]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/buffer_head.h> | |
6e9624b8 | 18 | #include <linux/smp_lock.h> |
1da177e4 LT |
19 | #include <linux/mempool.h> |
20 | #include <linux/slab.h> | |
21 | #include <linux/idr.h> | |
3ac51e74 | 22 | #include <linux/hdreg.h> |
3f77316d | 23 | #include <linux/delay.h> |
55782138 LZ |
24 | |
25 | #include <trace/events/block.h> | |
1da177e4 | 26 | |
72d94861 AK |
27 | #define DM_MSG_PREFIX "core" |
28 | ||
60935eb2 MB |
29 | /* |
30 | * Cookies are numeric values sent with CHANGE and REMOVE | |
31 | * uevents while resuming, removing or renaming the device. | |
32 | */ | |
33 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
34 | #define DM_COOKIE_LENGTH 24 | |
35 | ||
1da177e4 LT |
36 | static const char *_name = DM_NAME; |
37 | ||
38 | static unsigned int major = 0; | |
39 | static unsigned int _major = 0; | |
40 | ||
f32c10b0 | 41 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 42 | /* |
8fbf26ad | 43 | * For bio-based dm. |
1da177e4 LT |
44 | * One of these is allocated per bio. |
45 | */ | |
46 | struct dm_io { | |
47 | struct mapped_device *md; | |
48 | int error; | |
1da177e4 | 49 | atomic_t io_count; |
6ae2fa67 | 50 | struct bio *bio; |
3eaf840e | 51 | unsigned long start_time; |
f88fb981 | 52 | spinlock_t endio_lock; |
1da177e4 LT |
53 | }; |
54 | ||
55 | /* | |
8fbf26ad | 56 | * For bio-based dm. |
1da177e4 LT |
57 | * One of these is allocated per target within a bio. Hopefully |
58 | * this will be simplified out one day. | |
59 | */ | |
028867ac | 60 | struct dm_target_io { |
1da177e4 LT |
61 | struct dm_io *io; |
62 | struct dm_target *ti; | |
63 | union map_info info; | |
64 | }; | |
65 | ||
8fbf26ad KU |
66 | /* |
67 | * For request-based dm. | |
68 | * One of these is allocated per request. | |
69 | */ | |
70 | struct dm_rq_target_io { | |
71 | struct mapped_device *md; | |
72 | struct dm_target *ti; | |
73 | struct request *orig, clone; | |
74 | int error; | |
75 | union map_info info; | |
76 | }; | |
77 | ||
78 | /* | |
79 | * For request-based dm. | |
80 | * One of these is allocated per bio. | |
81 | */ | |
82 | struct dm_rq_clone_bio_info { | |
83 | struct bio *orig; | |
cec47e3d | 84 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
85 | }; |
86 | ||
1da177e4 LT |
87 | union map_info *dm_get_mapinfo(struct bio *bio) |
88 | { | |
17b2f66f | 89 | if (bio && bio->bi_private) |
028867ac | 90 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 91 | return NULL; |
1da177e4 LT |
92 | } |
93 | ||
cec47e3d KU |
94 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
95 | { | |
96 | if (rq && rq->end_io_data) | |
97 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
98 | return NULL; | |
99 | } | |
100 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
101 | ||
ba61fdd1 JM |
102 | #define MINOR_ALLOCED ((void *)-1) |
103 | ||
1da177e4 LT |
104 | /* |
105 | * Bits for the md->flags field. | |
106 | */ | |
1eb787ec | 107 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 108 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 109 | #define DMF_FROZEN 2 |
fba9f90e | 110 | #define DMF_FREEING 3 |
5c6bd75d | 111 | #define DMF_DELETING 4 |
2e93ccc1 | 112 | #define DMF_NOFLUSH_SUSPENDING 5 |
1eb787ec | 113 | #define DMF_QUEUE_IO_TO_THREAD 6 |
1da177e4 | 114 | |
304f3f6a MB |
115 | /* |
116 | * Work processed by per-device workqueue. | |
117 | */ | |
1da177e4 | 118 | struct mapped_device { |
2ca3310e | 119 | struct rw_semaphore io_lock; |
e61290a4 | 120 | struct mutex suspend_lock; |
1da177e4 LT |
121 | rwlock_t map_lock; |
122 | atomic_t holders; | |
5c6bd75d | 123 | atomic_t open_count; |
1da177e4 LT |
124 | |
125 | unsigned long flags; | |
126 | ||
165125e1 | 127 | struct request_queue *queue; |
1da177e4 | 128 | struct gendisk *disk; |
7e51f257 | 129 | char name[16]; |
1da177e4 LT |
130 | |
131 | void *interface_ptr; | |
132 | ||
133 | /* | |
134 | * A list of ios that arrived while we were suspended. | |
135 | */ | |
316d315b | 136 | atomic_t pending[2]; |
1da177e4 | 137 | wait_queue_head_t wait; |
53d5914f | 138 | struct work_struct work; |
74859364 | 139 | struct bio_list deferred; |
022c2611 | 140 | spinlock_t deferred_lock; |
1da177e4 | 141 | |
af7e466a MP |
142 | /* |
143 | * An error from the barrier request currently being processed. | |
144 | */ | |
145 | int barrier_error; | |
146 | ||
d0bcb878 KU |
147 | /* |
148 | * Protect barrier_error from concurrent endio processing | |
149 | * in request-based dm. | |
150 | */ | |
151 | spinlock_t barrier_error_lock; | |
152 | ||
304f3f6a MB |
153 | /* |
154 | * Processing queue (flush/barriers) | |
155 | */ | |
156 | struct workqueue_struct *wq; | |
d0bcb878 KU |
157 | struct work_struct barrier_work; |
158 | ||
159 | /* A pointer to the currently processing pre/post flush request */ | |
160 | struct request *flush_request; | |
304f3f6a | 161 | |
1da177e4 LT |
162 | /* |
163 | * The current mapping. | |
164 | */ | |
165 | struct dm_table *map; | |
166 | ||
167 | /* | |
168 | * io objects are allocated from here. | |
169 | */ | |
170 | mempool_t *io_pool; | |
171 | mempool_t *tio_pool; | |
172 | ||
9faf400f SB |
173 | struct bio_set *bs; |
174 | ||
1da177e4 LT |
175 | /* |
176 | * Event handling. | |
177 | */ | |
178 | atomic_t event_nr; | |
179 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
180 | atomic_t uevent_seq; |
181 | struct list_head uevent_list; | |
182 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
183 | |
184 | /* | |
185 | * freeze/thaw support require holding onto a super block | |
186 | */ | |
187 | struct super_block *frozen_sb; | |
db8fef4f | 188 | struct block_device *bdev; |
3ac51e74 DW |
189 | |
190 | /* forced geometry settings */ | |
191 | struct hd_geometry geometry; | |
784aae73 | 192 | |
cec47e3d KU |
193 | /* For saving the address of __make_request for request based dm */ |
194 | make_request_fn *saved_make_request_fn; | |
195 | ||
784aae73 MB |
196 | /* sysfs handle */ |
197 | struct kobject kobj; | |
52b1fd5a MP |
198 | |
199 | /* zero-length barrier that will be cloned and submitted to targets */ | |
200 | struct bio barrier_bio; | |
1da177e4 LT |
201 | }; |
202 | ||
e6ee8c0b KU |
203 | /* |
204 | * For mempools pre-allocation at the table loading time. | |
205 | */ | |
206 | struct dm_md_mempools { | |
207 | mempool_t *io_pool; | |
208 | mempool_t *tio_pool; | |
209 | struct bio_set *bs; | |
210 | }; | |
211 | ||
1da177e4 | 212 | #define MIN_IOS 256 |
e18b890b CL |
213 | static struct kmem_cache *_io_cache; |
214 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
215 | static struct kmem_cache *_rq_tio_cache; |
216 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 217 | |
1da177e4 LT |
218 | static int __init local_init(void) |
219 | { | |
51157b4a | 220 | int r = -ENOMEM; |
1da177e4 | 221 | |
1da177e4 | 222 | /* allocate a slab for the dm_ios */ |
028867ac | 223 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 224 | if (!_io_cache) |
51157b4a | 225 | return r; |
1da177e4 LT |
226 | |
227 | /* allocate a slab for the target ios */ | |
028867ac | 228 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
229 | if (!_tio_cache) |
230 | goto out_free_io_cache; | |
1da177e4 | 231 | |
8fbf26ad KU |
232 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
233 | if (!_rq_tio_cache) | |
234 | goto out_free_tio_cache; | |
235 | ||
236 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
237 | if (!_rq_bio_info_cache) | |
238 | goto out_free_rq_tio_cache; | |
239 | ||
51e5b2bd | 240 | r = dm_uevent_init(); |
51157b4a | 241 | if (r) |
8fbf26ad | 242 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 243 | |
1da177e4 LT |
244 | _major = major; |
245 | r = register_blkdev(_major, _name); | |
51157b4a KU |
246 | if (r < 0) |
247 | goto out_uevent_exit; | |
1da177e4 LT |
248 | |
249 | if (!_major) | |
250 | _major = r; | |
251 | ||
252 | return 0; | |
51157b4a KU |
253 | |
254 | out_uevent_exit: | |
255 | dm_uevent_exit(); | |
8fbf26ad KU |
256 | out_free_rq_bio_info_cache: |
257 | kmem_cache_destroy(_rq_bio_info_cache); | |
258 | out_free_rq_tio_cache: | |
259 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
260 | out_free_tio_cache: |
261 | kmem_cache_destroy(_tio_cache); | |
262 | out_free_io_cache: | |
263 | kmem_cache_destroy(_io_cache); | |
264 | ||
265 | return r; | |
1da177e4 LT |
266 | } |
267 | ||
268 | static void local_exit(void) | |
269 | { | |
8fbf26ad KU |
270 | kmem_cache_destroy(_rq_bio_info_cache); |
271 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
272 | kmem_cache_destroy(_tio_cache); |
273 | kmem_cache_destroy(_io_cache); | |
00d59405 | 274 | unregister_blkdev(_major, _name); |
51e5b2bd | 275 | dm_uevent_exit(); |
1da177e4 LT |
276 | |
277 | _major = 0; | |
278 | ||
279 | DMINFO("cleaned up"); | |
280 | } | |
281 | ||
b9249e55 | 282 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
283 | local_init, |
284 | dm_target_init, | |
285 | dm_linear_init, | |
286 | dm_stripe_init, | |
952b3557 | 287 | dm_io_init, |
945fa4d2 | 288 | dm_kcopyd_init, |
1da177e4 LT |
289 | dm_interface_init, |
290 | }; | |
291 | ||
b9249e55 | 292 | static void (*_exits[])(void) = { |
1da177e4 LT |
293 | local_exit, |
294 | dm_target_exit, | |
295 | dm_linear_exit, | |
296 | dm_stripe_exit, | |
952b3557 | 297 | dm_io_exit, |
945fa4d2 | 298 | dm_kcopyd_exit, |
1da177e4 LT |
299 | dm_interface_exit, |
300 | }; | |
301 | ||
302 | static int __init dm_init(void) | |
303 | { | |
304 | const int count = ARRAY_SIZE(_inits); | |
305 | ||
306 | int r, i; | |
307 | ||
308 | for (i = 0; i < count; i++) { | |
309 | r = _inits[i](); | |
310 | if (r) | |
311 | goto bad; | |
312 | } | |
313 | ||
314 | return 0; | |
315 | ||
316 | bad: | |
317 | while (i--) | |
318 | _exits[i](); | |
319 | ||
320 | return r; | |
321 | } | |
322 | ||
323 | static void __exit dm_exit(void) | |
324 | { | |
325 | int i = ARRAY_SIZE(_exits); | |
326 | ||
327 | while (i--) | |
328 | _exits[i](); | |
329 | } | |
330 | ||
331 | /* | |
332 | * Block device functions | |
333 | */ | |
432a212c MA |
334 | int dm_deleting_md(struct mapped_device *md) |
335 | { | |
336 | return test_bit(DMF_DELETING, &md->flags); | |
337 | } | |
338 | ||
fe5f9f2c | 339 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
340 | { |
341 | struct mapped_device *md; | |
342 | ||
6e9624b8 | 343 | lock_kernel(); |
fba9f90e JM |
344 | spin_lock(&_minor_lock); |
345 | ||
fe5f9f2c | 346 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
347 | if (!md) |
348 | goto out; | |
349 | ||
5c6bd75d | 350 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 351 | dm_deleting_md(md)) { |
fba9f90e JM |
352 | md = NULL; |
353 | goto out; | |
354 | } | |
355 | ||
1da177e4 | 356 | dm_get(md); |
5c6bd75d | 357 | atomic_inc(&md->open_count); |
fba9f90e JM |
358 | |
359 | out: | |
360 | spin_unlock(&_minor_lock); | |
6e9624b8 | 361 | unlock_kernel(); |
fba9f90e JM |
362 | |
363 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
364 | } |
365 | ||
fe5f9f2c | 366 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 367 | { |
fe5f9f2c | 368 | struct mapped_device *md = disk->private_data; |
6e9624b8 AB |
369 | |
370 | lock_kernel(); | |
5c6bd75d | 371 | atomic_dec(&md->open_count); |
1da177e4 | 372 | dm_put(md); |
6e9624b8 AB |
373 | unlock_kernel(); |
374 | ||
1da177e4 LT |
375 | return 0; |
376 | } | |
377 | ||
5c6bd75d AK |
378 | int dm_open_count(struct mapped_device *md) |
379 | { | |
380 | return atomic_read(&md->open_count); | |
381 | } | |
382 | ||
383 | /* | |
384 | * Guarantees nothing is using the device before it's deleted. | |
385 | */ | |
386 | int dm_lock_for_deletion(struct mapped_device *md) | |
387 | { | |
388 | int r = 0; | |
389 | ||
390 | spin_lock(&_minor_lock); | |
391 | ||
392 | if (dm_open_count(md)) | |
393 | r = -EBUSY; | |
394 | else | |
395 | set_bit(DMF_DELETING, &md->flags); | |
396 | ||
397 | spin_unlock(&_minor_lock); | |
398 | ||
399 | return r; | |
400 | } | |
401 | ||
3ac51e74 DW |
402 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
403 | { | |
404 | struct mapped_device *md = bdev->bd_disk->private_data; | |
405 | ||
406 | return dm_get_geometry(md, geo); | |
407 | } | |
408 | ||
fe5f9f2c | 409 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
410 | unsigned int cmd, unsigned long arg) |
411 | { | |
fe5f9f2c | 412 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 413 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
414 | struct dm_target *tgt; |
415 | int r = -ENOTTY; | |
416 | ||
aa129a22 MB |
417 | if (!map || !dm_table_get_size(map)) |
418 | goto out; | |
419 | ||
420 | /* We only support devices that have a single target */ | |
421 | if (dm_table_get_num_targets(map) != 1) | |
422 | goto out; | |
423 | ||
424 | tgt = dm_table_get_target(map, 0); | |
425 | ||
4f186f8b | 426 | if (dm_suspended_md(md)) { |
aa129a22 MB |
427 | r = -EAGAIN; |
428 | goto out; | |
429 | } | |
430 | ||
431 | if (tgt->type->ioctl) | |
647b3d00 | 432 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
433 | |
434 | out: | |
435 | dm_table_put(map); | |
436 | ||
aa129a22 MB |
437 | return r; |
438 | } | |
439 | ||
028867ac | 440 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
441 | { |
442 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
443 | } | |
444 | ||
028867ac | 445 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
446 | { |
447 | mempool_free(io, md->io_pool); | |
448 | } | |
449 | ||
028867ac | 450 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
451 | { |
452 | mempool_free(tio, md->tio_pool); | |
453 | } | |
454 | ||
08885643 KU |
455 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
456 | gfp_t gfp_mask) | |
cec47e3d | 457 | { |
08885643 | 458 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
459 | } |
460 | ||
461 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
462 | { | |
463 | mempool_free(tio, tio->md->tio_pool); | |
464 | } | |
465 | ||
466 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
467 | { | |
468 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
469 | } | |
470 | ||
471 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
472 | { | |
473 | mempool_free(info, info->tio->md->io_pool); | |
474 | } | |
475 | ||
90abb8c4 KU |
476 | static int md_in_flight(struct mapped_device *md) |
477 | { | |
478 | return atomic_read(&md->pending[READ]) + | |
479 | atomic_read(&md->pending[WRITE]); | |
480 | } | |
481 | ||
3eaf840e JNN |
482 | static void start_io_acct(struct dm_io *io) |
483 | { | |
484 | struct mapped_device *md = io->md; | |
c9959059 | 485 | int cpu; |
316d315b | 486 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
487 | |
488 | io->start_time = jiffies; | |
489 | ||
074a7aca TH |
490 | cpu = part_stat_lock(); |
491 | part_round_stats(cpu, &dm_disk(md)->part0); | |
492 | part_stat_unlock(); | |
316d315b | 493 | dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]); |
3eaf840e JNN |
494 | } |
495 | ||
d221d2e7 | 496 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
497 | { |
498 | struct mapped_device *md = io->md; | |
499 | struct bio *bio = io->bio; | |
500 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 501 | int pending, cpu; |
3eaf840e JNN |
502 | int rw = bio_data_dir(bio); |
503 | ||
074a7aca TH |
504 | cpu = part_stat_lock(); |
505 | part_round_stats(cpu, &dm_disk(md)->part0); | |
506 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
507 | part_stat_unlock(); | |
3eaf840e | 508 | |
af7e466a MP |
509 | /* |
510 | * After this is decremented the bio must not be touched if it is | |
511 | * a barrier. | |
512 | */ | |
316d315b NK |
513 | dm_disk(md)->part0.in_flight[rw] = pending = |
514 | atomic_dec_return(&md->pending[rw]); | |
515 | pending += atomic_read(&md->pending[rw^0x1]); | |
3eaf840e | 516 | |
d221d2e7 MP |
517 | /* nudge anyone waiting on suspend queue */ |
518 | if (!pending) | |
519 | wake_up(&md->wait); | |
3eaf840e JNN |
520 | } |
521 | ||
1da177e4 LT |
522 | /* |
523 | * Add the bio to the list of deferred io. | |
524 | */ | |
92c63902 | 525 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 526 | { |
2ca3310e | 527 | down_write(&md->io_lock); |
1da177e4 | 528 | |
022c2611 | 529 | spin_lock_irq(&md->deferred_lock); |
1da177e4 | 530 | bio_list_add(&md->deferred, bio); |
022c2611 | 531 | spin_unlock_irq(&md->deferred_lock); |
1da177e4 | 532 | |
92c63902 MP |
533 | if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) |
534 | queue_work(md->wq, &md->work); | |
535 | ||
2ca3310e | 536 | up_write(&md->io_lock); |
1da177e4 LT |
537 | } |
538 | ||
539 | /* | |
540 | * Everyone (including functions in this file), should use this | |
541 | * function to access the md->map field, and make sure they call | |
542 | * dm_table_put() when finished. | |
543 | */ | |
7c666411 | 544 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
545 | { |
546 | struct dm_table *t; | |
523d9297 | 547 | unsigned long flags; |
1da177e4 | 548 | |
523d9297 | 549 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
550 | t = md->map; |
551 | if (t) | |
552 | dm_table_get(t); | |
523d9297 | 553 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
554 | |
555 | return t; | |
556 | } | |
557 | ||
3ac51e74 DW |
558 | /* |
559 | * Get the geometry associated with a dm device | |
560 | */ | |
561 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
562 | { | |
563 | *geo = md->geometry; | |
564 | ||
565 | return 0; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Set the geometry of a device. | |
570 | */ | |
571 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
572 | { | |
573 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
574 | ||
575 | if (geo->start > sz) { | |
576 | DMWARN("Start sector is beyond the geometry limits."); | |
577 | return -EINVAL; | |
578 | } | |
579 | ||
580 | md->geometry = *geo; | |
581 | ||
582 | return 0; | |
583 | } | |
584 | ||
1da177e4 LT |
585 | /*----------------------------------------------------------------- |
586 | * CRUD START: | |
587 | * A more elegant soln is in the works that uses the queue | |
588 | * merge fn, unfortunately there are a couple of changes to | |
589 | * the block layer that I want to make for this. So in the | |
590 | * interests of getting something for people to use I give | |
591 | * you this clearly demarcated crap. | |
592 | *---------------------------------------------------------------*/ | |
593 | ||
2e93ccc1 KU |
594 | static int __noflush_suspending(struct mapped_device *md) |
595 | { | |
596 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
597 | } | |
598 | ||
1da177e4 LT |
599 | /* |
600 | * Decrements the number of outstanding ios that a bio has been | |
601 | * cloned into, completing the original io if necc. | |
602 | */ | |
858119e1 | 603 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 604 | { |
2e93ccc1 | 605 | unsigned long flags; |
b35f8caa MB |
606 | int io_error; |
607 | struct bio *bio; | |
608 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
609 | |
610 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
611 | if (unlikely(error)) { |
612 | spin_lock_irqsave(&io->endio_lock, flags); | |
613 | if (!(io->error > 0 && __noflush_suspending(md))) | |
614 | io->error = error; | |
615 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
616 | } | |
1da177e4 LT |
617 | |
618 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
619 | if (io->error == DM_ENDIO_REQUEUE) { |
620 | /* | |
621 | * Target requested pushing back the I/O. | |
2e93ccc1 | 622 | */ |
022c2611 | 623 | spin_lock_irqsave(&md->deferred_lock, flags); |
2761e95f | 624 | if (__noflush_suspending(md)) { |
7b6d91da | 625 | if (!(io->bio->bi_rw & REQ_HARDBARRIER)) |
2761e95f MP |
626 | bio_list_add_head(&md->deferred, |
627 | io->bio); | |
628 | } else | |
2e93ccc1 KU |
629 | /* noflush suspend was interrupted. */ |
630 | io->error = -EIO; | |
022c2611 | 631 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
632 | } |
633 | ||
b35f8caa MB |
634 | io_error = io->error; |
635 | bio = io->bio; | |
2e93ccc1 | 636 | |
7b6d91da | 637 | if (bio->bi_rw & REQ_HARDBARRIER) { |
af7e466a MP |
638 | /* |
639 | * There can be just one barrier request so we use | |
640 | * a per-device variable for error reporting. | |
641 | * Note that you can't touch the bio after end_io_acct | |
708e9295 MP |
642 | * |
643 | * We ignore -EOPNOTSUPP for empty flush reported by | |
644 | * underlying devices. We assume that if the device | |
645 | * doesn't support empty barriers, it doesn't need | |
646 | * cache flushing commands. | |
af7e466a | 647 | */ |
708e9295 MP |
648 | if (!md->barrier_error && |
649 | !(bio_empty_barrier(bio) && io_error == -EOPNOTSUPP)) | |
5aa2781d | 650 | md->barrier_error = io_error; |
af7e466a | 651 | end_io_acct(io); |
a97f925a | 652 | free_io(md, io); |
af7e466a MP |
653 | } else { |
654 | end_io_acct(io); | |
a97f925a | 655 | free_io(md, io); |
b35f8caa | 656 | |
af7e466a MP |
657 | if (io_error != DM_ENDIO_REQUEUE) { |
658 | trace_block_bio_complete(md->queue, bio); | |
2056a782 | 659 | |
af7e466a MP |
660 | bio_endio(bio, io_error); |
661 | } | |
b35f8caa | 662 | } |
1da177e4 LT |
663 | } |
664 | } | |
665 | ||
6712ecf8 | 666 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
667 | { |
668 | int r = 0; | |
028867ac | 669 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 670 | struct dm_io *io = tio->io; |
9faf400f | 671 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
672 | dm_endio_fn endio = tio->ti->type->end_io; |
673 | ||
1da177e4 LT |
674 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
675 | error = -EIO; | |
676 | ||
677 | if (endio) { | |
678 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
679 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
680 | /* | |
681 | * error and requeue request are handled | |
682 | * in dec_pending(). | |
683 | */ | |
1da177e4 | 684 | error = r; |
45cbcd79 KU |
685 | else if (r == DM_ENDIO_INCOMPLETE) |
686 | /* The target will handle the io */ | |
6712ecf8 | 687 | return; |
45cbcd79 KU |
688 | else if (r) { |
689 | DMWARN("unimplemented target endio return value: %d", r); | |
690 | BUG(); | |
691 | } | |
1da177e4 LT |
692 | } |
693 | ||
9faf400f SB |
694 | /* |
695 | * Store md for cleanup instead of tio which is about to get freed. | |
696 | */ | |
697 | bio->bi_private = md->bs; | |
698 | ||
9faf400f | 699 | free_tio(md, tio); |
b35f8caa MB |
700 | bio_put(bio); |
701 | dec_pending(io, error); | |
1da177e4 LT |
702 | } |
703 | ||
cec47e3d KU |
704 | /* |
705 | * Partial completion handling for request-based dm | |
706 | */ | |
707 | static void end_clone_bio(struct bio *clone, int error) | |
708 | { | |
709 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
710 | struct dm_rq_target_io *tio = info->tio; | |
711 | struct bio *bio = info->orig; | |
712 | unsigned int nr_bytes = info->orig->bi_size; | |
713 | ||
714 | bio_put(clone); | |
715 | ||
716 | if (tio->error) | |
717 | /* | |
718 | * An error has already been detected on the request. | |
719 | * Once error occurred, just let clone->end_io() handle | |
720 | * the remainder. | |
721 | */ | |
722 | return; | |
723 | else if (error) { | |
724 | /* | |
725 | * Don't notice the error to the upper layer yet. | |
726 | * The error handling decision is made by the target driver, | |
727 | * when the request is completed. | |
728 | */ | |
729 | tio->error = error; | |
730 | return; | |
731 | } | |
732 | ||
733 | /* | |
734 | * I/O for the bio successfully completed. | |
735 | * Notice the data completion to the upper layer. | |
736 | */ | |
737 | ||
738 | /* | |
739 | * bios are processed from the head of the list. | |
740 | * So the completing bio should always be rq->bio. | |
741 | * If it's not, something wrong is happening. | |
742 | */ | |
743 | if (tio->orig->bio != bio) | |
744 | DMERR("bio completion is going in the middle of the request"); | |
745 | ||
746 | /* | |
747 | * Update the original request. | |
748 | * Do not use blk_end_request() here, because it may complete | |
749 | * the original request before the clone, and break the ordering. | |
750 | */ | |
751 | blk_update_request(tio->orig, 0, nr_bytes); | |
752 | } | |
753 | ||
d0bcb878 KU |
754 | static void store_barrier_error(struct mapped_device *md, int error) |
755 | { | |
756 | unsigned long flags; | |
757 | ||
758 | spin_lock_irqsave(&md->barrier_error_lock, flags); | |
759 | /* | |
760 | * Basically, the first error is taken, but: | |
761 | * -EOPNOTSUPP supersedes any I/O error. | |
762 | * Requeue request supersedes any I/O error but -EOPNOTSUPP. | |
763 | */ | |
764 | if (!md->barrier_error || error == -EOPNOTSUPP || | |
765 | (md->barrier_error != -EOPNOTSUPP && | |
766 | error == DM_ENDIO_REQUEUE)) | |
767 | md->barrier_error = error; | |
768 | spin_unlock_irqrestore(&md->barrier_error_lock, flags); | |
769 | } | |
770 | ||
cec47e3d KU |
771 | /* |
772 | * Don't touch any member of the md after calling this function because | |
773 | * the md may be freed in dm_put() at the end of this function. | |
774 | * Or do dm_get() before calling this function and dm_put() later. | |
775 | */ | |
b4324fee | 776 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 777 | { |
b4324fee | 778 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
779 | |
780 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 781 | if (!md_in_flight(md)) |
cec47e3d KU |
782 | wake_up(&md->wait); |
783 | ||
784 | if (run_queue) | |
b4324fee | 785 | blk_run_queue(md->queue); |
cec47e3d KU |
786 | |
787 | /* | |
788 | * dm_put() must be at the end of this function. See the comment above | |
789 | */ | |
790 | dm_put(md); | |
791 | } | |
792 | ||
a77e28c7 KU |
793 | static void free_rq_clone(struct request *clone) |
794 | { | |
795 | struct dm_rq_target_io *tio = clone->end_io_data; | |
796 | ||
797 | blk_rq_unprep_clone(clone); | |
798 | free_rq_tio(tio); | |
799 | } | |
800 | ||
980691e5 KU |
801 | /* |
802 | * Complete the clone and the original request. | |
803 | * Must be called without queue lock. | |
804 | */ | |
805 | static void dm_end_request(struct request *clone, int error) | |
806 | { | |
807 | int rw = rq_data_dir(clone); | |
d0bcb878 | 808 | int run_queue = 1; |
33659ebb | 809 | bool is_barrier = clone->cmd_flags & REQ_HARDBARRIER; |
980691e5 KU |
810 | struct dm_rq_target_io *tio = clone->end_io_data; |
811 | struct mapped_device *md = tio->md; | |
812 | struct request *rq = tio->orig; | |
813 | ||
33659ebb | 814 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC && !is_barrier) { |
980691e5 KU |
815 | rq->errors = clone->errors; |
816 | rq->resid_len = clone->resid_len; | |
817 | ||
818 | if (rq->sense) | |
819 | /* | |
820 | * We are using the sense buffer of the original | |
821 | * request. | |
822 | * So setting the length of the sense data is enough. | |
823 | */ | |
824 | rq->sense_len = clone->sense_len; | |
825 | } | |
826 | ||
827 | free_rq_clone(clone); | |
828 | ||
d0bcb878 KU |
829 | if (unlikely(is_barrier)) { |
830 | if (unlikely(error)) | |
831 | store_barrier_error(md, error); | |
832 | run_queue = 0; | |
833 | } else | |
834 | blk_end_request_all(rq, error); | |
980691e5 | 835 | |
d0bcb878 | 836 | rq_completed(md, rw, run_queue); |
980691e5 KU |
837 | } |
838 | ||
cec47e3d KU |
839 | static void dm_unprep_request(struct request *rq) |
840 | { | |
841 | struct request *clone = rq->special; | |
cec47e3d KU |
842 | |
843 | rq->special = NULL; | |
844 | rq->cmd_flags &= ~REQ_DONTPREP; | |
845 | ||
a77e28c7 | 846 | free_rq_clone(clone); |
cec47e3d KU |
847 | } |
848 | ||
849 | /* | |
850 | * Requeue the original request of a clone. | |
851 | */ | |
852 | void dm_requeue_unmapped_request(struct request *clone) | |
853 | { | |
b4324fee | 854 | int rw = rq_data_dir(clone); |
cec47e3d KU |
855 | struct dm_rq_target_io *tio = clone->end_io_data; |
856 | struct mapped_device *md = tio->md; | |
857 | struct request *rq = tio->orig; | |
858 | struct request_queue *q = rq->q; | |
859 | unsigned long flags; | |
860 | ||
33659ebb | 861 | if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) { |
d0bcb878 KU |
862 | /* |
863 | * Barrier clones share an original request. | |
864 | * Leave it to dm_end_request(), which handles this special | |
865 | * case. | |
866 | */ | |
867 | dm_end_request(clone, DM_ENDIO_REQUEUE); | |
868 | return; | |
869 | } | |
870 | ||
cec47e3d KU |
871 | dm_unprep_request(rq); |
872 | ||
873 | spin_lock_irqsave(q->queue_lock, flags); | |
874 | if (elv_queue_empty(q)) | |
875 | blk_plug_device(q); | |
876 | blk_requeue_request(q, rq); | |
877 | spin_unlock_irqrestore(q->queue_lock, flags); | |
878 | ||
b4324fee | 879 | rq_completed(md, rw, 0); |
cec47e3d KU |
880 | } |
881 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
882 | ||
883 | static void __stop_queue(struct request_queue *q) | |
884 | { | |
885 | blk_stop_queue(q); | |
886 | } | |
887 | ||
888 | static void stop_queue(struct request_queue *q) | |
889 | { | |
890 | unsigned long flags; | |
891 | ||
892 | spin_lock_irqsave(q->queue_lock, flags); | |
893 | __stop_queue(q); | |
894 | spin_unlock_irqrestore(q->queue_lock, flags); | |
895 | } | |
896 | ||
897 | static void __start_queue(struct request_queue *q) | |
898 | { | |
899 | if (blk_queue_stopped(q)) | |
900 | blk_start_queue(q); | |
901 | } | |
902 | ||
903 | static void start_queue(struct request_queue *q) | |
904 | { | |
905 | unsigned long flags; | |
906 | ||
907 | spin_lock_irqsave(q->queue_lock, flags); | |
908 | __start_queue(q); | |
909 | spin_unlock_irqrestore(q->queue_lock, flags); | |
910 | } | |
911 | ||
11a68244 | 912 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 913 | { |
11a68244 | 914 | int r = error; |
cec47e3d KU |
915 | struct dm_rq_target_io *tio = clone->end_io_data; |
916 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
cec47e3d | 917 | |
11a68244 KU |
918 | if (mapped && rq_end_io) |
919 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
cec47e3d | 920 | |
11a68244 | 921 | if (r <= 0) |
cec47e3d | 922 | /* The target wants to complete the I/O */ |
11a68244 KU |
923 | dm_end_request(clone, r); |
924 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
925 | /* The target will handle the I/O */ |
926 | return; | |
11a68244 | 927 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
928 | /* The target wants to requeue the I/O */ |
929 | dm_requeue_unmapped_request(clone); | |
930 | else { | |
11a68244 | 931 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
932 | BUG(); |
933 | } | |
934 | } | |
935 | ||
11a68244 KU |
936 | /* |
937 | * Request completion handler for request-based dm | |
938 | */ | |
939 | static void dm_softirq_done(struct request *rq) | |
940 | { | |
941 | bool mapped = true; | |
942 | struct request *clone = rq->completion_data; | |
943 | struct dm_rq_target_io *tio = clone->end_io_data; | |
944 | ||
945 | if (rq->cmd_flags & REQ_FAILED) | |
946 | mapped = false; | |
947 | ||
948 | dm_done(clone, tio->error, mapped); | |
949 | } | |
950 | ||
cec47e3d KU |
951 | /* |
952 | * Complete the clone and the original request with the error status | |
953 | * through softirq context. | |
954 | */ | |
955 | static void dm_complete_request(struct request *clone, int error) | |
956 | { | |
957 | struct dm_rq_target_io *tio = clone->end_io_data; | |
958 | struct request *rq = tio->orig; | |
959 | ||
33659ebb | 960 | if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) { |
d0bcb878 KU |
961 | /* |
962 | * Barrier clones share an original request. So can't use | |
963 | * softirq_done with the original. | |
964 | * Pass the clone to dm_done() directly in this special case. | |
965 | * It is safe (even if clone->q->queue_lock is held here) | |
966 | * because there is no I/O dispatching during the completion | |
967 | * of barrier clone. | |
968 | */ | |
969 | dm_done(clone, error, true); | |
970 | return; | |
971 | } | |
972 | ||
cec47e3d KU |
973 | tio->error = error; |
974 | rq->completion_data = clone; | |
975 | blk_complete_request(rq); | |
976 | } | |
977 | ||
978 | /* | |
979 | * Complete the not-mapped clone and the original request with the error status | |
980 | * through softirq context. | |
981 | * Target's rq_end_io() function isn't called. | |
982 | * This may be used when the target's map_rq() function fails. | |
983 | */ | |
984 | void dm_kill_unmapped_request(struct request *clone, int error) | |
985 | { | |
986 | struct dm_rq_target_io *tio = clone->end_io_data; | |
987 | struct request *rq = tio->orig; | |
988 | ||
33659ebb | 989 | if (unlikely(clone->cmd_flags & REQ_HARDBARRIER)) { |
d0bcb878 KU |
990 | /* |
991 | * Barrier clones share an original request. | |
992 | * Leave it to dm_end_request(), which handles this special | |
993 | * case. | |
994 | */ | |
995 | BUG_ON(error > 0); | |
996 | dm_end_request(clone, error); | |
997 | return; | |
998 | } | |
999 | ||
cec47e3d KU |
1000 | rq->cmd_flags |= REQ_FAILED; |
1001 | dm_complete_request(clone, error); | |
1002 | } | |
1003 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1004 | ||
1005 | /* | |
1006 | * Called with the queue lock held | |
1007 | */ | |
1008 | static void end_clone_request(struct request *clone, int error) | |
1009 | { | |
1010 | /* | |
1011 | * For just cleaning up the information of the queue in which | |
1012 | * the clone was dispatched. | |
1013 | * The clone is *NOT* freed actually here because it is alloced from | |
1014 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1015 | */ | |
1016 | __blk_put_request(clone->q, clone); | |
1017 | ||
1018 | /* | |
1019 | * Actual request completion is done in a softirq context which doesn't | |
1020 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1021 | * - another request may be submitted by the upper level driver | |
1022 | * of the stacking during the completion | |
1023 | * - the submission which requires queue lock may be done | |
1024 | * against this queue | |
1025 | */ | |
1026 | dm_complete_request(clone, error); | |
1027 | } | |
1028 | ||
1da177e4 LT |
1029 | static sector_t max_io_len(struct mapped_device *md, |
1030 | sector_t sector, struct dm_target *ti) | |
1031 | { | |
1032 | sector_t offset = sector - ti->begin; | |
1033 | sector_t len = ti->len - offset; | |
1034 | ||
1035 | /* | |
1036 | * Does the target need to split even further ? | |
1037 | */ | |
1038 | if (ti->split_io) { | |
1039 | sector_t boundary; | |
1040 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
1041 | - offset; | |
1042 | if (len > boundary) | |
1043 | len = boundary; | |
1044 | } | |
1045 | ||
1046 | return len; | |
1047 | } | |
1048 | ||
1049 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 1050 | struct dm_target_io *tio) |
1da177e4 LT |
1051 | { |
1052 | int r; | |
2056a782 | 1053 | sector_t sector; |
9faf400f | 1054 | struct mapped_device *md; |
1da177e4 | 1055 | |
1da177e4 LT |
1056 | clone->bi_end_io = clone_endio; |
1057 | clone->bi_private = tio; | |
1058 | ||
1059 | /* | |
1060 | * Map the clone. If r == 0 we don't need to do | |
1061 | * anything, the target has assumed ownership of | |
1062 | * this io. | |
1063 | */ | |
1064 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1065 | sector = clone->bi_sector; |
1da177e4 | 1066 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 1067 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1068 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1069 | |
5f3ea37c | 1070 | trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, |
22a7c31a | 1071 | tio->io->bio->bi_bdev->bd_dev, sector); |
2056a782 | 1072 | |
1da177e4 | 1073 | generic_make_request(clone); |
2e93ccc1 KU |
1074 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1075 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1076 | md = tio->io->md; |
1077 | dec_pending(tio->io, r); | |
1078 | /* | |
1079 | * Store bio_set for cleanup. | |
1080 | */ | |
1081 | clone->bi_private = md->bs; | |
1da177e4 | 1082 | bio_put(clone); |
9faf400f | 1083 | free_tio(md, tio); |
45cbcd79 KU |
1084 | } else if (r) { |
1085 | DMWARN("unimplemented target map return value: %d", r); | |
1086 | BUG(); | |
1da177e4 LT |
1087 | } |
1088 | } | |
1089 | ||
1090 | struct clone_info { | |
1091 | struct mapped_device *md; | |
1092 | struct dm_table *map; | |
1093 | struct bio *bio; | |
1094 | struct dm_io *io; | |
1095 | sector_t sector; | |
1096 | sector_t sector_count; | |
1097 | unsigned short idx; | |
1098 | }; | |
1099 | ||
3676347a PO |
1100 | static void dm_bio_destructor(struct bio *bio) |
1101 | { | |
9faf400f SB |
1102 | struct bio_set *bs = bio->bi_private; |
1103 | ||
1104 | bio_free(bio, bs); | |
3676347a PO |
1105 | } |
1106 | ||
1da177e4 LT |
1107 | /* |
1108 | * Creates a little bio that is just does part of a bvec. | |
1109 | */ | |
1110 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1111 | unsigned short idx, unsigned int offset, | |
9faf400f | 1112 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1113 | { |
1114 | struct bio *clone; | |
1115 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1116 | ||
9faf400f | 1117 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1118 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1119 | *clone->bi_io_vec = *bv; |
1120 | ||
1121 | clone->bi_sector = sector; | |
1122 | clone->bi_bdev = bio->bi_bdev; | |
7b6d91da | 1123 | clone->bi_rw = bio->bi_rw & ~REQ_HARDBARRIER; |
1da177e4 LT |
1124 | clone->bi_vcnt = 1; |
1125 | clone->bi_size = to_bytes(len); | |
1126 | clone->bi_io_vec->bv_offset = offset; | |
1127 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1128 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1129 | |
9c47008d | 1130 | if (bio_integrity(bio)) { |
7878cba9 | 1131 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1132 | bio_integrity_trim(clone, |
1133 | bio_sector_offset(bio, idx, offset), len); | |
1134 | } | |
1135 | ||
1da177e4 LT |
1136 | return clone; |
1137 | } | |
1138 | ||
1139 | /* | |
1140 | * Creates a bio that consists of range of complete bvecs. | |
1141 | */ | |
1142 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1143 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1144 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1145 | { |
1146 | struct bio *clone; | |
1147 | ||
9faf400f SB |
1148 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1149 | __bio_clone(clone, bio); | |
7b6d91da | 1150 | clone->bi_rw &= ~REQ_HARDBARRIER; |
9faf400f | 1151 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1152 | clone->bi_sector = sector; |
1153 | clone->bi_idx = idx; | |
1154 | clone->bi_vcnt = idx + bv_count; | |
1155 | clone->bi_size = to_bytes(len); | |
1156 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1157 | ||
9c47008d | 1158 | if (bio_integrity(bio)) { |
7878cba9 | 1159 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1160 | |
1161 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1162 | bio_integrity_trim(clone, | |
1163 | bio_sector_offset(bio, idx, 0), len); | |
1164 | } | |
1165 | ||
1da177e4 LT |
1166 | return clone; |
1167 | } | |
1168 | ||
9015df24 AK |
1169 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1170 | struct dm_target *ti) | |
f9ab94ce | 1171 | { |
9015df24 | 1172 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1173 | |
1174 | tio->io = ci->io; | |
1175 | tio->ti = ti; | |
f9ab94ce | 1176 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1177 | |
1178 | return tio; | |
1179 | } | |
1180 | ||
1181 | static void __flush_target(struct clone_info *ci, struct dm_target *ti, | |
1182 | unsigned flush_nr) | |
1183 | { | |
1184 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1185 | struct bio *clone; | |
1186 | ||
f9ab94ce MP |
1187 | tio->info.flush_request = flush_nr; |
1188 | ||
1189 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); | |
1190 | __bio_clone(clone, ci->bio); | |
1191 | clone->bi_destructor = dm_bio_destructor; | |
1192 | ||
1193 | __map_bio(ti, clone, tio); | |
1194 | } | |
1195 | ||
1196 | static int __clone_and_map_empty_barrier(struct clone_info *ci) | |
1197 | { | |
1198 | unsigned target_nr = 0, flush_nr; | |
1199 | struct dm_target *ti; | |
1200 | ||
1201 | while ((ti = dm_table_get_target(ci->map, target_nr++))) | |
1202 | for (flush_nr = 0; flush_nr < ti->num_flush_requests; | |
1203 | flush_nr++) | |
1204 | __flush_target(ci, ti, flush_nr); | |
1205 | ||
1206 | ci->sector_count = 0; | |
1207 | ||
1208 | return 0; | |
1209 | } | |
1210 | ||
512875bd | 1211 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1212 | { |
1213 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1214 | struct dm_target *ti; |
1215 | sector_t len = 0, max; | |
028867ac | 1216 | struct dm_target_io *tio; |
1da177e4 | 1217 | |
f9ab94ce MP |
1218 | if (unlikely(bio_empty_barrier(bio))) |
1219 | return __clone_and_map_empty_barrier(ci); | |
1220 | ||
512875bd JN |
1221 | ti = dm_table_find_target(ci->map, ci->sector); |
1222 | if (!dm_target_is_valid(ti)) | |
1223 | return -EIO; | |
1224 | ||
1225 | max = max_io_len(ci->md, ci->sector, ti); | |
1226 | ||
1da177e4 LT |
1227 | /* |
1228 | * Allocate a target io object. | |
1229 | */ | |
9015df24 | 1230 | tio = alloc_tio(ci, ti); |
1da177e4 LT |
1231 | |
1232 | if (ci->sector_count <= max) { | |
1233 | /* | |
1234 | * Optimise for the simple case where we can do all of | |
1235 | * the remaining io with a single clone. | |
1236 | */ | |
1237 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
1238 | bio->bi_vcnt - ci->idx, ci->sector_count, |
1239 | ci->md->bs); | |
1da177e4 LT |
1240 | __map_bio(ti, clone, tio); |
1241 | ci->sector_count = 0; | |
1242 | ||
1243 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1244 | /* | |
1245 | * There are some bvecs that don't span targets. | |
1246 | * Do as many of these as possible. | |
1247 | */ | |
1248 | int i; | |
1249 | sector_t remaining = max; | |
1250 | sector_t bv_len; | |
1251 | ||
1252 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1253 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1254 | ||
1255 | if (bv_len > remaining) | |
1256 | break; | |
1257 | ||
1258 | remaining -= bv_len; | |
1259 | len += bv_len; | |
1260 | } | |
1261 | ||
9faf400f SB |
1262 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1263 | ci->md->bs); | |
1da177e4 LT |
1264 | __map_bio(ti, clone, tio); |
1265 | ||
1266 | ci->sector += len; | |
1267 | ci->sector_count -= len; | |
1268 | ci->idx = i; | |
1269 | ||
1270 | } else { | |
1271 | /* | |
d2044a94 | 1272 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1273 | */ |
1274 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1275 | sector_t remaining = to_sector(bv->bv_len); |
1276 | unsigned int offset = 0; | |
1da177e4 | 1277 | |
d2044a94 AK |
1278 | do { |
1279 | if (offset) { | |
1280 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1281 | if (!dm_target_is_valid(ti)) |
1282 | return -EIO; | |
1283 | ||
d2044a94 | 1284 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 1285 | |
9015df24 | 1286 | tio = alloc_tio(ci, ti); |
d2044a94 AK |
1287 | } |
1288 | ||
1289 | len = min(remaining, max); | |
1290 | ||
1291 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
1292 | bv->bv_offset + offset, len, |
1293 | ci->md->bs); | |
d2044a94 AK |
1294 | |
1295 | __map_bio(ti, clone, tio); | |
1296 | ||
1297 | ci->sector += len; | |
1298 | ci->sector_count -= len; | |
1299 | offset += to_bytes(len); | |
1300 | } while (remaining -= len); | |
1da177e4 | 1301 | |
1da177e4 LT |
1302 | ci->idx++; |
1303 | } | |
512875bd JN |
1304 | |
1305 | return 0; | |
1da177e4 LT |
1306 | } |
1307 | ||
1308 | /* | |
8a53c28d | 1309 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1310 | */ |
f0b9a450 | 1311 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1312 | { |
1313 | struct clone_info ci; | |
512875bd | 1314 | int error = 0; |
1da177e4 | 1315 | |
7c666411 | 1316 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1317 | if (unlikely(!ci.map)) { |
7b6d91da | 1318 | if (!(bio->bi_rw & REQ_HARDBARRIER)) |
af7e466a MP |
1319 | bio_io_error(bio); |
1320 | else | |
5aa2781d MP |
1321 | if (!md->barrier_error) |
1322 | md->barrier_error = -EIO; | |
f0b9a450 MP |
1323 | return; |
1324 | } | |
692d0eb9 | 1325 | |
1da177e4 LT |
1326 | ci.md = md; |
1327 | ci.bio = bio; | |
1328 | ci.io = alloc_io(md); | |
1329 | ci.io->error = 0; | |
1330 | atomic_set(&ci.io->io_count, 1); | |
1331 | ci.io->bio = bio; | |
1332 | ci.io->md = md; | |
f88fb981 | 1333 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 LT |
1334 | ci.sector = bio->bi_sector; |
1335 | ci.sector_count = bio_sectors(bio); | |
f9ab94ce MP |
1336 | if (unlikely(bio_empty_barrier(bio))) |
1337 | ci.sector_count = 1; | |
1da177e4 LT |
1338 | ci.idx = bio->bi_idx; |
1339 | ||
3eaf840e | 1340 | start_io_acct(ci.io); |
512875bd JN |
1341 | while (ci.sector_count && !error) |
1342 | error = __clone_and_map(&ci); | |
1da177e4 LT |
1343 | |
1344 | /* drop the extra reference count */ | |
512875bd | 1345 | dec_pending(ci.io, error); |
1da177e4 LT |
1346 | dm_table_put(ci.map); |
1347 | } | |
1348 | /*----------------------------------------------------------------- | |
1349 | * CRUD END | |
1350 | *---------------------------------------------------------------*/ | |
1351 | ||
f6fccb12 MB |
1352 | static int dm_merge_bvec(struct request_queue *q, |
1353 | struct bvec_merge_data *bvm, | |
1354 | struct bio_vec *biovec) | |
1355 | { | |
1356 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1357 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1358 | struct dm_target *ti; |
1359 | sector_t max_sectors; | |
5037108a | 1360 | int max_size = 0; |
f6fccb12 MB |
1361 | |
1362 | if (unlikely(!map)) | |
5037108a | 1363 | goto out; |
f6fccb12 MB |
1364 | |
1365 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1366 | if (!dm_target_is_valid(ti)) |
1367 | goto out_table; | |
f6fccb12 MB |
1368 | |
1369 | /* | |
1370 | * Find maximum amount of I/O that won't need splitting | |
1371 | */ | |
1372 | max_sectors = min(max_io_len(md, bvm->bi_sector, ti), | |
1373 | (sector_t) BIO_MAX_SECTORS); | |
1374 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1375 | if (max_size < 0) | |
1376 | max_size = 0; | |
1377 | ||
1378 | /* | |
1379 | * merge_bvec_fn() returns number of bytes | |
1380 | * it can accept at this offset | |
1381 | * max is precomputed maximal io size | |
1382 | */ | |
1383 | if (max_size && ti->type->merge) | |
1384 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1385 | /* |
1386 | * If the target doesn't support merge method and some of the devices | |
1387 | * provided their merge_bvec method (we know this by looking at | |
1388 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1389 | * entries. So always set max_size to 0, and the code below allows | |
1390 | * just one page. | |
1391 | */ | |
1392 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1393 | ||
1394 | max_size = 0; | |
f6fccb12 | 1395 | |
b01cd5ac | 1396 | out_table: |
5037108a MP |
1397 | dm_table_put(map); |
1398 | ||
1399 | out: | |
f6fccb12 MB |
1400 | /* |
1401 | * Always allow an entire first page | |
1402 | */ | |
1403 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1404 | max_size = biovec->bv_len; | |
1405 | ||
f6fccb12 MB |
1406 | return max_size; |
1407 | } | |
1408 | ||
1da177e4 LT |
1409 | /* |
1410 | * The request function that just remaps the bio built up by | |
1411 | * dm_merge_bvec. | |
1412 | */ | |
cec47e3d | 1413 | static int _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1414 | { |
12f03a49 | 1415 | int rw = bio_data_dir(bio); |
1da177e4 | 1416 | struct mapped_device *md = q->queuedata; |
c9959059 | 1417 | int cpu; |
1da177e4 | 1418 | |
2ca3310e | 1419 | down_read(&md->io_lock); |
1da177e4 | 1420 | |
074a7aca TH |
1421 | cpu = part_stat_lock(); |
1422 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1423 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1424 | part_stat_unlock(); | |
12f03a49 | 1425 | |
1da177e4 | 1426 | /* |
1eb787ec AK |
1427 | * If we're suspended or the thread is processing barriers |
1428 | * we have to queue this io for later. | |
1da177e4 | 1429 | */ |
af7e466a | 1430 | if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || |
7b6d91da | 1431 | unlikely(bio->bi_rw & REQ_HARDBARRIER)) { |
2ca3310e | 1432 | up_read(&md->io_lock); |
1da177e4 | 1433 | |
54d9a1b4 AK |
1434 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && |
1435 | bio_rw(bio) == READA) { | |
1436 | bio_io_error(bio); | |
1437 | return 0; | |
1438 | } | |
1da177e4 | 1439 | |
92c63902 | 1440 | queue_io(md, bio); |
1da177e4 | 1441 | |
92c63902 | 1442 | return 0; |
1da177e4 LT |
1443 | } |
1444 | ||
f0b9a450 | 1445 | __split_and_process_bio(md, bio); |
2ca3310e | 1446 | up_read(&md->io_lock); |
f0b9a450 | 1447 | return 0; |
1da177e4 LT |
1448 | } |
1449 | ||
cec47e3d KU |
1450 | static int dm_make_request(struct request_queue *q, struct bio *bio) |
1451 | { | |
1452 | struct mapped_device *md = q->queuedata; | |
1453 | ||
cec47e3d KU |
1454 | return md->saved_make_request_fn(q, bio); /* call __make_request() */ |
1455 | } | |
1456 | ||
1457 | static int dm_request_based(struct mapped_device *md) | |
1458 | { | |
1459 | return blk_queue_stackable(md->queue); | |
1460 | } | |
1461 | ||
1462 | static int dm_request(struct request_queue *q, struct bio *bio) | |
1463 | { | |
1464 | struct mapped_device *md = q->queuedata; | |
1465 | ||
1466 | if (dm_request_based(md)) | |
1467 | return dm_make_request(q, bio); | |
1468 | ||
1469 | return _dm_request(q, bio); | |
1470 | } | |
1471 | ||
d0bcb878 KU |
1472 | static bool dm_rq_is_flush_request(struct request *rq) |
1473 | { | |
144d6ed5 | 1474 | if (rq->cmd_flags & REQ_FLUSH) |
d0bcb878 KU |
1475 | return true; |
1476 | else | |
1477 | return false; | |
1478 | } | |
1479 | ||
cec47e3d KU |
1480 | void dm_dispatch_request(struct request *rq) |
1481 | { | |
1482 | int r; | |
1483 | ||
1484 | if (blk_queue_io_stat(rq->q)) | |
1485 | rq->cmd_flags |= REQ_IO_STAT; | |
1486 | ||
1487 | rq->start_time = jiffies; | |
1488 | r = blk_insert_cloned_request(rq->q, rq); | |
1489 | if (r) | |
1490 | dm_complete_request(rq, r); | |
1491 | } | |
1492 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1493 | ||
1494 | static void dm_rq_bio_destructor(struct bio *bio) | |
1495 | { | |
1496 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1497 | struct mapped_device *md = info->tio->md; | |
1498 | ||
1499 | free_bio_info(info); | |
1500 | bio_free(bio, md->bs); | |
1501 | } | |
1502 | ||
1503 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1504 | void *data) | |
1505 | { | |
1506 | struct dm_rq_target_io *tio = data; | |
1507 | struct mapped_device *md = tio->md; | |
1508 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1509 | ||
1510 | if (!info) | |
1511 | return -ENOMEM; | |
1512 | ||
1513 | info->orig = bio_orig; | |
1514 | info->tio = tio; | |
1515 | bio->bi_end_io = end_clone_bio; | |
1516 | bio->bi_private = info; | |
1517 | bio->bi_destructor = dm_rq_bio_destructor; | |
1518 | ||
1519 | return 0; | |
1520 | } | |
1521 | ||
1522 | static int setup_clone(struct request *clone, struct request *rq, | |
1523 | struct dm_rq_target_io *tio) | |
1524 | { | |
d0bcb878 | 1525 | int r; |
cec47e3d | 1526 | |
d0bcb878 KU |
1527 | if (dm_rq_is_flush_request(rq)) { |
1528 | blk_rq_init(NULL, clone); | |
1529 | clone->cmd_type = REQ_TYPE_FS; | |
1530 | clone->cmd_flags |= (REQ_HARDBARRIER | WRITE); | |
1531 | } else { | |
1532 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, | |
1533 | dm_rq_bio_constructor, tio); | |
1534 | if (r) | |
1535 | return r; | |
1536 | ||
1537 | clone->cmd = rq->cmd; | |
1538 | clone->cmd_len = rq->cmd_len; | |
1539 | clone->sense = rq->sense; | |
1540 | clone->buffer = rq->buffer; | |
1541 | } | |
cec47e3d | 1542 | |
cec47e3d KU |
1543 | clone->end_io = end_clone_request; |
1544 | clone->end_io_data = tio; | |
1545 | ||
1546 | return 0; | |
1547 | } | |
1548 | ||
6facdaff KU |
1549 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1550 | gfp_t gfp_mask) | |
1551 | { | |
1552 | struct request *clone; | |
1553 | struct dm_rq_target_io *tio; | |
1554 | ||
1555 | tio = alloc_rq_tio(md, gfp_mask); | |
1556 | if (!tio) | |
1557 | return NULL; | |
1558 | ||
1559 | tio->md = md; | |
1560 | tio->ti = NULL; | |
1561 | tio->orig = rq; | |
1562 | tio->error = 0; | |
1563 | memset(&tio->info, 0, sizeof(tio->info)); | |
1564 | ||
1565 | clone = &tio->clone; | |
1566 | if (setup_clone(clone, rq, tio)) { | |
1567 | /* -ENOMEM */ | |
1568 | free_rq_tio(tio); | |
1569 | return NULL; | |
1570 | } | |
1571 | ||
1572 | return clone; | |
1573 | } | |
1574 | ||
cec47e3d KU |
1575 | /* |
1576 | * Called with the queue lock held. | |
1577 | */ | |
1578 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1579 | { | |
1580 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1581 | struct request *clone; |
1582 | ||
d0bcb878 KU |
1583 | if (unlikely(dm_rq_is_flush_request(rq))) |
1584 | return BLKPREP_OK; | |
1585 | ||
cec47e3d KU |
1586 | if (unlikely(rq->special)) { |
1587 | DMWARN("Already has something in rq->special."); | |
1588 | return BLKPREP_KILL; | |
1589 | } | |
1590 | ||
6facdaff KU |
1591 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1592 | if (!clone) | |
cec47e3d | 1593 | return BLKPREP_DEFER; |
cec47e3d KU |
1594 | |
1595 | rq->special = clone; | |
1596 | rq->cmd_flags |= REQ_DONTPREP; | |
1597 | ||
1598 | return BLKPREP_OK; | |
1599 | } | |
1600 | ||
9eef87da KU |
1601 | /* |
1602 | * Returns: | |
1603 | * 0 : the request has been processed (not requeued) | |
1604 | * !0 : the request has been requeued | |
1605 | */ | |
1606 | static int map_request(struct dm_target *ti, struct request *clone, | |
1607 | struct mapped_device *md) | |
cec47e3d | 1608 | { |
9eef87da | 1609 | int r, requeued = 0; |
cec47e3d KU |
1610 | struct dm_rq_target_io *tio = clone->end_io_data; |
1611 | ||
1612 | /* | |
1613 | * Hold the md reference here for the in-flight I/O. | |
1614 | * We can't rely on the reference count by device opener, | |
1615 | * because the device may be closed during the request completion | |
1616 | * when all bios are completed. | |
1617 | * See the comment in rq_completed() too. | |
1618 | */ | |
1619 | dm_get(md); | |
1620 | ||
1621 | tio->ti = ti; | |
1622 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1623 | switch (r) { | |
1624 | case DM_MAPIO_SUBMITTED: | |
1625 | /* The target has taken the I/O to submit by itself later */ | |
1626 | break; | |
1627 | case DM_MAPIO_REMAPPED: | |
1628 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1629 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1630 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1631 | dm_dispatch_request(clone); |
1632 | break; | |
1633 | case DM_MAPIO_REQUEUE: | |
1634 | /* The target wants to requeue the I/O */ | |
1635 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1636 | requeued = 1; |
cec47e3d KU |
1637 | break; |
1638 | default: | |
1639 | if (r > 0) { | |
1640 | DMWARN("unimplemented target map return value: %d", r); | |
1641 | BUG(); | |
1642 | } | |
1643 | ||
1644 | /* The target wants to complete the I/O */ | |
1645 | dm_kill_unmapped_request(clone, r); | |
1646 | break; | |
1647 | } | |
9eef87da KU |
1648 | |
1649 | return requeued; | |
cec47e3d KU |
1650 | } |
1651 | ||
1652 | /* | |
1653 | * q->request_fn for request-based dm. | |
1654 | * Called with the queue lock held. | |
1655 | */ | |
1656 | static void dm_request_fn(struct request_queue *q) | |
1657 | { | |
1658 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1659 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1660 | struct dm_target *ti; |
b4324fee | 1661 | struct request *rq, *clone; |
cec47e3d KU |
1662 | |
1663 | /* | |
b4324fee KU |
1664 | * For suspend, check blk_queue_stopped() and increment |
1665 | * ->pending within a single queue_lock not to increment the | |
1666 | * number of in-flight I/Os after the queue is stopped in | |
1667 | * dm_suspend(). | |
cec47e3d KU |
1668 | */ |
1669 | while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) { | |
1670 | rq = blk_peek_request(q); | |
1671 | if (!rq) | |
1672 | goto plug_and_out; | |
1673 | ||
d0bcb878 KU |
1674 | if (unlikely(dm_rq_is_flush_request(rq))) { |
1675 | BUG_ON(md->flush_request); | |
1676 | md->flush_request = rq; | |
1677 | blk_start_request(rq); | |
1678 | queue_work(md->wq, &md->barrier_work); | |
1679 | goto out; | |
1680 | } | |
1681 | ||
cec47e3d KU |
1682 | ti = dm_table_find_target(map, blk_rq_pos(rq)); |
1683 | if (ti->type->busy && ti->type->busy(ti)) | |
1684 | goto plug_and_out; | |
1685 | ||
1686 | blk_start_request(rq); | |
b4324fee KU |
1687 | clone = rq->special; |
1688 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1689 | ||
cec47e3d | 1690 | spin_unlock(q->queue_lock); |
9eef87da KU |
1691 | if (map_request(ti, clone, md)) |
1692 | goto requeued; | |
1693 | ||
cec47e3d KU |
1694 | spin_lock_irq(q->queue_lock); |
1695 | } | |
1696 | ||
1697 | goto out; | |
1698 | ||
9eef87da KU |
1699 | requeued: |
1700 | spin_lock_irq(q->queue_lock); | |
1701 | ||
cec47e3d KU |
1702 | plug_and_out: |
1703 | if (!elv_queue_empty(q)) | |
1704 | /* Some requests still remain, retry later */ | |
1705 | blk_plug_device(q); | |
1706 | ||
1707 | out: | |
1708 | dm_table_put(map); | |
1709 | ||
1710 | return; | |
1711 | } | |
1712 | ||
1713 | int dm_underlying_device_busy(struct request_queue *q) | |
1714 | { | |
1715 | return blk_lld_busy(q); | |
1716 | } | |
1717 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1718 | ||
1719 | static int dm_lld_busy(struct request_queue *q) | |
1720 | { | |
1721 | int r; | |
1722 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1723 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1724 | |
1725 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1726 | r = 1; | |
1727 | else | |
1728 | r = dm_table_any_busy_target(map); | |
1729 | ||
1730 | dm_table_put(map); | |
1731 | ||
1732 | return r; | |
1733 | } | |
1734 | ||
165125e1 | 1735 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
1736 | { |
1737 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1738 | struct dm_table *map = dm_get_live_table(md); |
1da177e4 LT |
1739 | |
1740 | if (map) { | |
cec47e3d KU |
1741 | if (dm_request_based(md)) |
1742 | generic_unplug_device(q); | |
1743 | ||
1da177e4 LT |
1744 | dm_table_unplug_all(map); |
1745 | dm_table_put(map); | |
1746 | } | |
1747 | } | |
1748 | ||
1749 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
1750 | { | |
8a57dfc6 CS |
1751 | int r = bdi_bits; |
1752 | struct mapped_device *md = congested_data; | |
1753 | struct dm_table *map; | |
1da177e4 | 1754 | |
1eb787ec | 1755 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1756 | map = dm_get_live_table(md); |
8a57dfc6 | 1757 | if (map) { |
cec47e3d KU |
1758 | /* |
1759 | * Request-based dm cares about only own queue for | |
1760 | * the query about congestion status of request_queue | |
1761 | */ | |
1762 | if (dm_request_based(md)) | |
1763 | r = md->queue->backing_dev_info.state & | |
1764 | bdi_bits; | |
1765 | else | |
1766 | r = dm_table_any_congested(map, bdi_bits); | |
1767 | ||
8a57dfc6 CS |
1768 | dm_table_put(map); |
1769 | } | |
1770 | } | |
1771 | ||
1da177e4 LT |
1772 | return r; |
1773 | } | |
1774 | ||
1775 | /*----------------------------------------------------------------- | |
1776 | * An IDR is used to keep track of allocated minor numbers. | |
1777 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1778 | static DEFINE_IDR(_minor_idr); |
1779 | ||
2b06cfff | 1780 | static void free_minor(int minor) |
1da177e4 | 1781 | { |
f32c10b0 | 1782 | spin_lock(&_minor_lock); |
1da177e4 | 1783 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1784 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1785 | } |
1786 | ||
1787 | /* | |
1788 | * See if the device with a specific minor # is free. | |
1789 | */ | |
cf13ab8e | 1790 | static int specific_minor(int minor) |
1da177e4 LT |
1791 | { |
1792 | int r, m; | |
1793 | ||
1794 | if (minor >= (1 << MINORBITS)) | |
1795 | return -EINVAL; | |
1796 | ||
62f75c2f JM |
1797 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1798 | if (!r) | |
1799 | return -ENOMEM; | |
1800 | ||
f32c10b0 | 1801 | spin_lock(&_minor_lock); |
1da177e4 LT |
1802 | |
1803 | if (idr_find(&_minor_idr, minor)) { | |
1804 | r = -EBUSY; | |
1805 | goto out; | |
1806 | } | |
1807 | ||
ba61fdd1 | 1808 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1809 | if (r) |
1da177e4 | 1810 | goto out; |
1da177e4 LT |
1811 | |
1812 | if (m != minor) { | |
1813 | idr_remove(&_minor_idr, m); | |
1814 | r = -EBUSY; | |
1815 | goto out; | |
1816 | } | |
1817 | ||
1818 | out: | |
f32c10b0 | 1819 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1820 | return r; |
1821 | } | |
1822 | ||
cf13ab8e | 1823 | static int next_free_minor(int *minor) |
1da177e4 | 1824 | { |
2b06cfff | 1825 | int r, m; |
1da177e4 | 1826 | |
1da177e4 | 1827 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1828 | if (!r) |
1829 | return -ENOMEM; | |
1830 | ||
f32c10b0 | 1831 | spin_lock(&_minor_lock); |
1da177e4 | 1832 | |
ba61fdd1 | 1833 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1834 | if (r) |
1da177e4 | 1835 | goto out; |
1da177e4 LT |
1836 | |
1837 | if (m >= (1 << MINORBITS)) { | |
1838 | idr_remove(&_minor_idr, m); | |
1839 | r = -ENOSPC; | |
1840 | goto out; | |
1841 | } | |
1842 | ||
1843 | *minor = m; | |
1844 | ||
1845 | out: | |
f32c10b0 | 1846 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1847 | return r; |
1848 | } | |
1849 | ||
83d5cde4 | 1850 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1851 | |
53d5914f | 1852 | static void dm_wq_work(struct work_struct *work); |
d0bcb878 | 1853 | static void dm_rq_barrier_work(struct work_struct *work); |
53d5914f | 1854 | |
1da177e4 LT |
1855 | /* |
1856 | * Allocate and initialise a blank device with a given minor. | |
1857 | */ | |
2b06cfff | 1858 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1859 | { |
1860 | int r; | |
cf13ab8e | 1861 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1862 | void *old_md; |
1da177e4 LT |
1863 | |
1864 | if (!md) { | |
1865 | DMWARN("unable to allocate device, out of memory."); | |
1866 | return NULL; | |
1867 | } | |
1868 | ||
10da4f79 | 1869 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1870 | goto bad_module_get; |
10da4f79 | 1871 | |
1da177e4 | 1872 | /* get a minor number for the dev */ |
2b06cfff | 1873 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1874 | r = next_free_minor(&minor); |
2b06cfff | 1875 | else |
cf13ab8e | 1876 | r = specific_minor(minor); |
1da177e4 | 1877 | if (r < 0) |
6ed7ade8 | 1878 | goto bad_minor; |
1da177e4 | 1879 | |
2ca3310e | 1880 | init_rwsem(&md->io_lock); |
e61290a4 | 1881 | mutex_init(&md->suspend_lock); |
022c2611 | 1882 | spin_lock_init(&md->deferred_lock); |
d0bcb878 | 1883 | spin_lock_init(&md->barrier_error_lock); |
1da177e4 LT |
1884 | rwlock_init(&md->map_lock); |
1885 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1886 | atomic_set(&md->open_count, 0); |
1da177e4 | 1887 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1888 | atomic_set(&md->uevent_seq, 0); |
1889 | INIT_LIST_HEAD(&md->uevent_list); | |
1890 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1891 | |
e6ee8c0b | 1892 | md->queue = blk_init_queue(dm_request_fn, NULL); |
1da177e4 | 1893 | if (!md->queue) |
6ed7ade8 | 1894 | goto bad_queue; |
1da177e4 | 1895 | |
e6ee8c0b KU |
1896 | /* |
1897 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1898 | * devices. The type of this dm device has not been decided yet, | |
1899 | * although we initialized the queue using blk_init_queue(). | |
1900 | * The type is decided at the first table loading time. | |
1901 | * To prevent problematic device stacking, clear the queue flag | |
1902 | * for request stacking support until then. | |
1903 | * | |
1904 | * This queue is new, so no concurrency on the queue_flags. | |
1905 | */ | |
1906 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1907 | md->saved_make_request_fn = md->queue->make_request_fn; | |
1da177e4 LT |
1908 | md->queue->queuedata = md; |
1909 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1910 | md->queue->backing_dev_info.congested_data = md; | |
1911 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1912 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1913 | md->queue->unplug_fn = dm_unplug_all; |
f6fccb12 | 1914 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
e6ee8c0b KU |
1915 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
1916 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
1917 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
00fff265 | 1918 | blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN_FLUSH); |
9faf400f | 1919 | |
1da177e4 LT |
1920 | md->disk = alloc_disk(1); |
1921 | if (!md->disk) | |
6ed7ade8 | 1922 | goto bad_disk; |
1da177e4 | 1923 | |
316d315b NK |
1924 | atomic_set(&md->pending[0], 0); |
1925 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1926 | init_waitqueue_head(&md->wait); |
53d5914f | 1927 | INIT_WORK(&md->work, dm_wq_work); |
d0bcb878 | 1928 | INIT_WORK(&md->barrier_work, dm_rq_barrier_work); |
f0b04115 JM |
1929 | init_waitqueue_head(&md->eventq); |
1930 | ||
1da177e4 LT |
1931 | md->disk->major = _major; |
1932 | md->disk->first_minor = minor; | |
1933 | md->disk->fops = &dm_blk_dops; | |
1934 | md->disk->queue = md->queue; | |
1935 | md->disk->private_data = md; | |
1936 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1937 | add_disk(md->disk); | |
7e51f257 | 1938 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1939 | |
304f3f6a MB |
1940 | md->wq = create_singlethread_workqueue("kdmflush"); |
1941 | if (!md->wq) | |
1942 | goto bad_thread; | |
1943 | ||
32a926da MP |
1944 | md->bdev = bdget_disk(md->disk, 0); |
1945 | if (!md->bdev) | |
1946 | goto bad_bdev; | |
1947 | ||
ba61fdd1 | 1948 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1949 | spin_lock(&_minor_lock); |
ba61fdd1 | 1950 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1951 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1952 | |
1953 | BUG_ON(old_md != MINOR_ALLOCED); | |
1954 | ||
1da177e4 LT |
1955 | return md; |
1956 | ||
32a926da MP |
1957 | bad_bdev: |
1958 | destroy_workqueue(md->wq); | |
304f3f6a | 1959 | bad_thread: |
03022c54 | 1960 | del_gendisk(md->disk); |
304f3f6a | 1961 | put_disk(md->disk); |
6ed7ade8 | 1962 | bad_disk: |
1312f40e | 1963 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1964 | bad_queue: |
1da177e4 | 1965 | free_minor(minor); |
6ed7ade8 | 1966 | bad_minor: |
10da4f79 | 1967 | module_put(THIS_MODULE); |
6ed7ade8 | 1968 | bad_module_get: |
1da177e4 LT |
1969 | kfree(md); |
1970 | return NULL; | |
1971 | } | |
1972 | ||
ae9da83f JN |
1973 | static void unlock_fs(struct mapped_device *md); |
1974 | ||
1da177e4 LT |
1975 | static void free_dev(struct mapped_device *md) |
1976 | { | |
f331c029 | 1977 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1978 | |
32a926da MP |
1979 | unlock_fs(md); |
1980 | bdput(md->bdev); | |
304f3f6a | 1981 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1982 | if (md->tio_pool) |
1983 | mempool_destroy(md->tio_pool); | |
1984 | if (md->io_pool) | |
1985 | mempool_destroy(md->io_pool); | |
1986 | if (md->bs) | |
1987 | bioset_free(md->bs); | |
9c47008d | 1988 | blk_integrity_unregister(md->disk); |
1da177e4 | 1989 | del_gendisk(md->disk); |
63d94e48 | 1990 | free_minor(minor); |
fba9f90e JM |
1991 | |
1992 | spin_lock(&_minor_lock); | |
1993 | md->disk->private_data = NULL; | |
1994 | spin_unlock(&_minor_lock); | |
1995 | ||
1da177e4 | 1996 | put_disk(md->disk); |
1312f40e | 1997 | blk_cleanup_queue(md->queue); |
10da4f79 | 1998 | module_put(THIS_MODULE); |
1da177e4 LT |
1999 | kfree(md); |
2000 | } | |
2001 | ||
e6ee8c0b KU |
2002 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2003 | { | |
2004 | struct dm_md_mempools *p; | |
2005 | ||
2006 | if (md->io_pool && md->tio_pool && md->bs) | |
2007 | /* the md already has necessary mempools */ | |
2008 | goto out; | |
2009 | ||
2010 | p = dm_table_get_md_mempools(t); | |
2011 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
2012 | ||
2013 | md->io_pool = p->io_pool; | |
2014 | p->io_pool = NULL; | |
2015 | md->tio_pool = p->tio_pool; | |
2016 | p->tio_pool = NULL; | |
2017 | md->bs = p->bs; | |
2018 | p->bs = NULL; | |
2019 | ||
2020 | out: | |
2021 | /* mempool bind completed, now no need any mempools in the table */ | |
2022 | dm_table_free_md_mempools(t); | |
2023 | } | |
2024 | ||
1da177e4 LT |
2025 | /* |
2026 | * Bind a table to the device. | |
2027 | */ | |
2028 | static void event_callback(void *context) | |
2029 | { | |
7a8c3d3b MA |
2030 | unsigned long flags; |
2031 | LIST_HEAD(uevents); | |
1da177e4 LT |
2032 | struct mapped_device *md = (struct mapped_device *) context; |
2033 | ||
7a8c3d3b MA |
2034 | spin_lock_irqsave(&md->uevent_lock, flags); |
2035 | list_splice_init(&md->uevent_list, &uevents); | |
2036 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2037 | ||
ed9e1982 | 2038 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2039 | |
1da177e4 LT |
2040 | atomic_inc(&md->event_nr); |
2041 | wake_up(&md->eventq); | |
2042 | } | |
2043 | ||
4e90188b | 2044 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2045 | { |
4e90188b | 2046 | set_capacity(md->disk, size); |
1da177e4 | 2047 | |
db8fef4f MP |
2048 | mutex_lock(&md->bdev->bd_inode->i_mutex); |
2049 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
2050 | mutex_unlock(&md->bdev->bd_inode->i_mutex); | |
1da177e4 LT |
2051 | } |
2052 | ||
042d2a9b AK |
2053 | /* |
2054 | * Returns old map, which caller must destroy. | |
2055 | */ | |
2056 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2057 | struct queue_limits *limits) | |
1da177e4 | 2058 | { |
042d2a9b | 2059 | struct dm_table *old_map; |
165125e1 | 2060 | struct request_queue *q = md->queue; |
1da177e4 | 2061 | sector_t size; |
523d9297 | 2062 | unsigned long flags; |
1da177e4 LT |
2063 | |
2064 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2065 | |
2066 | /* | |
2067 | * Wipe any geometry if the size of the table changed. | |
2068 | */ | |
2069 | if (size != get_capacity(md->disk)) | |
2070 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2071 | ||
32a926da | 2072 | __set_size(md, size); |
d5816876 | 2073 | |
2ca3310e AK |
2074 | dm_table_event_callback(t, event_callback, md); |
2075 | ||
e6ee8c0b KU |
2076 | /* |
2077 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2078 | * for request-based during suspension. So stop it to prevent | |
2079 | * I/O mapping before resume. | |
2080 | * This must be done before setting the queue restrictions, | |
2081 | * because request-based dm may be run just after the setting. | |
2082 | */ | |
2083 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2084 | stop_queue(q); | |
2085 | ||
2086 | __bind_mempools(md, t); | |
2087 | ||
523d9297 | 2088 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2089 | old_map = md->map; |
1da177e4 | 2090 | md->map = t; |
754c5fc7 | 2091 | dm_table_set_restrictions(t, q, limits); |
523d9297 | 2092 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2093 | |
042d2a9b | 2094 | return old_map; |
1da177e4 LT |
2095 | } |
2096 | ||
a7940155 AK |
2097 | /* |
2098 | * Returns unbound table for the caller to free. | |
2099 | */ | |
2100 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2101 | { |
2102 | struct dm_table *map = md->map; | |
523d9297 | 2103 | unsigned long flags; |
1da177e4 LT |
2104 | |
2105 | if (!map) | |
a7940155 | 2106 | return NULL; |
1da177e4 LT |
2107 | |
2108 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2109 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2110 | md->map = NULL; |
523d9297 | 2111 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2112 | |
2113 | return map; | |
1da177e4 LT |
2114 | } |
2115 | ||
2116 | /* | |
2117 | * Constructor for a new device. | |
2118 | */ | |
2b06cfff | 2119 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2120 | { |
2121 | struct mapped_device *md; | |
2122 | ||
2b06cfff | 2123 | md = alloc_dev(minor); |
1da177e4 LT |
2124 | if (!md) |
2125 | return -ENXIO; | |
2126 | ||
784aae73 MB |
2127 | dm_sysfs_init(md); |
2128 | ||
1da177e4 LT |
2129 | *result = md; |
2130 | return 0; | |
2131 | } | |
2132 | ||
637842cf | 2133 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2134 | { |
2135 | struct mapped_device *md; | |
1da177e4 LT |
2136 | unsigned minor = MINOR(dev); |
2137 | ||
2138 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2139 | return NULL; | |
2140 | ||
f32c10b0 | 2141 | spin_lock(&_minor_lock); |
1da177e4 LT |
2142 | |
2143 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2144 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2145 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2146 | dm_deleting_md(md) || |
17b2f66f | 2147 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2148 | md = NULL; |
fba9f90e JM |
2149 | goto out; |
2150 | } | |
1da177e4 | 2151 | |
fba9f90e | 2152 | out: |
f32c10b0 | 2153 | spin_unlock(&_minor_lock); |
1da177e4 | 2154 | |
637842cf DT |
2155 | return md; |
2156 | } | |
2157 | ||
d229a958 DT |
2158 | struct mapped_device *dm_get_md(dev_t dev) |
2159 | { | |
2160 | struct mapped_device *md = dm_find_md(dev); | |
2161 | ||
2162 | if (md) | |
2163 | dm_get(md); | |
2164 | ||
2165 | return md; | |
2166 | } | |
2167 | ||
9ade92a9 | 2168 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2169 | { |
9ade92a9 | 2170 | return md->interface_ptr; |
1da177e4 LT |
2171 | } |
2172 | ||
2173 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2174 | { | |
2175 | md->interface_ptr = ptr; | |
2176 | } | |
2177 | ||
2178 | void dm_get(struct mapped_device *md) | |
2179 | { | |
2180 | atomic_inc(&md->holders); | |
3f77316d | 2181 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2182 | } |
2183 | ||
72d94861 AK |
2184 | const char *dm_device_name(struct mapped_device *md) |
2185 | { | |
2186 | return md->name; | |
2187 | } | |
2188 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2189 | ||
3f77316d | 2190 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2191 | { |
1134e5ae | 2192 | struct dm_table *map; |
1da177e4 | 2193 | |
3f77316d | 2194 | might_sleep(); |
fba9f90e | 2195 | |
3f77316d KU |
2196 | spin_lock(&_minor_lock); |
2197 | map = dm_get_live_table(md); | |
2198 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2199 | set_bit(DMF_FREEING, &md->flags); | |
2200 | spin_unlock(&_minor_lock); | |
2201 | ||
2202 | if (!dm_suspended_md(md)) { | |
2203 | dm_table_presuspend_targets(map); | |
2204 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2205 | } |
3f77316d KU |
2206 | |
2207 | /* | |
2208 | * Rare, but there may be I/O requests still going to complete, | |
2209 | * for example. Wait for all references to disappear. | |
2210 | * No one should increment the reference count of the mapped_device, | |
2211 | * after the mapped_device state becomes DMF_FREEING. | |
2212 | */ | |
2213 | if (wait) | |
2214 | while (atomic_read(&md->holders)) | |
2215 | msleep(1); | |
2216 | else if (atomic_read(&md->holders)) | |
2217 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2218 | dm_device_name(md), atomic_read(&md->holders)); | |
2219 | ||
2220 | dm_sysfs_exit(md); | |
2221 | dm_table_put(map); | |
2222 | dm_table_destroy(__unbind(md)); | |
2223 | free_dev(md); | |
2224 | } | |
2225 | ||
2226 | void dm_destroy(struct mapped_device *md) | |
2227 | { | |
2228 | __dm_destroy(md, true); | |
2229 | } | |
2230 | ||
2231 | void dm_destroy_immediate(struct mapped_device *md) | |
2232 | { | |
2233 | __dm_destroy(md, false); | |
2234 | } | |
2235 | ||
2236 | void dm_put(struct mapped_device *md) | |
2237 | { | |
2238 | atomic_dec(&md->holders); | |
1da177e4 | 2239 | } |
79eb885c | 2240 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2241 | |
401600df | 2242 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2243 | { |
2244 | int r = 0; | |
b44ebeb0 MP |
2245 | DECLARE_WAITQUEUE(wait, current); |
2246 | ||
2247 | dm_unplug_all(md->queue); | |
2248 | ||
2249 | add_wait_queue(&md->wait, &wait); | |
46125c1c MB |
2250 | |
2251 | while (1) { | |
401600df | 2252 | set_current_state(interruptible); |
46125c1c MB |
2253 | |
2254 | smp_mb(); | |
b4324fee | 2255 | if (!md_in_flight(md)) |
46125c1c MB |
2256 | break; |
2257 | ||
401600df MP |
2258 | if (interruptible == TASK_INTERRUPTIBLE && |
2259 | signal_pending(current)) { | |
46125c1c MB |
2260 | r = -EINTR; |
2261 | break; | |
2262 | } | |
2263 | ||
2264 | io_schedule(); | |
2265 | } | |
2266 | set_current_state(TASK_RUNNING); | |
2267 | ||
b44ebeb0 MP |
2268 | remove_wait_queue(&md->wait, &wait); |
2269 | ||
46125c1c MB |
2270 | return r; |
2271 | } | |
2272 | ||
531fe963 | 2273 | static void dm_flush(struct mapped_device *md) |
af7e466a MP |
2274 | { |
2275 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
52b1fd5a MP |
2276 | |
2277 | bio_init(&md->barrier_bio); | |
2278 | md->barrier_bio.bi_bdev = md->bdev; | |
2279 | md->barrier_bio.bi_rw = WRITE_BARRIER; | |
2280 | __split_and_process_bio(md, &md->barrier_bio); | |
2281 | ||
2282 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
af7e466a MP |
2283 | } |
2284 | ||
2285 | static void process_barrier(struct mapped_device *md, struct bio *bio) | |
2286 | { | |
5aa2781d MP |
2287 | md->barrier_error = 0; |
2288 | ||
531fe963 | 2289 | dm_flush(md); |
af7e466a | 2290 | |
5aa2781d MP |
2291 | if (!bio_empty_barrier(bio)) { |
2292 | __split_and_process_bio(md, bio); | |
708e9295 MP |
2293 | /* |
2294 | * If the request isn't supported, don't waste time with | |
2295 | * the second flush. | |
2296 | */ | |
2297 | if (md->barrier_error != -EOPNOTSUPP) | |
2298 | dm_flush(md); | |
af7e466a MP |
2299 | } |
2300 | ||
af7e466a | 2301 | if (md->barrier_error != DM_ENDIO_REQUEUE) |
531fe963 | 2302 | bio_endio(bio, md->barrier_error); |
2761e95f MP |
2303 | else { |
2304 | spin_lock_irq(&md->deferred_lock); | |
2305 | bio_list_add_head(&md->deferred, bio); | |
2306 | spin_unlock_irq(&md->deferred_lock); | |
2307 | } | |
af7e466a MP |
2308 | } |
2309 | ||
1da177e4 LT |
2310 | /* |
2311 | * Process the deferred bios | |
2312 | */ | |
ef208587 | 2313 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2314 | { |
ef208587 MP |
2315 | struct mapped_device *md = container_of(work, struct mapped_device, |
2316 | work); | |
6d6f10df | 2317 | struct bio *c; |
1da177e4 | 2318 | |
ef208587 MP |
2319 | down_write(&md->io_lock); |
2320 | ||
3b00b203 | 2321 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2322 | spin_lock_irq(&md->deferred_lock); |
2323 | c = bio_list_pop(&md->deferred); | |
2324 | spin_unlock_irq(&md->deferred_lock); | |
2325 | ||
2326 | if (!c) { | |
1eb787ec | 2327 | clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); |
df12ee99 AK |
2328 | break; |
2329 | } | |
022c2611 | 2330 | |
3b00b203 MP |
2331 | up_write(&md->io_lock); |
2332 | ||
e6ee8c0b KU |
2333 | if (dm_request_based(md)) |
2334 | generic_make_request(c); | |
2335 | else { | |
7b6d91da | 2336 | if (c->bi_rw & REQ_HARDBARRIER) |
e6ee8c0b KU |
2337 | process_barrier(md, c); |
2338 | else | |
2339 | __split_and_process_bio(md, c); | |
2340 | } | |
3b00b203 MP |
2341 | |
2342 | down_write(&md->io_lock); | |
022c2611 | 2343 | } |
73d410c0 | 2344 | |
ef208587 | 2345 | up_write(&md->io_lock); |
1da177e4 LT |
2346 | } |
2347 | ||
9a1fb464 | 2348 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2349 | { |
3b00b203 MP |
2350 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2351 | smp_mb__after_clear_bit(); | |
53d5914f | 2352 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2353 | } |
2354 | ||
d0bcb878 KU |
2355 | static void dm_rq_set_flush_nr(struct request *clone, unsigned flush_nr) |
2356 | { | |
2357 | struct dm_rq_target_io *tio = clone->end_io_data; | |
2358 | ||
2359 | tio->info.flush_request = flush_nr; | |
2360 | } | |
2361 | ||
2362 | /* Issue barrier requests to targets and wait for their completion. */ | |
2363 | static int dm_rq_barrier(struct mapped_device *md) | |
2364 | { | |
2365 | int i, j; | |
7c666411 | 2366 | struct dm_table *map = dm_get_live_table(md); |
d0bcb878 KU |
2367 | unsigned num_targets = dm_table_get_num_targets(map); |
2368 | struct dm_target *ti; | |
2369 | struct request *clone; | |
2370 | ||
2371 | md->barrier_error = 0; | |
2372 | ||
2373 | for (i = 0; i < num_targets; i++) { | |
2374 | ti = dm_table_get_target(map, i); | |
2375 | for (j = 0; j < ti->num_flush_requests; j++) { | |
2376 | clone = clone_rq(md->flush_request, md, GFP_NOIO); | |
2377 | dm_rq_set_flush_nr(clone, j); | |
2378 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
2379 | map_request(ti, clone, md); | |
2380 | } | |
2381 | } | |
2382 | ||
2383 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2384 | dm_table_put(map); | |
2385 | ||
2386 | return md->barrier_error; | |
2387 | } | |
2388 | ||
2389 | static void dm_rq_barrier_work(struct work_struct *work) | |
2390 | { | |
2391 | int error; | |
2392 | struct mapped_device *md = container_of(work, struct mapped_device, | |
2393 | barrier_work); | |
2394 | struct request_queue *q = md->queue; | |
2395 | struct request *rq; | |
2396 | unsigned long flags; | |
2397 | ||
2398 | /* | |
2399 | * Hold the md reference here and leave it at the last part so that | |
2400 | * the md can't be deleted by device opener when the barrier request | |
2401 | * completes. | |
2402 | */ | |
2403 | dm_get(md); | |
2404 | ||
2405 | error = dm_rq_barrier(md); | |
2406 | ||
2407 | rq = md->flush_request; | |
2408 | md->flush_request = NULL; | |
2409 | ||
2410 | if (error == DM_ENDIO_REQUEUE) { | |
2411 | spin_lock_irqsave(q->queue_lock, flags); | |
2412 | blk_requeue_request(q, rq); | |
2413 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2414 | } else | |
2415 | blk_end_request_all(rq, error); | |
2416 | ||
2417 | blk_run_queue(q); | |
2418 | ||
2419 | dm_put(md); | |
2420 | } | |
2421 | ||
1da177e4 | 2422 | /* |
042d2a9b | 2423 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2424 | */ |
042d2a9b | 2425 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2426 | { |
042d2a9b | 2427 | struct dm_table *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2428 | struct queue_limits limits; |
042d2a9b | 2429 | int r; |
1da177e4 | 2430 | |
e61290a4 | 2431 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2432 | |
2433 | /* device must be suspended */ | |
4f186f8b | 2434 | if (!dm_suspended_md(md)) |
93c534ae | 2435 | goto out; |
1da177e4 | 2436 | |
754c5fc7 | 2437 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2438 | if (r) { |
2439 | map = ERR_PTR(r); | |
754c5fc7 | 2440 | goto out; |
042d2a9b | 2441 | } |
754c5fc7 | 2442 | |
e6ee8c0b KU |
2443 | /* cannot change the device type, once a table is bound */ |
2444 | if (md->map && | |
2445 | (dm_table_get_type(md->map) != dm_table_get_type(table))) { | |
2446 | DMWARN("can't change the device type after a table is bound"); | |
2447 | goto out; | |
2448 | } | |
2449 | ||
042d2a9b | 2450 | map = __bind(md, table, &limits); |
1da177e4 | 2451 | |
93c534ae | 2452 | out: |
e61290a4 | 2453 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2454 | return map; |
1da177e4 LT |
2455 | } |
2456 | ||
2457 | /* | |
2458 | * Functions to lock and unlock any filesystem running on the | |
2459 | * device. | |
2460 | */ | |
2ca3310e | 2461 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2462 | { |
e39e2e95 | 2463 | int r; |
1da177e4 LT |
2464 | |
2465 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2466 | |
db8fef4f | 2467 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2468 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2469 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2470 | md->frozen_sb = NULL; |
2471 | return r; | |
dfbe03f6 AK |
2472 | } |
2473 | ||
aa8d7c2f AK |
2474 | set_bit(DMF_FROZEN, &md->flags); |
2475 | ||
1da177e4 LT |
2476 | return 0; |
2477 | } | |
2478 | ||
2ca3310e | 2479 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2480 | { |
aa8d7c2f AK |
2481 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2482 | return; | |
2483 | ||
db8fef4f | 2484 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2485 | md->frozen_sb = NULL; |
aa8d7c2f | 2486 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2487 | } |
2488 | ||
2489 | /* | |
2490 | * We need to be able to change a mapping table under a mounted | |
2491 | * filesystem. For example we might want to move some data in | |
2492 | * the background. Before the table can be swapped with | |
2493 | * dm_bind_table, dm_suspend must be called to flush any in | |
2494 | * flight bios and ensure that any further io gets deferred. | |
2495 | */ | |
cec47e3d KU |
2496 | /* |
2497 | * Suspend mechanism in request-based dm. | |
2498 | * | |
9f518b27 KU |
2499 | * 1. Flush all I/Os by lock_fs() if needed. |
2500 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2501 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2502 | * |
9f518b27 | 2503 | * To abort suspend, start the request_queue. |
cec47e3d | 2504 | */ |
a3d77d35 | 2505 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2506 | { |
2ca3310e | 2507 | struct dm_table *map = NULL; |
46125c1c | 2508 | int r = 0; |
a3d77d35 | 2509 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2510 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2511 | |
e61290a4 | 2512 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2513 | |
4f186f8b | 2514 | if (dm_suspended_md(md)) { |
73d410c0 | 2515 | r = -EINVAL; |
d287483d | 2516 | goto out_unlock; |
73d410c0 | 2517 | } |
1da177e4 | 2518 | |
7c666411 | 2519 | map = dm_get_live_table(md); |
1da177e4 | 2520 | |
2e93ccc1 KU |
2521 | /* |
2522 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2523 | * This flag is cleared before dm_suspend returns. | |
2524 | */ | |
2525 | if (noflush) | |
2526 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2527 | ||
cf222b37 AK |
2528 | /* This does not get reverted if there's an error later. */ |
2529 | dm_table_presuspend_targets(map); | |
2530 | ||
32a926da | 2531 | /* |
9f518b27 KU |
2532 | * Flush I/O to the device. |
2533 | * Any I/O submitted after lock_fs() may not be flushed. | |
2534 | * noflush takes precedence over do_lockfs. | |
2535 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2536 | */ |
2537 | if (!noflush && do_lockfs) { | |
2538 | r = lock_fs(md); | |
2539 | if (r) | |
f431d966 | 2540 | goto out; |
aa8d7c2f | 2541 | } |
1da177e4 LT |
2542 | |
2543 | /* | |
3b00b203 MP |
2544 | * Here we must make sure that no processes are submitting requests |
2545 | * to target drivers i.e. no one may be executing | |
2546 | * __split_and_process_bio. This is called from dm_request and | |
2547 | * dm_wq_work. | |
2548 | * | |
2549 | * To get all processes out of __split_and_process_bio in dm_request, | |
2550 | * we take the write lock. To prevent any process from reentering | |
2551 | * __split_and_process_bio from dm_request, we set | |
2552 | * DMF_QUEUE_IO_TO_THREAD. | |
2553 | * | |
2554 | * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND | |
2555 | * and call flush_workqueue(md->wq). flush_workqueue will wait until | |
2556 | * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any | |
2557 | * further calls to __split_and_process_bio from dm_wq_work. | |
1da177e4 | 2558 | */ |
2ca3310e | 2559 | down_write(&md->io_lock); |
1eb787ec AK |
2560 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2561 | set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); | |
2ca3310e | 2562 | up_write(&md->io_lock); |
1da177e4 | 2563 | |
d0bcb878 KU |
2564 | /* |
2565 | * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which | |
2566 | * can be kicked until md->queue is stopped. So stop md->queue before | |
2567 | * flushing md->wq. | |
2568 | */ | |
cec47e3d | 2569 | if (dm_request_based(md)) |
9f518b27 | 2570 | stop_queue(md->queue); |
cec47e3d | 2571 | |
d0bcb878 KU |
2572 | flush_workqueue(md->wq); |
2573 | ||
1da177e4 | 2574 | /* |
3b00b203 MP |
2575 | * At this point no more requests are entering target request routines. |
2576 | * We call dm_wait_for_completion to wait for all existing requests | |
2577 | * to finish. | |
1da177e4 | 2578 | */ |
401600df | 2579 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2580 | |
2ca3310e | 2581 | down_write(&md->io_lock); |
6d6f10df | 2582 | if (noflush) |
022c2611 | 2583 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2584 | up_write(&md->io_lock); |
2e93ccc1 | 2585 | |
1da177e4 | 2586 | /* were we interrupted ? */ |
46125c1c | 2587 | if (r < 0) { |
9a1fb464 | 2588 | dm_queue_flush(md); |
73d410c0 | 2589 | |
cec47e3d | 2590 | if (dm_request_based(md)) |
9f518b27 | 2591 | start_queue(md->queue); |
cec47e3d | 2592 | |
2ca3310e | 2593 | unlock_fs(md); |
2e93ccc1 | 2594 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2595 | } |
1da177e4 | 2596 | |
3b00b203 MP |
2597 | /* |
2598 | * If dm_wait_for_completion returned 0, the device is completely | |
2599 | * quiescent now. There is no request-processing activity. All new | |
2600 | * requests are being added to md->deferred list. | |
2601 | */ | |
2602 | ||
2ca3310e | 2603 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2604 | |
4d4471cb KU |
2605 | dm_table_postsuspend_targets(map); |
2606 | ||
2ca3310e AK |
2607 | out: |
2608 | dm_table_put(map); | |
d287483d AK |
2609 | |
2610 | out_unlock: | |
e61290a4 | 2611 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2612 | return r; |
1da177e4 LT |
2613 | } |
2614 | ||
2615 | int dm_resume(struct mapped_device *md) | |
2616 | { | |
cf222b37 | 2617 | int r = -EINVAL; |
cf222b37 | 2618 | struct dm_table *map = NULL; |
1da177e4 | 2619 | |
e61290a4 | 2620 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2621 | if (!dm_suspended_md(md)) |
cf222b37 | 2622 | goto out; |
cf222b37 | 2623 | |
7c666411 | 2624 | map = dm_get_live_table(md); |
2ca3310e | 2625 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2626 | goto out; |
1da177e4 | 2627 | |
8757b776 MB |
2628 | r = dm_table_resume_targets(map); |
2629 | if (r) | |
2630 | goto out; | |
2ca3310e | 2631 | |
9a1fb464 | 2632 | dm_queue_flush(md); |
2ca3310e | 2633 | |
cec47e3d KU |
2634 | /* |
2635 | * Flushing deferred I/Os must be done after targets are resumed | |
2636 | * so that mapping of targets can work correctly. | |
2637 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2638 | */ | |
2639 | if (dm_request_based(md)) | |
2640 | start_queue(md->queue); | |
2641 | ||
2ca3310e AK |
2642 | unlock_fs(md); |
2643 | ||
2644 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2645 | ||
1da177e4 | 2646 | dm_table_unplug_all(map); |
cf222b37 AK |
2647 | r = 0; |
2648 | out: | |
2649 | dm_table_put(map); | |
e61290a4 | 2650 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2651 | |
cf222b37 | 2652 | return r; |
1da177e4 LT |
2653 | } |
2654 | ||
2655 | /*----------------------------------------------------------------- | |
2656 | * Event notification. | |
2657 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2658 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2659 | unsigned cookie) |
69267a30 | 2660 | { |
60935eb2 MB |
2661 | char udev_cookie[DM_COOKIE_LENGTH]; |
2662 | char *envp[] = { udev_cookie, NULL }; | |
2663 | ||
2664 | if (!cookie) | |
3abf85b5 | 2665 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2666 | else { |
2667 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2668 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2669 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2670 | action, envp); | |
60935eb2 | 2671 | } |
69267a30 AK |
2672 | } |
2673 | ||
7a8c3d3b MA |
2674 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2675 | { | |
2676 | return atomic_add_return(1, &md->uevent_seq); | |
2677 | } | |
2678 | ||
1da177e4 LT |
2679 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2680 | { | |
2681 | return atomic_read(&md->event_nr); | |
2682 | } | |
2683 | ||
2684 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2685 | { | |
2686 | return wait_event_interruptible(md->eventq, | |
2687 | (event_nr != atomic_read(&md->event_nr))); | |
2688 | } | |
2689 | ||
7a8c3d3b MA |
2690 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2691 | { | |
2692 | unsigned long flags; | |
2693 | ||
2694 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2695 | list_add(elist, &md->uevent_list); | |
2696 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2697 | } | |
2698 | ||
1da177e4 LT |
2699 | /* |
2700 | * The gendisk is only valid as long as you have a reference | |
2701 | * count on 'md'. | |
2702 | */ | |
2703 | struct gendisk *dm_disk(struct mapped_device *md) | |
2704 | { | |
2705 | return md->disk; | |
2706 | } | |
2707 | ||
784aae73 MB |
2708 | struct kobject *dm_kobject(struct mapped_device *md) |
2709 | { | |
2710 | return &md->kobj; | |
2711 | } | |
2712 | ||
2713 | /* | |
2714 | * struct mapped_device should not be exported outside of dm.c | |
2715 | * so use this check to verify that kobj is part of md structure | |
2716 | */ | |
2717 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2718 | { | |
2719 | struct mapped_device *md; | |
2720 | ||
2721 | md = container_of(kobj, struct mapped_device, kobj); | |
2722 | if (&md->kobj != kobj) | |
2723 | return NULL; | |
2724 | ||
4d89b7b4 | 2725 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2726 | dm_deleting_md(md)) |
4d89b7b4 MB |
2727 | return NULL; |
2728 | ||
784aae73 MB |
2729 | dm_get(md); |
2730 | return md; | |
2731 | } | |
2732 | ||
4f186f8b | 2733 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2734 | { |
2735 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2736 | } | |
2737 | ||
64dbce58 KU |
2738 | int dm_suspended(struct dm_target *ti) |
2739 | { | |
ecdb2e25 | 2740 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2741 | } |
2742 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2743 | ||
2e93ccc1 KU |
2744 | int dm_noflush_suspending(struct dm_target *ti) |
2745 | { | |
ecdb2e25 | 2746 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2747 | } |
2748 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2749 | ||
e6ee8c0b KU |
2750 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type) |
2751 | { | |
2752 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
2753 | ||
2754 | if (!pools) | |
2755 | return NULL; | |
2756 | ||
2757 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2758 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2759 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2760 | if (!pools->io_pool) | |
2761 | goto free_pools_and_out; | |
2762 | ||
2763 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2764 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2765 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2766 | if (!pools->tio_pool) | |
2767 | goto free_io_pool_and_out; | |
2768 | ||
2769 | pools->bs = (type == DM_TYPE_BIO_BASED) ? | |
2770 | bioset_create(16, 0) : bioset_create(MIN_IOS, 0); | |
2771 | if (!pools->bs) | |
2772 | goto free_tio_pool_and_out; | |
2773 | ||
2774 | return pools; | |
2775 | ||
2776 | free_tio_pool_and_out: | |
2777 | mempool_destroy(pools->tio_pool); | |
2778 | ||
2779 | free_io_pool_and_out: | |
2780 | mempool_destroy(pools->io_pool); | |
2781 | ||
2782 | free_pools_and_out: | |
2783 | kfree(pools); | |
2784 | ||
2785 | return NULL; | |
2786 | } | |
2787 | ||
2788 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2789 | { | |
2790 | if (!pools) | |
2791 | return; | |
2792 | ||
2793 | if (pools->io_pool) | |
2794 | mempool_destroy(pools->io_pool); | |
2795 | ||
2796 | if (pools->tio_pool) | |
2797 | mempool_destroy(pools->tio_pool); | |
2798 | ||
2799 | if (pools->bs) | |
2800 | bioset_free(pools->bs); | |
2801 | ||
2802 | kfree(pools); | |
2803 | } | |
2804 | ||
83d5cde4 | 2805 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2806 | .open = dm_blk_open, |
2807 | .release = dm_blk_close, | |
aa129a22 | 2808 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2809 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2810 | .owner = THIS_MODULE |
2811 | }; | |
2812 | ||
2813 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2814 | ||
2815 | /* | |
2816 | * module hooks | |
2817 | */ | |
2818 | module_init(dm_init); | |
2819 | module_exit(dm_exit); | |
2820 | ||
2821 | module_param(major, uint, 0); | |
2822 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2823 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2824 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2825 | MODULE_LICENSE("GPL"); |