]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
17 | #include <linux/buffer_head.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
3ac51e74 | 21 | #include <linux/hdreg.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
60935eb2 MB |
27 | /* |
28 | * Cookies are numeric values sent with CHANGE and REMOVE | |
29 | * uevents while resuming, removing or renaming the device. | |
30 | */ | |
31 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
32 | #define DM_COOKIE_LENGTH 24 | |
33 | ||
1da177e4 LT |
34 | static const char *_name = DM_NAME; |
35 | ||
36 | static unsigned int major = 0; | |
37 | static unsigned int _major = 0; | |
38 | ||
f32c10b0 | 39 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 40 | /* |
8fbf26ad | 41 | * For bio-based dm. |
1da177e4 LT |
42 | * One of these is allocated per bio. |
43 | */ | |
44 | struct dm_io { | |
45 | struct mapped_device *md; | |
46 | int error; | |
1da177e4 | 47 | atomic_t io_count; |
6ae2fa67 | 48 | struct bio *bio; |
3eaf840e | 49 | unsigned long start_time; |
f88fb981 | 50 | spinlock_t endio_lock; |
1da177e4 LT |
51 | }; |
52 | ||
53 | /* | |
8fbf26ad | 54 | * For bio-based dm. |
1da177e4 LT |
55 | * One of these is allocated per target within a bio. Hopefully |
56 | * this will be simplified out one day. | |
57 | */ | |
028867ac | 58 | struct dm_target_io { |
1da177e4 LT |
59 | struct dm_io *io; |
60 | struct dm_target *ti; | |
61 | union map_info info; | |
62 | }; | |
63 | ||
8fbf26ad KU |
64 | /* |
65 | * For request-based dm. | |
66 | * One of these is allocated per request. | |
67 | */ | |
68 | struct dm_rq_target_io { | |
69 | struct mapped_device *md; | |
70 | struct dm_target *ti; | |
71 | struct request *orig, clone; | |
72 | int error; | |
73 | union map_info info; | |
74 | }; | |
75 | ||
76 | /* | |
77 | * For request-based dm. | |
78 | * One of these is allocated per bio. | |
79 | */ | |
80 | struct dm_rq_clone_bio_info { | |
81 | struct bio *orig; | |
cec47e3d | 82 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
83 | }; |
84 | ||
1da177e4 LT |
85 | union map_info *dm_get_mapinfo(struct bio *bio) |
86 | { | |
17b2f66f | 87 | if (bio && bio->bi_private) |
028867ac | 88 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 89 | return NULL; |
1da177e4 LT |
90 | } |
91 | ||
cec47e3d KU |
92 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
93 | { | |
94 | if (rq && rq->end_io_data) | |
95 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
96 | return NULL; | |
97 | } | |
98 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
99 | ||
ba61fdd1 JM |
100 | #define MINOR_ALLOCED ((void *)-1) |
101 | ||
1da177e4 LT |
102 | /* |
103 | * Bits for the md->flags field. | |
104 | */ | |
1eb787ec | 105 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 106 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 107 | #define DMF_FROZEN 2 |
fba9f90e | 108 | #define DMF_FREEING 3 |
5c6bd75d | 109 | #define DMF_DELETING 4 |
2e93ccc1 | 110 | #define DMF_NOFLUSH_SUSPENDING 5 |
1eb787ec | 111 | #define DMF_QUEUE_IO_TO_THREAD 6 |
1da177e4 | 112 | |
304f3f6a MB |
113 | /* |
114 | * Work processed by per-device workqueue. | |
115 | */ | |
1da177e4 | 116 | struct mapped_device { |
2ca3310e | 117 | struct rw_semaphore io_lock; |
e61290a4 | 118 | struct mutex suspend_lock; |
1da177e4 LT |
119 | rwlock_t map_lock; |
120 | atomic_t holders; | |
5c6bd75d | 121 | atomic_t open_count; |
1da177e4 LT |
122 | |
123 | unsigned long flags; | |
124 | ||
165125e1 | 125 | struct request_queue *queue; |
1da177e4 | 126 | struct gendisk *disk; |
7e51f257 | 127 | char name[16]; |
1da177e4 LT |
128 | |
129 | void *interface_ptr; | |
130 | ||
131 | /* | |
132 | * A list of ios that arrived while we were suspended. | |
133 | */ | |
316d315b | 134 | atomic_t pending[2]; |
1da177e4 | 135 | wait_queue_head_t wait; |
53d5914f | 136 | struct work_struct work; |
74859364 | 137 | struct bio_list deferred; |
022c2611 | 138 | spinlock_t deferred_lock; |
1da177e4 | 139 | |
af7e466a MP |
140 | /* |
141 | * An error from the barrier request currently being processed. | |
142 | */ | |
143 | int barrier_error; | |
144 | ||
d0bcb878 KU |
145 | /* |
146 | * Protect barrier_error from concurrent endio processing | |
147 | * in request-based dm. | |
148 | */ | |
149 | spinlock_t barrier_error_lock; | |
150 | ||
304f3f6a MB |
151 | /* |
152 | * Processing queue (flush/barriers) | |
153 | */ | |
154 | struct workqueue_struct *wq; | |
d0bcb878 KU |
155 | struct work_struct barrier_work; |
156 | ||
157 | /* A pointer to the currently processing pre/post flush request */ | |
158 | struct request *flush_request; | |
304f3f6a | 159 | |
1da177e4 LT |
160 | /* |
161 | * The current mapping. | |
162 | */ | |
163 | struct dm_table *map; | |
164 | ||
165 | /* | |
166 | * io objects are allocated from here. | |
167 | */ | |
168 | mempool_t *io_pool; | |
169 | mempool_t *tio_pool; | |
170 | ||
9faf400f SB |
171 | struct bio_set *bs; |
172 | ||
1da177e4 LT |
173 | /* |
174 | * Event handling. | |
175 | */ | |
176 | atomic_t event_nr; | |
177 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
178 | atomic_t uevent_seq; |
179 | struct list_head uevent_list; | |
180 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
181 | |
182 | /* | |
183 | * freeze/thaw support require holding onto a super block | |
184 | */ | |
185 | struct super_block *frozen_sb; | |
db8fef4f | 186 | struct block_device *bdev; |
3ac51e74 DW |
187 | |
188 | /* forced geometry settings */ | |
189 | struct hd_geometry geometry; | |
784aae73 | 190 | |
cec47e3d KU |
191 | /* For saving the address of __make_request for request based dm */ |
192 | make_request_fn *saved_make_request_fn; | |
193 | ||
784aae73 MB |
194 | /* sysfs handle */ |
195 | struct kobject kobj; | |
52b1fd5a MP |
196 | |
197 | /* zero-length barrier that will be cloned and submitted to targets */ | |
198 | struct bio barrier_bio; | |
1da177e4 LT |
199 | }; |
200 | ||
e6ee8c0b KU |
201 | /* |
202 | * For mempools pre-allocation at the table loading time. | |
203 | */ | |
204 | struct dm_md_mempools { | |
205 | mempool_t *io_pool; | |
206 | mempool_t *tio_pool; | |
207 | struct bio_set *bs; | |
208 | }; | |
209 | ||
1da177e4 | 210 | #define MIN_IOS 256 |
e18b890b CL |
211 | static struct kmem_cache *_io_cache; |
212 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
213 | static struct kmem_cache *_rq_tio_cache; |
214 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 215 | |
1da177e4 LT |
216 | static int __init local_init(void) |
217 | { | |
51157b4a | 218 | int r = -ENOMEM; |
1da177e4 | 219 | |
1da177e4 | 220 | /* allocate a slab for the dm_ios */ |
028867ac | 221 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 222 | if (!_io_cache) |
51157b4a | 223 | return r; |
1da177e4 LT |
224 | |
225 | /* allocate a slab for the target ios */ | |
028867ac | 226 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
227 | if (!_tio_cache) |
228 | goto out_free_io_cache; | |
1da177e4 | 229 | |
8fbf26ad KU |
230 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
231 | if (!_rq_tio_cache) | |
232 | goto out_free_tio_cache; | |
233 | ||
234 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
235 | if (!_rq_bio_info_cache) | |
236 | goto out_free_rq_tio_cache; | |
237 | ||
51e5b2bd | 238 | r = dm_uevent_init(); |
51157b4a | 239 | if (r) |
8fbf26ad | 240 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 241 | |
1da177e4 LT |
242 | _major = major; |
243 | r = register_blkdev(_major, _name); | |
51157b4a KU |
244 | if (r < 0) |
245 | goto out_uevent_exit; | |
1da177e4 LT |
246 | |
247 | if (!_major) | |
248 | _major = r; | |
249 | ||
250 | return 0; | |
51157b4a KU |
251 | |
252 | out_uevent_exit: | |
253 | dm_uevent_exit(); | |
8fbf26ad KU |
254 | out_free_rq_bio_info_cache: |
255 | kmem_cache_destroy(_rq_bio_info_cache); | |
256 | out_free_rq_tio_cache: | |
257 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
258 | out_free_tio_cache: |
259 | kmem_cache_destroy(_tio_cache); | |
260 | out_free_io_cache: | |
261 | kmem_cache_destroy(_io_cache); | |
262 | ||
263 | return r; | |
1da177e4 LT |
264 | } |
265 | ||
266 | static void local_exit(void) | |
267 | { | |
8fbf26ad KU |
268 | kmem_cache_destroy(_rq_bio_info_cache); |
269 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
270 | kmem_cache_destroy(_tio_cache); |
271 | kmem_cache_destroy(_io_cache); | |
00d59405 | 272 | unregister_blkdev(_major, _name); |
51e5b2bd | 273 | dm_uevent_exit(); |
1da177e4 LT |
274 | |
275 | _major = 0; | |
276 | ||
277 | DMINFO("cleaned up"); | |
278 | } | |
279 | ||
b9249e55 | 280 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
281 | local_init, |
282 | dm_target_init, | |
283 | dm_linear_init, | |
284 | dm_stripe_init, | |
952b3557 | 285 | dm_io_init, |
945fa4d2 | 286 | dm_kcopyd_init, |
1da177e4 LT |
287 | dm_interface_init, |
288 | }; | |
289 | ||
b9249e55 | 290 | static void (*_exits[])(void) = { |
1da177e4 LT |
291 | local_exit, |
292 | dm_target_exit, | |
293 | dm_linear_exit, | |
294 | dm_stripe_exit, | |
952b3557 | 295 | dm_io_exit, |
945fa4d2 | 296 | dm_kcopyd_exit, |
1da177e4 LT |
297 | dm_interface_exit, |
298 | }; | |
299 | ||
300 | static int __init dm_init(void) | |
301 | { | |
302 | const int count = ARRAY_SIZE(_inits); | |
303 | ||
304 | int r, i; | |
305 | ||
306 | for (i = 0; i < count; i++) { | |
307 | r = _inits[i](); | |
308 | if (r) | |
309 | goto bad; | |
310 | } | |
311 | ||
312 | return 0; | |
313 | ||
314 | bad: | |
315 | while (i--) | |
316 | _exits[i](); | |
317 | ||
318 | return r; | |
319 | } | |
320 | ||
321 | static void __exit dm_exit(void) | |
322 | { | |
323 | int i = ARRAY_SIZE(_exits); | |
324 | ||
325 | while (i--) | |
326 | _exits[i](); | |
327 | } | |
328 | ||
329 | /* | |
330 | * Block device functions | |
331 | */ | |
432a212c MA |
332 | int dm_deleting_md(struct mapped_device *md) |
333 | { | |
334 | return test_bit(DMF_DELETING, &md->flags); | |
335 | } | |
336 | ||
fe5f9f2c | 337 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
338 | { |
339 | struct mapped_device *md; | |
340 | ||
fba9f90e JM |
341 | spin_lock(&_minor_lock); |
342 | ||
fe5f9f2c | 343 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
344 | if (!md) |
345 | goto out; | |
346 | ||
5c6bd75d | 347 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 348 | dm_deleting_md(md)) { |
fba9f90e JM |
349 | md = NULL; |
350 | goto out; | |
351 | } | |
352 | ||
1da177e4 | 353 | dm_get(md); |
5c6bd75d | 354 | atomic_inc(&md->open_count); |
fba9f90e JM |
355 | |
356 | out: | |
357 | spin_unlock(&_minor_lock); | |
358 | ||
359 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
360 | } |
361 | ||
fe5f9f2c | 362 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 363 | { |
fe5f9f2c | 364 | struct mapped_device *md = disk->private_data; |
5c6bd75d | 365 | atomic_dec(&md->open_count); |
1da177e4 LT |
366 | dm_put(md); |
367 | return 0; | |
368 | } | |
369 | ||
5c6bd75d AK |
370 | int dm_open_count(struct mapped_device *md) |
371 | { | |
372 | return atomic_read(&md->open_count); | |
373 | } | |
374 | ||
375 | /* | |
376 | * Guarantees nothing is using the device before it's deleted. | |
377 | */ | |
378 | int dm_lock_for_deletion(struct mapped_device *md) | |
379 | { | |
380 | int r = 0; | |
381 | ||
382 | spin_lock(&_minor_lock); | |
383 | ||
384 | if (dm_open_count(md)) | |
385 | r = -EBUSY; | |
386 | else | |
387 | set_bit(DMF_DELETING, &md->flags); | |
388 | ||
389 | spin_unlock(&_minor_lock); | |
390 | ||
391 | return r; | |
392 | } | |
393 | ||
3ac51e74 DW |
394 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
395 | { | |
396 | struct mapped_device *md = bdev->bd_disk->private_data; | |
397 | ||
398 | return dm_get_geometry(md, geo); | |
399 | } | |
400 | ||
fe5f9f2c | 401 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
402 | unsigned int cmd, unsigned long arg) |
403 | { | |
fe5f9f2c | 404 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 405 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
406 | struct dm_target *tgt; |
407 | int r = -ENOTTY; | |
408 | ||
aa129a22 MB |
409 | if (!map || !dm_table_get_size(map)) |
410 | goto out; | |
411 | ||
412 | /* We only support devices that have a single target */ | |
413 | if (dm_table_get_num_targets(map) != 1) | |
414 | goto out; | |
415 | ||
416 | tgt = dm_table_get_target(map, 0); | |
417 | ||
4f186f8b | 418 | if (dm_suspended_md(md)) { |
aa129a22 MB |
419 | r = -EAGAIN; |
420 | goto out; | |
421 | } | |
422 | ||
423 | if (tgt->type->ioctl) | |
647b3d00 | 424 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
425 | |
426 | out: | |
427 | dm_table_put(map); | |
428 | ||
aa129a22 MB |
429 | return r; |
430 | } | |
431 | ||
028867ac | 432 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
433 | { |
434 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
435 | } | |
436 | ||
028867ac | 437 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
438 | { |
439 | mempool_free(io, md->io_pool); | |
440 | } | |
441 | ||
028867ac | 442 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
443 | { |
444 | mempool_free(tio, md->tio_pool); | |
445 | } | |
446 | ||
08885643 KU |
447 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
448 | gfp_t gfp_mask) | |
cec47e3d | 449 | { |
08885643 | 450 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
451 | } |
452 | ||
453 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
454 | { | |
455 | mempool_free(tio, tio->md->tio_pool); | |
456 | } | |
457 | ||
458 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
459 | { | |
460 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
461 | } | |
462 | ||
463 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
464 | { | |
465 | mempool_free(info, info->tio->md->io_pool); | |
466 | } | |
467 | ||
90abb8c4 KU |
468 | static int md_in_flight(struct mapped_device *md) |
469 | { | |
470 | return atomic_read(&md->pending[READ]) + | |
471 | atomic_read(&md->pending[WRITE]); | |
472 | } | |
473 | ||
3eaf840e JNN |
474 | static void start_io_acct(struct dm_io *io) |
475 | { | |
476 | struct mapped_device *md = io->md; | |
c9959059 | 477 | int cpu; |
316d315b | 478 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
479 | |
480 | io->start_time = jiffies; | |
481 | ||
074a7aca TH |
482 | cpu = part_stat_lock(); |
483 | part_round_stats(cpu, &dm_disk(md)->part0); | |
484 | part_stat_unlock(); | |
316d315b | 485 | dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]); |
3eaf840e JNN |
486 | } |
487 | ||
d221d2e7 | 488 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
489 | { |
490 | struct mapped_device *md = io->md; | |
491 | struct bio *bio = io->bio; | |
492 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 493 | int pending, cpu; |
3eaf840e JNN |
494 | int rw = bio_data_dir(bio); |
495 | ||
074a7aca TH |
496 | cpu = part_stat_lock(); |
497 | part_round_stats(cpu, &dm_disk(md)->part0); | |
498 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
499 | part_stat_unlock(); | |
3eaf840e | 500 | |
af7e466a MP |
501 | /* |
502 | * After this is decremented the bio must not be touched if it is | |
503 | * a barrier. | |
504 | */ | |
316d315b NK |
505 | dm_disk(md)->part0.in_flight[rw] = pending = |
506 | atomic_dec_return(&md->pending[rw]); | |
507 | pending += atomic_read(&md->pending[rw^0x1]); | |
3eaf840e | 508 | |
d221d2e7 MP |
509 | /* nudge anyone waiting on suspend queue */ |
510 | if (!pending) | |
511 | wake_up(&md->wait); | |
3eaf840e JNN |
512 | } |
513 | ||
1da177e4 LT |
514 | /* |
515 | * Add the bio to the list of deferred io. | |
516 | */ | |
92c63902 | 517 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 518 | { |
2ca3310e | 519 | down_write(&md->io_lock); |
1da177e4 | 520 | |
022c2611 | 521 | spin_lock_irq(&md->deferred_lock); |
1da177e4 | 522 | bio_list_add(&md->deferred, bio); |
022c2611 | 523 | spin_unlock_irq(&md->deferred_lock); |
1da177e4 | 524 | |
92c63902 MP |
525 | if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) |
526 | queue_work(md->wq, &md->work); | |
527 | ||
2ca3310e | 528 | up_write(&md->io_lock); |
1da177e4 LT |
529 | } |
530 | ||
531 | /* | |
532 | * Everyone (including functions in this file), should use this | |
533 | * function to access the md->map field, and make sure they call | |
534 | * dm_table_put() when finished. | |
535 | */ | |
7c666411 | 536 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
537 | { |
538 | struct dm_table *t; | |
523d9297 | 539 | unsigned long flags; |
1da177e4 | 540 | |
523d9297 | 541 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
542 | t = md->map; |
543 | if (t) | |
544 | dm_table_get(t); | |
523d9297 | 545 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
546 | |
547 | return t; | |
548 | } | |
549 | ||
3ac51e74 DW |
550 | /* |
551 | * Get the geometry associated with a dm device | |
552 | */ | |
553 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
554 | { | |
555 | *geo = md->geometry; | |
556 | ||
557 | return 0; | |
558 | } | |
559 | ||
560 | /* | |
561 | * Set the geometry of a device. | |
562 | */ | |
563 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
564 | { | |
565 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
566 | ||
567 | if (geo->start > sz) { | |
568 | DMWARN("Start sector is beyond the geometry limits."); | |
569 | return -EINVAL; | |
570 | } | |
571 | ||
572 | md->geometry = *geo; | |
573 | ||
574 | return 0; | |
575 | } | |
576 | ||
1da177e4 LT |
577 | /*----------------------------------------------------------------- |
578 | * CRUD START: | |
579 | * A more elegant soln is in the works that uses the queue | |
580 | * merge fn, unfortunately there are a couple of changes to | |
581 | * the block layer that I want to make for this. So in the | |
582 | * interests of getting something for people to use I give | |
583 | * you this clearly demarcated crap. | |
584 | *---------------------------------------------------------------*/ | |
585 | ||
2e93ccc1 KU |
586 | static int __noflush_suspending(struct mapped_device *md) |
587 | { | |
588 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
589 | } | |
590 | ||
1da177e4 LT |
591 | /* |
592 | * Decrements the number of outstanding ios that a bio has been | |
593 | * cloned into, completing the original io if necc. | |
594 | */ | |
858119e1 | 595 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 596 | { |
2e93ccc1 | 597 | unsigned long flags; |
b35f8caa MB |
598 | int io_error; |
599 | struct bio *bio; | |
600 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
601 | |
602 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
603 | if (unlikely(error)) { |
604 | spin_lock_irqsave(&io->endio_lock, flags); | |
605 | if (!(io->error > 0 && __noflush_suspending(md))) | |
606 | io->error = error; | |
607 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
608 | } | |
1da177e4 LT |
609 | |
610 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
611 | if (io->error == DM_ENDIO_REQUEUE) { |
612 | /* | |
613 | * Target requested pushing back the I/O. | |
2e93ccc1 | 614 | */ |
022c2611 | 615 | spin_lock_irqsave(&md->deferred_lock, flags); |
2761e95f | 616 | if (__noflush_suspending(md)) { |
1f98a13f | 617 | if (!bio_rw_flagged(io->bio, BIO_RW_BARRIER)) |
2761e95f MP |
618 | bio_list_add_head(&md->deferred, |
619 | io->bio); | |
620 | } else | |
2e93ccc1 KU |
621 | /* noflush suspend was interrupted. */ |
622 | io->error = -EIO; | |
022c2611 | 623 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
624 | } |
625 | ||
b35f8caa MB |
626 | io_error = io->error; |
627 | bio = io->bio; | |
2e93ccc1 | 628 | |
1f98a13f | 629 | if (bio_rw_flagged(bio, BIO_RW_BARRIER)) { |
af7e466a MP |
630 | /* |
631 | * There can be just one barrier request so we use | |
632 | * a per-device variable for error reporting. | |
633 | * Note that you can't touch the bio after end_io_acct | |
634 | */ | |
fdb9572b | 635 | if (!md->barrier_error && io_error != -EOPNOTSUPP) |
5aa2781d | 636 | md->barrier_error = io_error; |
af7e466a | 637 | end_io_acct(io); |
a97f925a | 638 | free_io(md, io); |
af7e466a MP |
639 | } else { |
640 | end_io_acct(io); | |
a97f925a | 641 | free_io(md, io); |
b35f8caa | 642 | |
af7e466a MP |
643 | if (io_error != DM_ENDIO_REQUEUE) { |
644 | trace_block_bio_complete(md->queue, bio); | |
2056a782 | 645 | |
af7e466a MP |
646 | bio_endio(bio, io_error); |
647 | } | |
b35f8caa | 648 | } |
1da177e4 LT |
649 | } |
650 | } | |
651 | ||
6712ecf8 | 652 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
653 | { |
654 | int r = 0; | |
028867ac | 655 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 656 | struct dm_io *io = tio->io; |
9faf400f | 657 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
658 | dm_endio_fn endio = tio->ti->type->end_io; |
659 | ||
1da177e4 LT |
660 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
661 | error = -EIO; | |
662 | ||
663 | if (endio) { | |
664 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
665 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
666 | /* | |
667 | * error and requeue request are handled | |
668 | * in dec_pending(). | |
669 | */ | |
1da177e4 | 670 | error = r; |
45cbcd79 KU |
671 | else if (r == DM_ENDIO_INCOMPLETE) |
672 | /* The target will handle the io */ | |
6712ecf8 | 673 | return; |
45cbcd79 KU |
674 | else if (r) { |
675 | DMWARN("unimplemented target endio return value: %d", r); | |
676 | BUG(); | |
677 | } | |
1da177e4 LT |
678 | } |
679 | ||
9faf400f SB |
680 | /* |
681 | * Store md for cleanup instead of tio which is about to get freed. | |
682 | */ | |
683 | bio->bi_private = md->bs; | |
684 | ||
9faf400f | 685 | free_tio(md, tio); |
b35f8caa MB |
686 | bio_put(bio); |
687 | dec_pending(io, error); | |
1da177e4 LT |
688 | } |
689 | ||
cec47e3d KU |
690 | /* |
691 | * Partial completion handling for request-based dm | |
692 | */ | |
693 | static void end_clone_bio(struct bio *clone, int error) | |
694 | { | |
695 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
696 | struct dm_rq_target_io *tio = info->tio; | |
697 | struct bio *bio = info->orig; | |
698 | unsigned int nr_bytes = info->orig->bi_size; | |
699 | ||
700 | bio_put(clone); | |
701 | ||
702 | if (tio->error) | |
703 | /* | |
704 | * An error has already been detected on the request. | |
705 | * Once error occurred, just let clone->end_io() handle | |
706 | * the remainder. | |
707 | */ | |
708 | return; | |
709 | else if (error) { | |
710 | /* | |
711 | * Don't notice the error to the upper layer yet. | |
712 | * The error handling decision is made by the target driver, | |
713 | * when the request is completed. | |
714 | */ | |
715 | tio->error = error; | |
716 | return; | |
717 | } | |
718 | ||
719 | /* | |
720 | * I/O for the bio successfully completed. | |
721 | * Notice the data completion to the upper layer. | |
722 | */ | |
723 | ||
724 | /* | |
725 | * bios are processed from the head of the list. | |
726 | * So the completing bio should always be rq->bio. | |
727 | * If it's not, something wrong is happening. | |
728 | */ | |
729 | if (tio->orig->bio != bio) | |
730 | DMERR("bio completion is going in the middle of the request"); | |
731 | ||
732 | /* | |
733 | * Update the original request. | |
734 | * Do not use blk_end_request() here, because it may complete | |
735 | * the original request before the clone, and break the ordering. | |
736 | */ | |
737 | blk_update_request(tio->orig, 0, nr_bytes); | |
738 | } | |
739 | ||
d0bcb878 KU |
740 | static void store_barrier_error(struct mapped_device *md, int error) |
741 | { | |
742 | unsigned long flags; | |
743 | ||
744 | spin_lock_irqsave(&md->barrier_error_lock, flags); | |
745 | /* | |
746 | * Basically, the first error is taken, but: | |
747 | * -EOPNOTSUPP supersedes any I/O error. | |
748 | * Requeue request supersedes any I/O error but -EOPNOTSUPP. | |
749 | */ | |
750 | if (!md->barrier_error || error == -EOPNOTSUPP || | |
751 | (md->barrier_error != -EOPNOTSUPP && | |
752 | error == DM_ENDIO_REQUEUE)) | |
753 | md->barrier_error = error; | |
754 | spin_unlock_irqrestore(&md->barrier_error_lock, flags); | |
755 | } | |
756 | ||
cec47e3d KU |
757 | /* |
758 | * Don't touch any member of the md after calling this function because | |
759 | * the md may be freed in dm_put() at the end of this function. | |
760 | * Or do dm_get() before calling this function and dm_put() later. | |
761 | */ | |
b4324fee | 762 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 763 | { |
b4324fee | 764 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
765 | |
766 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 767 | if (!md_in_flight(md)) |
cec47e3d KU |
768 | wake_up(&md->wait); |
769 | ||
770 | if (run_queue) | |
b4324fee | 771 | blk_run_queue(md->queue); |
cec47e3d KU |
772 | |
773 | /* | |
774 | * dm_put() must be at the end of this function. See the comment above | |
775 | */ | |
776 | dm_put(md); | |
777 | } | |
778 | ||
a77e28c7 KU |
779 | static void free_rq_clone(struct request *clone) |
780 | { | |
781 | struct dm_rq_target_io *tio = clone->end_io_data; | |
782 | ||
783 | blk_rq_unprep_clone(clone); | |
784 | free_rq_tio(tio); | |
785 | } | |
786 | ||
980691e5 KU |
787 | /* |
788 | * Complete the clone and the original request. | |
789 | * Must be called without queue lock. | |
790 | */ | |
791 | static void dm_end_request(struct request *clone, int error) | |
792 | { | |
793 | int rw = rq_data_dir(clone); | |
d0bcb878 KU |
794 | int run_queue = 1; |
795 | bool is_barrier = blk_barrier_rq(clone); | |
980691e5 KU |
796 | struct dm_rq_target_io *tio = clone->end_io_data; |
797 | struct mapped_device *md = tio->md; | |
798 | struct request *rq = tio->orig; | |
799 | ||
d0bcb878 | 800 | if (blk_pc_request(rq) && !is_barrier) { |
980691e5 KU |
801 | rq->errors = clone->errors; |
802 | rq->resid_len = clone->resid_len; | |
803 | ||
804 | if (rq->sense) | |
805 | /* | |
806 | * We are using the sense buffer of the original | |
807 | * request. | |
808 | * So setting the length of the sense data is enough. | |
809 | */ | |
810 | rq->sense_len = clone->sense_len; | |
811 | } | |
812 | ||
813 | free_rq_clone(clone); | |
814 | ||
d0bcb878 KU |
815 | if (unlikely(is_barrier)) { |
816 | if (unlikely(error)) | |
817 | store_barrier_error(md, error); | |
818 | run_queue = 0; | |
819 | } else | |
820 | blk_end_request_all(rq, error); | |
980691e5 | 821 | |
d0bcb878 | 822 | rq_completed(md, rw, run_queue); |
980691e5 KU |
823 | } |
824 | ||
cec47e3d KU |
825 | static void dm_unprep_request(struct request *rq) |
826 | { | |
827 | struct request *clone = rq->special; | |
cec47e3d KU |
828 | |
829 | rq->special = NULL; | |
830 | rq->cmd_flags &= ~REQ_DONTPREP; | |
831 | ||
a77e28c7 | 832 | free_rq_clone(clone); |
cec47e3d KU |
833 | } |
834 | ||
835 | /* | |
836 | * Requeue the original request of a clone. | |
837 | */ | |
838 | void dm_requeue_unmapped_request(struct request *clone) | |
839 | { | |
b4324fee | 840 | int rw = rq_data_dir(clone); |
cec47e3d KU |
841 | struct dm_rq_target_io *tio = clone->end_io_data; |
842 | struct mapped_device *md = tio->md; | |
843 | struct request *rq = tio->orig; | |
844 | struct request_queue *q = rq->q; | |
845 | unsigned long flags; | |
846 | ||
d0bcb878 KU |
847 | if (unlikely(blk_barrier_rq(clone))) { |
848 | /* | |
849 | * Barrier clones share an original request. | |
850 | * Leave it to dm_end_request(), which handles this special | |
851 | * case. | |
852 | */ | |
853 | dm_end_request(clone, DM_ENDIO_REQUEUE); | |
854 | return; | |
855 | } | |
856 | ||
cec47e3d KU |
857 | dm_unprep_request(rq); |
858 | ||
859 | spin_lock_irqsave(q->queue_lock, flags); | |
860 | if (elv_queue_empty(q)) | |
861 | blk_plug_device(q); | |
862 | blk_requeue_request(q, rq); | |
863 | spin_unlock_irqrestore(q->queue_lock, flags); | |
864 | ||
b4324fee | 865 | rq_completed(md, rw, 0); |
cec47e3d KU |
866 | } |
867 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
868 | ||
869 | static void __stop_queue(struct request_queue *q) | |
870 | { | |
871 | blk_stop_queue(q); | |
872 | } | |
873 | ||
874 | static void stop_queue(struct request_queue *q) | |
875 | { | |
876 | unsigned long flags; | |
877 | ||
878 | spin_lock_irqsave(q->queue_lock, flags); | |
879 | __stop_queue(q); | |
880 | spin_unlock_irqrestore(q->queue_lock, flags); | |
881 | } | |
882 | ||
883 | static void __start_queue(struct request_queue *q) | |
884 | { | |
885 | if (blk_queue_stopped(q)) | |
886 | blk_start_queue(q); | |
887 | } | |
888 | ||
889 | static void start_queue(struct request_queue *q) | |
890 | { | |
891 | unsigned long flags; | |
892 | ||
893 | spin_lock_irqsave(q->queue_lock, flags); | |
894 | __start_queue(q); | |
895 | spin_unlock_irqrestore(q->queue_lock, flags); | |
896 | } | |
897 | ||
11a68244 | 898 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 899 | { |
11a68244 | 900 | int r = error; |
cec47e3d KU |
901 | struct dm_rq_target_io *tio = clone->end_io_data; |
902 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
cec47e3d | 903 | |
11a68244 KU |
904 | if (mapped && rq_end_io) |
905 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
cec47e3d | 906 | |
11a68244 | 907 | if (r <= 0) |
cec47e3d | 908 | /* The target wants to complete the I/O */ |
11a68244 KU |
909 | dm_end_request(clone, r); |
910 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
911 | /* The target will handle the I/O */ |
912 | return; | |
11a68244 | 913 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
914 | /* The target wants to requeue the I/O */ |
915 | dm_requeue_unmapped_request(clone); | |
916 | else { | |
11a68244 | 917 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
918 | BUG(); |
919 | } | |
920 | } | |
921 | ||
11a68244 KU |
922 | /* |
923 | * Request completion handler for request-based dm | |
924 | */ | |
925 | static void dm_softirq_done(struct request *rq) | |
926 | { | |
927 | bool mapped = true; | |
928 | struct request *clone = rq->completion_data; | |
929 | struct dm_rq_target_io *tio = clone->end_io_data; | |
930 | ||
931 | if (rq->cmd_flags & REQ_FAILED) | |
932 | mapped = false; | |
933 | ||
934 | dm_done(clone, tio->error, mapped); | |
935 | } | |
936 | ||
cec47e3d KU |
937 | /* |
938 | * Complete the clone and the original request with the error status | |
939 | * through softirq context. | |
940 | */ | |
941 | static void dm_complete_request(struct request *clone, int error) | |
942 | { | |
943 | struct dm_rq_target_io *tio = clone->end_io_data; | |
944 | struct request *rq = tio->orig; | |
945 | ||
d0bcb878 KU |
946 | if (unlikely(blk_barrier_rq(clone))) { |
947 | /* | |
948 | * Barrier clones share an original request. So can't use | |
949 | * softirq_done with the original. | |
950 | * Pass the clone to dm_done() directly in this special case. | |
951 | * It is safe (even if clone->q->queue_lock is held here) | |
952 | * because there is no I/O dispatching during the completion | |
953 | * of barrier clone. | |
954 | */ | |
955 | dm_done(clone, error, true); | |
956 | return; | |
957 | } | |
958 | ||
cec47e3d KU |
959 | tio->error = error; |
960 | rq->completion_data = clone; | |
961 | blk_complete_request(rq); | |
962 | } | |
963 | ||
964 | /* | |
965 | * Complete the not-mapped clone and the original request with the error status | |
966 | * through softirq context. | |
967 | * Target's rq_end_io() function isn't called. | |
968 | * This may be used when the target's map_rq() function fails. | |
969 | */ | |
970 | void dm_kill_unmapped_request(struct request *clone, int error) | |
971 | { | |
972 | struct dm_rq_target_io *tio = clone->end_io_data; | |
973 | struct request *rq = tio->orig; | |
974 | ||
d0bcb878 KU |
975 | if (unlikely(blk_barrier_rq(clone))) { |
976 | /* | |
977 | * Barrier clones share an original request. | |
978 | * Leave it to dm_end_request(), which handles this special | |
979 | * case. | |
980 | */ | |
981 | BUG_ON(error > 0); | |
982 | dm_end_request(clone, error); | |
983 | return; | |
984 | } | |
985 | ||
cec47e3d KU |
986 | rq->cmd_flags |= REQ_FAILED; |
987 | dm_complete_request(clone, error); | |
988 | } | |
989 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
990 | ||
991 | /* | |
992 | * Called with the queue lock held | |
993 | */ | |
994 | static void end_clone_request(struct request *clone, int error) | |
995 | { | |
996 | /* | |
997 | * For just cleaning up the information of the queue in which | |
998 | * the clone was dispatched. | |
999 | * The clone is *NOT* freed actually here because it is alloced from | |
1000 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1001 | */ | |
1002 | __blk_put_request(clone->q, clone); | |
1003 | ||
1004 | /* | |
1005 | * Actual request completion is done in a softirq context which doesn't | |
1006 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1007 | * - another request may be submitted by the upper level driver | |
1008 | * of the stacking during the completion | |
1009 | * - the submission which requires queue lock may be done | |
1010 | * against this queue | |
1011 | */ | |
1012 | dm_complete_request(clone, error); | |
1013 | } | |
1014 | ||
1da177e4 LT |
1015 | static sector_t max_io_len(struct mapped_device *md, |
1016 | sector_t sector, struct dm_target *ti) | |
1017 | { | |
1018 | sector_t offset = sector - ti->begin; | |
1019 | sector_t len = ti->len - offset; | |
1020 | ||
1021 | /* | |
1022 | * Does the target need to split even further ? | |
1023 | */ | |
1024 | if (ti->split_io) { | |
1025 | sector_t boundary; | |
1026 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
1027 | - offset; | |
1028 | if (len > boundary) | |
1029 | len = boundary; | |
1030 | } | |
1031 | ||
1032 | return len; | |
1033 | } | |
1034 | ||
1035 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 1036 | struct dm_target_io *tio) |
1da177e4 LT |
1037 | { |
1038 | int r; | |
2056a782 | 1039 | sector_t sector; |
9faf400f | 1040 | struct mapped_device *md; |
1da177e4 | 1041 | |
1da177e4 LT |
1042 | clone->bi_end_io = clone_endio; |
1043 | clone->bi_private = tio; | |
1044 | ||
1045 | /* | |
1046 | * Map the clone. If r == 0 we don't need to do | |
1047 | * anything, the target has assumed ownership of | |
1048 | * this io. | |
1049 | */ | |
1050 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1051 | sector = clone->bi_sector; |
1da177e4 | 1052 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 1053 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1054 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1055 | |
5f3ea37c | 1056 | trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, |
22a7c31a | 1057 | tio->io->bio->bi_bdev->bd_dev, sector); |
2056a782 | 1058 | |
1da177e4 | 1059 | generic_make_request(clone); |
2e93ccc1 KU |
1060 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1061 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1062 | md = tio->io->md; |
1063 | dec_pending(tio->io, r); | |
1064 | /* | |
1065 | * Store bio_set for cleanup. | |
1066 | */ | |
1067 | clone->bi_private = md->bs; | |
1da177e4 | 1068 | bio_put(clone); |
9faf400f | 1069 | free_tio(md, tio); |
45cbcd79 KU |
1070 | } else if (r) { |
1071 | DMWARN("unimplemented target map return value: %d", r); | |
1072 | BUG(); | |
1da177e4 LT |
1073 | } |
1074 | } | |
1075 | ||
1076 | struct clone_info { | |
1077 | struct mapped_device *md; | |
1078 | struct dm_table *map; | |
1079 | struct bio *bio; | |
1080 | struct dm_io *io; | |
1081 | sector_t sector; | |
1082 | sector_t sector_count; | |
1083 | unsigned short idx; | |
1084 | }; | |
1085 | ||
3676347a PO |
1086 | static void dm_bio_destructor(struct bio *bio) |
1087 | { | |
9faf400f SB |
1088 | struct bio_set *bs = bio->bi_private; |
1089 | ||
1090 | bio_free(bio, bs); | |
3676347a PO |
1091 | } |
1092 | ||
1da177e4 LT |
1093 | /* |
1094 | * Creates a little bio that is just does part of a bvec. | |
1095 | */ | |
1096 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1097 | unsigned short idx, unsigned int offset, | |
9faf400f | 1098 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1099 | { |
1100 | struct bio *clone; | |
1101 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1102 | ||
9faf400f | 1103 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1104 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1105 | *clone->bi_io_vec = *bv; |
1106 | ||
1107 | clone->bi_sector = sector; | |
1108 | clone->bi_bdev = bio->bi_bdev; | |
af7e466a | 1109 | clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER); |
1da177e4 LT |
1110 | clone->bi_vcnt = 1; |
1111 | clone->bi_size = to_bytes(len); | |
1112 | clone->bi_io_vec->bv_offset = offset; | |
1113 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1114 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1115 | |
9c47008d | 1116 | if (bio_integrity(bio)) { |
7878cba9 | 1117 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1118 | bio_integrity_trim(clone, |
1119 | bio_sector_offset(bio, idx, offset), len); | |
1120 | } | |
1121 | ||
1da177e4 LT |
1122 | return clone; |
1123 | } | |
1124 | ||
1125 | /* | |
1126 | * Creates a bio that consists of range of complete bvecs. | |
1127 | */ | |
1128 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1129 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1130 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1131 | { |
1132 | struct bio *clone; | |
1133 | ||
9faf400f SB |
1134 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1135 | __bio_clone(clone, bio); | |
af7e466a | 1136 | clone->bi_rw &= ~(1 << BIO_RW_BARRIER); |
9faf400f | 1137 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1138 | clone->bi_sector = sector; |
1139 | clone->bi_idx = idx; | |
1140 | clone->bi_vcnt = idx + bv_count; | |
1141 | clone->bi_size = to_bytes(len); | |
1142 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1143 | ||
9c47008d | 1144 | if (bio_integrity(bio)) { |
7878cba9 | 1145 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1146 | |
1147 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1148 | bio_integrity_trim(clone, | |
1149 | bio_sector_offset(bio, idx, 0), len); | |
1150 | } | |
1151 | ||
1da177e4 LT |
1152 | return clone; |
1153 | } | |
1154 | ||
9015df24 AK |
1155 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1156 | struct dm_target *ti) | |
f9ab94ce | 1157 | { |
9015df24 | 1158 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1159 | |
1160 | tio->io = ci->io; | |
1161 | tio->ti = ti; | |
f9ab94ce | 1162 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1163 | |
1164 | return tio; | |
1165 | } | |
1166 | ||
1167 | static void __flush_target(struct clone_info *ci, struct dm_target *ti, | |
1168 | unsigned flush_nr) | |
1169 | { | |
1170 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1171 | struct bio *clone; | |
1172 | ||
f9ab94ce MP |
1173 | tio->info.flush_request = flush_nr; |
1174 | ||
1175 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); | |
1176 | __bio_clone(clone, ci->bio); | |
1177 | clone->bi_destructor = dm_bio_destructor; | |
1178 | ||
1179 | __map_bio(ti, clone, tio); | |
1180 | } | |
1181 | ||
1182 | static int __clone_and_map_empty_barrier(struct clone_info *ci) | |
1183 | { | |
1184 | unsigned target_nr = 0, flush_nr; | |
1185 | struct dm_target *ti; | |
1186 | ||
1187 | while ((ti = dm_table_get_target(ci->map, target_nr++))) | |
1188 | for (flush_nr = 0; flush_nr < ti->num_flush_requests; | |
1189 | flush_nr++) | |
1190 | __flush_target(ci, ti, flush_nr); | |
1191 | ||
1192 | ci->sector_count = 0; | |
1193 | ||
1194 | return 0; | |
1195 | } | |
1196 | ||
512875bd | 1197 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1198 | { |
1199 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1200 | struct dm_target *ti; |
1201 | sector_t len = 0, max; | |
028867ac | 1202 | struct dm_target_io *tio; |
1da177e4 | 1203 | |
f9ab94ce MP |
1204 | if (unlikely(bio_empty_barrier(bio))) |
1205 | return __clone_and_map_empty_barrier(ci); | |
1206 | ||
512875bd JN |
1207 | ti = dm_table_find_target(ci->map, ci->sector); |
1208 | if (!dm_target_is_valid(ti)) | |
1209 | return -EIO; | |
1210 | ||
1211 | max = max_io_len(ci->md, ci->sector, ti); | |
1212 | ||
1da177e4 LT |
1213 | /* |
1214 | * Allocate a target io object. | |
1215 | */ | |
9015df24 | 1216 | tio = alloc_tio(ci, ti); |
1da177e4 LT |
1217 | |
1218 | if (ci->sector_count <= max) { | |
1219 | /* | |
1220 | * Optimise for the simple case where we can do all of | |
1221 | * the remaining io with a single clone. | |
1222 | */ | |
1223 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
1224 | bio->bi_vcnt - ci->idx, ci->sector_count, |
1225 | ci->md->bs); | |
1da177e4 LT |
1226 | __map_bio(ti, clone, tio); |
1227 | ci->sector_count = 0; | |
1228 | ||
1229 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1230 | /* | |
1231 | * There are some bvecs that don't span targets. | |
1232 | * Do as many of these as possible. | |
1233 | */ | |
1234 | int i; | |
1235 | sector_t remaining = max; | |
1236 | sector_t bv_len; | |
1237 | ||
1238 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1239 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1240 | ||
1241 | if (bv_len > remaining) | |
1242 | break; | |
1243 | ||
1244 | remaining -= bv_len; | |
1245 | len += bv_len; | |
1246 | } | |
1247 | ||
9faf400f SB |
1248 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1249 | ci->md->bs); | |
1da177e4 LT |
1250 | __map_bio(ti, clone, tio); |
1251 | ||
1252 | ci->sector += len; | |
1253 | ci->sector_count -= len; | |
1254 | ci->idx = i; | |
1255 | ||
1256 | } else { | |
1257 | /* | |
d2044a94 | 1258 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1259 | */ |
1260 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1261 | sector_t remaining = to_sector(bv->bv_len); |
1262 | unsigned int offset = 0; | |
1da177e4 | 1263 | |
d2044a94 AK |
1264 | do { |
1265 | if (offset) { | |
1266 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1267 | if (!dm_target_is_valid(ti)) |
1268 | return -EIO; | |
1269 | ||
d2044a94 | 1270 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 1271 | |
9015df24 | 1272 | tio = alloc_tio(ci, ti); |
d2044a94 AK |
1273 | } |
1274 | ||
1275 | len = min(remaining, max); | |
1276 | ||
1277 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
1278 | bv->bv_offset + offset, len, |
1279 | ci->md->bs); | |
d2044a94 AK |
1280 | |
1281 | __map_bio(ti, clone, tio); | |
1282 | ||
1283 | ci->sector += len; | |
1284 | ci->sector_count -= len; | |
1285 | offset += to_bytes(len); | |
1286 | } while (remaining -= len); | |
1da177e4 | 1287 | |
1da177e4 LT |
1288 | ci->idx++; |
1289 | } | |
512875bd JN |
1290 | |
1291 | return 0; | |
1da177e4 LT |
1292 | } |
1293 | ||
1294 | /* | |
8a53c28d | 1295 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1296 | */ |
f0b9a450 | 1297 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1298 | { |
1299 | struct clone_info ci; | |
512875bd | 1300 | int error = 0; |
1da177e4 | 1301 | |
7c666411 | 1302 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1303 | if (unlikely(!ci.map)) { |
1f98a13f | 1304 | if (!bio_rw_flagged(bio, BIO_RW_BARRIER)) |
af7e466a MP |
1305 | bio_io_error(bio); |
1306 | else | |
5aa2781d MP |
1307 | if (!md->barrier_error) |
1308 | md->barrier_error = -EIO; | |
f0b9a450 MP |
1309 | return; |
1310 | } | |
692d0eb9 | 1311 | |
1da177e4 LT |
1312 | ci.md = md; |
1313 | ci.bio = bio; | |
1314 | ci.io = alloc_io(md); | |
1315 | ci.io->error = 0; | |
1316 | atomic_set(&ci.io->io_count, 1); | |
1317 | ci.io->bio = bio; | |
1318 | ci.io->md = md; | |
f88fb981 | 1319 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 LT |
1320 | ci.sector = bio->bi_sector; |
1321 | ci.sector_count = bio_sectors(bio); | |
f9ab94ce MP |
1322 | if (unlikely(bio_empty_barrier(bio))) |
1323 | ci.sector_count = 1; | |
1da177e4 LT |
1324 | ci.idx = bio->bi_idx; |
1325 | ||
3eaf840e | 1326 | start_io_acct(ci.io); |
512875bd JN |
1327 | while (ci.sector_count && !error) |
1328 | error = __clone_and_map(&ci); | |
1da177e4 LT |
1329 | |
1330 | /* drop the extra reference count */ | |
512875bd | 1331 | dec_pending(ci.io, error); |
1da177e4 LT |
1332 | dm_table_put(ci.map); |
1333 | } | |
1334 | /*----------------------------------------------------------------- | |
1335 | * CRUD END | |
1336 | *---------------------------------------------------------------*/ | |
1337 | ||
f6fccb12 MB |
1338 | static int dm_merge_bvec(struct request_queue *q, |
1339 | struct bvec_merge_data *bvm, | |
1340 | struct bio_vec *biovec) | |
1341 | { | |
1342 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1343 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1344 | struct dm_target *ti; |
1345 | sector_t max_sectors; | |
5037108a | 1346 | int max_size = 0; |
f6fccb12 MB |
1347 | |
1348 | if (unlikely(!map)) | |
5037108a | 1349 | goto out; |
f6fccb12 MB |
1350 | |
1351 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1352 | if (!dm_target_is_valid(ti)) |
1353 | goto out_table; | |
f6fccb12 MB |
1354 | |
1355 | /* | |
1356 | * Find maximum amount of I/O that won't need splitting | |
1357 | */ | |
1358 | max_sectors = min(max_io_len(md, bvm->bi_sector, ti), | |
1359 | (sector_t) BIO_MAX_SECTORS); | |
1360 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1361 | if (max_size < 0) | |
1362 | max_size = 0; | |
1363 | ||
1364 | /* | |
1365 | * merge_bvec_fn() returns number of bytes | |
1366 | * it can accept at this offset | |
1367 | * max is precomputed maximal io size | |
1368 | */ | |
1369 | if (max_size && ti->type->merge) | |
1370 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1371 | /* |
1372 | * If the target doesn't support merge method and some of the devices | |
1373 | * provided their merge_bvec method (we know this by looking at | |
1374 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1375 | * entries. So always set max_size to 0, and the code below allows | |
1376 | * just one page. | |
1377 | */ | |
1378 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1379 | ||
1380 | max_size = 0; | |
f6fccb12 | 1381 | |
b01cd5ac | 1382 | out_table: |
5037108a MP |
1383 | dm_table_put(map); |
1384 | ||
1385 | out: | |
f6fccb12 MB |
1386 | /* |
1387 | * Always allow an entire first page | |
1388 | */ | |
1389 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1390 | max_size = biovec->bv_len; | |
1391 | ||
f6fccb12 MB |
1392 | return max_size; |
1393 | } | |
1394 | ||
1da177e4 LT |
1395 | /* |
1396 | * The request function that just remaps the bio built up by | |
1397 | * dm_merge_bvec. | |
1398 | */ | |
cec47e3d | 1399 | static int _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1400 | { |
12f03a49 | 1401 | int rw = bio_data_dir(bio); |
1da177e4 | 1402 | struct mapped_device *md = q->queuedata; |
c9959059 | 1403 | int cpu; |
1da177e4 | 1404 | |
2ca3310e | 1405 | down_read(&md->io_lock); |
1da177e4 | 1406 | |
074a7aca TH |
1407 | cpu = part_stat_lock(); |
1408 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1409 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1410 | part_stat_unlock(); | |
12f03a49 | 1411 | |
1da177e4 | 1412 | /* |
1eb787ec AK |
1413 | * If we're suspended or the thread is processing barriers |
1414 | * we have to queue this io for later. | |
1da177e4 | 1415 | */ |
af7e466a | 1416 | if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || |
1f98a13f | 1417 | unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) { |
2ca3310e | 1418 | up_read(&md->io_lock); |
1da177e4 | 1419 | |
54d9a1b4 AK |
1420 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && |
1421 | bio_rw(bio) == READA) { | |
1422 | bio_io_error(bio); | |
1423 | return 0; | |
1424 | } | |
1da177e4 | 1425 | |
92c63902 | 1426 | queue_io(md, bio); |
1da177e4 | 1427 | |
92c63902 | 1428 | return 0; |
1da177e4 LT |
1429 | } |
1430 | ||
f0b9a450 | 1431 | __split_and_process_bio(md, bio); |
2ca3310e | 1432 | up_read(&md->io_lock); |
f0b9a450 | 1433 | return 0; |
1da177e4 LT |
1434 | } |
1435 | ||
cec47e3d KU |
1436 | static int dm_make_request(struct request_queue *q, struct bio *bio) |
1437 | { | |
1438 | struct mapped_device *md = q->queuedata; | |
1439 | ||
cec47e3d KU |
1440 | return md->saved_make_request_fn(q, bio); /* call __make_request() */ |
1441 | } | |
1442 | ||
1443 | static int dm_request_based(struct mapped_device *md) | |
1444 | { | |
1445 | return blk_queue_stackable(md->queue); | |
1446 | } | |
1447 | ||
1448 | static int dm_request(struct request_queue *q, struct bio *bio) | |
1449 | { | |
1450 | struct mapped_device *md = q->queuedata; | |
1451 | ||
1452 | if (dm_request_based(md)) | |
1453 | return dm_make_request(q, bio); | |
1454 | ||
1455 | return _dm_request(q, bio); | |
1456 | } | |
1457 | ||
d0bcb878 KU |
1458 | /* |
1459 | * Mark this request as flush request, so that dm_request_fn() can | |
1460 | * recognize. | |
1461 | */ | |
1462 | static void dm_rq_prepare_flush(struct request_queue *q, struct request *rq) | |
1463 | { | |
1464 | rq->cmd_type = REQ_TYPE_LINUX_BLOCK; | |
1465 | rq->cmd[0] = REQ_LB_OP_FLUSH; | |
1466 | } | |
1467 | ||
1468 | static bool dm_rq_is_flush_request(struct request *rq) | |
1469 | { | |
1470 | if (rq->cmd_type == REQ_TYPE_LINUX_BLOCK && | |
1471 | rq->cmd[0] == REQ_LB_OP_FLUSH) | |
1472 | return true; | |
1473 | else | |
1474 | return false; | |
1475 | } | |
1476 | ||
cec47e3d KU |
1477 | void dm_dispatch_request(struct request *rq) |
1478 | { | |
1479 | int r; | |
1480 | ||
1481 | if (blk_queue_io_stat(rq->q)) | |
1482 | rq->cmd_flags |= REQ_IO_STAT; | |
1483 | ||
1484 | rq->start_time = jiffies; | |
1485 | r = blk_insert_cloned_request(rq->q, rq); | |
1486 | if (r) | |
1487 | dm_complete_request(rq, r); | |
1488 | } | |
1489 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1490 | ||
1491 | static void dm_rq_bio_destructor(struct bio *bio) | |
1492 | { | |
1493 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1494 | struct mapped_device *md = info->tio->md; | |
1495 | ||
1496 | free_bio_info(info); | |
1497 | bio_free(bio, md->bs); | |
1498 | } | |
1499 | ||
1500 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1501 | void *data) | |
1502 | { | |
1503 | struct dm_rq_target_io *tio = data; | |
1504 | struct mapped_device *md = tio->md; | |
1505 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1506 | ||
1507 | if (!info) | |
1508 | return -ENOMEM; | |
1509 | ||
1510 | info->orig = bio_orig; | |
1511 | info->tio = tio; | |
1512 | bio->bi_end_io = end_clone_bio; | |
1513 | bio->bi_private = info; | |
1514 | bio->bi_destructor = dm_rq_bio_destructor; | |
1515 | ||
1516 | return 0; | |
1517 | } | |
1518 | ||
1519 | static int setup_clone(struct request *clone, struct request *rq, | |
1520 | struct dm_rq_target_io *tio) | |
1521 | { | |
d0bcb878 | 1522 | int r; |
cec47e3d | 1523 | |
d0bcb878 KU |
1524 | if (dm_rq_is_flush_request(rq)) { |
1525 | blk_rq_init(NULL, clone); | |
1526 | clone->cmd_type = REQ_TYPE_FS; | |
1527 | clone->cmd_flags |= (REQ_HARDBARRIER | WRITE); | |
1528 | } else { | |
1529 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, | |
1530 | dm_rq_bio_constructor, tio); | |
1531 | if (r) | |
1532 | return r; | |
1533 | ||
1534 | clone->cmd = rq->cmd; | |
1535 | clone->cmd_len = rq->cmd_len; | |
1536 | clone->sense = rq->sense; | |
1537 | clone->buffer = rq->buffer; | |
1538 | } | |
cec47e3d | 1539 | |
cec47e3d KU |
1540 | clone->end_io = end_clone_request; |
1541 | clone->end_io_data = tio; | |
1542 | ||
1543 | return 0; | |
1544 | } | |
1545 | ||
6facdaff KU |
1546 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1547 | gfp_t gfp_mask) | |
1548 | { | |
1549 | struct request *clone; | |
1550 | struct dm_rq_target_io *tio; | |
1551 | ||
1552 | tio = alloc_rq_tio(md, gfp_mask); | |
1553 | if (!tio) | |
1554 | return NULL; | |
1555 | ||
1556 | tio->md = md; | |
1557 | tio->ti = NULL; | |
1558 | tio->orig = rq; | |
1559 | tio->error = 0; | |
1560 | memset(&tio->info, 0, sizeof(tio->info)); | |
1561 | ||
1562 | clone = &tio->clone; | |
1563 | if (setup_clone(clone, rq, tio)) { | |
1564 | /* -ENOMEM */ | |
1565 | free_rq_tio(tio); | |
1566 | return NULL; | |
1567 | } | |
1568 | ||
1569 | return clone; | |
1570 | } | |
1571 | ||
cec47e3d KU |
1572 | /* |
1573 | * Called with the queue lock held. | |
1574 | */ | |
1575 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1576 | { | |
1577 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1578 | struct request *clone; |
1579 | ||
d0bcb878 KU |
1580 | if (unlikely(dm_rq_is_flush_request(rq))) |
1581 | return BLKPREP_OK; | |
1582 | ||
cec47e3d KU |
1583 | if (unlikely(rq->special)) { |
1584 | DMWARN("Already has something in rq->special."); | |
1585 | return BLKPREP_KILL; | |
1586 | } | |
1587 | ||
6facdaff KU |
1588 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1589 | if (!clone) | |
cec47e3d | 1590 | return BLKPREP_DEFER; |
cec47e3d KU |
1591 | |
1592 | rq->special = clone; | |
1593 | rq->cmd_flags |= REQ_DONTPREP; | |
1594 | ||
1595 | return BLKPREP_OK; | |
1596 | } | |
1597 | ||
9eef87da KU |
1598 | /* |
1599 | * Returns: | |
1600 | * 0 : the request has been processed (not requeued) | |
1601 | * !0 : the request has been requeued | |
1602 | */ | |
1603 | static int map_request(struct dm_target *ti, struct request *clone, | |
1604 | struct mapped_device *md) | |
cec47e3d | 1605 | { |
9eef87da | 1606 | int r, requeued = 0; |
cec47e3d KU |
1607 | struct dm_rq_target_io *tio = clone->end_io_data; |
1608 | ||
1609 | /* | |
1610 | * Hold the md reference here for the in-flight I/O. | |
1611 | * We can't rely on the reference count by device opener, | |
1612 | * because the device may be closed during the request completion | |
1613 | * when all bios are completed. | |
1614 | * See the comment in rq_completed() too. | |
1615 | */ | |
1616 | dm_get(md); | |
1617 | ||
1618 | tio->ti = ti; | |
1619 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1620 | switch (r) { | |
1621 | case DM_MAPIO_SUBMITTED: | |
1622 | /* The target has taken the I/O to submit by itself later */ | |
1623 | break; | |
1624 | case DM_MAPIO_REMAPPED: | |
1625 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1626 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1627 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1628 | dm_dispatch_request(clone); |
1629 | break; | |
1630 | case DM_MAPIO_REQUEUE: | |
1631 | /* The target wants to requeue the I/O */ | |
1632 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1633 | requeued = 1; |
cec47e3d KU |
1634 | break; |
1635 | default: | |
1636 | if (r > 0) { | |
1637 | DMWARN("unimplemented target map return value: %d", r); | |
1638 | BUG(); | |
1639 | } | |
1640 | ||
1641 | /* The target wants to complete the I/O */ | |
1642 | dm_kill_unmapped_request(clone, r); | |
1643 | break; | |
1644 | } | |
9eef87da KU |
1645 | |
1646 | return requeued; | |
cec47e3d KU |
1647 | } |
1648 | ||
1649 | /* | |
1650 | * q->request_fn for request-based dm. | |
1651 | * Called with the queue lock held. | |
1652 | */ | |
1653 | static void dm_request_fn(struct request_queue *q) | |
1654 | { | |
1655 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1656 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1657 | struct dm_target *ti; |
b4324fee | 1658 | struct request *rq, *clone; |
cec47e3d KU |
1659 | |
1660 | /* | |
b4324fee KU |
1661 | * For suspend, check blk_queue_stopped() and increment |
1662 | * ->pending within a single queue_lock not to increment the | |
1663 | * number of in-flight I/Os after the queue is stopped in | |
1664 | * dm_suspend(). | |
cec47e3d KU |
1665 | */ |
1666 | while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) { | |
1667 | rq = blk_peek_request(q); | |
1668 | if (!rq) | |
1669 | goto plug_and_out; | |
1670 | ||
d0bcb878 KU |
1671 | if (unlikely(dm_rq_is_flush_request(rq))) { |
1672 | BUG_ON(md->flush_request); | |
1673 | md->flush_request = rq; | |
1674 | blk_start_request(rq); | |
1675 | queue_work(md->wq, &md->barrier_work); | |
1676 | goto out; | |
1677 | } | |
1678 | ||
cec47e3d KU |
1679 | ti = dm_table_find_target(map, blk_rq_pos(rq)); |
1680 | if (ti->type->busy && ti->type->busy(ti)) | |
1681 | goto plug_and_out; | |
1682 | ||
1683 | blk_start_request(rq); | |
b4324fee KU |
1684 | clone = rq->special; |
1685 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1686 | ||
cec47e3d | 1687 | spin_unlock(q->queue_lock); |
9eef87da KU |
1688 | if (map_request(ti, clone, md)) |
1689 | goto requeued; | |
1690 | ||
cec47e3d KU |
1691 | spin_lock_irq(q->queue_lock); |
1692 | } | |
1693 | ||
1694 | goto out; | |
1695 | ||
9eef87da KU |
1696 | requeued: |
1697 | spin_lock_irq(q->queue_lock); | |
1698 | ||
cec47e3d KU |
1699 | plug_and_out: |
1700 | if (!elv_queue_empty(q)) | |
1701 | /* Some requests still remain, retry later */ | |
1702 | blk_plug_device(q); | |
1703 | ||
1704 | out: | |
1705 | dm_table_put(map); | |
1706 | ||
1707 | return; | |
1708 | } | |
1709 | ||
1710 | int dm_underlying_device_busy(struct request_queue *q) | |
1711 | { | |
1712 | return blk_lld_busy(q); | |
1713 | } | |
1714 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1715 | ||
1716 | static int dm_lld_busy(struct request_queue *q) | |
1717 | { | |
1718 | int r; | |
1719 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1720 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1721 | |
1722 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1723 | r = 1; | |
1724 | else | |
1725 | r = dm_table_any_busy_target(map); | |
1726 | ||
1727 | dm_table_put(map); | |
1728 | ||
1729 | return r; | |
1730 | } | |
1731 | ||
165125e1 | 1732 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
1733 | { |
1734 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1735 | struct dm_table *map = dm_get_live_table(md); |
1da177e4 LT |
1736 | |
1737 | if (map) { | |
cec47e3d KU |
1738 | if (dm_request_based(md)) |
1739 | generic_unplug_device(q); | |
1740 | ||
1da177e4 LT |
1741 | dm_table_unplug_all(map); |
1742 | dm_table_put(map); | |
1743 | } | |
1744 | } | |
1745 | ||
1746 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
1747 | { | |
8a57dfc6 CS |
1748 | int r = bdi_bits; |
1749 | struct mapped_device *md = congested_data; | |
1750 | struct dm_table *map; | |
1da177e4 | 1751 | |
1eb787ec | 1752 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1753 | map = dm_get_live_table(md); |
8a57dfc6 | 1754 | if (map) { |
cec47e3d KU |
1755 | /* |
1756 | * Request-based dm cares about only own queue for | |
1757 | * the query about congestion status of request_queue | |
1758 | */ | |
1759 | if (dm_request_based(md)) | |
1760 | r = md->queue->backing_dev_info.state & | |
1761 | bdi_bits; | |
1762 | else | |
1763 | r = dm_table_any_congested(map, bdi_bits); | |
1764 | ||
8a57dfc6 CS |
1765 | dm_table_put(map); |
1766 | } | |
1767 | } | |
1768 | ||
1da177e4 LT |
1769 | return r; |
1770 | } | |
1771 | ||
1772 | /*----------------------------------------------------------------- | |
1773 | * An IDR is used to keep track of allocated minor numbers. | |
1774 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1775 | static DEFINE_IDR(_minor_idr); |
1776 | ||
2b06cfff | 1777 | static void free_minor(int minor) |
1da177e4 | 1778 | { |
f32c10b0 | 1779 | spin_lock(&_minor_lock); |
1da177e4 | 1780 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1781 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1782 | } |
1783 | ||
1784 | /* | |
1785 | * See if the device with a specific minor # is free. | |
1786 | */ | |
cf13ab8e | 1787 | static int specific_minor(int minor) |
1da177e4 LT |
1788 | { |
1789 | int r, m; | |
1790 | ||
1791 | if (minor >= (1 << MINORBITS)) | |
1792 | return -EINVAL; | |
1793 | ||
62f75c2f JM |
1794 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1795 | if (!r) | |
1796 | return -ENOMEM; | |
1797 | ||
f32c10b0 | 1798 | spin_lock(&_minor_lock); |
1da177e4 LT |
1799 | |
1800 | if (idr_find(&_minor_idr, minor)) { | |
1801 | r = -EBUSY; | |
1802 | goto out; | |
1803 | } | |
1804 | ||
ba61fdd1 | 1805 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1806 | if (r) |
1da177e4 | 1807 | goto out; |
1da177e4 LT |
1808 | |
1809 | if (m != minor) { | |
1810 | idr_remove(&_minor_idr, m); | |
1811 | r = -EBUSY; | |
1812 | goto out; | |
1813 | } | |
1814 | ||
1815 | out: | |
f32c10b0 | 1816 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1817 | return r; |
1818 | } | |
1819 | ||
cf13ab8e | 1820 | static int next_free_minor(int *minor) |
1da177e4 | 1821 | { |
2b06cfff | 1822 | int r, m; |
1da177e4 | 1823 | |
1da177e4 | 1824 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1825 | if (!r) |
1826 | return -ENOMEM; | |
1827 | ||
f32c10b0 | 1828 | spin_lock(&_minor_lock); |
1da177e4 | 1829 | |
ba61fdd1 | 1830 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1831 | if (r) |
1da177e4 | 1832 | goto out; |
1da177e4 LT |
1833 | |
1834 | if (m >= (1 << MINORBITS)) { | |
1835 | idr_remove(&_minor_idr, m); | |
1836 | r = -ENOSPC; | |
1837 | goto out; | |
1838 | } | |
1839 | ||
1840 | *minor = m; | |
1841 | ||
1842 | out: | |
f32c10b0 | 1843 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1844 | return r; |
1845 | } | |
1846 | ||
83d5cde4 | 1847 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1848 | |
53d5914f | 1849 | static void dm_wq_work(struct work_struct *work); |
d0bcb878 | 1850 | static void dm_rq_barrier_work(struct work_struct *work); |
53d5914f | 1851 | |
1da177e4 LT |
1852 | /* |
1853 | * Allocate and initialise a blank device with a given minor. | |
1854 | */ | |
2b06cfff | 1855 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1856 | { |
1857 | int r; | |
cf13ab8e | 1858 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1859 | void *old_md; |
1da177e4 LT |
1860 | |
1861 | if (!md) { | |
1862 | DMWARN("unable to allocate device, out of memory."); | |
1863 | return NULL; | |
1864 | } | |
1865 | ||
10da4f79 | 1866 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1867 | goto bad_module_get; |
10da4f79 | 1868 | |
1da177e4 | 1869 | /* get a minor number for the dev */ |
2b06cfff | 1870 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1871 | r = next_free_minor(&minor); |
2b06cfff | 1872 | else |
cf13ab8e | 1873 | r = specific_minor(minor); |
1da177e4 | 1874 | if (r < 0) |
6ed7ade8 | 1875 | goto bad_minor; |
1da177e4 | 1876 | |
2ca3310e | 1877 | init_rwsem(&md->io_lock); |
e61290a4 | 1878 | mutex_init(&md->suspend_lock); |
022c2611 | 1879 | spin_lock_init(&md->deferred_lock); |
d0bcb878 | 1880 | spin_lock_init(&md->barrier_error_lock); |
1da177e4 LT |
1881 | rwlock_init(&md->map_lock); |
1882 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1883 | atomic_set(&md->open_count, 0); |
1da177e4 | 1884 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1885 | atomic_set(&md->uevent_seq, 0); |
1886 | INIT_LIST_HEAD(&md->uevent_list); | |
1887 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1888 | |
e6ee8c0b | 1889 | md->queue = blk_init_queue(dm_request_fn, NULL); |
1da177e4 | 1890 | if (!md->queue) |
6ed7ade8 | 1891 | goto bad_queue; |
1da177e4 | 1892 | |
e6ee8c0b KU |
1893 | /* |
1894 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1895 | * devices. The type of this dm device has not been decided yet, | |
1896 | * although we initialized the queue using blk_init_queue(). | |
1897 | * The type is decided at the first table loading time. | |
1898 | * To prevent problematic device stacking, clear the queue flag | |
1899 | * for request stacking support until then. | |
1900 | * | |
1901 | * This queue is new, so no concurrency on the queue_flags. | |
1902 | */ | |
1903 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1904 | md->saved_make_request_fn = md->queue->make_request_fn; | |
1da177e4 LT |
1905 | md->queue->queuedata = md; |
1906 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1907 | md->queue->backing_dev_info.congested_data = md; | |
1908 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1909 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1910 | md->queue->unplug_fn = dm_unplug_all; |
f6fccb12 | 1911 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
e6ee8c0b KU |
1912 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
1913 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
1914 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
d0bcb878 KU |
1915 | blk_queue_ordered(md->queue, QUEUE_ORDERED_DRAIN_FLUSH, |
1916 | dm_rq_prepare_flush); | |
9faf400f | 1917 | |
1da177e4 LT |
1918 | md->disk = alloc_disk(1); |
1919 | if (!md->disk) | |
6ed7ade8 | 1920 | goto bad_disk; |
1da177e4 | 1921 | |
316d315b NK |
1922 | atomic_set(&md->pending[0], 0); |
1923 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1924 | init_waitqueue_head(&md->wait); |
53d5914f | 1925 | INIT_WORK(&md->work, dm_wq_work); |
d0bcb878 | 1926 | INIT_WORK(&md->barrier_work, dm_rq_barrier_work); |
f0b04115 JM |
1927 | init_waitqueue_head(&md->eventq); |
1928 | ||
1da177e4 LT |
1929 | md->disk->major = _major; |
1930 | md->disk->first_minor = minor; | |
1931 | md->disk->fops = &dm_blk_dops; | |
1932 | md->disk->queue = md->queue; | |
1933 | md->disk->private_data = md; | |
1934 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1935 | add_disk(md->disk); | |
7e51f257 | 1936 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1937 | |
304f3f6a MB |
1938 | md->wq = create_singlethread_workqueue("kdmflush"); |
1939 | if (!md->wq) | |
1940 | goto bad_thread; | |
1941 | ||
32a926da MP |
1942 | md->bdev = bdget_disk(md->disk, 0); |
1943 | if (!md->bdev) | |
1944 | goto bad_bdev; | |
1945 | ||
ba61fdd1 | 1946 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1947 | spin_lock(&_minor_lock); |
ba61fdd1 | 1948 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1949 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1950 | |
1951 | BUG_ON(old_md != MINOR_ALLOCED); | |
1952 | ||
1da177e4 LT |
1953 | return md; |
1954 | ||
32a926da MP |
1955 | bad_bdev: |
1956 | destroy_workqueue(md->wq); | |
304f3f6a | 1957 | bad_thread: |
03022c54 | 1958 | del_gendisk(md->disk); |
304f3f6a | 1959 | put_disk(md->disk); |
6ed7ade8 | 1960 | bad_disk: |
1312f40e | 1961 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1962 | bad_queue: |
1da177e4 | 1963 | free_minor(minor); |
6ed7ade8 | 1964 | bad_minor: |
10da4f79 | 1965 | module_put(THIS_MODULE); |
6ed7ade8 | 1966 | bad_module_get: |
1da177e4 LT |
1967 | kfree(md); |
1968 | return NULL; | |
1969 | } | |
1970 | ||
ae9da83f JN |
1971 | static void unlock_fs(struct mapped_device *md); |
1972 | ||
1da177e4 LT |
1973 | static void free_dev(struct mapped_device *md) |
1974 | { | |
f331c029 | 1975 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1976 | |
32a926da MP |
1977 | unlock_fs(md); |
1978 | bdput(md->bdev); | |
304f3f6a | 1979 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1980 | if (md->tio_pool) |
1981 | mempool_destroy(md->tio_pool); | |
1982 | if (md->io_pool) | |
1983 | mempool_destroy(md->io_pool); | |
1984 | if (md->bs) | |
1985 | bioset_free(md->bs); | |
9c47008d | 1986 | blk_integrity_unregister(md->disk); |
1da177e4 | 1987 | del_gendisk(md->disk); |
63d94e48 | 1988 | free_minor(minor); |
fba9f90e JM |
1989 | |
1990 | spin_lock(&_minor_lock); | |
1991 | md->disk->private_data = NULL; | |
1992 | spin_unlock(&_minor_lock); | |
1993 | ||
1da177e4 | 1994 | put_disk(md->disk); |
1312f40e | 1995 | blk_cleanup_queue(md->queue); |
10da4f79 | 1996 | module_put(THIS_MODULE); |
1da177e4 LT |
1997 | kfree(md); |
1998 | } | |
1999 | ||
e6ee8c0b KU |
2000 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2001 | { | |
2002 | struct dm_md_mempools *p; | |
2003 | ||
2004 | if (md->io_pool && md->tio_pool && md->bs) | |
2005 | /* the md already has necessary mempools */ | |
2006 | goto out; | |
2007 | ||
2008 | p = dm_table_get_md_mempools(t); | |
2009 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
2010 | ||
2011 | md->io_pool = p->io_pool; | |
2012 | p->io_pool = NULL; | |
2013 | md->tio_pool = p->tio_pool; | |
2014 | p->tio_pool = NULL; | |
2015 | md->bs = p->bs; | |
2016 | p->bs = NULL; | |
2017 | ||
2018 | out: | |
2019 | /* mempool bind completed, now no need any mempools in the table */ | |
2020 | dm_table_free_md_mempools(t); | |
2021 | } | |
2022 | ||
1da177e4 LT |
2023 | /* |
2024 | * Bind a table to the device. | |
2025 | */ | |
2026 | static void event_callback(void *context) | |
2027 | { | |
7a8c3d3b MA |
2028 | unsigned long flags; |
2029 | LIST_HEAD(uevents); | |
1da177e4 LT |
2030 | struct mapped_device *md = (struct mapped_device *) context; |
2031 | ||
7a8c3d3b MA |
2032 | spin_lock_irqsave(&md->uevent_lock, flags); |
2033 | list_splice_init(&md->uevent_list, &uevents); | |
2034 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2035 | ||
ed9e1982 | 2036 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2037 | |
1da177e4 LT |
2038 | atomic_inc(&md->event_nr); |
2039 | wake_up(&md->eventq); | |
2040 | } | |
2041 | ||
4e90188b | 2042 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2043 | { |
4e90188b | 2044 | set_capacity(md->disk, size); |
1da177e4 | 2045 | |
db8fef4f MP |
2046 | mutex_lock(&md->bdev->bd_inode->i_mutex); |
2047 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
2048 | mutex_unlock(&md->bdev->bd_inode->i_mutex); | |
1da177e4 LT |
2049 | } |
2050 | ||
042d2a9b AK |
2051 | /* |
2052 | * Returns old map, which caller must destroy. | |
2053 | */ | |
2054 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2055 | struct queue_limits *limits) | |
1da177e4 | 2056 | { |
042d2a9b | 2057 | struct dm_table *old_map; |
165125e1 | 2058 | struct request_queue *q = md->queue; |
1da177e4 | 2059 | sector_t size; |
523d9297 | 2060 | unsigned long flags; |
1da177e4 LT |
2061 | |
2062 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2063 | |
2064 | /* | |
2065 | * Wipe any geometry if the size of the table changed. | |
2066 | */ | |
2067 | if (size != get_capacity(md->disk)) | |
2068 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2069 | ||
32a926da | 2070 | __set_size(md, size); |
d5816876 | 2071 | |
2ca3310e AK |
2072 | dm_table_event_callback(t, event_callback, md); |
2073 | ||
e6ee8c0b KU |
2074 | /* |
2075 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2076 | * for request-based during suspension. So stop it to prevent | |
2077 | * I/O mapping before resume. | |
2078 | * This must be done before setting the queue restrictions, | |
2079 | * because request-based dm may be run just after the setting. | |
2080 | */ | |
2081 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2082 | stop_queue(q); | |
2083 | ||
2084 | __bind_mempools(md, t); | |
2085 | ||
523d9297 | 2086 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2087 | old_map = md->map; |
1da177e4 | 2088 | md->map = t; |
754c5fc7 | 2089 | dm_table_set_restrictions(t, q, limits); |
523d9297 | 2090 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2091 | |
042d2a9b | 2092 | return old_map; |
1da177e4 LT |
2093 | } |
2094 | ||
a7940155 AK |
2095 | /* |
2096 | * Returns unbound table for the caller to free. | |
2097 | */ | |
2098 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2099 | { |
2100 | struct dm_table *map = md->map; | |
523d9297 | 2101 | unsigned long flags; |
1da177e4 LT |
2102 | |
2103 | if (!map) | |
a7940155 | 2104 | return NULL; |
1da177e4 LT |
2105 | |
2106 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2107 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2108 | md->map = NULL; |
523d9297 | 2109 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2110 | |
2111 | return map; | |
1da177e4 LT |
2112 | } |
2113 | ||
2114 | /* | |
2115 | * Constructor for a new device. | |
2116 | */ | |
2b06cfff | 2117 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2118 | { |
2119 | struct mapped_device *md; | |
2120 | ||
2b06cfff | 2121 | md = alloc_dev(minor); |
1da177e4 LT |
2122 | if (!md) |
2123 | return -ENXIO; | |
2124 | ||
784aae73 MB |
2125 | dm_sysfs_init(md); |
2126 | ||
1da177e4 LT |
2127 | *result = md; |
2128 | return 0; | |
2129 | } | |
2130 | ||
637842cf | 2131 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2132 | { |
2133 | struct mapped_device *md; | |
1da177e4 LT |
2134 | unsigned minor = MINOR(dev); |
2135 | ||
2136 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2137 | return NULL; | |
2138 | ||
f32c10b0 | 2139 | spin_lock(&_minor_lock); |
1da177e4 LT |
2140 | |
2141 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2142 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2143 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
17b2f66f | 2144 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2145 | md = NULL; |
fba9f90e JM |
2146 | goto out; |
2147 | } | |
1da177e4 | 2148 | |
fba9f90e | 2149 | out: |
f32c10b0 | 2150 | spin_unlock(&_minor_lock); |
1da177e4 | 2151 | |
637842cf DT |
2152 | return md; |
2153 | } | |
2154 | ||
d229a958 DT |
2155 | struct mapped_device *dm_get_md(dev_t dev) |
2156 | { | |
2157 | struct mapped_device *md = dm_find_md(dev); | |
2158 | ||
2159 | if (md) | |
2160 | dm_get(md); | |
2161 | ||
2162 | return md; | |
2163 | } | |
2164 | ||
9ade92a9 | 2165 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2166 | { |
9ade92a9 | 2167 | return md->interface_ptr; |
1da177e4 LT |
2168 | } |
2169 | ||
2170 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2171 | { | |
2172 | md->interface_ptr = ptr; | |
2173 | } | |
2174 | ||
2175 | void dm_get(struct mapped_device *md) | |
2176 | { | |
2177 | atomic_inc(&md->holders); | |
2178 | } | |
2179 | ||
72d94861 AK |
2180 | const char *dm_device_name(struct mapped_device *md) |
2181 | { | |
2182 | return md->name; | |
2183 | } | |
2184 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2185 | ||
1da177e4 LT |
2186 | void dm_put(struct mapped_device *md) |
2187 | { | |
1134e5ae | 2188 | struct dm_table *map; |
1da177e4 | 2189 | |
fba9f90e JM |
2190 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
2191 | ||
f32c10b0 | 2192 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
7c666411 | 2193 | map = dm_get_live_table(md); |
f331c029 TH |
2194 | idr_replace(&_minor_idr, MINOR_ALLOCED, |
2195 | MINOR(disk_devt(dm_disk(md)))); | |
fba9f90e | 2196 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 2197 | spin_unlock(&_minor_lock); |
4f186f8b | 2198 | if (!dm_suspended_md(md)) { |
1da177e4 LT |
2199 | dm_table_presuspend_targets(map); |
2200 | dm_table_postsuspend_targets(map); | |
2201 | } | |
784aae73 | 2202 | dm_sysfs_exit(md); |
1134e5ae | 2203 | dm_table_put(map); |
a7940155 | 2204 | dm_table_destroy(__unbind(md)); |
1da177e4 LT |
2205 | free_dev(md); |
2206 | } | |
1da177e4 | 2207 | } |
79eb885c | 2208 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2209 | |
401600df | 2210 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2211 | { |
2212 | int r = 0; | |
b44ebeb0 MP |
2213 | DECLARE_WAITQUEUE(wait, current); |
2214 | ||
2215 | dm_unplug_all(md->queue); | |
2216 | ||
2217 | add_wait_queue(&md->wait, &wait); | |
46125c1c MB |
2218 | |
2219 | while (1) { | |
401600df | 2220 | set_current_state(interruptible); |
46125c1c MB |
2221 | |
2222 | smp_mb(); | |
b4324fee | 2223 | if (!md_in_flight(md)) |
46125c1c MB |
2224 | break; |
2225 | ||
401600df MP |
2226 | if (interruptible == TASK_INTERRUPTIBLE && |
2227 | signal_pending(current)) { | |
46125c1c MB |
2228 | r = -EINTR; |
2229 | break; | |
2230 | } | |
2231 | ||
2232 | io_schedule(); | |
2233 | } | |
2234 | set_current_state(TASK_RUNNING); | |
2235 | ||
b44ebeb0 MP |
2236 | remove_wait_queue(&md->wait, &wait); |
2237 | ||
46125c1c MB |
2238 | return r; |
2239 | } | |
2240 | ||
531fe963 | 2241 | static void dm_flush(struct mapped_device *md) |
af7e466a MP |
2242 | { |
2243 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
52b1fd5a MP |
2244 | |
2245 | bio_init(&md->barrier_bio); | |
2246 | md->barrier_bio.bi_bdev = md->bdev; | |
2247 | md->barrier_bio.bi_rw = WRITE_BARRIER; | |
2248 | __split_and_process_bio(md, &md->barrier_bio); | |
2249 | ||
2250 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
af7e466a MP |
2251 | } |
2252 | ||
2253 | static void process_barrier(struct mapped_device *md, struct bio *bio) | |
2254 | { | |
5aa2781d MP |
2255 | md->barrier_error = 0; |
2256 | ||
531fe963 | 2257 | dm_flush(md); |
af7e466a | 2258 | |
5aa2781d MP |
2259 | if (!bio_empty_barrier(bio)) { |
2260 | __split_and_process_bio(md, bio); | |
2261 | dm_flush(md); | |
af7e466a MP |
2262 | } |
2263 | ||
af7e466a | 2264 | if (md->barrier_error != DM_ENDIO_REQUEUE) |
531fe963 | 2265 | bio_endio(bio, md->barrier_error); |
2761e95f MP |
2266 | else { |
2267 | spin_lock_irq(&md->deferred_lock); | |
2268 | bio_list_add_head(&md->deferred, bio); | |
2269 | spin_unlock_irq(&md->deferred_lock); | |
2270 | } | |
af7e466a MP |
2271 | } |
2272 | ||
1da177e4 LT |
2273 | /* |
2274 | * Process the deferred bios | |
2275 | */ | |
ef208587 | 2276 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2277 | { |
ef208587 MP |
2278 | struct mapped_device *md = container_of(work, struct mapped_device, |
2279 | work); | |
6d6f10df | 2280 | struct bio *c; |
1da177e4 | 2281 | |
ef208587 MP |
2282 | down_write(&md->io_lock); |
2283 | ||
3b00b203 | 2284 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2285 | spin_lock_irq(&md->deferred_lock); |
2286 | c = bio_list_pop(&md->deferred); | |
2287 | spin_unlock_irq(&md->deferred_lock); | |
2288 | ||
2289 | if (!c) { | |
1eb787ec | 2290 | clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); |
df12ee99 AK |
2291 | break; |
2292 | } | |
022c2611 | 2293 | |
3b00b203 MP |
2294 | up_write(&md->io_lock); |
2295 | ||
e6ee8c0b KU |
2296 | if (dm_request_based(md)) |
2297 | generic_make_request(c); | |
2298 | else { | |
1f98a13f | 2299 | if (bio_rw_flagged(c, BIO_RW_BARRIER)) |
e6ee8c0b KU |
2300 | process_barrier(md, c); |
2301 | else | |
2302 | __split_and_process_bio(md, c); | |
2303 | } | |
3b00b203 MP |
2304 | |
2305 | down_write(&md->io_lock); | |
022c2611 | 2306 | } |
73d410c0 | 2307 | |
ef208587 | 2308 | up_write(&md->io_lock); |
1da177e4 LT |
2309 | } |
2310 | ||
9a1fb464 | 2311 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2312 | { |
3b00b203 MP |
2313 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2314 | smp_mb__after_clear_bit(); | |
53d5914f | 2315 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2316 | } |
2317 | ||
d0bcb878 KU |
2318 | static void dm_rq_set_flush_nr(struct request *clone, unsigned flush_nr) |
2319 | { | |
2320 | struct dm_rq_target_io *tio = clone->end_io_data; | |
2321 | ||
2322 | tio->info.flush_request = flush_nr; | |
2323 | } | |
2324 | ||
2325 | /* Issue barrier requests to targets and wait for their completion. */ | |
2326 | static int dm_rq_barrier(struct mapped_device *md) | |
2327 | { | |
2328 | int i, j; | |
7c666411 | 2329 | struct dm_table *map = dm_get_live_table(md); |
d0bcb878 KU |
2330 | unsigned num_targets = dm_table_get_num_targets(map); |
2331 | struct dm_target *ti; | |
2332 | struct request *clone; | |
2333 | ||
2334 | md->barrier_error = 0; | |
2335 | ||
2336 | for (i = 0; i < num_targets; i++) { | |
2337 | ti = dm_table_get_target(map, i); | |
2338 | for (j = 0; j < ti->num_flush_requests; j++) { | |
2339 | clone = clone_rq(md->flush_request, md, GFP_NOIO); | |
2340 | dm_rq_set_flush_nr(clone, j); | |
2341 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
2342 | map_request(ti, clone, md); | |
2343 | } | |
2344 | } | |
2345 | ||
2346 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2347 | dm_table_put(map); | |
2348 | ||
2349 | return md->barrier_error; | |
2350 | } | |
2351 | ||
2352 | static void dm_rq_barrier_work(struct work_struct *work) | |
2353 | { | |
2354 | int error; | |
2355 | struct mapped_device *md = container_of(work, struct mapped_device, | |
2356 | barrier_work); | |
2357 | struct request_queue *q = md->queue; | |
2358 | struct request *rq; | |
2359 | unsigned long flags; | |
2360 | ||
2361 | /* | |
2362 | * Hold the md reference here and leave it at the last part so that | |
2363 | * the md can't be deleted by device opener when the barrier request | |
2364 | * completes. | |
2365 | */ | |
2366 | dm_get(md); | |
2367 | ||
2368 | error = dm_rq_barrier(md); | |
2369 | ||
2370 | rq = md->flush_request; | |
2371 | md->flush_request = NULL; | |
2372 | ||
2373 | if (error == DM_ENDIO_REQUEUE) { | |
2374 | spin_lock_irqsave(q->queue_lock, flags); | |
2375 | blk_requeue_request(q, rq); | |
2376 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2377 | } else | |
2378 | blk_end_request_all(rq, error); | |
2379 | ||
2380 | blk_run_queue(q); | |
2381 | ||
2382 | dm_put(md); | |
2383 | } | |
2384 | ||
1da177e4 | 2385 | /* |
042d2a9b | 2386 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2387 | */ |
042d2a9b | 2388 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2389 | { |
042d2a9b | 2390 | struct dm_table *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2391 | struct queue_limits limits; |
042d2a9b | 2392 | int r; |
1da177e4 | 2393 | |
e61290a4 | 2394 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2395 | |
2396 | /* device must be suspended */ | |
4f186f8b | 2397 | if (!dm_suspended_md(md)) |
93c534ae | 2398 | goto out; |
1da177e4 | 2399 | |
754c5fc7 | 2400 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2401 | if (r) { |
2402 | map = ERR_PTR(r); | |
754c5fc7 | 2403 | goto out; |
042d2a9b | 2404 | } |
754c5fc7 | 2405 | |
e6ee8c0b KU |
2406 | /* cannot change the device type, once a table is bound */ |
2407 | if (md->map && | |
2408 | (dm_table_get_type(md->map) != dm_table_get_type(table))) { | |
2409 | DMWARN("can't change the device type after a table is bound"); | |
2410 | goto out; | |
2411 | } | |
2412 | ||
042d2a9b | 2413 | map = __bind(md, table, &limits); |
1da177e4 | 2414 | |
93c534ae | 2415 | out: |
e61290a4 | 2416 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2417 | return map; |
1da177e4 LT |
2418 | } |
2419 | ||
2420 | /* | |
2421 | * Functions to lock and unlock any filesystem running on the | |
2422 | * device. | |
2423 | */ | |
2ca3310e | 2424 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2425 | { |
e39e2e95 | 2426 | int r; |
1da177e4 LT |
2427 | |
2428 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2429 | |
db8fef4f | 2430 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2431 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2432 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2433 | md->frozen_sb = NULL; |
2434 | return r; | |
dfbe03f6 AK |
2435 | } |
2436 | ||
aa8d7c2f AK |
2437 | set_bit(DMF_FROZEN, &md->flags); |
2438 | ||
1da177e4 LT |
2439 | return 0; |
2440 | } | |
2441 | ||
2ca3310e | 2442 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2443 | { |
aa8d7c2f AK |
2444 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2445 | return; | |
2446 | ||
db8fef4f | 2447 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2448 | md->frozen_sb = NULL; |
aa8d7c2f | 2449 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2450 | } |
2451 | ||
2452 | /* | |
2453 | * We need to be able to change a mapping table under a mounted | |
2454 | * filesystem. For example we might want to move some data in | |
2455 | * the background. Before the table can be swapped with | |
2456 | * dm_bind_table, dm_suspend must be called to flush any in | |
2457 | * flight bios and ensure that any further io gets deferred. | |
2458 | */ | |
cec47e3d KU |
2459 | /* |
2460 | * Suspend mechanism in request-based dm. | |
2461 | * | |
9f518b27 KU |
2462 | * 1. Flush all I/Os by lock_fs() if needed. |
2463 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2464 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2465 | * |
9f518b27 | 2466 | * To abort suspend, start the request_queue. |
cec47e3d | 2467 | */ |
a3d77d35 | 2468 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2469 | { |
2ca3310e | 2470 | struct dm_table *map = NULL; |
46125c1c | 2471 | int r = 0; |
a3d77d35 | 2472 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2473 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2474 | |
e61290a4 | 2475 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2476 | |
4f186f8b | 2477 | if (dm_suspended_md(md)) { |
73d410c0 | 2478 | r = -EINVAL; |
d287483d | 2479 | goto out_unlock; |
73d410c0 | 2480 | } |
1da177e4 | 2481 | |
7c666411 | 2482 | map = dm_get_live_table(md); |
1da177e4 | 2483 | |
2e93ccc1 KU |
2484 | /* |
2485 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2486 | * This flag is cleared before dm_suspend returns. | |
2487 | */ | |
2488 | if (noflush) | |
2489 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2490 | ||
cf222b37 AK |
2491 | /* This does not get reverted if there's an error later. */ |
2492 | dm_table_presuspend_targets(map); | |
2493 | ||
32a926da | 2494 | /* |
9f518b27 KU |
2495 | * Flush I/O to the device. |
2496 | * Any I/O submitted after lock_fs() may not be flushed. | |
2497 | * noflush takes precedence over do_lockfs. | |
2498 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2499 | */ |
2500 | if (!noflush && do_lockfs) { | |
2501 | r = lock_fs(md); | |
2502 | if (r) | |
f431d966 | 2503 | goto out; |
aa8d7c2f | 2504 | } |
1da177e4 LT |
2505 | |
2506 | /* | |
3b00b203 MP |
2507 | * Here we must make sure that no processes are submitting requests |
2508 | * to target drivers i.e. no one may be executing | |
2509 | * __split_and_process_bio. This is called from dm_request and | |
2510 | * dm_wq_work. | |
2511 | * | |
2512 | * To get all processes out of __split_and_process_bio in dm_request, | |
2513 | * we take the write lock. To prevent any process from reentering | |
2514 | * __split_and_process_bio from dm_request, we set | |
2515 | * DMF_QUEUE_IO_TO_THREAD. | |
2516 | * | |
2517 | * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND | |
2518 | * and call flush_workqueue(md->wq). flush_workqueue will wait until | |
2519 | * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any | |
2520 | * further calls to __split_and_process_bio from dm_wq_work. | |
1da177e4 | 2521 | */ |
2ca3310e | 2522 | down_write(&md->io_lock); |
1eb787ec AK |
2523 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2524 | set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); | |
2ca3310e | 2525 | up_write(&md->io_lock); |
1da177e4 | 2526 | |
d0bcb878 KU |
2527 | /* |
2528 | * Request-based dm uses md->wq for barrier (dm_rq_barrier_work) which | |
2529 | * can be kicked until md->queue is stopped. So stop md->queue before | |
2530 | * flushing md->wq. | |
2531 | */ | |
cec47e3d | 2532 | if (dm_request_based(md)) |
9f518b27 | 2533 | stop_queue(md->queue); |
cec47e3d | 2534 | |
d0bcb878 KU |
2535 | flush_workqueue(md->wq); |
2536 | ||
1da177e4 | 2537 | /* |
3b00b203 MP |
2538 | * At this point no more requests are entering target request routines. |
2539 | * We call dm_wait_for_completion to wait for all existing requests | |
2540 | * to finish. | |
1da177e4 | 2541 | */ |
401600df | 2542 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2543 | |
2ca3310e | 2544 | down_write(&md->io_lock); |
6d6f10df | 2545 | if (noflush) |
022c2611 | 2546 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2547 | up_write(&md->io_lock); |
2e93ccc1 | 2548 | |
1da177e4 | 2549 | /* were we interrupted ? */ |
46125c1c | 2550 | if (r < 0) { |
9a1fb464 | 2551 | dm_queue_flush(md); |
73d410c0 | 2552 | |
cec47e3d | 2553 | if (dm_request_based(md)) |
9f518b27 | 2554 | start_queue(md->queue); |
cec47e3d | 2555 | |
2ca3310e | 2556 | unlock_fs(md); |
2e93ccc1 | 2557 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2558 | } |
1da177e4 | 2559 | |
3b00b203 MP |
2560 | /* |
2561 | * If dm_wait_for_completion returned 0, the device is completely | |
2562 | * quiescent now. There is no request-processing activity. All new | |
2563 | * requests are being added to md->deferred list. | |
2564 | */ | |
2565 | ||
2ca3310e | 2566 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2567 | |
4d4471cb KU |
2568 | dm_table_postsuspend_targets(map); |
2569 | ||
2ca3310e AK |
2570 | out: |
2571 | dm_table_put(map); | |
d287483d AK |
2572 | |
2573 | out_unlock: | |
e61290a4 | 2574 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2575 | return r; |
1da177e4 LT |
2576 | } |
2577 | ||
2578 | int dm_resume(struct mapped_device *md) | |
2579 | { | |
cf222b37 | 2580 | int r = -EINVAL; |
cf222b37 | 2581 | struct dm_table *map = NULL; |
1da177e4 | 2582 | |
e61290a4 | 2583 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2584 | if (!dm_suspended_md(md)) |
cf222b37 | 2585 | goto out; |
cf222b37 | 2586 | |
7c666411 | 2587 | map = dm_get_live_table(md); |
2ca3310e | 2588 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2589 | goto out; |
1da177e4 | 2590 | |
8757b776 MB |
2591 | r = dm_table_resume_targets(map); |
2592 | if (r) | |
2593 | goto out; | |
2ca3310e | 2594 | |
9a1fb464 | 2595 | dm_queue_flush(md); |
2ca3310e | 2596 | |
cec47e3d KU |
2597 | /* |
2598 | * Flushing deferred I/Os must be done after targets are resumed | |
2599 | * so that mapping of targets can work correctly. | |
2600 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2601 | */ | |
2602 | if (dm_request_based(md)) | |
2603 | start_queue(md->queue); | |
2604 | ||
2ca3310e AK |
2605 | unlock_fs(md); |
2606 | ||
2607 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2608 | ||
1da177e4 | 2609 | dm_table_unplug_all(map); |
cf222b37 AK |
2610 | r = 0; |
2611 | out: | |
2612 | dm_table_put(map); | |
e61290a4 | 2613 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2614 | |
cf222b37 | 2615 | return r; |
1da177e4 LT |
2616 | } |
2617 | ||
2618 | /*----------------------------------------------------------------- | |
2619 | * Event notification. | |
2620 | *---------------------------------------------------------------*/ | |
60935eb2 MB |
2621 | void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
2622 | unsigned cookie) | |
69267a30 | 2623 | { |
60935eb2 MB |
2624 | char udev_cookie[DM_COOKIE_LENGTH]; |
2625 | char *envp[] = { udev_cookie, NULL }; | |
2626 | ||
2627 | if (!cookie) | |
2628 | kobject_uevent(&disk_to_dev(md->disk)->kobj, action); | |
2629 | else { | |
2630 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2631 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
2632 | kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp); | |
2633 | } | |
69267a30 AK |
2634 | } |
2635 | ||
7a8c3d3b MA |
2636 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2637 | { | |
2638 | return atomic_add_return(1, &md->uevent_seq); | |
2639 | } | |
2640 | ||
1da177e4 LT |
2641 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2642 | { | |
2643 | return atomic_read(&md->event_nr); | |
2644 | } | |
2645 | ||
2646 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2647 | { | |
2648 | return wait_event_interruptible(md->eventq, | |
2649 | (event_nr != atomic_read(&md->event_nr))); | |
2650 | } | |
2651 | ||
7a8c3d3b MA |
2652 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2653 | { | |
2654 | unsigned long flags; | |
2655 | ||
2656 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2657 | list_add(elist, &md->uevent_list); | |
2658 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2659 | } | |
2660 | ||
1da177e4 LT |
2661 | /* |
2662 | * The gendisk is only valid as long as you have a reference | |
2663 | * count on 'md'. | |
2664 | */ | |
2665 | struct gendisk *dm_disk(struct mapped_device *md) | |
2666 | { | |
2667 | return md->disk; | |
2668 | } | |
2669 | ||
784aae73 MB |
2670 | struct kobject *dm_kobject(struct mapped_device *md) |
2671 | { | |
2672 | return &md->kobj; | |
2673 | } | |
2674 | ||
2675 | /* | |
2676 | * struct mapped_device should not be exported outside of dm.c | |
2677 | * so use this check to verify that kobj is part of md structure | |
2678 | */ | |
2679 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2680 | { | |
2681 | struct mapped_device *md; | |
2682 | ||
2683 | md = container_of(kobj, struct mapped_device, kobj); | |
2684 | if (&md->kobj != kobj) | |
2685 | return NULL; | |
2686 | ||
4d89b7b4 | 2687 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2688 | dm_deleting_md(md)) |
4d89b7b4 MB |
2689 | return NULL; |
2690 | ||
784aae73 MB |
2691 | dm_get(md); |
2692 | return md; | |
2693 | } | |
2694 | ||
4f186f8b | 2695 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2696 | { |
2697 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2698 | } | |
2699 | ||
64dbce58 KU |
2700 | int dm_suspended(struct dm_target *ti) |
2701 | { | |
ecdb2e25 | 2702 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2703 | } |
2704 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2705 | ||
2e93ccc1 KU |
2706 | int dm_noflush_suspending(struct dm_target *ti) |
2707 | { | |
ecdb2e25 | 2708 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2709 | } |
2710 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2711 | ||
e6ee8c0b KU |
2712 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type) |
2713 | { | |
2714 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
2715 | ||
2716 | if (!pools) | |
2717 | return NULL; | |
2718 | ||
2719 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2720 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2721 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2722 | if (!pools->io_pool) | |
2723 | goto free_pools_and_out; | |
2724 | ||
2725 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2726 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2727 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2728 | if (!pools->tio_pool) | |
2729 | goto free_io_pool_and_out; | |
2730 | ||
2731 | pools->bs = (type == DM_TYPE_BIO_BASED) ? | |
2732 | bioset_create(16, 0) : bioset_create(MIN_IOS, 0); | |
2733 | if (!pools->bs) | |
2734 | goto free_tio_pool_and_out; | |
2735 | ||
2736 | return pools; | |
2737 | ||
2738 | free_tio_pool_and_out: | |
2739 | mempool_destroy(pools->tio_pool); | |
2740 | ||
2741 | free_io_pool_and_out: | |
2742 | mempool_destroy(pools->io_pool); | |
2743 | ||
2744 | free_pools_and_out: | |
2745 | kfree(pools); | |
2746 | ||
2747 | return NULL; | |
2748 | } | |
2749 | ||
2750 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2751 | { | |
2752 | if (!pools) | |
2753 | return; | |
2754 | ||
2755 | if (pools->io_pool) | |
2756 | mempool_destroy(pools->io_pool); | |
2757 | ||
2758 | if (pools->tio_pool) | |
2759 | mempool_destroy(pools->tio_pool); | |
2760 | ||
2761 | if (pools->bs) | |
2762 | bioset_free(pools->bs); | |
2763 | ||
2764 | kfree(pools); | |
2765 | } | |
2766 | ||
83d5cde4 | 2767 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2768 | .open = dm_blk_open, |
2769 | .release = dm_blk_close, | |
aa129a22 | 2770 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2771 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2772 | .owner = THIS_MODULE |
2773 | }; | |
2774 | ||
2775 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2776 | ||
2777 | /* | |
2778 | * module hooks | |
2779 | */ | |
2780 | module_init(dm_init); | |
2781 | module_exit(dm_exit); | |
2782 | ||
2783 | module_param(major, uint, 0); | |
2784 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2785 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2786 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2787 | MODULE_LICENSE("GPL"); |