177e4 | 1398 | |
074a7aca TH |
1399 | cpu = part_stat_lock(); |
1400 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1401 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1402 | part_stat_unlock(); | |
12f03a49 | 1403 | |
6a8736d1 TH |
1404 | /* if we're suspended, we have to queue this io for later */ |
1405 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1406 | up_read(&md->io_lock); |
1da177e4 | 1407 | |
6a8736d1 TH |
1408 | if (bio_rw(bio) != READA) |
1409 | queue_io(md, bio); | |
1410 | else | |
54d9a1b4 | 1411 | bio_io_error(bio); |
5a7bbad2 | 1412 | return; |
1da177e4 LT |
1413 | } |
1414 | ||
f0b9a450 | 1415 | __split_and_process_bio(md, bio); |
2ca3310e | 1416 | up_read(&md->io_lock); |
5a7bbad2 | 1417 | return; |
1da177e4 LT |
1418 | } |
1419 | ||
cec47e3d KU |
1420 | static int dm_request_based(struct mapped_device *md) |
1421 | { | |
1422 | return blk_queue_stackable(md->queue); | |
1423 | } | |
1424 | ||
5a7bbad2 | 1425 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1426 | { |
1427 | struct mapped_device *md = q->queuedata; | |
1428 | ||
1429 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1430 | blk_queue_bio(q, bio); |
1431 | else | |
1432 | _dm_request(q, bio); | |
cec47e3d KU |
1433 | } |
1434 | ||
1435 | void dm_dispatch_request(struct request *rq) | |
1436 | { | |
1437 | int r; | |
1438 | ||
1439 | if (blk_queue_io_stat(rq->q)) | |
1440 | rq->cmd_flags |= REQ_IO_STAT; | |
1441 | ||
1442 | rq->start_time = jiffies; | |
1443 | r = blk_insert_cloned_request(rq->q, rq); | |
1444 | if (r) | |
1445 | dm_complete_request(rq, r); | |
1446 | } | |
1447 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1448 | ||
1449 | static void dm_rq_bio_destructor(struct bio *bio) | |
1450 | { | |
1451 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1452 | struct mapped_device *md = info->tio->md; | |
1453 | ||
1454 | free_bio_info(info); | |
1455 | bio_free(bio, md->bs); | |
1456 | } | |
1457 | ||
1458 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1459 | void *data) | |
1460 | { | |
1461 | struct dm_rq_target_io *tio = data; | |
1462 | struct mapped_device *md = tio->md; | |
1463 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1464 | ||
1465 | if (!info) | |
1466 | return -ENOMEM; | |
1467 | ||
1468 | info->orig = bio_orig; | |
1469 | info->tio = tio; | |
1470 | bio->bi_end_io = end_clone_bio; | |
1471 | bio->bi_private = info; | |
1472 | bio->bi_destructor = dm_rq_bio_destructor; | |
1473 | ||
1474 | return 0; | |
1475 | } | |
1476 | ||
1477 | static int setup_clone(struct request *clone, struct request *rq, | |
1478 | struct dm_rq_target_io *tio) | |
1479 | { | |
d0bcb878 | 1480 | int r; |
cec47e3d | 1481 | |
29e4013d TH |
1482 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1483 | dm_rq_bio_constructor, tio); | |
1484 | if (r) | |
1485 | return r; | |
cec47e3d | 1486 | |
29e4013d TH |
1487 | clone->cmd = rq->cmd; |
1488 | clone->cmd_len = rq->cmd_len; | |
1489 | clone->sense = rq->sense; | |
1490 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1491 | clone->end_io = end_clone_request; |
1492 | clone->end_io_data = tio; | |
1493 | ||
1494 | return 0; | |
1495 | } | |
1496 | ||
6facdaff KU |
1497 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1498 | gfp_t gfp_mask) | |
1499 | { | |
1500 | struct request *clone; | |
1501 | struct dm_rq_target_io *tio; | |
1502 | ||
1503 | tio = alloc_rq_tio(md, gfp_mask); | |
1504 | if (!tio) | |
1505 | return NULL; | |
1506 | ||
1507 | tio->md = md; | |
1508 | tio->ti = NULL; | |
1509 | tio->orig = rq; | |
1510 | tio->error = 0; | |
1511 | memset(&tio->info, 0, sizeof(tio->info)); | |
1512 | ||
1513 | clone = &tio->clone; | |
1514 | if (setup_clone(clone, rq, tio)) { | |
1515 | /* -ENOMEM */ | |
1516 | free_rq_tio(tio); | |
1517 | return NULL; | |
1518 | } | |
1519 | ||
1520 | return clone; | |
1521 | } | |
1522 | ||
cec47e3d KU |
1523 | /* |
1524 | * Called with the queue lock held. | |
1525 | */ | |
1526 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1527 | { | |
1528 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1529 | struct request *clone; |
1530 | ||
cec47e3d KU |
1531 | if (unlikely(rq->special)) { |
1532 | DMWARN("Already has something in rq->special."); | |
1533 | return BLKPREP_KILL; | |
1534 | } | |
1535 | ||
6facdaff KU |
1536 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1537 | if (!clone) | |
cec47e3d | 1538 | return BLKPREP_DEFER; |
cec47e3d KU |
1539 | |
1540 | rq->special = clone; | |
1541 | rq->cmd_flags |= REQ_DONTPREP; | |
1542 | ||
1543 | return BLKPREP_OK; | |
1544 | } | |
1545 | ||
9eef87da KU |
1546 | /* |
1547 | * Returns: | |
1548 | * 0 : the request has been processed (not requeued) | |
1549 | * !0 : the request has been requeued | |
1550 | */ | |
1551 | static int map_request(struct dm_target *ti, struct request *clone, | |
1552 | struct mapped_device *md) | |
cec47e3d | 1553 | { |
9eef87da | 1554 | int r, requeued = 0; |
cec47e3d KU |
1555 | struct dm_rq_target_io *tio = clone->end_io_data; |
1556 | ||
1557 | /* | |
1558 | * Hold the md reference here for the in-flight I/O. | |
1559 | * We can't rely on the reference count by device opener, | |
1560 | * because the device may be closed during the request completion | |
1561 | * when all bios are completed. | |
1562 | * See the comment in rq_completed() too. | |
1563 | */ | |
1564 | dm_get(md); | |
1565 | ||
1566 | tio->ti = ti; | |
1567 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1568 | switch (r) { | |
1569 | case DM_MAPIO_SUBMITTED: | |
1570 | /* The target has taken the I/O to submit by itself later */ | |
1571 | break; | |
1572 | case DM_MAPIO_REMAPPED: | |
1573 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1574 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1575 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1576 | dm_dispatch_request(clone); |
1577 | break; | |
1578 | case DM_MAPIO_REQUEUE: | |
1579 | /* The target wants to requeue the I/O */ | |
1580 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1581 | requeued = 1; |
cec47e3d KU |
1582 | break; |
1583 | default: | |
1584 | if (r > 0) { | |
1585 | DMWARN("unimplemented target map return value: %d", r); | |
1586 | BUG(); | |
1587 | } | |
1588 | ||
1589 | /* The target wants to complete the I/O */ | |
1590 | dm_kill_unmapped_request(clone, r); | |
1591 | break; | |
1592 | } | |
9eef87da KU |
1593 | |
1594 | return requeued; | |
cec47e3d KU |
1595 | } |
1596 | ||
1597 | /* | |
1598 | * q->request_fn for request-based dm. | |
1599 | * Called with the queue lock held. | |
1600 | */ | |
1601 | static void dm_request_fn(struct request_queue *q) | |
1602 | { | |
1603 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1604 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1605 | struct dm_target *ti; |
b4324fee | 1606 | struct request *rq, *clone; |
29e4013d | 1607 | sector_t pos; |
cec47e3d KU |
1608 | |
1609 | /* | |
b4324fee KU |
1610 | * For suspend, check blk_queue_stopped() and increment |
1611 | * ->pending within a single queue_lock not to increment the | |
1612 | * number of in-flight I/Os after the queue is stopped in | |
1613 | * dm_suspend(). | |
cec47e3d | 1614 | */ |
7eaceacc | 1615 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1616 | rq = blk_peek_request(q); |
1617 | if (!rq) | |
7eaceacc | 1618 | goto delay_and_out; |
cec47e3d | 1619 | |
29e4013d TH |
1620 | /* always use block 0 to find the target for flushes for now */ |
1621 | pos = 0; | |
1622 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1623 | pos = blk_rq_pos(rq); | |
1624 | ||
1625 | ti = dm_table_find_target(map, pos); | |
1626 | BUG_ON(!dm_target_is_valid(ti)); | |
d0bcb878 | 1627 | |
cec47e3d | 1628 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1629 | goto delay_and_out; |
cec47e3d KU |
1630 | |
1631 | blk_start_request(rq); | |
b4324fee KU |
1632 | clone = rq->special; |
1633 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1634 | ||
cec47e3d | 1635 | spin_unlock(q->queue_lock); |
9eef87da KU |
1636 | if (map_request(ti, clone, md)) |
1637 | goto requeued; | |
1638 | ||
052189a2 KU |
1639 | BUG_ON(!irqs_disabled()); |
1640 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1641 | } |
1642 | ||
1643 | goto out; | |
1644 | ||
9eef87da | 1645 | requeued: |
052189a2 KU |
1646 | BUG_ON(!irqs_disabled()); |
1647 | spin_lock(q->queue_lock); | |
9eef87da | 1648 | |
7eaceacc JA |
1649 | delay_and_out: |
1650 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1651 | out: |
1652 | dm_table_put(map); | |
1653 | ||
1654 | return; | |
1655 | } | |
1656 | ||
1657 | int dm_underlying_device_busy(struct request_queue *q) | |
1658 | { | |
1659 | return blk_lld_busy(q); | |
1660 | } | |
1661 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1662 | ||
1663 | static int dm_lld_busy(struct request_queue *q) | |
1664 | { | |
1665 | int r; | |
1666 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1667 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1668 | |
1669 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1670 | r = 1; | |
1671 | else | |
1672 | r = dm_table_any_busy_target(map); | |
1673 | ||
1674 | dm_table_put(map); | |
1675 | ||
1676 | return r; | |
1677 | } | |
1678 | ||
1da177e4 LT |
1679 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1680 | { | |
8a57dfc6 CS |
1681 | int r = bdi_bits; |
1682 | struct mapped_device *md = congested_data; | |
1683 | struct dm_table *map; | |
1da177e4 | 1684 | |
1eb787ec | 1685 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1686 | map = dm_get_live_table(md); |
8a57dfc6 | 1687 | if (map) { |
cec47e3d KU |
1688 | /* |
1689 | * Request-based dm cares about only own queue for | |
1690 | * the query about congestion status of request_queue | |
1691 | */ | |
1692 | if (dm_request_based(md)) | |
1693 | r = md->queue->backing_dev_info.state & | |
1694 | bdi_bits; | |
1695 | else | |
1696 | r = dm_table_any_congested(map, bdi_bits); | |
1697 | ||
8a57dfc6 CS |
1698 | dm_table_put(map); |
1699 | } | |
1700 | } | |
1701 | ||
1da177e4 LT |
1702 | return r; |
1703 | } | |
1704 | ||
1705 | /*----------------------------------------------------------------- | |
1706 | * An IDR is used to keep track of allocated minor numbers. | |
1707 | *---------------------------------------------------------------*/ | |
2b06cfff | 1708 | static void free_minor(int minor) |
1da177e4 | 1709 | { |
f32c10b0 | 1710 | spin_lock(&_minor_lock); |
1da177e4 | 1711 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1712 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1713 | } |
1714 | ||
1715 | /* | |
1716 | * See if the device with a specific minor # is free. | |
1717 | */ | |
cf13ab8e | 1718 | static int specific_minor(int minor) |
1da177e4 LT |
1719 | { |
1720 | int r, m; | |
1721 | ||
1722 | if (minor >= (1 << MINORBITS)) | |
1723 | return -EINVAL; | |
1724 | ||
62f75c2f JM |
1725 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1726 | if (!r) | |
1727 | return -ENOMEM; | |
1728 | ||
f32c10b0 | 1729 | spin_lock(&_minor_lock); |
1da177e4 LT |
1730 | |
1731 | if (idr_find(&_minor_idr, minor)) { | |
1732 | r = -EBUSY; | |
1733 | goto out; | |
1734 | } | |
1735 | ||
ba61fdd1 | 1736 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1737 | if (r) |
1da177e4 | 1738 | goto out; |
1da177e4 LT |
1739 | |
1740 | if (m != minor) { | |
1741 | idr_remove(&_minor_idr, m); | |
1742 | r = -EBUSY; | |
1743 | goto out; | |
1744 | } | |
1745 | ||
1746 | out: | |
f32c10b0 | 1747 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1748 | return r; |
1749 | } | |
1750 | ||
cf13ab8e | 1751 | static int next_free_minor(int *minor) |
1da177e4 | 1752 | { |
2b06cfff | 1753 | int r, m; |
1da177e4 | 1754 | |
1da177e4 | 1755 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1756 | if (!r) |
1757 | return -ENOMEM; | |
1758 | ||
f32c10b0 | 1759 | spin_lock(&_minor_lock); |
1da177e4 | 1760 | |
ba61fdd1 | 1761 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1762 | if (r) |
1da177e4 | 1763 | goto out; |
1da177e4 LT |
1764 | |
1765 | if (m >= (1 << MINORBITS)) { | |
1766 | idr_remove(&_minor_idr, m); | |
1767 | r = -ENOSPC; | |
1768 | goto out; | |
1769 | } | |
1770 | ||
1771 | *minor = m; | |
1772 | ||
1773 | out: | |
f32c10b0 | 1774 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1775 | return r; |
1776 | } | |
1777 | ||
83d5cde4 | 1778 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1779 | |
53d5914f MP |
1780 | static void dm_wq_work(struct work_struct *work); |
1781 | ||
4a0b4ddf MS |
1782 | static void dm_init_md_queue(struct mapped_device *md) |
1783 | { | |
1784 | /* | |
1785 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1786 | * devices. The type of this dm device has not been decided yet. | |
1787 | * The type is decided at the first table loading time. | |
1788 | * To prevent problematic device stacking, clear the queue flag | |
1789 | * for request stacking support until then. | |
1790 | * | |
1791 | * This queue is new, so no concurrency on the queue_flags. | |
1792 | */ | |
1793 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1794 | ||
1795 | md->queue->queuedata = md; | |
1796 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1797 | md->queue->backing_dev_info.congested_data = md; | |
1798 | blk_queue_make_request(md->queue, dm_request); | |
1799 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1800 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1801 | } | |
1802 | ||
1da177e4 LT |
1803 | /* |
1804 | * Allocate and initialise a blank device with a given minor. | |
1805 | */ | |
2b06cfff | 1806 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1807 | { |
1808 | int r; | |
cf13ab8e | 1809 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1810 | void *old_md; |
1da177e4 LT |
1811 | |
1812 | if (!md) { | |
1813 | DMWARN("unable to allocate device, out of memory."); | |
1814 | return NULL; | |
1815 | } | |
1816 | ||
10da4f79 | 1817 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1818 | goto bad_module_get; |
10da4f79 | 1819 | |
1da177e4 | 1820 | /* get a minor number for the dev */ |
2b06cfff | 1821 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1822 | r = next_free_minor(&minor); |
2b06cfff | 1823 | else |
cf13ab8e | 1824 | r = specific_minor(minor); |
1da177e4 | 1825 | if (r < 0) |
6ed7ade8 | 1826 | goto bad_minor; |
1da177e4 | 1827 | |
a5664dad | 1828 | md->type = DM_TYPE_NONE; |
2ca3310e | 1829 | init_rwsem(&md->io_lock); |
e61290a4 | 1830 | mutex_init(&md->suspend_lock); |
a5664dad | 1831 | mutex_init(&md->type_lock); |
022c2611 | 1832 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1833 | rwlock_init(&md->map_lock); |
1834 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1835 | atomic_set(&md->open_count, 0); |
1da177e4 | 1836 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1837 | atomic_set(&md->uevent_seq, 0); |
1838 | INIT_LIST_HEAD(&md->uevent_list); | |
1839 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1840 | |
4a0b4ddf | 1841 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1842 | if (!md->queue) |
6ed7ade8 | 1843 | goto bad_queue; |
1da177e4 | 1844 | |
4a0b4ddf | 1845 | dm_init_md_queue(md); |
9faf400f | 1846 | |
1da177e4 LT |
1847 | md->disk = alloc_disk(1); |
1848 | if (!md->disk) | |
6ed7ade8 | 1849 | goto bad_disk; |
1da177e4 | 1850 | |
316d315b NK |
1851 | atomic_set(&md->pending[0], 0); |
1852 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1853 | init_waitqueue_head(&md->wait); |
53d5914f | 1854 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1855 | init_waitqueue_head(&md->eventq); |
1856 | ||
1da177e4 LT |
1857 | md->disk->major = _major; |
1858 | md->disk->first_minor = minor; | |
1859 | md->disk->fops = &dm_blk_dops; | |
1860 | md->disk->queue = md->queue; | |
1861 | md->disk->private_data = md; | |
1862 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1863 | add_disk(md->disk); | |
7e51f257 | 1864 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1865 | |
9c4376de TH |
1866 | md->wq = alloc_workqueue("kdmflush", |
1867 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1868 | if (!md->wq) |
1869 | goto bad_thread; | |
1870 | ||
32a926da MP |
1871 | md->bdev = bdget_disk(md->disk, 0); |
1872 | if (!md->bdev) | |
1873 | goto bad_bdev; | |
1874 | ||
6a8736d1 TH |
1875 | bio_init(&md->flush_bio); |
1876 | md->flush_bio.bi_bdev = md->bdev; | |
1877 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1878 | ||
ba61fdd1 | 1879 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1880 | spin_lock(&_minor_lock); |
ba61fdd1 | 1881 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1882 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1883 | |
1884 | BUG_ON(old_md != MINOR_ALLOCED); | |
1885 | ||
1da177e4 LT |
1886 | return md; |
1887 | ||
32a926da MP |
1888 | bad_bdev: |
1889 | destroy_workqueue(md->wq); | |
304f3f6a | 1890 | bad_thread: |
03022c54 | 1891 | del_gendisk(md->disk); |
304f3f6a | 1892 | put_disk(md->disk); |
6ed7ade8 | 1893 | bad_disk: |
1312f40e | 1894 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1895 | bad_queue: |
1da177e4 | 1896 | free_minor(minor); |
6ed7ade8 | 1897 | bad_minor: |
10da4f79 | 1898 | module_put(THIS_MODULE); |
6ed7ade8 | 1899 | bad_module_get: |
1da177e4 LT |
1900 | kfree(md); |
1901 | return NULL; | |
1902 | } | |
1903 | ||
ae9da83f JN |
1904 | static void unlock_fs(struct mapped_device *md); |
1905 | ||
1da177e4 LT |
1906 | static void free_dev(struct mapped_device *md) |
1907 | { | |
f331c029 | 1908 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1909 | |
32a926da MP |
1910 | unlock_fs(md); |
1911 | bdput(md->bdev); | |
304f3f6a | 1912 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1913 | if (md->tio_pool) |
1914 | mempool_destroy(md->tio_pool); | |
1915 | if (md->io_pool) | |
1916 | mempool_destroy(md->io_pool); | |
1917 | if (md->bs) | |
1918 | bioset_free(md->bs); | |
9c47008d | 1919 | blk_integrity_unregister(md->disk); |
1da177e4 | 1920 | del_gendisk(md->disk); |
63d94e48 | 1921 | free_minor(minor); |
fba9f90e JM |
1922 | |
1923 | spin_lock(&_minor_lock); | |
1924 | md->disk->private_data = NULL; | |
1925 | spin_unlock(&_minor_lock); | |
1926 | ||
1da177e4 | 1927 | put_disk(md->disk); |
1312f40e | 1928 | blk_cleanup_queue(md->queue); |
10da4f79 | 1929 | module_put(THIS_MODULE); |
1da177e4 LT |
1930 | kfree(md); |
1931 | } | |
1932 | ||
e6ee8c0b KU |
1933 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1934 | { | |
1935 | struct dm_md_mempools *p; | |
1936 | ||
1937 | if (md->io_pool && md->tio_pool && md->bs) | |
1938 | /* the md already has necessary mempools */ | |
1939 | goto out; | |
1940 | ||
1941 | p = dm_table_get_md_mempools(t); | |
1942 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1943 | ||
1944 | md->io_pool = p->io_pool; | |
1945 | p->io_pool = NULL; | |
1946 | md->tio_pool = p->tio_pool; | |
1947 | p->tio_pool = NULL; | |
1948 | md->bs = p->bs; | |
1949 | p->bs = NULL; | |
1950 | ||
1951 | out: | |
1952 | /* mempool bind completed, now no need any mempools in the table */ | |
1953 | dm_table_free_md_mempools(t); | |
1954 | } | |
1955 | ||
1da177e4 LT |
1956 | /* |
1957 | * Bind a table to the device. | |
1958 | */ | |
1959 | static void event_callback(void *context) | |
1960 | { | |
7a8c3d3b MA |
1961 | unsigned long flags; |
1962 | LIST_HEAD(uevents); | |
1da177e4 LT |
1963 | struct mapped_device *md = (struct mapped_device *) context; |
1964 | ||
7a8c3d3b MA |
1965 | spin_lock_irqsave(&md->uevent_lock, flags); |
1966 | list_splice_init(&md->uevent_list, &uevents); | |
1967 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1968 | ||
ed9e1982 | 1969 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1970 | |
1da177e4 LT |
1971 | atomic_inc(&md->event_nr); |
1972 | wake_up(&md->eventq); | |
1973 | } | |
1974 | ||
c217649b MS |
1975 | /* |
1976 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1977 | */ | |
4e90188b | 1978 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1979 | { |
4e90188b | 1980 | set_capacity(md->disk, size); |
1da177e4 | 1981 | |
db8fef4f | 1982 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1983 | } |
1984 | ||
d5b9dd04 MP |
1985 | /* |
1986 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
1987 | * | |
1988 | * If this function returns 0, then the device is either a non-dm | |
1989 | * device without a merge_bvec_fn, or it is a dm device that is | |
1990 | * able to split any bios it receives that are too big. | |
1991 | */ | |
1992 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
1993 | { | |
1994 | struct mapped_device *dev_md; | |
1995 | ||
1996 | if (!q->merge_bvec_fn) | |
1997 | return 0; | |
1998 | ||
1999 | if (q->make_request_fn == dm_request) { | |
2000 | dev_md = q->queuedata; | |
2001 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2002 | return 0; | |
2003 | } | |
2004 | ||
2005 | return 1; | |
2006 | } | |
2007 | ||
2008 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2009 | struct dm_dev *dev, sector_t start, | |
2010 | sector_t len, void *data) | |
2011 | { | |
2012 | struct block_device *bdev = dev->bdev; | |
2013 | struct request_queue *q = bdev_get_queue(bdev); | |
2014 | ||
2015 | return dm_queue_merge_is_compulsory(q); | |
2016 | } | |
2017 | ||
2018 | /* | |
2019 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2020 | * on the properties of the underlying devices. | |
2021 | */ | |
2022 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2023 | { | |
2024 | unsigned i = 0; | |
2025 | struct dm_target *ti; | |
2026 | ||
2027 | while (i < dm_table_get_num_targets(table)) { | |
2028 | ti = dm_table_get_target(table, i++); | |
2029 | ||
2030 | if (ti->type->iterate_devices && | |
2031 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2032 | return 0; | |
2033 | } | |
2034 | ||
2035 | return 1; | |
2036 | } | |
2037 | ||
042d2a9b AK |
2038 | /* |
2039 | * Returns old map, which caller must destroy. | |
2040 | */ | |
2041 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2042 | struct queue_limits *limits) | |
1da177e4 | 2043 | { |
042d2a9b | 2044 | struct dm_table *old_map; |
165125e1 | 2045 | struct request_queue *q = md->queue; |
1da177e4 | 2046 | sector_t size; |
523d9297 | 2047 | unsigned long flags; |
d5b9dd04 | 2048 | int merge_is_optional; |
1da177e4 LT |
2049 | |
2050 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2051 | |
2052 | /* | |
2053 | * Wipe any geometry if the size of the table changed. | |
2054 | */ | |
2055 | if (size != get_capacity(md->disk)) | |
2056 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2057 | ||
32a926da | 2058 | __set_size(md, size); |
d5816876 | 2059 | |
2ca3310e AK |
2060 | dm_table_event_callback(t, event_callback, md); |
2061 | ||
e6ee8c0b KU |
2062 | /* |
2063 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2064 | * for request-based during suspension. So stop it to prevent | |
2065 | * I/O mapping before resume. | |
2066 | * This must be done before setting the queue restrictions, | |
2067 | * because request-based dm may be run just after the setting. | |
2068 | */ | |
2069 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2070 | stop_queue(q); | |
2071 | ||
2072 | __bind_mempools(md, t); | |
2073 | ||
d5b9dd04 MP |
2074 | merge_is_optional = dm_table_merge_is_optional(t); |
2075 | ||
523d9297 | 2076 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2077 | old_map = md->map; |
1da177e4 | 2078 | md->map = t; |
754c5fc7 | 2079 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2080 | if (merge_is_optional) |
2081 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2082 | else | |
2083 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2084 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2085 | |
042d2a9b | 2086 | return old_map; |
1da177e4 LT |
2087 | } |
2088 | ||
a7940155 AK |
2089 | /* |
2090 | * Returns unbound table for the caller to free. | |
2091 | */ | |
2092 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2093 | { |
2094 | struct dm_table *map = md->map; | |
523d9297 | 2095 | unsigned long flags; |
1da177e4 LT |
2096 | |
2097 | if (!map) | |
a7940155 | 2098 | return NULL; |
1da177e4 LT |
2099 | |
2100 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2101 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2102 | md->map = NULL; |
523d9297 | 2103 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2104 | |
2105 | return map; | |
1da177e4 LT |
2106 | } |
2107 | ||
2108 | /* | |
2109 | * Constructor for a new device. | |
2110 | */ | |
2b06cfff | 2111 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2112 | { |
2113 | struct mapped_device *md; | |
2114 | ||
2b06cfff | 2115 | md = alloc_dev(minor); |
1da177e4 LT |
2116 | if (!md) |
2117 | return -ENXIO; | |
2118 | ||
784aae73 MB |
2119 | dm_sysfs_init(md); |
2120 | ||
1da177e4 LT |
2121 | *result = md; |
2122 | return 0; | |
2123 | } | |
2124 | ||
a5664dad MS |
2125 | /* |
2126 | * Functions to manage md->type. | |
2127 | * All are required to hold md->type_lock. | |
2128 | */ | |
2129 | void dm_lock_md_type(struct mapped_device *md) | |
2130 | { | |
2131 | mutex_lock(&md->type_lock); | |
2132 | } | |
2133 | ||
2134 | void dm_unlock_md_type(struct mapped_device *md) | |
2135 | { | |
2136 | mutex_unlock(&md->type_lock); | |
2137 | } | |
2138 | ||
2139 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2140 | { | |
2141 | md->type = type; | |
2142 | } | |
2143 | ||
2144 | unsigned dm_get_md_type(struct mapped_device *md) | |
2145 | { | |
2146 | return md->type; | |
2147 | } | |
2148 | ||
4a0b4ddf MS |
2149 | /* |
2150 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2151 | */ | |
2152 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2153 | { | |
2154 | struct request_queue *q = NULL; | |
2155 | ||
2156 | if (md->queue->elevator) | |
2157 | return 1; | |
2158 | ||
2159 | /* Fully initialize the queue */ | |
2160 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2161 | if (!q) | |
2162 | return 0; | |
2163 | ||
2164 | md->queue = q; | |
4a0b4ddf MS |
2165 | dm_init_md_queue(md); |
2166 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2167 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2168 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2169 | |
2170 | elv_register_queue(md->queue); | |
2171 | ||
2172 | return 1; | |
2173 | } | |
2174 | ||
2175 | /* | |
2176 | * Setup the DM device's queue based on md's type | |
2177 | */ | |
2178 | int dm_setup_md_queue(struct mapped_device *md) | |
2179 | { | |
2180 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2181 | !dm_init_request_based_queue(md)) { | |
2182 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2183 | return -EINVAL; | |
2184 | } | |
2185 | ||
2186 | return 0; | |
2187 | } | |
2188 | ||
637842cf | 2189 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2190 | { |
2191 | struct mapped_device *md; | |
1da177e4 LT |
2192 | unsigned minor = MINOR(dev); |
2193 | ||
2194 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2195 | return NULL; | |
2196 | ||
f32c10b0 | 2197 | spin_lock(&_minor_lock); |
1da177e4 LT |
2198 | |
2199 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2200 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2201 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2202 | dm_deleting_md(md) || |
17b2f66f | 2203 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2204 | md = NULL; |
fba9f90e JM |
2205 | goto out; |
2206 | } | |
1da177e4 | 2207 | |
fba9f90e | 2208 | out: |
f32c10b0 | 2209 | spin_unlock(&_minor_lock); |
1da177e4 | 2210 | |
637842cf DT |
2211 | return md; |
2212 | } | |
2213 | ||
d229a958 DT |
2214 | struct mapped_device *dm_get_md(dev_t dev) |
2215 | { | |
2216 | struct mapped_device *md = dm_find_md(dev); | |
2217 | ||
2218 | if (md) | |
2219 | dm_get(md); | |
2220 | ||
2221 | return md; | |
2222 | } | |
2223 | ||
9ade92a9 | 2224 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2225 | { |
9ade92a9 | 2226 | return md->interface_ptr; |
1da177e4 LT |
2227 | } |
2228 | ||
2229 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2230 | { | |
2231 | md->interface_ptr = ptr; | |
2232 | } | |
2233 | ||
2234 | void dm_get(struct mapped_device *md) | |
2235 | { | |
2236 | atomic_inc(&md->holders); | |
3f77316d | 2237 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2238 | } |
2239 | ||
72d94861 AK |
2240 | const char *dm_device_name(struct mapped_device *md) |
2241 | { | |
2242 | return md->name; | |
2243 | } | |
2244 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2245 | ||
3f77316d | 2246 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2247 | { |
1134e5ae | 2248 | struct dm_table *map; |
1da177e4 | 2249 | |
3f77316d | 2250 | might_sleep(); |
fba9f90e | 2251 | |
3f77316d KU |
2252 | spin_lock(&_minor_lock); |
2253 | map = dm_get_live_table(md); | |
2254 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2255 | set_bit(DMF_FREEING, &md->flags); | |
2256 | spin_unlock(&_minor_lock); | |
2257 | ||
2258 | if (!dm_suspended_md(md)) { | |
2259 | dm_table_presuspend_targets(map); | |
2260 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2261 | } |
3f77316d KU |
2262 | |
2263 | /* | |
2264 | * Rare, but there may be I/O requests still going to complete, | |
2265 | * for example. Wait for all references to disappear. | |
2266 | * No one should increment the reference count of the mapped_device, | |
2267 | * after the mapped_device state becomes DMF_FREEING. | |
2268 | */ | |
2269 | if (wait) | |
2270 | while (atomic_read(&md->holders)) | |
2271 | msleep(1); | |
2272 | else if (atomic_read(&md->holders)) | |
2273 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2274 | dm_device_name(md), atomic_read(&md->holders)); | |
2275 | ||
2276 | dm_sysfs_exit(md); | |
2277 | dm_table_put(map); | |
2278 | dm_table_destroy(__unbind(md)); | |
2279 | free_dev(md); | |
2280 | } | |
2281 | ||
2282 | void dm_destroy(struct mapped_device *md) | |
2283 | { | |
2284 | __dm_destroy(md, true); | |
2285 | } | |
2286 | ||
2287 | void dm_destroy_immediate(struct mapped_device *md) | |
2288 | { | |
2289 | __dm_destroy(md, false); | |
2290 | } | |
2291 | ||
2292 | void dm_put(struct mapped_device *md) | |
2293 | { | |
2294 | atomic_dec(&md->holders); | |
1da177e4 | 2295 | } |
79eb885c | 2296 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2297 | |
401600df | 2298 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2299 | { |
2300 | int r = 0; | |
b44ebeb0 MP |
2301 | DECLARE_WAITQUEUE(wait, current); |
2302 | ||
b44ebeb0 | 2303 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2304 | |
2305 | while (1) { | |
401600df | 2306 | set_current_state(interruptible); |
46125c1c MB |
2307 | |
2308 | smp_mb(); | |
b4324fee | 2309 | if (!md_in_flight(md)) |
46125c1c MB |
2310 | break; |
2311 | ||
401600df MP |
2312 | if (interruptible == TASK_INTERRUPTIBLE && |
2313 | signal_pending(current)) { | |
46125c1c MB |
2314 | r = -EINTR; |
2315 | break; | |
2316 | } | |
2317 | ||
2318 | io_schedule(); | |
2319 | } | |
2320 | set_current_state(TASK_RUNNING); | |
2321 | ||
b44ebeb0 MP |
2322 | remove_wait_queue(&md->wait, &wait); |
2323 | ||
46125c1c MB |
2324 | return r; |
2325 | } | |
2326 | ||
1da177e4 LT |
2327 | /* |
2328 | * Process the deferred bios | |
2329 | */ | |
ef208587 | 2330 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2331 | { |
ef208587 MP |
2332 | struct mapped_device *md = container_of(work, struct mapped_device, |
2333 | work); | |
6d6f10df | 2334 | struct bio *c; |
1da177e4 | 2335 | |
6a8736d1 | 2336 | down_read(&md->io_lock); |
ef208587 | 2337 | |
3b00b203 | 2338 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2339 | spin_lock_irq(&md->deferred_lock); |
2340 | c = bio_list_pop(&md->deferred); | |
2341 | spin_unlock_irq(&md->deferred_lock); | |
2342 | ||
6a8736d1 | 2343 | if (!c) |
df12ee99 | 2344 | break; |
022c2611 | 2345 | |
6a8736d1 | 2346 | up_read(&md->io_lock); |
3b00b203 | 2347 | |
e6ee8c0b KU |
2348 | if (dm_request_based(md)) |
2349 | generic_make_request(c); | |
6a8736d1 TH |
2350 | else |
2351 | __split_and_process_bio(md, c); | |
3b00b203 | 2352 | |
6a8736d1 | 2353 | down_read(&md->io_lock); |
022c2611 | 2354 | } |
73d410c0 | 2355 | |
6a8736d1 | 2356 | up_read(&md->io_lock); |
1da177e4 LT |
2357 | } |
2358 | ||
9a1fb464 | 2359 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2360 | { |
3b00b203 MP |
2361 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2362 | smp_mb__after_clear_bit(); | |
53d5914f | 2363 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2364 | } |
2365 | ||
1da177e4 | 2366 | /* |
042d2a9b | 2367 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2368 | */ |
042d2a9b | 2369 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2370 | { |
042d2a9b | 2371 | struct dm_table *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2372 | struct queue_limits limits; |
042d2a9b | 2373 | int r; |
1da177e4 | 2374 | |
e61290a4 | 2375 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2376 | |
2377 | /* device must be suspended */ | |
4f186f8b | 2378 | if (!dm_suspended_md(md)) |
93c534ae | 2379 | goto out; |
1da177e4 | 2380 | |
754c5fc7 | 2381 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2382 | if (r) { |
2383 | map = ERR_PTR(r); | |
754c5fc7 | 2384 | goto out; |
042d2a9b | 2385 | } |
754c5fc7 | 2386 | |
042d2a9b | 2387 | map = __bind(md, table, &limits); |
1da177e4 | 2388 | |
93c534ae | 2389 | out: |
e61290a4 | 2390 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2391 | return map; |
1da177e4 LT |
2392 | } |
2393 | ||
2394 | /* | |
2395 | * Functions to lock and unlock any filesystem running on the | |
2396 | * device. | |
2397 | */ | |
2ca3310e | 2398 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2399 | { |
e39e2e95 | 2400 | int r; |
1da177e4 LT |
2401 | |
2402 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2403 | |
db8fef4f | 2404 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2405 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2406 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2407 | md->frozen_sb = NULL; |
2408 | return r; | |
dfbe03f6 AK |
2409 | } |
2410 | ||
aa8d7c2f AK |
2411 | set_bit(DMF_FROZEN, &md->flags); |
2412 | ||
1da177e4 LT |
2413 | return 0; |
2414 | } | |
2415 | ||
2ca3310e | 2416 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2417 | { |
aa8d7c2f AK |
2418 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2419 | return; | |
2420 | ||
db8fef4f | 2421 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2422 | md->frozen_sb = NULL; |
aa8d7c2f | 2423 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2424 | } |
2425 | ||
2426 | /* | |
2427 | * We need to be able to change a mapping table under a mounted | |
2428 | * filesystem. For example we might want to move some data in | |
2429 | * the background. Before the table can be swapped with | |
2430 | * dm_bind_table, dm_suspend must be called to flush any in | |
2431 | * flight bios and ensure that any further io gets deferred. | |
2432 | */ | |
cec47e3d KU |
2433 | /* |
2434 | * Suspend mechanism in request-based dm. | |
2435 | * | |
9f518b27 KU |
2436 | * 1. Flush all I/Os by lock_fs() if needed. |
2437 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2438 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2439 | * |
9f518b27 | 2440 | * To abort suspend, start the request_queue. |
cec47e3d | 2441 | */ |
a3d77d35 | 2442 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2443 | { |
2ca3310e | 2444 | struct dm_table *map = NULL; |
46125c1c | 2445 | int r = 0; |
a3d77d35 | 2446 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2447 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2448 | |
e61290a4 | 2449 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2450 | |
4f186f8b | 2451 | if (dm_suspended_md(md)) { |
73d410c0 | 2452 | r = -EINVAL; |
d287483d | 2453 | goto out_unlock; |
73d410c0 | 2454 | } |
1da177e4 | 2455 | |
7c666411 | 2456 | map = dm_get_live_table(md); |
1da177e4 | 2457 | |
2e93ccc1 KU |
2458 | /* |
2459 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2460 | * This flag is cleared before dm_suspend returns. | |
2461 | */ | |
2462 | if (noflush) | |
2463 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2464 | ||
cf222b37 AK |
2465 | /* This does not get reverted if there's an error later. */ |
2466 | dm_table_presuspend_targets(map); | |
2467 | ||
32a926da | 2468 | /* |
9f518b27 KU |
2469 | * Flush I/O to the device. |
2470 | * Any I/O submitted after lock_fs() may not be flushed. | |
2471 | * noflush takes precedence over do_lockfs. | |
2472 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2473 | */ |
2474 | if (!noflush && do_lockfs) { | |
2475 | r = lock_fs(md); | |
2476 | if (r) | |
f431d966 | 2477 | goto out; |
aa8d7c2f | 2478 | } |
1da177e4 LT |
2479 | |
2480 | /* | |
3b00b203 MP |
2481 | * Here we must make sure that no processes are submitting requests |
2482 | * to target drivers i.e. no one may be executing | |
2483 | * __split_and_process_bio. This is called from dm_request and | |
2484 | * dm_wq_work. | |
2485 | * | |
2486 | * To get all processes out of __split_and_process_bio in dm_request, | |
2487 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2488 | * __split_and_process_bio from dm_request and quiesce the thread |
2489 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2490 | * flush_workqueue(md->wq). | |
1da177e4 | 2491 | */ |
2ca3310e | 2492 | down_write(&md->io_lock); |
1eb787ec | 2493 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2494 | up_write(&md->io_lock); |
1da177e4 | 2495 | |
d0bcb878 | 2496 | /* |
29e4013d TH |
2497 | * Stop md->queue before flushing md->wq in case request-based |
2498 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2499 | */ |
cec47e3d | 2500 | if (dm_request_based(md)) |
9f518b27 | 2501 | stop_queue(md->queue); |
cec47e3d | 2502 | |
d0bcb878 KU |
2503 | flush_workqueue(md->wq); |
2504 | ||
1da177e4 | 2505 | /* |
3b00b203 MP |
2506 | * At this point no more requests are entering target request routines. |
2507 | * We call dm_wait_for_completion to wait for all existing requests | |
2508 | * to finish. | |
1da177e4 | 2509 | */ |
401600df | 2510 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2511 | |
2ca3310e | 2512 | down_write(&md->io_lock); |
6d6f10df | 2513 | if (noflush) |
022c2611 | 2514 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2515 | up_write(&md->io_lock); |
2e93ccc1 | 2516 | |
1da177e4 | 2517 | /* were we interrupted ? */ |
46125c1c | 2518 | if (r < 0) { |
9a1fb464 | 2519 | dm_queue_flush(md); |
73d410c0 | 2520 | |
cec47e3d | 2521 | if (dm_request_based(md)) |
9f518b27 | 2522 | start_queue(md->queue); |
cec47e3d | 2523 | |
2ca3310e | 2524 | unlock_fs(md); |
2e93ccc1 | 2525 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2526 | } |
1da177e4 | 2527 | |
3b00b203 MP |
2528 | /* |
2529 | * If dm_wait_for_completion returned 0, the device is completely | |
2530 | * quiescent now. There is no request-processing activity. All new | |
2531 | * requests are being added to md->deferred list. | |
2532 | */ | |
2533 | ||
2ca3310e | 2534 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2535 | |
4d4471cb KU |
2536 | dm_table_postsuspend_targets(map); |
2537 | ||
2ca3310e AK |
2538 | out: |
2539 | dm_table_put(map); | |
d287483d AK |
2540 | |
2541 | out_unlock: | |
e61290a4 | 2542 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2543 | return r; |
1da177e4 LT |
2544 | } |
2545 | ||
2546 | int dm_resume(struct mapped_device *md) | |
2547 | { | |
cf222b37 | 2548 | int r = -EINVAL; |
cf222b37 | 2549 | struct dm_table *map = NULL; |
1da177e4 | 2550 | |
e61290a4 | 2551 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2552 | if (!dm_suspended_md(md)) |
cf222b37 | 2553 | goto out; |
cf222b37 | 2554 | |
7c666411 | 2555 | map = dm_get_live_table(md); |
2ca3310e | 2556 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2557 | goto out; |
1da177e4 | 2558 | |
8757b776 MB |
2559 | r = dm_table_resume_targets(map); |
2560 | if (r) | |
2561 | goto out; | |
2ca3310e | 2562 | |
9a1fb464 | 2563 | dm_queue_flush(md); |
2ca3310e | 2564 | |
cec47e3d KU |
2565 | /* |
2566 | * Flushing deferred I/Os must be done after targets are resumed | |
2567 | * so that mapping of targets can work correctly. | |
2568 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2569 | */ | |
2570 | if (dm_request_based(md)) | |
2571 | start_queue(md->queue); | |
2572 | ||
2ca3310e AK |
2573 | unlock_fs(md); |
2574 | ||
2575 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2576 | ||
cf222b37 AK |
2577 | r = 0; |
2578 | out: | |
2579 | dm_table_put(map); | |
e61290a4 | 2580 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2581 | |
cf222b37 | 2582 | return r; |
1da177e4 LT |
2583 | } |
2584 | ||
2585 | /*----------------------------------------------------------------- | |
2586 | * Event notification. | |
2587 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2588 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2589 | unsigned cookie) |
69267a30 | 2590 | { |
60935eb2 MB |
2591 | char udev_cookie[DM_COOKIE_LENGTH]; |
2592 | char *envp[] = { udev_cookie, NULL }; | |
2593 | ||
2594 | if (!cookie) | |
3abf85b5 | 2595 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2596 | else { |
2597 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2598 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2599 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2600 | action, envp); | |
60935eb2 | 2601 | } |
69267a30 AK |
2602 | } |
2603 | ||
7a8c3d3b MA |
2604 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2605 | { | |
2606 | return atomic_add_return(1, &md->uevent_seq); | |
2607 | } | |
2608 | ||
1da177e4 LT |
2609 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2610 | { | |
2611 | return atomic_read(&md->event_nr); | |
2612 | } | |
2613 | ||
2614 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2615 | { | |
2616 | return wait_event_interruptible(md->eventq, | |
2617 | (event_nr != atomic_read(&md->event_nr))); | |
2618 | } | |
2619 | ||
7a8c3d3b MA |
2620 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2621 | { | |
2622 | unsigned long flags; | |
2623 | ||
2624 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2625 | list_add(elist, &md->uevent_list); | |
2626 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2627 | } | |
2628 | ||
1da177e4 LT |
2629 | /* |
2630 | * The gendisk is only valid as long as you have a reference | |
2631 | * count on 'md'. | |
2632 | */ | |
2633 | struct gendisk *dm_disk(struct mapped_device *md) | |
2634 | { | |
2635 | return md->disk; | |
2636 | } | |
2637 | ||
784aae73 MB |
2638 | struct kobject *dm_kobject(struct mapped_device *md) |
2639 | { | |
2640 | return &md->kobj; | |
2641 | } | |
2642 | ||
2643 | /* | |
2644 | * struct mapped_device should not be exported outside of dm.c | |
2645 | * so use this check to verify that kobj is part of md structure | |
2646 | */ | |
2647 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2648 | { | |
2649 | struct mapped_device *md; | |
2650 | ||
2651 | md = container_of(kobj, struct mapped_device, kobj); | |
2652 | if (&md->kobj != kobj) | |
2653 | return NULL; | |
2654 | ||
4d89b7b4 | 2655 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2656 | dm_deleting_md(md)) |
4d89b7b4 MB |
2657 | return NULL; |
2658 | ||
784aae73 MB |
2659 | dm_get(md); |
2660 | return md; | |
2661 | } | |
2662 | ||
4f186f8b | 2663 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2664 | { |
2665 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2666 | } | |
2667 | ||
64dbce58 KU |
2668 | int dm_suspended(struct dm_target *ti) |
2669 | { | |
ecdb2e25 | 2670 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2671 | } |
2672 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2673 | ||
2e93ccc1 KU |
2674 | int dm_noflush_suspending(struct dm_target *ti) |
2675 | { | |
ecdb2e25 | 2676 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2677 | } |
2678 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2679 | ||
a91a2785 | 2680 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity) |
e6ee8c0b KU |
2681 | { |
2682 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2683 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2684 | |
2685 | if (!pools) | |
2686 | return NULL; | |
2687 | ||
2688 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2689 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2690 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2691 | if (!pools->io_pool) | |
2692 | goto free_pools_and_out; | |
2693 | ||
2694 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2695 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2696 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2697 | if (!pools->tio_pool) | |
2698 | goto free_io_pool_and_out; | |
2699 | ||
a91a2785 | 2700 | pools->bs = bioset_create(pool_size, 0); |
e6ee8c0b KU |
2701 | if (!pools->bs) |
2702 | goto free_tio_pool_and_out; | |
2703 | ||
a91a2785 MP |
2704 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2705 | goto free_bioset_and_out; | |
2706 | ||
e6ee8c0b KU |
2707 | return pools; |
2708 | ||
a91a2785 MP |
2709 | free_bioset_and_out: |
2710 | bioset_free(pools->bs); | |
2711 | ||
e6ee8c0b KU |
2712 | free_tio_pool_and_out: |
2713 | mempool_destroy(pools->tio_pool); | |
2714 | ||
2715 | free_io_pool_and_out: | |
2716 | mempool_destroy(pools->io_pool); | |
2717 | ||
2718 | free_pools_and_out: | |
2719 | kfree(pools); | |
2720 | ||
2721 | return NULL; | |
2722 | } | |
2723 | ||
2724 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2725 | { | |
2726 | if (!pools) | |
2727 | return; | |
2728 | ||
2729 | if (pools->io_pool) | |
2730 | mempool_destroy(pools->io_pool); | |
2731 | ||
2732 | if (pools->tio_pool) | |
2733 | mempool_destroy(pools->tio_pool); | |
2734 | ||
2735 | if (pools->bs) | |
2736 | bioset_free(pools->bs); | |
2737 | ||
2738 | kfree(pools); | |
2739 | } | |
2740 | ||
83d5cde4 | 2741 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2742 | .open = dm_blk_open, |
2743 | .release = dm_blk_close, | |
aa129a22 | 2744 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2745 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2746 | .owner = THIS_MODULE |
2747 | }; | |
2748 | ||
2749 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2750 | ||
2751 | /* | |
2752 | * module hooks | |
2753 | */ | |
2754 | module_init(dm_init); | |
2755 | module_exit(dm_exit); | |
2756 | ||
2757 | module_param(major, uint, 0); | |
2758 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2759 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2760 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2761 | MODULE_LICENSE("GPL"); |