Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
1da177e4 LT |
63 | }; |
64 | ||
65 | /* | |
8fbf26ad | 66 | * For bio-based dm. |
1da177e4 LT |
67 | * One of these is allocated per target within a bio. Hopefully |
68 | * this will be simplified out one day. | |
69 | */ | |
028867ac | 70 | struct dm_target_io { |
1da177e4 LT |
71 | struct dm_io *io; |
72 | struct dm_target *ti; | |
73 | union map_info info; | |
74 | }; | |
75 | ||
8fbf26ad KU |
76 | /* |
77 | * For request-based dm. | |
78 | * One of these is allocated per request. | |
79 | */ | |
80 | struct dm_rq_target_io { | |
81 | struct mapped_device *md; | |
82 | struct dm_target *ti; | |
83 | struct request *orig, clone; | |
84 | int error; | |
85 | union map_info info; | |
86 | }; | |
87 | ||
88 | /* | |
94818742 KO |
89 | * For request-based dm - the bio clones we allocate are embedded in these |
90 | * structs. | |
91 | * | |
92 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
93 | * the bioset is created - this means the bio has to come at the end of the | |
94 | * struct. | |
8fbf26ad KU |
95 | */ |
96 | struct dm_rq_clone_bio_info { | |
97 | struct bio *orig; | |
cec47e3d | 98 | struct dm_rq_target_io *tio; |
94818742 | 99 | struct bio clone; |
8fbf26ad KU |
100 | }; |
101 | ||
1da177e4 LT |
102 | union map_info *dm_get_mapinfo(struct bio *bio) |
103 | { | |
17b2f66f | 104 | if (bio && bio->bi_private) |
028867ac | 105 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 106 | return NULL; |
1da177e4 LT |
107 | } |
108 | ||
cec47e3d KU |
109 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
110 | { | |
111 | if (rq && rq->end_io_data) | |
112 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
113 | return NULL; | |
114 | } | |
115 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
116 | ||
ba61fdd1 JM |
117 | #define MINOR_ALLOCED ((void *)-1) |
118 | ||
1da177e4 LT |
119 | /* |
120 | * Bits for the md->flags field. | |
121 | */ | |
1eb787ec | 122 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 123 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 124 | #define DMF_FROZEN 2 |
fba9f90e | 125 | #define DMF_FREEING 3 |
5c6bd75d | 126 | #define DMF_DELETING 4 |
2e93ccc1 | 127 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 128 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 129 | |
304f3f6a MB |
130 | /* |
131 | * Work processed by per-device workqueue. | |
132 | */ | |
1da177e4 | 133 | struct mapped_device { |
2ca3310e | 134 | struct rw_semaphore io_lock; |
e61290a4 | 135 | struct mutex suspend_lock; |
1da177e4 LT |
136 | rwlock_t map_lock; |
137 | atomic_t holders; | |
5c6bd75d | 138 | atomic_t open_count; |
1da177e4 LT |
139 | |
140 | unsigned long flags; | |
141 | ||
165125e1 | 142 | struct request_queue *queue; |
a5664dad | 143 | unsigned type; |
4a0b4ddf | 144 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
145 | struct mutex type_lock; |
146 | ||
36a0456f AK |
147 | struct target_type *immutable_target_type; |
148 | ||
1da177e4 | 149 | struct gendisk *disk; |
7e51f257 | 150 | char name[16]; |
1da177e4 LT |
151 | |
152 | void *interface_ptr; | |
153 | ||
154 | /* | |
155 | * A list of ios that arrived while we were suspended. | |
156 | */ | |
316d315b | 157 | atomic_t pending[2]; |
1da177e4 | 158 | wait_queue_head_t wait; |
53d5914f | 159 | struct work_struct work; |
74859364 | 160 | struct bio_list deferred; |
022c2611 | 161 | spinlock_t deferred_lock; |
1da177e4 | 162 | |
af7e466a | 163 | /* |
29e4013d | 164 | * Processing queue (flush) |
304f3f6a MB |
165 | */ |
166 | struct workqueue_struct *wq; | |
167 | ||
1da177e4 LT |
168 | /* |
169 | * The current mapping. | |
170 | */ | |
171 | struct dm_table *map; | |
172 | ||
173 | /* | |
174 | * io objects are allocated from here. | |
175 | */ | |
176 | mempool_t *io_pool; | |
177 | mempool_t *tio_pool; | |
178 | ||
9faf400f SB |
179 | struct bio_set *bs; |
180 | ||
1da177e4 LT |
181 | /* |
182 | * Event handling. | |
183 | */ | |
184 | atomic_t event_nr; | |
185 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
186 | atomic_t uevent_seq; |
187 | struct list_head uevent_list; | |
188 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
189 | |
190 | /* | |
191 | * freeze/thaw support require holding onto a super block | |
192 | */ | |
193 | struct super_block *frozen_sb; | |
db8fef4f | 194 | struct block_device *bdev; |
3ac51e74 DW |
195 | |
196 | /* forced geometry settings */ | |
197 | struct hd_geometry geometry; | |
784aae73 MB |
198 | |
199 | /* sysfs handle */ | |
200 | struct kobject kobj; | |
52b1fd5a | 201 | |
d87f4c14 TH |
202 | /* zero-length flush that will be cloned and submitted to targets */ |
203 | struct bio flush_bio; | |
1da177e4 LT |
204 | }; |
205 | ||
e6ee8c0b KU |
206 | /* |
207 | * For mempools pre-allocation at the table loading time. | |
208 | */ | |
209 | struct dm_md_mempools { | |
210 | mempool_t *io_pool; | |
211 | mempool_t *tio_pool; | |
212 | struct bio_set *bs; | |
213 | }; | |
214 | ||
1da177e4 | 215 | #define MIN_IOS 256 |
e18b890b CL |
216 | static struct kmem_cache *_io_cache; |
217 | static struct kmem_cache *_tio_cache; | |
8fbf26ad | 218 | static struct kmem_cache *_rq_tio_cache; |
94818742 KO |
219 | |
220 | /* | |
221 | * Unused now, and needs to be deleted. But since io_pool is overloaded and it's | |
222 | * still used for _io_cache, I'm leaving this for a later cleanup | |
223 | */ | |
8fbf26ad | 224 | static struct kmem_cache *_rq_bio_info_cache; |
1da177e4 | 225 | |
1da177e4 LT |
226 | static int __init local_init(void) |
227 | { | |
51157b4a | 228 | int r = -ENOMEM; |
1da177e4 | 229 | |
1da177e4 | 230 | /* allocate a slab for the dm_ios */ |
028867ac | 231 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 232 | if (!_io_cache) |
51157b4a | 233 | return r; |
1da177e4 LT |
234 | |
235 | /* allocate a slab for the target ios */ | |
028867ac | 236 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
237 | if (!_tio_cache) |
238 | goto out_free_io_cache; | |
1da177e4 | 239 | |
8fbf26ad KU |
240 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
241 | if (!_rq_tio_cache) | |
242 | goto out_free_tio_cache; | |
243 | ||
244 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
245 | if (!_rq_bio_info_cache) | |
246 | goto out_free_rq_tio_cache; | |
247 | ||
51e5b2bd | 248 | r = dm_uevent_init(); |
51157b4a | 249 | if (r) |
8fbf26ad | 250 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 251 | |
1da177e4 LT |
252 | _major = major; |
253 | r = register_blkdev(_major, _name); | |
51157b4a KU |
254 | if (r < 0) |
255 | goto out_uevent_exit; | |
1da177e4 LT |
256 | |
257 | if (!_major) | |
258 | _major = r; | |
259 | ||
260 | return 0; | |
51157b4a KU |
261 | |
262 | out_uevent_exit: | |
263 | dm_uevent_exit(); | |
8fbf26ad KU |
264 | out_free_rq_bio_info_cache: |
265 | kmem_cache_destroy(_rq_bio_info_cache); | |
266 | out_free_rq_tio_cache: | |
267 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
268 | out_free_tio_cache: |
269 | kmem_cache_destroy(_tio_cache); | |
270 | out_free_io_cache: | |
271 | kmem_cache_destroy(_io_cache); | |
272 | ||
273 | return r; | |
1da177e4 LT |
274 | } |
275 | ||
276 | static void local_exit(void) | |
277 | { | |
8fbf26ad KU |
278 | kmem_cache_destroy(_rq_bio_info_cache); |
279 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
280 | kmem_cache_destroy(_tio_cache); |
281 | kmem_cache_destroy(_io_cache); | |
00d59405 | 282 | unregister_blkdev(_major, _name); |
51e5b2bd | 283 | dm_uevent_exit(); |
1da177e4 LT |
284 | |
285 | _major = 0; | |
286 | ||
287 | DMINFO("cleaned up"); | |
288 | } | |
289 | ||
b9249e55 | 290 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
291 | local_init, |
292 | dm_target_init, | |
293 | dm_linear_init, | |
294 | dm_stripe_init, | |
952b3557 | 295 | dm_io_init, |
945fa4d2 | 296 | dm_kcopyd_init, |
1da177e4 LT |
297 | dm_interface_init, |
298 | }; | |
299 | ||
b9249e55 | 300 | static void (*_exits[])(void) = { |
1da177e4 LT |
301 | local_exit, |
302 | dm_target_exit, | |
303 | dm_linear_exit, | |
304 | dm_stripe_exit, | |
952b3557 | 305 | dm_io_exit, |
945fa4d2 | 306 | dm_kcopyd_exit, |
1da177e4 LT |
307 | dm_interface_exit, |
308 | }; | |
309 | ||
310 | static int __init dm_init(void) | |
311 | { | |
312 | const int count = ARRAY_SIZE(_inits); | |
313 | ||
314 | int r, i; | |
315 | ||
316 | for (i = 0; i < count; i++) { | |
317 | r = _inits[i](); | |
318 | if (r) | |
319 | goto bad; | |
320 | } | |
321 | ||
322 | return 0; | |
323 | ||
324 | bad: | |
325 | while (i--) | |
326 | _exits[i](); | |
327 | ||
328 | return r; | |
329 | } | |
330 | ||
331 | static void __exit dm_exit(void) | |
332 | { | |
333 | int i = ARRAY_SIZE(_exits); | |
334 | ||
335 | while (i--) | |
336 | _exits[i](); | |
d15b774c AK |
337 | |
338 | /* | |
339 | * Should be empty by this point. | |
340 | */ | |
341 | idr_remove_all(&_minor_idr); | |
342 | idr_destroy(&_minor_idr); | |
1da177e4 LT |
343 | } |
344 | ||
345 | /* | |
346 | * Block device functions | |
347 | */ | |
432a212c MA |
348 | int dm_deleting_md(struct mapped_device *md) |
349 | { | |
350 | return test_bit(DMF_DELETING, &md->flags); | |
351 | } | |
352 | ||
fe5f9f2c | 353 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
354 | { |
355 | struct mapped_device *md; | |
356 | ||
fba9f90e JM |
357 | spin_lock(&_minor_lock); |
358 | ||
fe5f9f2c | 359 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
360 | if (!md) |
361 | goto out; | |
362 | ||
5c6bd75d | 363 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 364 | dm_deleting_md(md)) { |
fba9f90e JM |
365 | md = NULL; |
366 | goto out; | |
367 | } | |
368 | ||
1da177e4 | 369 | dm_get(md); |
5c6bd75d | 370 | atomic_inc(&md->open_count); |
fba9f90e JM |
371 | |
372 | out: | |
373 | spin_unlock(&_minor_lock); | |
374 | ||
375 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
376 | } |
377 | ||
fe5f9f2c | 378 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 379 | { |
fe5f9f2c | 380 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 381 | |
4a1aeb98 MB |
382 | spin_lock(&_minor_lock); |
383 | ||
5c6bd75d | 384 | atomic_dec(&md->open_count); |
1da177e4 | 385 | dm_put(md); |
4a1aeb98 MB |
386 | |
387 | spin_unlock(&_minor_lock); | |
6e9624b8 | 388 | |
1da177e4 LT |
389 | return 0; |
390 | } | |
391 | ||
5c6bd75d AK |
392 | int dm_open_count(struct mapped_device *md) |
393 | { | |
394 | return atomic_read(&md->open_count); | |
395 | } | |
396 | ||
397 | /* | |
398 | * Guarantees nothing is using the device before it's deleted. | |
399 | */ | |
400 | int dm_lock_for_deletion(struct mapped_device *md) | |
401 | { | |
402 | int r = 0; | |
403 | ||
404 | spin_lock(&_minor_lock); | |
405 | ||
406 | if (dm_open_count(md)) | |
407 | r = -EBUSY; | |
408 | else | |
409 | set_bit(DMF_DELETING, &md->flags); | |
410 | ||
411 | spin_unlock(&_minor_lock); | |
412 | ||
413 | return r; | |
414 | } | |
415 | ||
3ac51e74 DW |
416 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
417 | { | |
418 | struct mapped_device *md = bdev->bd_disk->private_data; | |
419 | ||
420 | return dm_get_geometry(md, geo); | |
421 | } | |
422 | ||
fe5f9f2c | 423 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
424 | unsigned int cmd, unsigned long arg) |
425 | { | |
fe5f9f2c | 426 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 427 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
428 | struct dm_target *tgt; |
429 | int r = -ENOTTY; | |
430 | ||
aa129a22 MB |
431 | if (!map || !dm_table_get_size(map)) |
432 | goto out; | |
433 | ||
434 | /* We only support devices that have a single target */ | |
435 | if (dm_table_get_num_targets(map) != 1) | |
436 | goto out; | |
437 | ||
438 | tgt = dm_table_get_target(map, 0); | |
439 | ||
4f186f8b | 440 | if (dm_suspended_md(md)) { |
aa129a22 MB |
441 | r = -EAGAIN; |
442 | goto out; | |
443 | } | |
444 | ||
445 | if (tgt->type->ioctl) | |
647b3d00 | 446 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
447 | |
448 | out: | |
449 | dm_table_put(map); | |
450 | ||
aa129a22 MB |
451 | return r; |
452 | } | |
453 | ||
028867ac | 454 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
455 | { |
456 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
457 | } | |
458 | ||
028867ac | 459 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
460 | { |
461 | mempool_free(io, md->io_pool); | |
462 | } | |
463 | ||
028867ac | 464 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
465 | { |
466 | mempool_free(tio, md->tio_pool); | |
467 | } | |
468 | ||
08885643 KU |
469 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
470 | gfp_t gfp_mask) | |
cec47e3d | 471 | { |
08885643 | 472 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
473 | } |
474 | ||
475 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
476 | { | |
477 | mempool_free(tio, tio->md->tio_pool); | |
478 | } | |
479 | ||
90abb8c4 KU |
480 | static int md_in_flight(struct mapped_device *md) |
481 | { | |
482 | return atomic_read(&md->pending[READ]) + | |
483 | atomic_read(&md->pending[WRITE]); | |
484 | } | |
485 | ||
3eaf840e JNN |
486 | static void start_io_acct(struct dm_io *io) |
487 | { | |
488 | struct mapped_device *md = io->md; | |
c9959059 | 489 | int cpu; |
316d315b | 490 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
491 | |
492 | io->start_time = jiffies; | |
493 | ||
074a7aca TH |
494 | cpu = part_stat_lock(); |
495 | part_round_stats(cpu, &dm_disk(md)->part0); | |
496 | part_stat_unlock(); | |
1e9bb880 SL |
497 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
498 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
499 | } |
500 | ||
d221d2e7 | 501 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
502 | { |
503 | struct mapped_device *md = io->md; | |
504 | struct bio *bio = io->bio; | |
505 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 506 | int pending, cpu; |
3eaf840e JNN |
507 | int rw = bio_data_dir(bio); |
508 | ||
074a7aca TH |
509 | cpu = part_stat_lock(); |
510 | part_round_stats(cpu, &dm_disk(md)->part0); | |
511 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
512 | part_stat_unlock(); | |
3eaf840e | 513 | |
af7e466a MP |
514 | /* |
515 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 516 | * a flush. |
af7e466a | 517 | */ |
1e9bb880 SL |
518 | pending = atomic_dec_return(&md->pending[rw]); |
519 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 520 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 521 | |
d221d2e7 MP |
522 | /* nudge anyone waiting on suspend queue */ |
523 | if (!pending) | |
524 | wake_up(&md->wait); | |
3eaf840e JNN |
525 | } |
526 | ||
1da177e4 LT |
527 | /* |
528 | * Add the bio to the list of deferred io. | |
529 | */ | |
92c63902 | 530 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 531 | { |
05447420 | 532 | unsigned long flags; |
1da177e4 | 533 | |
05447420 | 534 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 535 | bio_list_add(&md->deferred, bio); |
05447420 | 536 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 537 | queue_work(md->wq, &md->work); |
1da177e4 LT |
538 | } |
539 | ||
540 | /* | |
541 | * Everyone (including functions in this file), should use this | |
542 | * function to access the md->map field, and make sure they call | |
543 | * dm_table_put() when finished. | |
544 | */ | |
7c666411 | 545 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
546 | { |
547 | struct dm_table *t; | |
523d9297 | 548 | unsigned long flags; |
1da177e4 | 549 | |
523d9297 | 550 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
551 | t = md->map; |
552 | if (t) | |
553 | dm_table_get(t); | |
523d9297 | 554 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
555 | |
556 | return t; | |
557 | } | |
558 | ||
3ac51e74 DW |
559 | /* |
560 | * Get the geometry associated with a dm device | |
561 | */ | |
562 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
563 | { | |
564 | *geo = md->geometry; | |
565 | ||
566 | return 0; | |
567 | } | |
568 | ||
569 | /* | |
570 | * Set the geometry of a device. | |
571 | */ | |
572 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
573 | { | |
574 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
575 | ||
576 | if (geo->start > sz) { | |
577 | DMWARN("Start sector is beyond the geometry limits."); | |
578 | return -EINVAL; | |
579 | } | |
580 | ||
581 | md->geometry = *geo; | |
582 | ||
583 | return 0; | |
584 | } | |
585 | ||
1da177e4 LT |
586 | /*----------------------------------------------------------------- |
587 | * CRUD START: | |
588 | * A more elegant soln is in the works that uses the queue | |
589 | * merge fn, unfortunately there are a couple of changes to | |
590 | * the block layer that I want to make for this. So in the | |
591 | * interests of getting something for people to use I give | |
592 | * you this clearly demarcated crap. | |
593 | *---------------------------------------------------------------*/ | |
594 | ||
2e93ccc1 KU |
595 | static int __noflush_suspending(struct mapped_device *md) |
596 | { | |
597 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
598 | } | |
599 | ||
1da177e4 LT |
600 | /* |
601 | * Decrements the number of outstanding ios that a bio has been | |
602 | * cloned into, completing the original io if necc. | |
603 | */ | |
858119e1 | 604 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 605 | { |
2e93ccc1 | 606 | unsigned long flags; |
b35f8caa MB |
607 | int io_error; |
608 | struct bio *bio; | |
609 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
610 | |
611 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
612 | if (unlikely(error)) { |
613 | spin_lock_irqsave(&io->endio_lock, flags); | |
614 | if (!(io->error > 0 && __noflush_suspending(md))) | |
615 | io->error = error; | |
616 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
617 | } | |
1da177e4 LT |
618 | |
619 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
620 | if (io->error == DM_ENDIO_REQUEUE) { |
621 | /* | |
622 | * Target requested pushing back the I/O. | |
2e93ccc1 | 623 | */ |
022c2611 | 624 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
625 | if (__noflush_suspending(md)) |
626 | bio_list_add_head(&md->deferred, io->bio); | |
627 | else | |
2e93ccc1 KU |
628 | /* noflush suspend was interrupted. */ |
629 | io->error = -EIO; | |
022c2611 | 630 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
631 | } |
632 | ||
b35f8caa MB |
633 | io_error = io->error; |
634 | bio = io->bio; | |
6a8736d1 TH |
635 | end_io_acct(io); |
636 | free_io(md, io); | |
637 | ||
638 | if (io_error == DM_ENDIO_REQUEUE) | |
639 | return; | |
2e93ccc1 | 640 | |
b372d360 | 641 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 642 | /* |
6a8736d1 TH |
643 | * Preflush done for flush with data, reissue |
644 | * without REQ_FLUSH. | |
af7e466a | 645 | */ |
6a8736d1 TH |
646 | bio->bi_rw &= ~REQ_FLUSH; |
647 | queue_io(md, bio); | |
af7e466a | 648 | } else { |
b372d360 | 649 | /* done with normal IO or empty flush */ |
b7908c10 | 650 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 651 | bio_endio(bio, io_error); |
b35f8caa | 652 | } |
1da177e4 LT |
653 | } |
654 | } | |
655 | ||
6712ecf8 | 656 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
657 | { |
658 | int r = 0; | |
028867ac | 659 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 660 | struct dm_io *io = tio->io; |
9faf400f | 661 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
662 | dm_endio_fn endio = tio->ti->type->end_io; |
663 | ||
1da177e4 LT |
664 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
665 | error = -EIO; | |
666 | ||
667 | if (endio) { | |
668 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
669 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
670 | /* | |
671 | * error and requeue request are handled | |
672 | * in dec_pending(). | |
673 | */ | |
1da177e4 | 674 | error = r; |
45cbcd79 KU |
675 | else if (r == DM_ENDIO_INCOMPLETE) |
676 | /* The target will handle the io */ | |
6712ecf8 | 677 | return; |
45cbcd79 KU |
678 | else if (r) { |
679 | DMWARN("unimplemented target endio return value: %d", r); | |
680 | BUG(); | |
681 | } | |
1da177e4 LT |
682 | } |
683 | ||
9faf400f | 684 | free_tio(md, tio); |
b35f8caa MB |
685 | bio_put(bio); |
686 | dec_pending(io, error); | |
1da177e4 LT |
687 | } |
688 | ||
cec47e3d KU |
689 | /* |
690 | * Partial completion handling for request-based dm | |
691 | */ | |
692 | static void end_clone_bio(struct bio *clone, int error) | |
693 | { | |
694 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
695 | struct dm_rq_target_io *tio = info->tio; | |
696 | struct bio *bio = info->orig; | |
697 | unsigned int nr_bytes = info->orig->bi_size; | |
698 | ||
699 | bio_put(clone); | |
700 | ||
701 | if (tio->error) | |
702 | /* | |
703 | * An error has already been detected on the request. | |
704 | * Once error occurred, just let clone->end_io() handle | |
705 | * the remainder. | |
706 | */ | |
707 | return; | |
708 | else if (error) { | |
709 | /* | |
710 | * Don't notice the error to the upper layer yet. | |
711 | * The error handling decision is made by the target driver, | |
712 | * when the request is completed. | |
713 | */ | |
714 | tio->error = error; | |
715 | return; | |
716 | } | |
717 | ||
718 | /* | |
719 | * I/O for the bio successfully completed. | |
720 | * Notice the data completion to the upper layer. | |
721 | */ | |
722 | ||
723 | /* | |
724 | * bios are processed from the head of the list. | |
725 | * So the completing bio should always be rq->bio. | |
726 | * If it's not, something wrong is happening. | |
727 | */ | |
728 | if (tio->orig->bio != bio) | |
729 | DMERR("bio completion is going in the middle of the request"); | |
730 | ||
731 | /* | |
732 | * Update the original request. | |
733 | * Do not use blk_end_request() here, because it may complete | |
734 | * the original request before the clone, and break the ordering. | |
735 | */ | |
736 | blk_update_request(tio->orig, 0, nr_bytes); | |
737 | } | |
738 | ||
739 | /* | |
740 | * Don't touch any member of the md after calling this function because | |
741 | * the md may be freed in dm_put() at the end of this function. | |
742 | * Or do dm_get() before calling this function and dm_put() later. | |
743 | */ | |
b4324fee | 744 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 745 | { |
b4324fee | 746 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
747 | |
748 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 749 | if (!md_in_flight(md)) |
cec47e3d KU |
750 | wake_up(&md->wait); |
751 | ||
752 | if (run_queue) | |
b4324fee | 753 | blk_run_queue(md->queue); |
cec47e3d KU |
754 | |
755 | /* | |
756 | * dm_put() must be at the end of this function. See the comment above | |
757 | */ | |
758 | dm_put(md); | |
759 | } | |
760 | ||
a77e28c7 KU |
761 | static void free_rq_clone(struct request *clone) |
762 | { | |
763 | struct dm_rq_target_io *tio = clone->end_io_data; | |
764 | ||
765 | blk_rq_unprep_clone(clone); | |
766 | free_rq_tio(tio); | |
767 | } | |
768 | ||
980691e5 KU |
769 | /* |
770 | * Complete the clone and the original request. | |
771 | * Must be called without queue lock. | |
772 | */ | |
773 | static void dm_end_request(struct request *clone, int error) | |
774 | { | |
775 | int rw = rq_data_dir(clone); | |
776 | struct dm_rq_target_io *tio = clone->end_io_data; | |
777 | struct mapped_device *md = tio->md; | |
778 | struct request *rq = tio->orig; | |
779 | ||
29e4013d | 780 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
781 | rq->errors = clone->errors; |
782 | rq->resid_len = clone->resid_len; | |
783 | ||
784 | if (rq->sense) | |
785 | /* | |
786 | * We are using the sense buffer of the original | |
787 | * request. | |
788 | * So setting the length of the sense data is enough. | |
789 | */ | |
790 | rq->sense_len = clone->sense_len; | |
791 | } | |
792 | ||
793 | free_rq_clone(clone); | |
29e4013d TH |
794 | blk_end_request_all(rq, error); |
795 | rq_completed(md, rw, true); | |
980691e5 KU |
796 | } |
797 | ||
cec47e3d KU |
798 | static void dm_unprep_request(struct request *rq) |
799 | { | |
800 | struct request *clone = rq->special; | |
cec47e3d KU |
801 | |
802 | rq->special = NULL; | |
803 | rq->cmd_flags &= ~REQ_DONTPREP; | |
804 | ||
a77e28c7 | 805 | free_rq_clone(clone); |
cec47e3d KU |
806 | } |
807 | ||
808 | /* | |
809 | * Requeue the original request of a clone. | |
810 | */ | |
811 | void dm_requeue_unmapped_request(struct request *clone) | |
812 | { | |
b4324fee | 813 | int rw = rq_data_dir(clone); |
cec47e3d KU |
814 | struct dm_rq_target_io *tio = clone->end_io_data; |
815 | struct mapped_device *md = tio->md; | |
816 | struct request *rq = tio->orig; | |
817 | struct request_queue *q = rq->q; | |
818 | unsigned long flags; | |
819 | ||
820 | dm_unprep_request(rq); | |
821 | ||
822 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
823 | blk_requeue_request(q, rq); |
824 | spin_unlock_irqrestore(q->queue_lock, flags); | |
825 | ||
b4324fee | 826 | rq_completed(md, rw, 0); |
cec47e3d KU |
827 | } |
828 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
829 | ||
830 | static void __stop_queue(struct request_queue *q) | |
831 | { | |
832 | blk_stop_queue(q); | |
833 | } | |
834 | ||
835 | static void stop_queue(struct request_queue *q) | |
836 | { | |
837 | unsigned long flags; | |
838 | ||
839 | spin_lock_irqsave(q->queue_lock, flags); | |
840 | __stop_queue(q); | |
841 | spin_unlock_irqrestore(q->queue_lock, flags); | |
842 | } | |
843 | ||
844 | static void __start_queue(struct request_queue *q) | |
845 | { | |
846 | if (blk_queue_stopped(q)) | |
847 | blk_start_queue(q); | |
848 | } | |
849 | ||
850 | static void start_queue(struct request_queue *q) | |
851 | { | |
852 | unsigned long flags; | |
853 | ||
854 | spin_lock_irqsave(q->queue_lock, flags); | |
855 | __start_queue(q); | |
856 | spin_unlock_irqrestore(q->queue_lock, flags); | |
857 | } | |
858 | ||
11a68244 | 859 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 860 | { |
11a68244 | 861 | int r = error; |
cec47e3d | 862 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 863 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 864 | |
ba1cbad9 MS |
865 | if (tio->ti) { |
866 | rq_end_io = tio->ti->type->rq_end_io; | |
867 | ||
868 | if (mapped && rq_end_io) | |
869 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
870 | } | |
cec47e3d | 871 | |
11a68244 | 872 | if (r <= 0) |
cec47e3d | 873 | /* The target wants to complete the I/O */ |
11a68244 KU |
874 | dm_end_request(clone, r); |
875 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
876 | /* The target will handle the I/O */ |
877 | return; | |
11a68244 | 878 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
879 | /* The target wants to requeue the I/O */ |
880 | dm_requeue_unmapped_request(clone); | |
881 | else { | |
11a68244 | 882 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
883 | BUG(); |
884 | } | |
885 | } | |
886 | ||
11a68244 KU |
887 | /* |
888 | * Request completion handler for request-based dm | |
889 | */ | |
890 | static void dm_softirq_done(struct request *rq) | |
891 | { | |
892 | bool mapped = true; | |
893 | struct request *clone = rq->completion_data; | |
894 | struct dm_rq_target_io *tio = clone->end_io_data; | |
895 | ||
896 | if (rq->cmd_flags & REQ_FAILED) | |
897 | mapped = false; | |
898 | ||
899 | dm_done(clone, tio->error, mapped); | |
900 | } | |
901 | ||
cec47e3d KU |
902 | /* |
903 | * Complete the clone and the original request with the error status | |
904 | * through softirq context. | |
905 | */ | |
906 | static void dm_complete_request(struct request *clone, int error) | |
907 | { | |
908 | struct dm_rq_target_io *tio = clone->end_io_data; | |
909 | struct request *rq = tio->orig; | |
910 | ||
911 | tio->error = error; | |
912 | rq->completion_data = clone; | |
913 | blk_complete_request(rq); | |
914 | } | |
915 | ||
916 | /* | |
917 | * Complete the not-mapped clone and the original request with the error status | |
918 | * through softirq context. | |
919 | * Target's rq_end_io() function isn't called. | |
920 | * This may be used when the target's map_rq() function fails. | |
921 | */ | |
922 | void dm_kill_unmapped_request(struct request *clone, int error) | |
923 | { | |
924 | struct dm_rq_target_io *tio = clone->end_io_data; | |
925 | struct request *rq = tio->orig; | |
926 | ||
927 | rq->cmd_flags |= REQ_FAILED; | |
928 | dm_complete_request(clone, error); | |
929 | } | |
930 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
931 | ||
932 | /* | |
933 | * Called with the queue lock held | |
934 | */ | |
935 | static void end_clone_request(struct request *clone, int error) | |
936 | { | |
937 | /* | |
938 | * For just cleaning up the information of the queue in which | |
939 | * the clone was dispatched. | |
940 | * The clone is *NOT* freed actually here because it is alloced from | |
941 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
942 | */ | |
943 | __blk_put_request(clone->q, clone); | |
944 | ||
945 | /* | |
946 | * Actual request completion is done in a softirq context which doesn't | |
947 | * hold the queue lock. Otherwise, deadlock could occur because: | |
948 | * - another request may be submitted by the upper level driver | |
949 | * of the stacking during the completion | |
950 | * - the submission which requires queue lock may be done | |
951 | * against this queue | |
952 | */ | |
953 | dm_complete_request(clone, error); | |
954 | } | |
955 | ||
56a67df7 MS |
956 | /* |
957 | * Return maximum size of I/O possible at the supplied sector up to the current | |
958 | * target boundary. | |
959 | */ | |
960 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
961 | { | |
962 | sector_t target_offset = dm_target_offset(ti, sector); | |
963 | ||
964 | return ti->len - target_offset; | |
965 | } | |
966 | ||
967 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 968 | { |
56a67df7 | 969 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 970 | sector_t offset, max_len; |
1da177e4 LT |
971 | |
972 | /* | |
542f9038 | 973 | * Does the target need to split even further? |
1da177e4 | 974 | */ |
542f9038 MS |
975 | if (ti->max_io_len) { |
976 | offset = dm_target_offset(ti, sector); | |
977 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
978 | max_len = sector_div(offset, ti->max_io_len); | |
979 | else | |
980 | max_len = offset & (ti->max_io_len - 1); | |
981 | max_len = ti->max_io_len - max_len; | |
982 | ||
983 | if (len > max_len) | |
984 | len = max_len; | |
1da177e4 LT |
985 | } |
986 | ||
987 | return len; | |
988 | } | |
989 | ||
542f9038 MS |
990 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
991 | { | |
992 | if (len > UINT_MAX) { | |
993 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
994 | (unsigned long long)len, UINT_MAX); | |
995 | ti->error = "Maximum size of target IO is too large"; | |
996 | return -EINVAL; | |
997 | } | |
998 | ||
999 | ti->max_io_len = (uint32_t) len; | |
1000 | ||
1001 | return 0; | |
1002 | } | |
1003 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1004 | ||
1da177e4 | 1005 | static void __map_bio(struct dm_target *ti, struct bio *clone, |
028867ac | 1006 | struct dm_target_io *tio) |
1da177e4 LT |
1007 | { |
1008 | int r; | |
2056a782 | 1009 | sector_t sector; |
9faf400f | 1010 | struct mapped_device *md; |
1da177e4 | 1011 | |
1da177e4 LT |
1012 | clone->bi_end_io = clone_endio; |
1013 | clone->bi_private = tio; | |
1014 | ||
1015 | /* | |
1016 | * Map the clone. If r == 0 we don't need to do | |
1017 | * anything, the target has assumed ownership of | |
1018 | * this io. | |
1019 | */ | |
1020 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1021 | sector = clone->bi_sector; |
1da177e4 | 1022 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 1023 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1024 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1025 | |
d07335e5 MS |
1026 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1027 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1028 | |
1da177e4 | 1029 | generic_make_request(clone); |
2e93ccc1 KU |
1030 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1031 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1032 | md = tio->io->md; |
1033 | dec_pending(tio->io, r); | |
1da177e4 | 1034 | bio_put(clone); |
9faf400f | 1035 | free_tio(md, tio); |
45cbcd79 KU |
1036 | } else if (r) { |
1037 | DMWARN("unimplemented target map return value: %d", r); | |
1038 | BUG(); | |
1da177e4 LT |
1039 | } |
1040 | } | |
1041 | ||
1042 | struct clone_info { | |
1043 | struct mapped_device *md; | |
1044 | struct dm_table *map; | |
1045 | struct bio *bio; | |
1046 | struct dm_io *io; | |
1047 | sector_t sector; | |
1048 | sector_t sector_count; | |
1049 | unsigned short idx; | |
1050 | }; | |
1051 | ||
1052 | /* | |
d87f4c14 | 1053 | * Creates a little bio that just does part of a bvec. |
1da177e4 LT |
1054 | */ |
1055 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1056 | unsigned short idx, unsigned int offset, | |
9faf400f | 1057 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1058 | { |
1059 | struct bio *clone; | |
1060 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1061 | ||
9faf400f | 1062 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
1da177e4 LT |
1063 | *clone->bi_io_vec = *bv; |
1064 | ||
1065 | clone->bi_sector = sector; | |
1066 | clone->bi_bdev = bio->bi_bdev; | |
d87f4c14 | 1067 | clone->bi_rw = bio->bi_rw; |
1da177e4 LT |
1068 | clone->bi_vcnt = 1; |
1069 | clone->bi_size = to_bytes(len); | |
1070 | clone->bi_io_vec->bv_offset = offset; | |
1071 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1072 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1073 | |
9c47008d | 1074 | if (bio_integrity(bio)) { |
1e2a410f | 1075 | bio_integrity_clone(clone, bio, GFP_NOIO); |
9c47008d MP |
1076 | bio_integrity_trim(clone, |
1077 | bio_sector_offset(bio, idx, offset), len); | |
1078 | } | |
1079 | ||
1da177e4 LT |
1080 | return clone; |
1081 | } | |
1082 | ||
1083 | /* | |
1084 | * Creates a bio that consists of range of complete bvecs. | |
1085 | */ | |
1086 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1087 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1088 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1089 | { |
1090 | struct bio *clone; | |
1091 | ||
9faf400f SB |
1092 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1093 | __bio_clone(clone, bio); | |
1da177e4 LT |
1094 | clone->bi_sector = sector; |
1095 | clone->bi_idx = idx; | |
1096 | clone->bi_vcnt = idx + bv_count; | |
1097 | clone->bi_size = to_bytes(len); | |
1098 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1099 | ||
9c47008d | 1100 | if (bio_integrity(bio)) { |
1e2a410f | 1101 | bio_integrity_clone(clone, bio, GFP_NOIO); |
9c47008d MP |
1102 | |
1103 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1104 | bio_integrity_trim(clone, | |
1105 | bio_sector_offset(bio, idx, 0), len); | |
1106 | } | |
1107 | ||
1da177e4 LT |
1108 | return clone; |
1109 | } | |
1110 | ||
9015df24 AK |
1111 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1112 | struct dm_target *ti) | |
f9ab94ce | 1113 | { |
9015df24 | 1114 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1115 | |
1116 | tio->io = ci->io; | |
1117 | tio->ti = ti; | |
f9ab94ce | 1118 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1119 | |
1120 | return tio; | |
1121 | } | |
1122 | ||
06a426ce | 1123 | static void __issue_target_request(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1124 | unsigned request_nr, sector_t len) |
9015df24 AK |
1125 | { |
1126 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1127 | struct bio *clone; | |
1128 | ||
57cba5d3 | 1129 | tio->info.target_request_nr = request_nr; |
f9ab94ce | 1130 | |
06a426ce MS |
1131 | /* |
1132 | * Discard requests require the bio's inline iovecs be initialized. | |
1133 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1134 | * and discard, so no need for concern about wasted bvec allocations. | |
1135 | */ | |
bf800ef1 KO |
1136 | clone = bio_clone_bioset(ci->bio, GFP_NOIO, ci->md->bs); |
1137 | ||
a79245b3 MS |
1138 | if (len) { |
1139 | clone->bi_sector = ci->sector; | |
1140 | clone->bi_size = to_bytes(len); | |
1141 | } | |
f9ab94ce MP |
1142 | |
1143 | __map_bio(ti, clone, tio); | |
1144 | } | |
1145 | ||
06a426ce | 1146 | static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1147 | unsigned num_requests, sector_t len) |
06a426ce MS |
1148 | { |
1149 | unsigned request_nr; | |
1150 | ||
1151 | for (request_nr = 0; request_nr < num_requests; request_nr++) | |
a79245b3 | 1152 | __issue_target_request(ci, ti, request_nr, len); |
06a426ce MS |
1153 | } |
1154 | ||
b372d360 | 1155 | static int __clone_and_map_empty_flush(struct clone_info *ci) |
f9ab94ce | 1156 | { |
06a426ce | 1157 | unsigned target_nr = 0; |
f9ab94ce MP |
1158 | struct dm_target *ti; |
1159 | ||
b372d360 | 1160 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1161 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
a79245b3 | 1162 | __issue_target_requests(ci, ti, ti->num_flush_requests, 0); |
f9ab94ce | 1163 | |
f9ab94ce MP |
1164 | return 0; |
1165 | } | |
1166 | ||
5ae89a87 MS |
1167 | /* |
1168 | * Perform all io with a single clone. | |
1169 | */ | |
1170 | static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti) | |
1171 | { | |
1172 | struct bio *clone, *bio = ci->bio; | |
1173 | struct dm_target_io *tio; | |
1174 | ||
1175 | tio = alloc_tio(ci, ti); | |
1176 | clone = clone_bio(bio, ci->sector, ci->idx, | |
1177 | bio->bi_vcnt - ci->idx, ci->sector_count, | |
1178 | ci->md->bs); | |
1179 | __map_bio(ti, clone, tio); | |
1180 | ci->sector_count = 0; | |
1181 | } | |
1182 | ||
1183 | static int __clone_and_map_discard(struct clone_info *ci) | |
1184 | { | |
1185 | struct dm_target *ti; | |
a79245b3 | 1186 | sector_t len; |
5ae89a87 | 1187 | |
a79245b3 MS |
1188 | do { |
1189 | ti = dm_table_find_target(ci->map, ci->sector); | |
1190 | if (!dm_target_is_valid(ti)) | |
1191 | return -EIO; | |
5ae89a87 | 1192 | |
5ae89a87 | 1193 | /* |
a79245b3 | 1194 | * Even though the device advertised discard support, |
936688d7 MS |
1195 | * that does not mean every target supports it, and |
1196 | * reconfiguration might also have changed that since the | |
a79245b3 | 1197 | * check was performed. |
5ae89a87 | 1198 | */ |
a79245b3 MS |
1199 | if (!ti->num_discard_requests) |
1200 | return -EOPNOTSUPP; | |
5ae89a87 | 1201 | |
7acf0277 MP |
1202 | if (!ti->split_discard_requests) |
1203 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1204 | else | |
1205 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1206 | |
a79245b3 MS |
1207 | __issue_target_requests(ci, ti, ti->num_discard_requests, len); |
1208 | ||
1209 | ci->sector += len; | |
1210 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1211 | |
1212 | return 0; | |
1213 | } | |
1214 | ||
512875bd | 1215 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1216 | { |
1217 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1218 | struct dm_target *ti; |
1219 | sector_t len = 0, max; | |
028867ac | 1220 | struct dm_target_io *tio; |
1da177e4 | 1221 | |
5ae89a87 MS |
1222 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
1223 | return __clone_and_map_discard(ci); | |
1224 | ||
512875bd JN |
1225 | ti = dm_table_find_target(ci->map, ci->sector); |
1226 | if (!dm_target_is_valid(ti)) | |
1227 | return -EIO; | |
1228 | ||
56a67df7 | 1229 | max = max_io_len(ci->sector, ti); |
512875bd | 1230 | |
1da177e4 LT |
1231 | if (ci->sector_count <= max) { |
1232 | /* | |
1233 | * Optimise for the simple case where we can do all of | |
1234 | * the remaining io with a single clone. | |
1235 | */ | |
5ae89a87 | 1236 | __clone_and_map_simple(ci, ti); |
1da177e4 LT |
1237 | |
1238 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1239 | /* | |
1240 | * There are some bvecs that don't span targets. | |
1241 | * Do as many of these as possible. | |
1242 | */ | |
1243 | int i; | |
1244 | sector_t remaining = max; | |
1245 | sector_t bv_len; | |
1246 | ||
1247 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1248 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1249 | ||
1250 | if (bv_len > remaining) | |
1251 | break; | |
1252 | ||
1253 | remaining -= bv_len; | |
1254 | len += bv_len; | |
1255 | } | |
1256 | ||
5ae89a87 | 1257 | tio = alloc_tio(ci, ti); |
9faf400f SB |
1258 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1259 | ci->md->bs); | |
1da177e4 LT |
1260 | __map_bio(ti, clone, tio); |
1261 | ||
1262 | ci->sector += len; | |
1263 | ci->sector_count -= len; | |
1264 | ci->idx = i; | |
1265 | ||
1266 | } else { | |
1267 | /* | |
d2044a94 | 1268 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1269 | */ |
1270 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1271 | sector_t remaining = to_sector(bv->bv_len); |
1272 | unsigned int offset = 0; | |
1da177e4 | 1273 | |
d2044a94 AK |
1274 | do { |
1275 | if (offset) { | |
1276 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1277 | if (!dm_target_is_valid(ti)) |
1278 | return -EIO; | |
1279 | ||
56a67df7 | 1280 | max = max_io_len(ci->sector, ti); |
d2044a94 AK |
1281 | } |
1282 | ||
1283 | len = min(remaining, max); | |
1284 | ||
5ae89a87 | 1285 | tio = alloc_tio(ci, ti); |
d2044a94 | 1286 | clone = split_bvec(bio, ci->sector, ci->idx, |
9faf400f SB |
1287 | bv->bv_offset + offset, len, |
1288 | ci->md->bs); | |
d2044a94 AK |
1289 | |
1290 | __map_bio(ti, clone, tio); | |
1291 | ||
1292 | ci->sector += len; | |
1293 | ci->sector_count -= len; | |
1294 | offset += to_bytes(len); | |
1295 | } while (remaining -= len); | |
1da177e4 | 1296 | |
1da177e4 LT |
1297 | ci->idx++; |
1298 | } | |
512875bd JN |
1299 | |
1300 | return 0; | |
1da177e4 LT |
1301 | } |
1302 | ||
1303 | /* | |
8a53c28d | 1304 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1305 | */ |
f0b9a450 | 1306 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1307 | { |
1308 | struct clone_info ci; | |
512875bd | 1309 | int error = 0; |
1da177e4 | 1310 | |
7c666411 | 1311 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1312 | if (unlikely(!ci.map)) { |
6a8736d1 | 1313 | bio_io_error(bio); |
f0b9a450 MP |
1314 | return; |
1315 | } | |
692d0eb9 | 1316 | |
1da177e4 | 1317 | ci.md = md; |
1da177e4 LT |
1318 | ci.io = alloc_io(md); |
1319 | ci.io->error = 0; | |
1320 | atomic_set(&ci.io->io_count, 1); | |
1321 | ci.io->bio = bio; | |
1322 | ci.io->md = md; | |
f88fb981 | 1323 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1324 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1325 | ci.idx = bio->bi_idx; |
1326 | ||
3eaf840e | 1327 | start_io_acct(ci.io); |
b372d360 MS |
1328 | if (bio->bi_rw & REQ_FLUSH) { |
1329 | ci.bio = &ci.md->flush_bio; | |
1330 | ci.sector_count = 0; | |
1331 | error = __clone_and_map_empty_flush(&ci); | |
1332 | /* dec_pending submits any data associated with flush */ | |
1333 | } else { | |
6a8736d1 | 1334 | ci.bio = bio; |
d87f4c14 | 1335 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1336 | while (ci.sector_count && !error) |
d87f4c14 | 1337 | error = __clone_and_map(&ci); |
d87f4c14 | 1338 | } |
1da177e4 LT |
1339 | |
1340 | /* drop the extra reference count */ | |
512875bd | 1341 | dec_pending(ci.io, error); |
1da177e4 LT |
1342 | dm_table_put(ci.map); |
1343 | } | |
1344 | /*----------------------------------------------------------------- | |
1345 | * CRUD END | |
1346 | *---------------------------------------------------------------*/ | |
1347 | ||
f6fccb12 MB |
1348 | static int dm_merge_bvec(struct request_queue *q, |
1349 | struct bvec_merge_data *bvm, | |
1350 | struct bio_vec *biovec) | |
1351 | { | |
1352 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1353 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1354 | struct dm_target *ti; |
1355 | sector_t max_sectors; | |
5037108a | 1356 | int max_size = 0; |
f6fccb12 MB |
1357 | |
1358 | if (unlikely(!map)) | |
5037108a | 1359 | goto out; |
f6fccb12 MB |
1360 | |
1361 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1362 | if (!dm_target_is_valid(ti)) |
1363 | goto out_table; | |
f6fccb12 MB |
1364 | |
1365 | /* | |
1366 | * Find maximum amount of I/O that won't need splitting | |
1367 | */ | |
56a67df7 | 1368 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1369 | (sector_t) BIO_MAX_SECTORS); |
1370 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1371 | if (max_size < 0) | |
1372 | max_size = 0; | |
1373 | ||
1374 | /* | |
1375 | * merge_bvec_fn() returns number of bytes | |
1376 | * it can accept at this offset | |
1377 | * max is precomputed maximal io size | |
1378 | */ | |
1379 | if (max_size && ti->type->merge) | |
1380 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1381 | /* |
1382 | * If the target doesn't support merge method and some of the devices | |
1383 | * provided their merge_bvec method (we know this by looking at | |
1384 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1385 | * entries. So always set max_size to 0, and the code below allows | |
1386 | * just one page. | |
1387 | */ | |
1388 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1389 | ||
1390 | max_size = 0; | |
f6fccb12 | 1391 | |
b01cd5ac | 1392 | out_table: |
5037108a MP |
1393 | dm_table_put(map); |
1394 | ||
1395 | out: | |
f6fccb12 MB |
1396 | /* |
1397 | * Always allow an entire first page | |
1398 | */ | |
1399 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1400 | max_size = biovec->bv_len; | |
1401 | ||
f6fccb12 MB |
1402 | return max_size; |
1403 | } | |
1404 | ||
1da177e4 LT |
1405 | /* |
1406 | * The request function that just remaps the bio built up by | |
1407 | * dm_merge_bvec. | |
1408 | */ | |
5a7bbad2 | 1409 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1410 | { |
12f03a49 | 1411 | int rw = bio_data_dir(bio); |
1da177e4 | 1412 | struct mapped_device *md = q->queuedata; |
c9959059 | 1413 | int cpu; |
1da177e4 | 1414 | |
2ca3310e | 1415 | down_read(&md->io_lock); |
1da177e4 | 1416 | |
074a7aca TH |
1417 | cpu = part_stat_lock(); |
1418 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1419 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1420 | part_stat_unlock(); | |
12f03a49 | 1421 | |
6a8736d1 TH |
1422 | /* if we're suspended, we have to queue this io for later */ |
1423 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1424 | up_read(&md->io_lock); |
1da177e4 | 1425 | |
6a8736d1 TH |
1426 | if (bio_rw(bio) != READA) |
1427 | queue_io(md, bio); | |
1428 | else | |
54d9a1b4 | 1429 | bio_io_error(bio); |
5a7bbad2 | 1430 | return; |
1da177e4 LT |
1431 | } |
1432 | ||
f0b9a450 | 1433 | __split_and_process_bio(md, bio); |
2ca3310e | 1434 | up_read(&md->io_lock); |
5a7bbad2 | 1435 | return; |
cec47e3d KU |
1436 | } |
1437 | ||
1438 | static int dm_request_based(struct mapped_device *md) | |
1439 | { | |
1440 | return blk_queue_stackable(md->queue); | |
1441 | } | |
1442 | ||
5a7bbad2 | 1443 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1444 | { |
1445 | struct mapped_device *md = q->queuedata; | |
1446 | ||
1447 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1448 | blk_queue_bio(q, bio); |
1449 | else | |
1450 | _dm_request(q, bio); | |
cec47e3d KU |
1451 | } |
1452 | ||
1453 | void dm_dispatch_request(struct request *rq) | |
1454 | { | |
1455 | int r; | |
1456 | ||
1457 | if (blk_queue_io_stat(rq->q)) | |
1458 | rq->cmd_flags |= REQ_IO_STAT; | |
1459 | ||
1460 | rq->start_time = jiffies; | |
1461 | r = blk_insert_cloned_request(rq->q, rq); | |
1462 | if (r) | |
1463 | dm_complete_request(rq, r); | |
1464 | } | |
1465 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1466 | ||
cec47e3d KU |
1467 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1468 | void *data) | |
1469 | { | |
1470 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1471 | struct dm_rq_clone_bio_info *info = |
1472 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1473 | |
1474 | info->orig = bio_orig; | |
1475 | info->tio = tio; | |
1476 | bio->bi_end_io = end_clone_bio; | |
1477 | bio->bi_private = info; | |
cec47e3d KU |
1478 | |
1479 | return 0; | |
1480 | } | |
1481 | ||
1482 | static int setup_clone(struct request *clone, struct request *rq, | |
1483 | struct dm_rq_target_io *tio) | |
1484 | { | |
d0bcb878 | 1485 | int r; |
cec47e3d | 1486 | |
29e4013d TH |
1487 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1488 | dm_rq_bio_constructor, tio); | |
1489 | if (r) | |
1490 | return r; | |
cec47e3d | 1491 | |
29e4013d TH |
1492 | clone->cmd = rq->cmd; |
1493 | clone->cmd_len = rq->cmd_len; | |
1494 | clone->sense = rq->sense; | |
1495 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1496 | clone->end_io = end_clone_request; |
1497 | clone->end_io_data = tio; | |
1498 | ||
1499 | return 0; | |
1500 | } | |
1501 | ||
6facdaff KU |
1502 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1503 | gfp_t gfp_mask) | |
1504 | { | |
1505 | struct request *clone; | |
1506 | struct dm_rq_target_io *tio; | |
1507 | ||
1508 | tio = alloc_rq_tio(md, gfp_mask); | |
1509 | if (!tio) | |
1510 | return NULL; | |
1511 | ||
1512 | tio->md = md; | |
1513 | tio->ti = NULL; | |
1514 | tio->orig = rq; | |
1515 | tio->error = 0; | |
1516 | memset(&tio->info, 0, sizeof(tio->info)); | |
1517 | ||
1518 | clone = &tio->clone; | |
1519 | if (setup_clone(clone, rq, tio)) { | |
1520 | /* -ENOMEM */ | |
1521 | free_rq_tio(tio); | |
1522 | return NULL; | |
1523 | } | |
1524 | ||
1525 | return clone; | |
1526 | } | |
1527 | ||
cec47e3d KU |
1528 | /* |
1529 | * Called with the queue lock held. | |
1530 | */ | |
1531 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1532 | { | |
1533 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1534 | struct request *clone; |
1535 | ||
cec47e3d KU |
1536 | if (unlikely(rq->special)) { |
1537 | DMWARN("Already has something in rq->special."); | |
1538 | return BLKPREP_KILL; | |
1539 | } | |
1540 | ||
6facdaff KU |
1541 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1542 | if (!clone) | |
cec47e3d | 1543 | return BLKPREP_DEFER; |
cec47e3d KU |
1544 | |
1545 | rq->special = clone; | |
1546 | rq->cmd_flags |= REQ_DONTPREP; | |
1547 | ||
1548 | return BLKPREP_OK; | |
1549 | } | |
1550 | ||
9eef87da KU |
1551 | /* |
1552 | * Returns: | |
1553 | * 0 : the request has been processed (not requeued) | |
1554 | * !0 : the request has been requeued | |
1555 | */ | |
1556 | static int map_request(struct dm_target *ti, struct request *clone, | |
1557 | struct mapped_device *md) | |
cec47e3d | 1558 | { |
9eef87da | 1559 | int r, requeued = 0; |
cec47e3d KU |
1560 | struct dm_rq_target_io *tio = clone->end_io_data; |
1561 | ||
cec47e3d KU |
1562 | tio->ti = ti; |
1563 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1564 | switch (r) { | |
1565 | case DM_MAPIO_SUBMITTED: | |
1566 | /* The target has taken the I/O to submit by itself later */ | |
1567 | break; | |
1568 | case DM_MAPIO_REMAPPED: | |
1569 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1570 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1571 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1572 | dm_dispatch_request(clone); |
1573 | break; | |
1574 | case DM_MAPIO_REQUEUE: | |
1575 | /* The target wants to requeue the I/O */ | |
1576 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1577 | requeued = 1; |
cec47e3d KU |
1578 | break; |
1579 | default: | |
1580 | if (r > 0) { | |
1581 | DMWARN("unimplemented target map return value: %d", r); | |
1582 | BUG(); | |
1583 | } | |
1584 | ||
1585 | /* The target wants to complete the I/O */ | |
1586 | dm_kill_unmapped_request(clone, r); | |
1587 | break; | |
1588 | } | |
9eef87da KU |
1589 | |
1590 | return requeued; | |
cec47e3d KU |
1591 | } |
1592 | ||
ba1cbad9 MS |
1593 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1594 | { | |
1595 | struct request *clone; | |
1596 | ||
1597 | blk_start_request(orig); | |
1598 | clone = orig->special; | |
1599 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1600 | ||
1601 | /* | |
1602 | * Hold the md reference here for the in-flight I/O. | |
1603 | * We can't rely on the reference count by device opener, | |
1604 | * because the device may be closed during the request completion | |
1605 | * when all bios are completed. | |
1606 | * See the comment in rq_completed() too. | |
1607 | */ | |
1608 | dm_get(md); | |
1609 | ||
1610 | return clone; | |
1611 | } | |
1612 | ||
cec47e3d KU |
1613 | /* |
1614 | * q->request_fn for request-based dm. | |
1615 | * Called with the queue lock held. | |
1616 | */ | |
1617 | static void dm_request_fn(struct request_queue *q) | |
1618 | { | |
1619 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1620 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1621 | struct dm_target *ti; |
b4324fee | 1622 | struct request *rq, *clone; |
29e4013d | 1623 | sector_t pos; |
cec47e3d KU |
1624 | |
1625 | /* | |
b4324fee KU |
1626 | * For suspend, check blk_queue_stopped() and increment |
1627 | * ->pending within a single queue_lock not to increment the | |
1628 | * number of in-flight I/Os after the queue is stopped in | |
1629 | * dm_suspend(). | |
cec47e3d | 1630 | */ |
7eaceacc | 1631 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1632 | rq = blk_peek_request(q); |
1633 | if (!rq) | |
7eaceacc | 1634 | goto delay_and_out; |
cec47e3d | 1635 | |
29e4013d TH |
1636 | /* always use block 0 to find the target for flushes for now */ |
1637 | pos = 0; | |
1638 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1639 | pos = blk_rq_pos(rq); | |
1640 | ||
1641 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1642 | if (!dm_target_is_valid(ti)) { |
1643 | /* | |
1644 | * Must perform setup, that dm_done() requires, | |
1645 | * before calling dm_kill_unmapped_request | |
1646 | */ | |
1647 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1648 | clone = dm_start_request(md, rq); | |
1649 | dm_kill_unmapped_request(clone, -EIO); | |
1650 | continue; | |
1651 | } | |
d0bcb878 | 1652 | |
cec47e3d | 1653 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1654 | goto delay_and_out; |
cec47e3d | 1655 | |
ba1cbad9 | 1656 | clone = dm_start_request(md, rq); |
b4324fee | 1657 | |
cec47e3d | 1658 | spin_unlock(q->queue_lock); |
9eef87da KU |
1659 | if (map_request(ti, clone, md)) |
1660 | goto requeued; | |
1661 | ||
052189a2 KU |
1662 | BUG_ON(!irqs_disabled()); |
1663 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1664 | } |
1665 | ||
1666 | goto out; | |
1667 | ||
9eef87da | 1668 | requeued: |
052189a2 KU |
1669 | BUG_ON(!irqs_disabled()); |
1670 | spin_lock(q->queue_lock); | |
9eef87da | 1671 | |
7eaceacc JA |
1672 | delay_and_out: |
1673 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1674 | out: |
1675 | dm_table_put(map); | |
cec47e3d KU |
1676 | } |
1677 | ||
1678 | int dm_underlying_device_busy(struct request_queue *q) | |
1679 | { | |
1680 | return blk_lld_busy(q); | |
1681 | } | |
1682 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1683 | ||
1684 | static int dm_lld_busy(struct request_queue *q) | |
1685 | { | |
1686 | int r; | |
1687 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1688 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1689 | |
1690 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1691 | r = 1; | |
1692 | else | |
1693 | r = dm_table_any_busy_target(map); | |
1694 | ||
1695 | dm_table_put(map); | |
1696 | ||
1697 | return r; | |
1698 | } | |
1699 | ||
1da177e4 LT |
1700 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1701 | { | |
8a57dfc6 CS |
1702 | int r = bdi_bits; |
1703 | struct mapped_device *md = congested_data; | |
1704 | struct dm_table *map; | |
1da177e4 | 1705 | |
1eb787ec | 1706 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1707 | map = dm_get_live_table(md); |
8a57dfc6 | 1708 | if (map) { |
cec47e3d KU |
1709 | /* |
1710 | * Request-based dm cares about only own queue for | |
1711 | * the query about congestion status of request_queue | |
1712 | */ | |
1713 | if (dm_request_based(md)) | |
1714 | r = md->queue->backing_dev_info.state & | |
1715 | bdi_bits; | |
1716 | else | |
1717 | r = dm_table_any_congested(map, bdi_bits); | |
1718 | ||
8a57dfc6 CS |
1719 | dm_table_put(map); |
1720 | } | |
1721 | } | |
1722 | ||
1da177e4 LT |
1723 | return r; |
1724 | } | |
1725 | ||
1726 | /*----------------------------------------------------------------- | |
1727 | * An IDR is used to keep track of allocated minor numbers. | |
1728 | *---------------------------------------------------------------*/ | |
2b06cfff | 1729 | static void free_minor(int minor) |
1da177e4 | 1730 | { |
f32c10b0 | 1731 | spin_lock(&_minor_lock); |
1da177e4 | 1732 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1733 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1734 | } |
1735 | ||
1736 | /* | |
1737 | * See if the device with a specific minor # is free. | |
1738 | */ | |
cf13ab8e | 1739 | static int specific_minor(int minor) |
1da177e4 LT |
1740 | { |
1741 | int r, m; | |
1742 | ||
1743 | if (minor >= (1 << MINORBITS)) | |
1744 | return -EINVAL; | |
1745 | ||
62f75c2f JM |
1746 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1747 | if (!r) | |
1748 | return -ENOMEM; | |
1749 | ||
f32c10b0 | 1750 | spin_lock(&_minor_lock); |
1da177e4 LT |
1751 | |
1752 | if (idr_find(&_minor_idr, minor)) { | |
1753 | r = -EBUSY; | |
1754 | goto out; | |
1755 | } | |
1756 | ||
ba61fdd1 | 1757 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1758 | if (r) |
1da177e4 | 1759 | goto out; |
1da177e4 LT |
1760 | |
1761 | if (m != minor) { | |
1762 | idr_remove(&_minor_idr, m); | |
1763 | r = -EBUSY; | |
1764 | goto out; | |
1765 | } | |
1766 | ||
1767 | out: | |
f32c10b0 | 1768 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1769 | return r; |
1770 | } | |
1771 | ||
cf13ab8e | 1772 | static int next_free_minor(int *minor) |
1da177e4 | 1773 | { |
2b06cfff | 1774 | int r, m; |
1da177e4 | 1775 | |
1da177e4 | 1776 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1777 | if (!r) |
1778 | return -ENOMEM; | |
1779 | ||
f32c10b0 | 1780 | spin_lock(&_minor_lock); |
1da177e4 | 1781 | |
ba61fdd1 | 1782 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1783 | if (r) |
1da177e4 | 1784 | goto out; |
1da177e4 LT |
1785 | |
1786 | if (m >= (1 << MINORBITS)) { | |
1787 | idr_remove(&_minor_idr, m); | |
1788 | r = -ENOSPC; | |
1789 | goto out; | |
1790 | } | |
1791 | ||
1792 | *minor = m; | |
1793 | ||
1794 | out: | |
f32c10b0 | 1795 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1796 | return r; |
1797 | } | |
1798 | ||
83d5cde4 | 1799 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1800 | |
53d5914f MP |
1801 | static void dm_wq_work(struct work_struct *work); |
1802 | ||
4a0b4ddf MS |
1803 | static void dm_init_md_queue(struct mapped_device *md) |
1804 | { | |
1805 | /* | |
1806 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1807 | * devices. The type of this dm device has not been decided yet. | |
1808 | * The type is decided at the first table loading time. | |
1809 | * To prevent problematic device stacking, clear the queue flag | |
1810 | * for request stacking support until then. | |
1811 | * | |
1812 | * This queue is new, so no concurrency on the queue_flags. | |
1813 | */ | |
1814 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1815 | ||
1816 | md->queue->queuedata = md; | |
1817 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1818 | md->queue->backing_dev_info.congested_data = md; | |
1819 | blk_queue_make_request(md->queue, dm_request); | |
1820 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1821 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1822 | } | |
1823 | ||
1da177e4 LT |
1824 | /* |
1825 | * Allocate and initialise a blank device with a given minor. | |
1826 | */ | |
2b06cfff | 1827 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1828 | { |
1829 | int r; | |
cf13ab8e | 1830 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1831 | void *old_md; |
1da177e4 LT |
1832 | |
1833 | if (!md) { | |
1834 | DMWARN("unable to allocate device, out of memory."); | |
1835 | return NULL; | |
1836 | } | |
1837 | ||
10da4f79 | 1838 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1839 | goto bad_module_get; |
10da4f79 | 1840 | |
1da177e4 | 1841 | /* get a minor number for the dev */ |
2b06cfff | 1842 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1843 | r = next_free_minor(&minor); |
2b06cfff | 1844 | else |
cf13ab8e | 1845 | r = specific_minor(minor); |
1da177e4 | 1846 | if (r < 0) |
6ed7ade8 | 1847 | goto bad_minor; |
1da177e4 | 1848 | |
a5664dad | 1849 | md->type = DM_TYPE_NONE; |
2ca3310e | 1850 | init_rwsem(&md->io_lock); |
e61290a4 | 1851 | mutex_init(&md->suspend_lock); |
a5664dad | 1852 | mutex_init(&md->type_lock); |
022c2611 | 1853 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1854 | rwlock_init(&md->map_lock); |
1855 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1856 | atomic_set(&md->open_count, 0); |
1da177e4 | 1857 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1858 | atomic_set(&md->uevent_seq, 0); |
1859 | INIT_LIST_HEAD(&md->uevent_list); | |
1860 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1861 | |
4a0b4ddf | 1862 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1863 | if (!md->queue) |
6ed7ade8 | 1864 | goto bad_queue; |
1da177e4 | 1865 | |
4a0b4ddf | 1866 | dm_init_md_queue(md); |
9faf400f | 1867 | |
1da177e4 LT |
1868 | md->disk = alloc_disk(1); |
1869 | if (!md->disk) | |
6ed7ade8 | 1870 | goto bad_disk; |
1da177e4 | 1871 | |
316d315b NK |
1872 | atomic_set(&md->pending[0], 0); |
1873 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1874 | init_waitqueue_head(&md->wait); |
53d5914f | 1875 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1876 | init_waitqueue_head(&md->eventq); |
1877 | ||
1da177e4 LT |
1878 | md->disk->major = _major; |
1879 | md->disk->first_minor = minor; | |
1880 | md->disk->fops = &dm_blk_dops; | |
1881 | md->disk->queue = md->queue; | |
1882 | md->disk->private_data = md; | |
1883 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1884 | add_disk(md->disk); | |
7e51f257 | 1885 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1886 | |
9c4376de TH |
1887 | md->wq = alloc_workqueue("kdmflush", |
1888 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1889 | if (!md->wq) |
1890 | goto bad_thread; | |
1891 | ||
32a926da MP |
1892 | md->bdev = bdget_disk(md->disk, 0); |
1893 | if (!md->bdev) | |
1894 | goto bad_bdev; | |
1895 | ||
6a8736d1 TH |
1896 | bio_init(&md->flush_bio); |
1897 | md->flush_bio.bi_bdev = md->bdev; | |
1898 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1899 | ||
ba61fdd1 | 1900 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1901 | spin_lock(&_minor_lock); |
ba61fdd1 | 1902 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1903 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1904 | |
1905 | BUG_ON(old_md != MINOR_ALLOCED); | |
1906 | ||
1da177e4 LT |
1907 | return md; |
1908 | ||
32a926da MP |
1909 | bad_bdev: |
1910 | destroy_workqueue(md->wq); | |
304f3f6a | 1911 | bad_thread: |
03022c54 | 1912 | del_gendisk(md->disk); |
304f3f6a | 1913 | put_disk(md->disk); |
6ed7ade8 | 1914 | bad_disk: |
1312f40e | 1915 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1916 | bad_queue: |
1da177e4 | 1917 | free_minor(minor); |
6ed7ade8 | 1918 | bad_minor: |
10da4f79 | 1919 | module_put(THIS_MODULE); |
6ed7ade8 | 1920 | bad_module_get: |
1da177e4 LT |
1921 | kfree(md); |
1922 | return NULL; | |
1923 | } | |
1924 | ||
ae9da83f JN |
1925 | static void unlock_fs(struct mapped_device *md); |
1926 | ||
1da177e4 LT |
1927 | static void free_dev(struct mapped_device *md) |
1928 | { | |
f331c029 | 1929 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1930 | |
32a926da MP |
1931 | unlock_fs(md); |
1932 | bdput(md->bdev); | |
304f3f6a | 1933 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1934 | if (md->tio_pool) |
1935 | mempool_destroy(md->tio_pool); | |
1936 | if (md->io_pool) | |
1937 | mempool_destroy(md->io_pool); | |
1938 | if (md->bs) | |
1939 | bioset_free(md->bs); | |
9c47008d | 1940 | blk_integrity_unregister(md->disk); |
1da177e4 | 1941 | del_gendisk(md->disk); |
63d94e48 | 1942 | free_minor(minor); |
fba9f90e JM |
1943 | |
1944 | spin_lock(&_minor_lock); | |
1945 | md->disk->private_data = NULL; | |
1946 | spin_unlock(&_minor_lock); | |
1947 | ||
1da177e4 | 1948 | put_disk(md->disk); |
1312f40e | 1949 | blk_cleanup_queue(md->queue); |
10da4f79 | 1950 | module_put(THIS_MODULE); |
1da177e4 LT |
1951 | kfree(md); |
1952 | } | |
1953 | ||
e6ee8c0b KU |
1954 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1955 | { | |
1956 | struct dm_md_mempools *p; | |
1957 | ||
1958 | if (md->io_pool && md->tio_pool && md->bs) | |
1959 | /* the md already has necessary mempools */ | |
1960 | goto out; | |
1961 | ||
1962 | p = dm_table_get_md_mempools(t); | |
1963 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1964 | ||
1965 | md->io_pool = p->io_pool; | |
1966 | p->io_pool = NULL; | |
1967 | md->tio_pool = p->tio_pool; | |
1968 | p->tio_pool = NULL; | |
1969 | md->bs = p->bs; | |
1970 | p->bs = NULL; | |
1971 | ||
1972 | out: | |
1973 | /* mempool bind completed, now no need any mempools in the table */ | |
1974 | dm_table_free_md_mempools(t); | |
1975 | } | |
1976 | ||
1da177e4 LT |
1977 | /* |
1978 | * Bind a table to the device. | |
1979 | */ | |
1980 | static void event_callback(void *context) | |
1981 | { | |
7a8c3d3b MA |
1982 | unsigned long flags; |
1983 | LIST_HEAD(uevents); | |
1da177e4 LT |
1984 | struct mapped_device *md = (struct mapped_device *) context; |
1985 | ||
7a8c3d3b MA |
1986 | spin_lock_irqsave(&md->uevent_lock, flags); |
1987 | list_splice_init(&md->uevent_list, &uevents); | |
1988 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1989 | ||
ed9e1982 | 1990 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1991 | |
1da177e4 LT |
1992 | atomic_inc(&md->event_nr); |
1993 | wake_up(&md->eventq); | |
1994 | } | |
1995 | ||
c217649b MS |
1996 | /* |
1997 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1998 | */ | |
4e90188b | 1999 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2000 | { |
4e90188b | 2001 | set_capacity(md->disk, size); |
1da177e4 | 2002 | |
db8fef4f | 2003 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2004 | } |
2005 | ||
d5b9dd04 MP |
2006 | /* |
2007 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2008 | * | |
2009 | * If this function returns 0, then the device is either a non-dm | |
2010 | * device without a merge_bvec_fn, or it is a dm device that is | |
2011 | * able to split any bios it receives that are too big. | |
2012 | */ | |
2013 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2014 | { | |
2015 | struct mapped_device *dev_md; | |
2016 | ||
2017 | if (!q->merge_bvec_fn) | |
2018 | return 0; | |
2019 | ||
2020 | if (q->make_request_fn == dm_request) { | |
2021 | dev_md = q->queuedata; | |
2022 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2023 | return 0; | |
2024 | } | |
2025 | ||
2026 | return 1; | |
2027 | } | |
2028 | ||
2029 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2030 | struct dm_dev *dev, sector_t start, | |
2031 | sector_t len, void *data) | |
2032 | { | |
2033 | struct block_device *bdev = dev->bdev; | |
2034 | struct request_queue *q = bdev_get_queue(bdev); | |
2035 | ||
2036 | return dm_queue_merge_is_compulsory(q); | |
2037 | } | |
2038 | ||
2039 | /* | |
2040 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2041 | * on the properties of the underlying devices. | |
2042 | */ | |
2043 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2044 | { | |
2045 | unsigned i = 0; | |
2046 | struct dm_target *ti; | |
2047 | ||
2048 | while (i < dm_table_get_num_targets(table)) { | |
2049 | ti = dm_table_get_target(table, i++); | |
2050 | ||
2051 | if (ti->type->iterate_devices && | |
2052 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2053 | return 0; | |
2054 | } | |
2055 | ||
2056 | return 1; | |
2057 | } | |
2058 | ||
042d2a9b AK |
2059 | /* |
2060 | * Returns old map, which caller must destroy. | |
2061 | */ | |
2062 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2063 | struct queue_limits *limits) | |
1da177e4 | 2064 | { |
042d2a9b | 2065 | struct dm_table *old_map; |
165125e1 | 2066 | struct request_queue *q = md->queue; |
1da177e4 | 2067 | sector_t size; |
523d9297 | 2068 | unsigned long flags; |
d5b9dd04 | 2069 | int merge_is_optional; |
1da177e4 LT |
2070 | |
2071 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2072 | |
2073 | /* | |
2074 | * Wipe any geometry if the size of the table changed. | |
2075 | */ | |
2076 | if (size != get_capacity(md->disk)) | |
2077 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2078 | ||
32a926da | 2079 | __set_size(md, size); |
d5816876 | 2080 | |
2ca3310e AK |
2081 | dm_table_event_callback(t, event_callback, md); |
2082 | ||
e6ee8c0b KU |
2083 | /* |
2084 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2085 | * for request-based during suspension. So stop it to prevent | |
2086 | * I/O mapping before resume. | |
2087 | * This must be done before setting the queue restrictions, | |
2088 | * because request-based dm may be run just after the setting. | |
2089 | */ | |
2090 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2091 | stop_queue(q); | |
2092 | ||
2093 | __bind_mempools(md, t); | |
2094 | ||
d5b9dd04 MP |
2095 | merge_is_optional = dm_table_merge_is_optional(t); |
2096 | ||
523d9297 | 2097 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2098 | old_map = md->map; |
1da177e4 | 2099 | md->map = t; |
36a0456f AK |
2100 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2101 | ||
754c5fc7 | 2102 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2103 | if (merge_is_optional) |
2104 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2105 | else | |
2106 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2107 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2108 | |
042d2a9b | 2109 | return old_map; |
1da177e4 LT |
2110 | } |
2111 | ||
a7940155 AK |
2112 | /* |
2113 | * Returns unbound table for the caller to free. | |
2114 | */ | |
2115 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2116 | { |
2117 | struct dm_table *map = md->map; | |
523d9297 | 2118 | unsigned long flags; |
1da177e4 LT |
2119 | |
2120 | if (!map) | |
a7940155 | 2121 | return NULL; |
1da177e4 LT |
2122 | |
2123 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2124 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2125 | md->map = NULL; |
523d9297 | 2126 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2127 | |
2128 | return map; | |
1da177e4 LT |
2129 | } |
2130 | ||
2131 | /* | |
2132 | * Constructor for a new device. | |
2133 | */ | |
2b06cfff | 2134 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2135 | { |
2136 | struct mapped_device *md; | |
2137 | ||
2b06cfff | 2138 | md = alloc_dev(minor); |
1da177e4 LT |
2139 | if (!md) |
2140 | return -ENXIO; | |
2141 | ||
784aae73 MB |
2142 | dm_sysfs_init(md); |
2143 | ||
1da177e4 LT |
2144 | *result = md; |
2145 | return 0; | |
2146 | } | |
2147 | ||
a5664dad MS |
2148 | /* |
2149 | * Functions to manage md->type. | |
2150 | * All are required to hold md->type_lock. | |
2151 | */ | |
2152 | void dm_lock_md_type(struct mapped_device *md) | |
2153 | { | |
2154 | mutex_lock(&md->type_lock); | |
2155 | } | |
2156 | ||
2157 | void dm_unlock_md_type(struct mapped_device *md) | |
2158 | { | |
2159 | mutex_unlock(&md->type_lock); | |
2160 | } | |
2161 | ||
2162 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2163 | { | |
2164 | md->type = type; | |
2165 | } | |
2166 | ||
2167 | unsigned dm_get_md_type(struct mapped_device *md) | |
2168 | { | |
2169 | return md->type; | |
2170 | } | |
2171 | ||
36a0456f AK |
2172 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2173 | { | |
2174 | return md->immutable_target_type; | |
2175 | } | |
2176 | ||
4a0b4ddf MS |
2177 | /* |
2178 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2179 | */ | |
2180 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2181 | { | |
2182 | struct request_queue *q = NULL; | |
2183 | ||
2184 | if (md->queue->elevator) | |
2185 | return 1; | |
2186 | ||
2187 | /* Fully initialize the queue */ | |
2188 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2189 | if (!q) | |
2190 | return 0; | |
2191 | ||
2192 | md->queue = q; | |
4a0b4ddf MS |
2193 | dm_init_md_queue(md); |
2194 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2195 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2196 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2197 | |
2198 | elv_register_queue(md->queue); | |
2199 | ||
2200 | return 1; | |
2201 | } | |
2202 | ||
2203 | /* | |
2204 | * Setup the DM device's queue based on md's type | |
2205 | */ | |
2206 | int dm_setup_md_queue(struct mapped_device *md) | |
2207 | { | |
2208 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2209 | !dm_init_request_based_queue(md)) { | |
2210 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2211 | return -EINVAL; | |
2212 | } | |
2213 | ||
2214 | return 0; | |
2215 | } | |
2216 | ||
637842cf | 2217 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2218 | { |
2219 | struct mapped_device *md; | |
1da177e4 LT |
2220 | unsigned minor = MINOR(dev); |
2221 | ||
2222 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2223 | return NULL; | |
2224 | ||
f32c10b0 | 2225 | spin_lock(&_minor_lock); |
1da177e4 LT |
2226 | |
2227 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2228 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2229 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2230 | dm_deleting_md(md) || |
17b2f66f | 2231 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2232 | md = NULL; |
fba9f90e JM |
2233 | goto out; |
2234 | } | |
1da177e4 | 2235 | |
fba9f90e | 2236 | out: |
f32c10b0 | 2237 | spin_unlock(&_minor_lock); |
1da177e4 | 2238 | |
637842cf DT |
2239 | return md; |
2240 | } | |
2241 | ||
d229a958 DT |
2242 | struct mapped_device *dm_get_md(dev_t dev) |
2243 | { | |
2244 | struct mapped_device *md = dm_find_md(dev); | |
2245 | ||
2246 | if (md) | |
2247 | dm_get(md); | |
2248 | ||
2249 | return md; | |
2250 | } | |
3cf2e4ba | 2251 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2252 | |
9ade92a9 | 2253 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2254 | { |
9ade92a9 | 2255 | return md->interface_ptr; |
1da177e4 LT |
2256 | } |
2257 | ||
2258 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2259 | { | |
2260 | md->interface_ptr = ptr; | |
2261 | } | |
2262 | ||
2263 | void dm_get(struct mapped_device *md) | |
2264 | { | |
2265 | atomic_inc(&md->holders); | |
3f77316d | 2266 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2267 | } |
2268 | ||
72d94861 AK |
2269 | const char *dm_device_name(struct mapped_device *md) |
2270 | { | |
2271 | return md->name; | |
2272 | } | |
2273 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2274 | ||
3f77316d | 2275 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2276 | { |
1134e5ae | 2277 | struct dm_table *map; |
1da177e4 | 2278 | |
3f77316d | 2279 | might_sleep(); |
fba9f90e | 2280 | |
3f77316d KU |
2281 | spin_lock(&_minor_lock); |
2282 | map = dm_get_live_table(md); | |
2283 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2284 | set_bit(DMF_FREEING, &md->flags); | |
2285 | spin_unlock(&_minor_lock); | |
2286 | ||
2287 | if (!dm_suspended_md(md)) { | |
2288 | dm_table_presuspend_targets(map); | |
2289 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2290 | } |
3f77316d KU |
2291 | |
2292 | /* | |
2293 | * Rare, but there may be I/O requests still going to complete, | |
2294 | * for example. Wait for all references to disappear. | |
2295 | * No one should increment the reference count of the mapped_device, | |
2296 | * after the mapped_device state becomes DMF_FREEING. | |
2297 | */ | |
2298 | if (wait) | |
2299 | while (atomic_read(&md->holders)) | |
2300 | msleep(1); | |
2301 | else if (atomic_read(&md->holders)) | |
2302 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2303 | dm_device_name(md), atomic_read(&md->holders)); | |
2304 | ||
2305 | dm_sysfs_exit(md); | |
2306 | dm_table_put(map); | |
2307 | dm_table_destroy(__unbind(md)); | |
2308 | free_dev(md); | |
2309 | } | |
2310 | ||
2311 | void dm_destroy(struct mapped_device *md) | |
2312 | { | |
2313 | __dm_destroy(md, true); | |
2314 | } | |
2315 | ||
2316 | void dm_destroy_immediate(struct mapped_device *md) | |
2317 | { | |
2318 | __dm_destroy(md, false); | |
2319 | } | |
2320 | ||
2321 | void dm_put(struct mapped_device *md) | |
2322 | { | |
2323 | atomic_dec(&md->holders); | |
1da177e4 | 2324 | } |
79eb885c | 2325 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2326 | |
401600df | 2327 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2328 | { |
2329 | int r = 0; | |
b44ebeb0 MP |
2330 | DECLARE_WAITQUEUE(wait, current); |
2331 | ||
b44ebeb0 | 2332 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2333 | |
2334 | while (1) { | |
401600df | 2335 | set_current_state(interruptible); |
46125c1c | 2336 | |
b4324fee | 2337 | if (!md_in_flight(md)) |
46125c1c MB |
2338 | break; |
2339 | ||
401600df MP |
2340 | if (interruptible == TASK_INTERRUPTIBLE && |
2341 | signal_pending(current)) { | |
46125c1c MB |
2342 | r = -EINTR; |
2343 | break; | |
2344 | } | |
2345 | ||
2346 | io_schedule(); | |
2347 | } | |
2348 | set_current_state(TASK_RUNNING); | |
2349 | ||
b44ebeb0 MP |
2350 | remove_wait_queue(&md->wait, &wait); |
2351 | ||
46125c1c MB |
2352 | return r; |
2353 | } | |
2354 | ||
1da177e4 LT |
2355 | /* |
2356 | * Process the deferred bios | |
2357 | */ | |
ef208587 | 2358 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2359 | { |
ef208587 MP |
2360 | struct mapped_device *md = container_of(work, struct mapped_device, |
2361 | work); | |
6d6f10df | 2362 | struct bio *c; |
1da177e4 | 2363 | |
6a8736d1 | 2364 | down_read(&md->io_lock); |
ef208587 | 2365 | |
3b00b203 | 2366 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2367 | spin_lock_irq(&md->deferred_lock); |
2368 | c = bio_list_pop(&md->deferred); | |
2369 | spin_unlock_irq(&md->deferred_lock); | |
2370 | ||
6a8736d1 | 2371 | if (!c) |
df12ee99 | 2372 | break; |
022c2611 | 2373 | |
6a8736d1 | 2374 | up_read(&md->io_lock); |
3b00b203 | 2375 | |
e6ee8c0b KU |
2376 | if (dm_request_based(md)) |
2377 | generic_make_request(c); | |
6a8736d1 TH |
2378 | else |
2379 | __split_and_process_bio(md, c); | |
3b00b203 | 2380 | |
6a8736d1 | 2381 | down_read(&md->io_lock); |
022c2611 | 2382 | } |
73d410c0 | 2383 | |
6a8736d1 | 2384 | up_read(&md->io_lock); |
1da177e4 LT |
2385 | } |
2386 | ||
9a1fb464 | 2387 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2388 | { |
3b00b203 MP |
2389 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2390 | smp_mb__after_clear_bit(); | |
53d5914f | 2391 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2392 | } |
2393 | ||
1da177e4 | 2394 | /* |
042d2a9b | 2395 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2396 | */ |
042d2a9b | 2397 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2398 | { |
3ae70656 | 2399 | struct dm_table *live_map, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2400 | struct queue_limits limits; |
042d2a9b | 2401 | int r; |
1da177e4 | 2402 | |
e61290a4 | 2403 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2404 | |
2405 | /* device must be suspended */ | |
4f186f8b | 2406 | if (!dm_suspended_md(md)) |
93c534ae | 2407 | goto out; |
1da177e4 | 2408 | |
3ae70656 MS |
2409 | /* |
2410 | * If the new table has no data devices, retain the existing limits. | |
2411 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2412 | * then new I/O is queued based on these limits, and then some paths | |
2413 | * reappear. | |
2414 | */ | |
2415 | if (dm_table_has_no_data_devices(table)) { | |
2416 | live_map = dm_get_live_table(md); | |
2417 | if (live_map) | |
2418 | limits = md->queue->limits; | |
2419 | dm_table_put(live_map); | |
2420 | } | |
2421 | ||
754c5fc7 | 2422 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2423 | if (r) { |
2424 | map = ERR_PTR(r); | |
754c5fc7 | 2425 | goto out; |
042d2a9b | 2426 | } |
754c5fc7 | 2427 | |
042d2a9b | 2428 | map = __bind(md, table, &limits); |
1da177e4 | 2429 | |
93c534ae | 2430 | out: |
e61290a4 | 2431 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2432 | return map; |
1da177e4 LT |
2433 | } |
2434 | ||
2435 | /* | |
2436 | * Functions to lock and unlock any filesystem running on the | |
2437 | * device. | |
2438 | */ | |
2ca3310e | 2439 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2440 | { |
e39e2e95 | 2441 | int r; |
1da177e4 LT |
2442 | |
2443 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2444 | |
db8fef4f | 2445 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2446 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2447 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2448 | md->frozen_sb = NULL; |
2449 | return r; | |
dfbe03f6 AK |
2450 | } |
2451 | ||
aa8d7c2f AK |
2452 | set_bit(DMF_FROZEN, &md->flags); |
2453 | ||
1da177e4 LT |
2454 | return 0; |
2455 | } | |
2456 | ||
2ca3310e | 2457 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2458 | { |
aa8d7c2f AK |
2459 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2460 | return; | |
2461 | ||
db8fef4f | 2462 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2463 | md->frozen_sb = NULL; |
aa8d7c2f | 2464 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2465 | } |
2466 | ||
2467 | /* | |
2468 | * We need to be able to change a mapping table under a mounted | |
2469 | * filesystem. For example we might want to move some data in | |
2470 | * the background. Before the table can be swapped with | |
2471 | * dm_bind_table, dm_suspend must be called to flush any in | |
2472 | * flight bios and ensure that any further io gets deferred. | |
2473 | */ | |
cec47e3d KU |
2474 | /* |
2475 | * Suspend mechanism in request-based dm. | |
2476 | * | |
9f518b27 KU |
2477 | * 1. Flush all I/Os by lock_fs() if needed. |
2478 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2479 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2480 | * |
9f518b27 | 2481 | * To abort suspend, start the request_queue. |
cec47e3d | 2482 | */ |
a3d77d35 | 2483 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2484 | { |
2ca3310e | 2485 | struct dm_table *map = NULL; |
46125c1c | 2486 | int r = 0; |
a3d77d35 | 2487 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2488 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2489 | |
e61290a4 | 2490 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2491 | |
4f186f8b | 2492 | if (dm_suspended_md(md)) { |
73d410c0 | 2493 | r = -EINVAL; |
d287483d | 2494 | goto out_unlock; |
73d410c0 | 2495 | } |
1da177e4 | 2496 | |
7c666411 | 2497 | map = dm_get_live_table(md); |
1da177e4 | 2498 | |
2e93ccc1 KU |
2499 | /* |
2500 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2501 | * This flag is cleared before dm_suspend returns. | |
2502 | */ | |
2503 | if (noflush) | |
2504 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2505 | ||
cf222b37 AK |
2506 | /* This does not get reverted if there's an error later. */ |
2507 | dm_table_presuspend_targets(map); | |
2508 | ||
32a926da | 2509 | /* |
9f518b27 KU |
2510 | * Flush I/O to the device. |
2511 | * Any I/O submitted after lock_fs() may not be flushed. | |
2512 | * noflush takes precedence over do_lockfs. | |
2513 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2514 | */ |
2515 | if (!noflush && do_lockfs) { | |
2516 | r = lock_fs(md); | |
2517 | if (r) | |
f431d966 | 2518 | goto out; |
aa8d7c2f | 2519 | } |
1da177e4 LT |
2520 | |
2521 | /* | |
3b00b203 MP |
2522 | * Here we must make sure that no processes are submitting requests |
2523 | * to target drivers i.e. no one may be executing | |
2524 | * __split_and_process_bio. This is called from dm_request and | |
2525 | * dm_wq_work. | |
2526 | * | |
2527 | * To get all processes out of __split_and_process_bio in dm_request, | |
2528 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2529 | * __split_and_process_bio from dm_request and quiesce the thread |
2530 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2531 | * flush_workqueue(md->wq). | |
1da177e4 | 2532 | */ |
2ca3310e | 2533 | down_write(&md->io_lock); |
1eb787ec | 2534 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2535 | up_write(&md->io_lock); |
1da177e4 | 2536 | |
d0bcb878 | 2537 | /* |
29e4013d TH |
2538 | * Stop md->queue before flushing md->wq in case request-based |
2539 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2540 | */ |
cec47e3d | 2541 | if (dm_request_based(md)) |
9f518b27 | 2542 | stop_queue(md->queue); |
cec47e3d | 2543 | |
d0bcb878 KU |
2544 | flush_workqueue(md->wq); |
2545 | ||
1da177e4 | 2546 | /* |
3b00b203 MP |
2547 | * At this point no more requests are entering target request routines. |
2548 | * We call dm_wait_for_completion to wait for all existing requests | |
2549 | * to finish. | |
1da177e4 | 2550 | */ |
401600df | 2551 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2552 | |
2ca3310e | 2553 | down_write(&md->io_lock); |
6d6f10df | 2554 | if (noflush) |
022c2611 | 2555 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2556 | up_write(&md->io_lock); |
2e93ccc1 | 2557 | |
1da177e4 | 2558 | /* were we interrupted ? */ |
46125c1c | 2559 | if (r < 0) { |
9a1fb464 | 2560 | dm_queue_flush(md); |
73d410c0 | 2561 | |
cec47e3d | 2562 | if (dm_request_based(md)) |
9f518b27 | 2563 | start_queue(md->queue); |
cec47e3d | 2564 | |
2ca3310e | 2565 | unlock_fs(md); |
2e93ccc1 | 2566 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2567 | } |
1da177e4 | 2568 | |
3b00b203 MP |
2569 | /* |
2570 | * If dm_wait_for_completion returned 0, the device is completely | |
2571 | * quiescent now. There is no request-processing activity. All new | |
2572 | * requests are being added to md->deferred list. | |
2573 | */ | |
2574 | ||
2ca3310e | 2575 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2576 | |
4d4471cb KU |
2577 | dm_table_postsuspend_targets(map); |
2578 | ||
2ca3310e AK |
2579 | out: |
2580 | dm_table_put(map); | |
d287483d AK |
2581 | |
2582 | out_unlock: | |
e61290a4 | 2583 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2584 | return r; |
1da177e4 LT |
2585 | } |
2586 | ||
2587 | int dm_resume(struct mapped_device *md) | |
2588 | { | |
cf222b37 | 2589 | int r = -EINVAL; |
cf222b37 | 2590 | struct dm_table *map = NULL; |
1da177e4 | 2591 | |
e61290a4 | 2592 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2593 | if (!dm_suspended_md(md)) |
cf222b37 | 2594 | goto out; |
cf222b37 | 2595 | |
7c666411 | 2596 | map = dm_get_live_table(md); |
2ca3310e | 2597 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2598 | goto out; |
1da177e4 | 2599 | |
8757b776 MB |
2600 | r = dm_table_resume_targets(map); |
2601 | if (r) | |
2602 | goto out; | |
2ca3310e | 2603 | |
9a1fb464 | 2604 | dm_queue_flush(md); |
2ca3310e | 2605 | |
cec47e3d KU |
2606 | /* |
2607 | * Flushing deferred I/Os must be done after targets are resumed | |
2608 | * so that mapping of targets can work correctly. | |
2609 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2610 | */ | |
2611 | if (dm_request_based(md)) | |
2612 | start_queue(md->queue); | |
2613 | ||
2ca3310e AK |
2614 | unlock_fs(md); |
2615 | ||
2616 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2617 | ||
cf222b37 AK |
2618 | r = 0; |
2619 | out: | |
2620 | dm_table_put(map); | |
e61290a4 | 2621 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2622 | |
cf222b37 | 2623 | return r; |
1da177e4 LT |
2624 | } |
2625 | ||
2626 | /*----------------------------------------------------------------- | |
2627 | * Event notification. | |
2628 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2629 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2630 | unsigned cookie) |
69267a30 | 2631 | { |
60935eb2 MB |
2632 | char udev_cookie[DM_COOKIE_LENGTH]; |
2633 | char *envp[] = { udev_cookie, NULL }; | |
2634 | ||
2635 | if (!cookie) | |
3abf85b5 | 2636 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2637 | else { |
2638 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2639 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2640 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2641 | action, envp); | |
60935eb2 | 2642 | } |
69267a30 AK |
2643 | } |
2644 | ||
7a8c3d3b MA |
2645 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2646 | { | |
2647 | return atomic_add_return(1, &md->uevent_seq); | |
2648 | } | |
2649 | ||
1da177e4 LT |
2650 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2651 | { | |
2652 | return atomic_read(&md->event_nr); | |
2653 | } | |
2654 | ||
2655 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2656 | { | |
2657 | return wait_event_interruptible(md->eventq, | |
2658 | (event_nr != atomic_read(&md->event_nr))); | |
2659 | } | |
2660 | ||
7a8c3d3b MA |
2661 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2662 | { | |
2663 | unsigned long flags; | |
2664 | ||
2665 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2666 | list_add(elist, &md->uevent_list); | |
2667 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2668 | } | |
2669 | ||
1da177e4 LT |
2670 | /* |
2671 | * The gendisk is only valid as long as you have a reference | |
2672 | * count on 'md'. | |
2673 | */ | |
2674 | struct gendisk *dm_disk(struct mapped_device *md) | |
2675 | { | |
2676 | return md->disk; | |
2677 | } | |
2678 | ||
784aae73 MB |
2679 | struct kobject *dm_kobject(struct mapped_device *md) |
2680 | { | |
2681 | return &md->kobj; | |
2682 | } | |
2683 | ||
2684 | /* | |
2685 | * struct mapped_device should not be exported outside of dm.c | |
2686 | * so use this check to verify that kobj is part of md structure | |
2687 | */ | |
2688 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2689 | { | |
2690 | struct mapped_device *md; | |
2691 | ||
2692 | md = container_of(kobj, struct mapped_device, kobj); | |
2693 | if (&md->kobj != kobj) | |
2694 | return NULL; | |
2695 | ||
4d89b7b4 | 2696 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2697 | dm_deleting_md(md)) |
4d89b7b4 MB |
2698 | return NULL; |
2699 | ||
784aae73 MB |
2700 | dm_get(md); |
2701 | return md; | |
2702 | } | |
2703 | ||
4f186f8b | 2704 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2705 | { |
2706 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2707 | } | |
2708 | ||
64dbce58 KU |
2709 | int dm_suspended(struct dm_target *ti) |
2710 | { | |
ecdb2e25 | 2711 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2712 | } |
2713 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2714 | ||
2e93ccc1 KU |
2715 | int dm_noflush_suspending(struct dm_target *ti) |
2716 | { | |
ecdb2e25 | 2717 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2718 | } |
2719 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2720 | ||
a91a2785 | 2721 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity) |
e6ee8c0b KU |
2722 | { |
2723 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2724 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2725 | |
2726 | if (!pools) | |
2727 | return NULL; | |
2728 | ||
2729 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2730 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2731 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2732 | if (!pools->io_pool) | |
2733 | goto free_pools_and_out; | |
2734 | ||
2735 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2736 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2737 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2738 | if (!pools->tio_pool) | |
2739 | goto free_io_pool_and_out; | |
2740 | ||
94818742 KO |
2741 | pools->bs = (type == DM_TYPE_BIO_BASED) ? |
2742 | bioset_create(pool_size, 0) : | |
2743 | bioset_create(pool_size, | |
2744 | offsetof(struct dm_rq_clone_bio_info, clone)); | |
e6ee8c0b KU |
2745 | if (!pools->bs) |
2746 | goto free_tio_pool_and_out; | |
2747 | ||
a91a2785 MP |
2748 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2749 | goto free_bioset_and_out; | |
2750 | ||
e6ee8c0b KU |
2751 | return pools; |
2752 | ||
a91a2785 MP |
2753 | free_bioset_and_out: |
2754 | bioset_free(pools->bs); | |
2755 | ||
e6ee8c0b KU |
2756 | free_tio_pool_and_out: |
2757 | mempool_destroy(pools->tio_pool); | |
2758 | ||
2759 | free_io_pool_and_out: | |
2760 | mempool_destroy(pools->io_pool); | |
2761 | ||
2762 | free_pools_and_out: | |
2763 | kfree(pools); | |
2764 | ||
2765 | return NULL; | |
2766 | } | |
2767 | ||
2768 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2769 | { | |
2770 | if (!pools) | |
2771 | return; | |
2772 | ||
2773 | if (pools->io_pool) | |
2774 | mempool_destroy(pools->io_pool); | |
2775 | ||
2776 | if (pools->tio_pool) | |
2777 | mempool_destroy(pools->tio_pool); | |
2778 | ||
2779 | if (pools->bs) | |
2780 | bioset_free(pools->bs); | |
2781 | ||
2782 | kfree(pools); | |
2783 | } | |
2784 | ||
83d5cde4 | 2785 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2786 | .open = dm_blk_open, |
2787 | .release = dm_blk_close, | |
aa129a22 | 2788 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2789 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2790 | .owner = THIS_MODULE |
2791 | }; | |
2792 | ||
2793 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2794 | ||
2795 | /* | |
2796 | * module hooks | |
2797 | */ | |
2798 | module_init(dm_init); | |
2799 | module_exit(dm_exit); | |
2800 | ||
2801 | module_param(major, uint, 0); | |
2802 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2803 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2804 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2805 | MODULE_LICENSE("GPL"); |