Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
1da177e4 LT |
63 | }; |
64 | ||
8fbf26ad KU |
65 | /* |
66 | * For request-based dm. | |
67 | * One of these is allocated per request. | |
68 | */ | |
69 | struct dm_rq_target_io { | |
70 | struct mapped_device *md; | |
71 | struct dm_target *ti; | |
72 | struct request *orig, clone; | |
73 | int error; | |
74 | union map_info info; | |
75 | }; | |
76 | ||
77 | /* | |
94818742 KO |
78 | * For request-based dm - the bio clones we allocate are embedded in these |
79 | * structs. | |
80 | * | |
81 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
82 | * the bioset is created - this means the bio has to come at the end of the | |
83 | * struct. | |
8fbf26ad KU |
84 | */ |
85 | struct dm_rq_clone_bio_info { | |
86 | struct bio *orig; | |
cec47e3d | 87 | struct dm_rq_target_io *tio; |
94818742 | 88 | struct bio clone; |
8fbf26ad KU |
89 | }; |
90 | ||
1da177e4 LT |
91 | union map_info *dm_get_mapinfo(struct bio *bio) |
92 | { | |
17b2f66f | 93 | if (bio && bio->bi_private) |
028867ac | 94 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 95 | return NULL; |
1da177e4 LT |
96 | } |
97 | ||
cec47e3d KU |
98 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
99 | { | |
100 | if (rq && rq->end_io_data) | |
101 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
102 | return NULL; | |
103 | } | |
104 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
105 | ||
ba61fdd1 JM |
106 | #define MINOR_ALLOCED ((void *)-1) |
107 | ||
1da177e4 LT |
108 | /* |
109 | * Bits for the md->flags field. | |
110 | */ | |
1eb787ec | 111 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 112 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 113 | #define DMF_FROZEN 2 |
fba9f90e | 114 | #define DMF_FREEING 3 |
5c6bd75d | 115 | #define DMF_DELETING 4 |
2e93ccc1 | 116 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 117 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 118 | |
304f3f6a MB |
119 | /* |
120 | * Work processed by per-device workqueue. | |
121 | */ | |
1da177e4 | 122 | struct mapped_device { |
2ca3310e | 123 | struct rw_semaphore io_lock; |
e61290a4 | 124 | struct mutex suspend_lock; |
1da177e4 LT |
125 | rwlock_t map_lock; |
126 | atomic_t holders; | |
5c6bd75d | 127 | atomic_t open_count; |
1da177e4 LT |
128 | |
129 | unsigned long flags; | |
130 | ||
165125e1 | 131 | struct request_queue *queue; |
a5664dad | 132 | unsigned type; |
4a0b4ddf | 133 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
134 | struct mutex type_lock; |
135 | ||
36a0456f AK |
136 | struct target_type *immutable_target_type; |
137 | ||
1da177e4 | 138 | struct gendisk *disk; |
7e51f257 | 139 | char name[16]; |
1da177e4 LT |
140 | |
141 | void *interface_ptr; | |
142 | ||
143 | /* | |
144 | * A list of ios that arrived while we were suspended. | |
145 | */ | |
316d315b | 146 | atomic_t pending[2]; |
1da177e4 | 147 | wait_queue_head_t wait; |
53d5914f | 148 | struct work_struct work; |
74859364 | 149 | struct bio_list deferred; |
022c2611 | 150 | spinlock_t deferred_lock; |
1da177e4 | 151 | |
af7e466a | 152 | /* |
29e4013d | 153 | * Processing queue (flush) |
304f3f6a MB |
154 | */ |
155 | struct workqueue_struct *wq; | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * The current mapping. | |
159 | */ | |
160 | struct dm_table *map; | |
161 | ||
162 | /* | |
163 | * io objects are allocated from here. | |
164 | */ | |
165 | mempool_t *io_pool; | |
166 | mempool_t *tio_pool; | |
167 | ||
9faf400f SB |
168 | struct bio_set *bs; |
169 | ||
1da177e4 LT |
170 | /* |
171 | * Event handling. | |
172 | */ | |
173 | atomic_t event_nr; | |
174 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
175 | atomic_t uevent_seq; |
176 | struct list_head uevent_list; | |
177 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
178 | |
179 | /* | |
180 | * freeze/thaw support require holding onto a super block | |
181 | */ | |
182 | struct super_block *frozen_sb; | |
db8fef4f | 183 | struct block_device *bdev; |
3ac51e74 DW |
184 | |
185 | /* forced geometry settings */ | |
186 | struct hd_geometry geometry; | |
784aae73 MB |
187 | |
188 | /* sysfs handle */ | |
189 | struct kobject kobj; | |
52b1fd5a | 190 | |
d87f4c14 TH |
191 | /* zero-length flush that will be cloned and submitted to targets */ |
192 | struct bio flush_bio; | |
1da177e4 LT |
193 | }; |
194 | ||
e6ee8c0b KU |
195 | /* |
196 | * For mempools pre-allocation at the table loading time. | |
197 | */ | |
198 | struct dm_md_mempools { | |
199 | mempool_t *io_pool; | |
200 | mempool_t *tio_pool; | |
201 | struct bio_set *bs; | |
202 | }; | |
203 | ||
1da177e4 | 204 | #define MIN_IOS 256 |
e18b890b | 205 | static struct kmem_cache *_io_cache; |
8fbf26ad | 206 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 207 | |
1da177e4 LT |
208 | static int __init local_init(void) |
209 | { | |
51157b4a | 210 | int r = -ENOMEM; |
1da177e4 | 211 | |
1da177e4 | 212 | /* allocate a slab for the dm_ios */ |
028867ac | 213 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 214 | if (!_io_cache) |
51157b4a | 215 | return r; |
1da177e4 | 216 | |
8fbf26ad KU |
217 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
218 | if (!_rq_tio_cache) | |
dba14160 | 219 | goto out_free_io_cache; |
8fbf26ad | 220 | |
51e5b2bd | 221 | r = dm_uevent_init(); |
51157b4a | 222 | if (r) |
23e5083b | 223 | goto out_free_rq_tio_cache; |
51e5b2bd | 224 | |
1da177e4 LT |
225 | _major = major; |
226 | r = register_blkdev(_major, _name); | |
51157b4a KU |
227 | if (r < 0) |
228 | goto out_uevent_exit; | |
1da177e4 LT |
229 | |
230 | if (!_major) | |
231 | _major = r; | |
232 | ||
233 | return 0; | |
51157b4a KU |
234 | |
235 | out_uevent_exit: | |
236 | dm_uevent_exit(); | |
8fbf26ad KU |
237 | out_free_rq_tio_cache: |
238 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
239 | out_free_io_cache: |
240 | kmem_cache_destroy(_io_cache); | |
241 | ||
242 | return r; | |
1da177e4 LT |
243 | } |
244 | ||
245 | static void local_exit(void) | |
246 | { | |
8fbf26ad | 247 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 248 | kmem_cache_destroy(_io_cache); |
00d59405 | 249 | unregister_blkdev(_major, _name); |
51e5b2bd | 250 | dm_uevent_exit(); |
1da177e4 LT |
251 | |
252 | _major = 0; | |
253 | ||
254 | DMINFO("cleaned up"); | |
255 | } | |
256 | ||
b9249e55 | 257 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
258 | local_init, |
259 | dm_target_init, | |
260 | dm_linear_init, | |
261 | dm_stripe_init, | |
952b3557 | 262 | dm_io_init, |
945fa4d2 | 263 | dm_kcopyd_init, |
1da177e4 LT |
264 | dm_interface_init, |
265 | }; | |
266 | ||
b9249e55 | 267 | static void (*_exits[])(void) = { |
1da177e4 LT |
268 | local_exit, |
269 | dm_target_exit, | |
270 | dm_linear_exit, | |
271 | dm_stripe_exit, | |
952b3557 | 272 | dm_io_exit, |
945fa4d2 | 273 | dm_kcopyd_exit, |
1da177e4 LT |
274 | dm_interface_exit, |
275 | }; | |
276 | ||
277 | static int __init dm_init(void) | |
278 | { | |
279 | const int count = ARRAY_SIZE(_inits); | |
280 | ||
281 | int r, i; | |
282 | ||
283 | for (i = 0; i < count; i++) { | |
284 | r = _inits[i](); | |
285 | if (r) | |
286 | goto bad; | |
287 | } | |
288 | ||
289 | return 0; | |
290 | ||
291 | bad: | |
292 | while (i--) | |
293 | _exits[i](); | |
294 | ||
295 | return r; | |
296 | } | |
297 | ||
298 | static void __exit dm_exit(void) | |
299 | { | |
300 | int i = ARRAY_SIZE(_exits); | |
301 | ||
302 | while (i--) | |
303 | _exits[i](); | |
d15b774c AK |
304 | |
305 | /* | |
306 | * Should be empty by this point. | |
307 | */ | |
d15b774c | 308 | idr_destroy(&_minor_idr); |
1da177e4 LT |
309 | } |
310 | ||
311 | /* | |
312 | * Block device functions | |
313 | */ | |
432a212c MA |
314 | int dm_deleting_md(struct mapped_device *md) |
315 | { | |
316 | return test_bit(DMF_DELETING, &md->flags); | |
317 | } | |
318 | ||
fe5f9f2c | 319 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
320 | { |
321 | struct mapped_device *md; | |
322 | ||
fba9f90e JM |
323 | spin_lock(&_minor_lock); |
324 | ||
fe5f9f2c | 325 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
326 | if (!md) |
327 | goto out; | |
328 | ||
5c6bd75d | 329 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 330 | dm_deleting_md(md)) { |
fba9f90e JM |
331 | md = NULL; |
332 | goto out; | |
333 | } | |
334 | ||
1da177e4 | 335 | dm_get(md); |
5c6bd75d | 336 | atomic_inc(&md->open_count); |
fba9f90e JM |
337 | |
338 | out: | |
339 | spin_unlock(&_minor_lock); | |
340 | ||
341 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
342 | } |
343 | ||
fe5f9f2c | 344 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 345 | { |
fe5f9f2c | 346 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 347 | |
4a1aeb98 MB |
348 | spin_lock(&_minor_lock); |
349 | ||
5c6bd75d | 350 | atomic_dec(&md->open_count); |
1da177e4 | 351 | dm_put(md); |
4a1aeb98 MB |
352 | |
353 | spin_unlock(&_minor_lock); | |
6e9624b8 | 354 | |
1da177e4 LT |
355 | return 0; |
356 | } | |
357 | ||
5c6bd75d AK |
358 | int dm_open_count(struct mapped_device *md) |
359 | { | |
360 | return atomic_read(&md->open_count); | |
361 | } | |
362 | ||
363 | /* | |
364 | * Guarantees nothing is using the device before it's deleted. | |
365 | */ | |
366 | int dm_lock_for_deletion(struct mapped_device *md) | |
367 | { | |
368 | int r = 0; | |
369 | ||
370 | spin_lock(&_minor_lock); | |
371 | ||
372 | if (dm_open_count(md)) | |
373 | r = -EBUSY; | |
374 | else | |
375 | set_bit(DMF_DELETING, &md->flags); | |
376 | ||
377 | spin_unlock(&_minor_lock); | |
378 | ||
379 | return r; | |
380 | } | |
381 | ||
3ac51e74 DW |
382 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
383 | { | |
384 | struct mapped_device *md = bdev->bd_disk->private_data; | |
385 | ||
386 | return dm_get_geometry(md, geo); | |
387 | } | |
388 | ||
fe5f9f2c | 389 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
390 | unsigned int cmd, unsigned long arg) |
391 | { | |
fe5f9f2c | 392 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 393 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
394 | struct dm_target *tgt; |
395 | int r = -ENOTTY; | |
396 | ||
aa129a22 MB |
397 | if (!map || !dm_table_get_size(map)) |
398 | goto out; | |
399 | ||
400 | /* We only support devices that have a single target */ | |
401 | if (dm_table_get_num_targets(map) != 1) | |
402 | goto out; | |
403 | ||
404 | tgt = dm_table_get_target(map, 0); | |
405 | ||
4f186f8b | 406 | if (dm_suspended_md(md)) { |
aa129a22 MB |
407 | r = -EAGAIN; |
408 | goto out; | |
409 | } | |
410 | ||
411 | if (tgt->type->ioctl) | |
647b3d00 | 412 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
413 | |
414 | out: | |
415 | dm_table_put(map); | |
416 | ||
aa129a22 MB |
417 | return r; |
418 | } | |
419 | ||
028867ac | 420 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
421 | { |
422 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
423 | } | |
424 | ||
028867ac | 425 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
426 | { |
427 | mempool_free(io, md->io_pool); | |
428 | } | |
429 | ||
028867ac | 430 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 431 | { |
dba14160 | 432 | bio_put(&tio->clone); |
1da177e4 LT |
433 | } |
434 | ||
08885643 KU |
435 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
436 | gfp_t gfp_mask) | |
cec47e3d | 437 | { |
08885643 | 438 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
439 | } |
440 | ||
441 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
442 | { | |
443 | mempool_free(tio, tio->md->tio_pool); | |
444 | } | |
445 | ||
90abb8c4 KU |
446 | static int md_in_flight(struct mapped_device *md) |
447 | { | |
448 | return atomic_read(&md->pending[READ]) + | |
449 | atomic_read(&md->pending[WRITE]); | |
450 | } | |
451 | ||
3eaf840e JNN |
452 | static void start_io_acct(struct dm_io *io) |
453 | { | |
454 | struct mapped_device *md = io->md; | |
c9959059 | 455 | int cpu; |
316d315b | 456 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
457 | |
458 | io->start_time = jiffies; | |
459 | ||
074a7aca TH |
460 | cpu = part_stat_lock(); |
461 | part_round_stats(cpu, &dm_disk(md)->part0); | |
462 | part_stat_unlock(); | |
1e9bb880 SL |
463 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
464 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
465 | } |
466 | ||
d221d2e7 | 467 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
468 | { |
469 | struct mapped_device *md = io->md; | |
470 | struct bio *bio = io->bio; | |
471 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 472 | int pending, cpu; |
3eaf840e JNN |
473 | int rw = bio_data_dir(bio); |
474 | ||
074a7aca TH |
475 | cpu = part_stat_lock(); |
476 | part_round_stats(cpu, &dm_disk(md)->part0); | |
477 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
478 | part_stat_unlock(); | |
3eaf840e | 479 | |
af7e466a MP |
480 | /* |
481 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 482 | * a flush. |
af7e466a | 483 | */ |
1e9bb880 SL |
484 | pending = atomic_dec_return(&md->pending[rw]); |
485 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 486 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 487 | |
d221d2e7 MP |
488 | /* nudge anyone waiting on suspend queue */ |
489 | if (!pending) | |
490 | wake_up(&md->wait); | |
3eaf840e JNN |
491 | } |
492 | ||
1da177e4 LT |
493 | /* |
494 | * Add the bio to the list of deferred io. | |
495 | */ | |
92c63902 | 496 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 497 | { |
05447420 | 498 | unsigned long flags; |
1da177e4 | 499 | |
05447420 | 500 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 501 | bio_list_add(&md->deferred, bio); |
05447420 | 502 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 503 | queue_work(md->wq, &md->work); |
1da177e4 LT |
504 | } |
505 | ||
506 | /* | |
507 | * Everyone (including functions in this file), should use this | |
508 | * function to access the md->map field, and make sure they call | |
509 | * dm_table_put() when finished. | |
510 | */ | |
7c666411 | 511 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
512 | { |
513 | struct dm_table *t; | |
523d9297 | 514 | unsigned long flags; |
1da177e4 | 515 | |
523d9297 | 516 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
517 | t = md->map; |
518 | if (t) | |
519 | dm_table_get(t); | |
523d9297 | 520 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
521 | |
522 | return t; | |
523 | } | |
524 | ||
3ac51e74 DW |
525 | /* |
526 | * Get the geometry associated with a dm device | |
527 | */ | |
528 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
529 | { | |
530 | *geo = md->geometry; | |
531 | ||
532 | return 0; | |
533 | } | |
534 | ||
535 | /* | |
536 | * Set the geometry of a device. | |
537 | */ | |
538 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
539 | { | |
540 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
541 | ||
542 | if (geo->start > sz) { | |
543 | DMWARN("Start sector is beyond the geometry limits."); | |
544 | return -EINVAL; | |
545 | } | |
546 | ||
547 | md->geometry = *geo; | |
548 | ||
549 | return 0; | |
550 | } | |
551 | ||
1da177e4 LT |
552 | /*----------------------------------------------------------------- |
553 | * CRUD START: | |
554 | * A more elegant soln is in the works that uses the queue | |
555 | * merge fn, unfortunately there are a couple of changes to | |
556 | * the block layer that I want to make for this. So in the | |
557 | * interests of getting something for people to use I give | |
558 | * you this clearly demarcated crap. | |
559 | *---------------------------------------------------------------*/ | |
560 | ||
2e93ccc1 KU |
561 | static int __noflush_suspending(struct mapped_device *md) |
562 | { | |
563 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
564 | } | |
565 | ||
1da177e4 LT |
566 | /* |
567 | * Decrements the number of outstanding ios that a bio has been | |
568 | * cloned into, completing the original io if necc. | |
569 | */ | |
858119e1 | 570 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 571 | { |
2e93ccc1 | 572 | unsigned long flags; |
b35f8caa MB |
573 | int io_error; |
574 | struct bio *bio; | |
575 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
576 | |
577 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
578 | if (unlikely(error)) { |
579 | spin_lock_irqsave(&io->endio_lock, flags); | |
580 | if (!(io->error > 0 && __noflush_suspending(md))) | |
581 | io->error = error; | |
582 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
583 | } | |
1da177e4 LT |
584 | |
585 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
586 | if (io->error == DM_ENDIO_REQUEUE) { |
587 | /* | |
588 | * Target requested pushing back the I/O. | |
2e93ccc1 | 589 | */ |
022c2611 | 590 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
591 | if (__noflush_suspending(md)) |
592 | bio_list_add_head(&md->deferred, io->bio); | |
593 | else | |
2e93ccc1 KU |
594 | /* noflush suspend was interrupted. */ |
595 | io->error = -EIO; | |
022c2611 | 596 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
597 | } |
598 | ||
b35f8caa MB |
599 | io_error = io->error; |
600 | bio = io->bio; | |
6a8736d1 TH |
601 | end_io_acct(io); |
602 | free_io(md, io); | |
603 | ||
604 | if (io_error == DM_ENDIO_REQUEUE) | |
605 | return; | |
2e93ccc1 | 606 | |
b372d360 | 607 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 608 | /* |
6a8736d1 TH |
609 | * Preflush done for flush with data, reissue |
610 | * without REQ_FLUSH. | |
af7e466a | 611 | */ |
6a8736d1 TH |
612 | bio->bi_rw &= ~REQ_FLUSH; |
613 | queue_io(md, bio); | |
af7e466a | 614 | } else { |
b372d360 | 615 | /* done with normal IO or empty flush */ |
b372d360 | 616 | bio_endio(bio, io_error); |
b35f8caa | 617 | } |
1da177e4 LT |
618 | } |
619 | } | |
620 | ||
6712ecf8 | 621 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
622 | { |
623 | int r = 0; | |
028867ac | 624 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 625 | struct dm_io *io = tio->io; |
9faf400f | 626 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
627 | dm_endio_fn endio = tio->ti->type->end_io; |
628 | ||
1da177e4 LT |
629 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
630 | error = -EIO; | |
631 | ||
632 | if (endio) { | |
7de3ee57 | 633 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
634 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
635 | /* | |
636 | * error and requeue request are handled | |
637 | * in dec_pending(). | |
638 | */ | |
1da177e4 | 639 | error = r; |
45cbcd79 KU |
640 | else if (r == DM_ENDIO_INCOMPLETE) |
641 | /* The target will handle the io */ | |
6712ecf8 | 642 | return; |
45cbcd79 KU |
643 | else if (r) { |
644 | DMWARN("unimplemented target endio return value: %d", r); | |
645 | BUG(); | |
646 | } | |
1da177e4 LT |
647 | } |
648 | ||
9faf400f | 649 | free_tio(md, tio); |
b35f8caa | 650 | dec_pending(io, error); |
1da177e4 LT |
651 | } |
652 | ||
cec47e3d KU |
653 | /* |
654 | * Partial completion handling for request-based dm | |
655 | */ | |
656 | static void end_clone_bio(struct bio *clone, int error) | |
657 | { | |
658 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
659 | struct dm_rq_target_io *tio = info->tio; | |
660 | struct bio *bio = info->orig; | |
661 | unsigned int nr_bytes = info->orig->bi_size; | |
662 | ||
663 | bio_put(clone); | |
664 | ||
665 | if (tio->error) | |
666 | /* | |
667 | * An error has already been detected on the request. | |
668 | * Once error occurred, just let clone->end_io() handle | |
669 | * the remainder. | |
670 | */ | |
671 | return; | |
672 | else if (error) { | |
673 | /* | |
674 | * Don't notice the error to the upper layer yet. | |
675 | * The error handling decision is made by the target driver, | |
676 | * when the request is completed. | |
677 | */ | |
678 | tio->error = error; | |
679 | return; | |
680 | } | |
681 | ||
682 | /* | |
683 | * I/O for the bio successfully completed. | |
684 | * Notice the data completion to the upper layer. | |
685 | */ | |
686 | ||
687 | /* | |
688 | * bios are processed from the head of the list. | |
689 | * So the completing bio should always be rq->bio. | |
690 | * If it's not, something wrong is happening. | |
691 | */ | |
692 | if (tio->orig->bio != bio) | |
693 | DMERR("bio completion is going in the middle of the request"); | |
694 | ||
695 | /* | |
696 | * Update the original request. | |
697 | * Do not use blk_end_request() here, because it may complete | |
698 | * the original request before the clone, and break the ordering. | |
699 | */ | |
700 | blk_update_request(tio->orig, 0, nr_bytes); | |
701 | } | |
702 | ||
703 | /* | |
704 | * Don't touch any member of the md after calling this function because | |
705 | * the md may be freed in dm_put() at the end of this function. | |
706 | * Or do dm_get() before calling this function and dm_put() later. | |
707 | */ | |
b4324fee | 708 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 709 | { |
b4324fee | 710 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
711 | |
712 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 713 | if (!md_in_flight(md)) |
cec47e3d KU |
714 | wake_up(&md->wait); |
715 | ||
a8c32a5c JA |
716 | /* |
717 | * Run this off this callpath, as drivers could invoke end_io while | |
718 | * inside their request_fn (and holding the queue lock). Calling | |
719 | * back into ->request_fn() could deadlock attempting to grab the | |
720 | * queue lock again. | |
721 | */ | |
cec47e3d | 722 | if (run_queue) |
a8c32a5c | 723 | blk_run_queue_async(md->queue); |
cec47e3d KU |
724 | |
725 | /* | |
726 | * dm_put() must be at the end of this function. See the comment above | |
727 | */ | |
728 | dm_put(md); | |
729 | } | |
730 | ||
a77e28c7 KU |
731 | static void free_rq_clone(struct request *clone) |
732 | { | |
733 | struct dm_rq_target_io *tio = clone->end_io_data; | |
734 | ||
735 | blk_rq_unprep_clone(clone); | |
736 | free_rq_tio(tio); | |
737 | } | |
738 | ||
980691e5 KU |
739 | /* |
740 | * Complete the clone and the original request. | |
741 | * Must be called without queue lock. | |
742 | */ | |
743 | static void dm_end_request(struct request *clone, int error) | |
744 | { | |
745 | int rw = rq_data_dir(clone); | |
746 | struct dm_rq_target_io *tio = clone->end_io_data; | |
747 | struct mapped_device *md = tio->md; | |
748 | struct request *rq = tio->orig; | |
749 | ||
29e4013d | 750 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
751 | rq->errors = clone->errors; |
752 | rq->resid_len = clone->resid_len; | |
753 | ||
754 | if (rq->sense) | |
755 | /* | |
756 | * We are using the sense buffer of the original | |
757 | * request. | |
758 | * So setting the length of the sense data is enough. | |
759 | */ | |
760 | rq->sense_len = clone->sense_len; | |
761 | } | |
762 | ||
763 | free_rq_clone(clone); | |
29e4013d TH |
764 | blk_end_request_all(rq, error); |
765 | rq_completed(md, rw, true); | |
980691e5 KU |
766 | } |
767 | ||
cec47e3d KU |
768 | static void dm_unprep_request(struct request *rq) |
769 | { | |
770 | struct request *clone = rq->special; | |
cec47e3d KU |
771 | |
772 | rq->special = NULL; | |
773 | rq->cmd_flags &= ~REQ_DONTPREP; | |
774 | ||
a77e28c7 | 775 | free_rq_clone(clone); |
cec47e3d KU |
776 | } |
777 | ||
778 | /* | |
779 | * Requeue the original request of a clone. | |
780 | */ | |
781 | void dm_requeue_unmapped_request(struct request *clone) | |
782 | { | |
b4324fee | 783 | int rw = rq_data_dir(clone); |
cec47e3d KU |
784 | struct dm_rq_target_io *tio = clone->end_io_data; |
785 | struct mapped_device *md = tio->md; | |
786 | struct request *rq = tio->orig; | |
787 | struct request_queue *q = rq->q; | |
788 | unsigned long flags; | |
789 | ||
790 | dm_unprep_request(rq); | |
791 | ||
792 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
793 | blk_requeue_request(q, rq); |
794 | spin_unlock_irqrestore(q->queue_lock, flags); | |
795 | ||
b4324fee | 796 | rq_completed(md, rw, 0); |
cec47e3d KU |
797 | } |
798 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
799 | ||
800 | static void __stop_queue(struct request_queue *q) | |
801 | { | |
802 | blk_stop_queue(q); | |
803 | } | |
804 | ||
805 | static void stop_queue(struct request_queue *q) | |
806 | { | |
807 | unsigned long flags; | |
808 | ||
809 | spin_lock_irqsave(q->queue_lock, flags); | |
810 | __stop_queue(q); | |
811 | spin_unlock_irqrestore(q->queue_lock, flags); | |
812 | } | |
813 | ||
814 | static void __start_queue(struct request_queue *q) | |
815 | { | |
816 | if (blk_queue_stopped(q)) | |
817 | blk_start_queue(q); | |
818 | } | |
819 | ||
820 | static void start_queue(struct request_queue *q) | |
821 | { | |
822 | unsigned long flags; | |
823 | ||
824 | spin_lock_irqsave(q->queue_lock, flags); | |
825 | __start_queue(q); | |
826 | spin_unlock_irqrestore(q->queue_lock, flags); | |
827 | } | |
828 | ||
11a68244 | 829 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 830 | { |
11a68244 | 831 | int r = error; |
cec47e3d | 832 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 833 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 834 | |
ba1cbad9 MS |
835 | if (tio->ti) { |
836 | rq_end_io = tio->ti->type->rq_end_io; | |
837 | ||
838 | if (mapped && rq_end_io) | |
839 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
840 | } | |
cec47e3d | 841 | |
11a68244 | 842 | if (r <= 0) |
cec47e3d | 843 | /* The target wants to complete the I/O */ |
11a68244 KU |
844 | dm_end_request(clone, r); |
845 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
846 | /* The target will handle the I/O */ |
847 | return; | |
11a68244 | 848 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
849 | /* The target wants to requeue the I/O */ |
850 | dm_requeue_unmapped_request(clone); | |
851 | else { | |
11a68244 | 852 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
853 | BUG(); |
854 | } | |
855 | } | |
856 | ||
11a68244 KU |
857 | /* |
858 | * Request completion handler for request-based dm | |
859 | */ | |
860 | static void dm_softirq_done(struct request *rq) | |
861 | { | |
862 | bool mapped = true; | |
863 | struct request *clone = rq->completion_data; | |
864 | struct dm_rq_target_io *tio = clone->end_io_data; | |
865 | ||
866 | if (rq->cmd_flags & REQ_FAILED) | |
867 | mapped = false; | |
868 | ||
869 | dm_done(clone, tio->error, mapped); | |
870 | } | |
871 | ||
cec47e3d KU |
872 | /* |
873 | * Complete the clone and the original request with the error status | |
874 | * through softirq context. | |
875 | */ | |
876 | static void dm_complete_request(struct request *clone, int error) | |
877 | { | |
878 | struct dm_rq_target_io *tio = clone->end_io_data; | |
879 | struct request *rq = tio->orig; | |
880 | ||
881 | tio->error = error; | |
882 | rq->completion_data = clone; | |
883 | blk_complete_request(rq); | |
884 | } | |
885 | ||
886 | /* | |
887 | * Complete the not-mapped clone and the original request with the error status | |
888 | * through softirq context. | |
889 | * Target's rq_end_io() function isn't called. | |
890 | * This may be used when the target's map_rq() function fails. | |
891 | */ | |
892 | void dm_kill_unmapped_request(struct request *clone, int error) | |
893 | { | |
894 | struct dm_rq_target_io *tio = clone->end_io_data; | |
895 | struct request *rq = tio->orig; | |
896 | ||
897 | rq->cmd_flags |= REQ_FAILED; | |
898 | dm_complete_request(clone, error); | |
899 | } | |
900 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
901 | ||
902 | /* | |
903 | * Called with the queue lock held | |
904 | */ | |
905 | static void end_clone_request(struct request *clone, int error) | |
906 | { | |
907 | /* | |
908 | * For just cleaning up the information of the queue in which | |
909 | * the clone was dispatched. | |
910 | * The clone is *NOT* freed actually here because it is alloced from | |
911 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
912 | */ | |
913 | __blk_put_request(clone->q, clone); | |
914 | ||
915 | /* | |
916 | * Actual request completion is done in a softirq context which doesn't | |
917 | * hold the queue lock. Otherwise, deadlock could occur because: | |
918 | * - another request may be submitted by the upper level driver | |
919 | * of the stacking during the completion | |
920 | * - the submission which requires queue lock may be done | |
921 | * against this queue | |
922 | */ | |
923 | dm_complete_request(clone, error); | |
924 | } | |
925 | ||
56a67df7 MS |
926 | /* |
927 | * Return maximum size of I/O possible at the supplied sector up to the current | |
928 | * target boundary. | |
929 | */ | |
930 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
931 | { | |
932 | sector_t target_offset = dm_target_offset(ti, sector); | |
933 | ||
934 | return ti->len - target_offset; | |
935 | } | |
936 | ||
937 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 938 | { |
56a67df7 | 939 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 940 | sector_t offset, max_len; |
1da177e4 LT |
941 | |
942 | /* | |
542f9038 | 943 | * Does the target need to split even further? |
1da177e4 | 944 | */ |
542f9038 MS |
945 | if (ti->max_io_len) { |
946 | offset = dm_target_offset(ti, sector); | |
947 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
948 | max_len = sector_div(offset, ti->max_io_len); | |
949 | else | |
950 | max_len = offset & (ti->max_io_len - 1); | |
951 | max_len = ti->max_io_len - max_len; | |
952 | ||
953 | if (len > max_len) | |
954 | len = max_len; | |
1da177e4 LT |
955 | } |
956 | ||
957 | return len; | |
958 | } | |
959 | ||
542f9038 MS |
960 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
961 | { | |
962 | if (len > UINT_MAX) { | |
963 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
964 | (unsigned long long)len, UINT_MAX); | |
965 | ti->error = "Maximum size of target IO is too large"; | |
966 | return -EINVAL; | |
967 | } | |
968 | ||
969 | ti->max_io_len = (uint32_t) len; | |
970 | ||
971 | return 0; | |
972 | } | |
973 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
974 | ||
bd2a49b8 | 975 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
976 | { |
977 | int r; | |
2056a782 | 978 | sector_t sector; |
9faf400f | 979 | struct mapped_device *md; |
dba14160 | 980 | struct bio *clone = &tio->clone; |
bd2a49b8 | 981 | struct dm_target *ti = tio->ti; |
1da177e4 | 982 | |
1da177e4 LT |
983 | clone->bi_end_io = clone_endio; |
984 | clone->bi_private = tio; | |
985 | ||
986 | /* | |
987 | * Map the clone. If r == 0 we don't need to do | |
988 | * anything, the target has assumed ownership of | |
989 | * this io. | |
990 | */ | |
991 | atomic_inc(&tio->io->io_count); | |
2056a782 | 992 | sector = clone->bi_sector; |
7de3ee57 | 993 | r = ti->type->map(ti, clone); |
45cbcd79 | 994 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 995 | /* the bio has been remapped so dispatch it */ |
2056a782 | 996 | |
d07335e5 MS |
997 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
998 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 999 | |
1da177e4 | 1000 | generic_make_request(clone); |
2e93ccc1 KU |
1001 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1002 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1003 | md = tio->io->md; |
1004 | dec_pending(tio->io, r); | |
9faf400f | 1005 | free_tio(md, tio); |
45cbcd79 KU |
1006 | } else if (r) { |
1007 | DMWARN("unimplemented target map return value: %d", r); | |
1008 | BUG(); | |
1da177e4 LT |
1009 | } |
1010 | } | |
1011 | ||
1012 | struct clone_info { | |
1013 | struct mapped_device *md; | |
1014 | struct dm_table *map; | |
1015 | struct bio *bio; | |
1016 | struct dm_io *io; | |
1017 | sector_t sector; | |
1018 | sector_t sector_count; | |
1019 | unsigned short idx; | |
1020 | }; | |
1021 | ||
bd2a49b8 AK |
1022 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1023 | { | |
1024 | bio->bi_sector = sector; | |
1025 | bio->bi_size = to_bytes(len); | |
1026 | } | |
1027 | ||
1028 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1029 | { | |
1030 | bio->bi_idx = idx; | |
1031 | bio->bi_vcnt = idx + bv_count; | |
1032 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1033 | } | |
1034 | ||
1035 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1036 | unsigned short idx, unsigned len, unsigned offset, | |
1037 | unsigned trim) | |
1038 | { | |
1039 | if (!bio_integrity(bio)) | |
1040 | return; | |
1041 | ||
1042 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1043 | ||
1044 | if (trim) | |
1045 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1046 | } | |
1047 | ||
1da177e4 | 1048 | /* |
d87f4c14 | 1049 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1050 | */ |
14fe594d AK |
1051 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1052 | sector_t sector, unsigned short idx, | |
1053 | unsigned offset, unsigned len) | |
1da177e4 | 1054 | { |
dba14160 | 1055 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1056 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1057 | ||
1da177e4 LT |
1058 | *clone->bi_io_vec = *bv; |
1059 | ||
bd2a49b8 AK |
1060 | bio_setup_sector(clone, sector, len); |
1061 | ||
1da177e4 | 1062 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1063 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1064 | clone->bi_vcnt = 1; |
1da177e4 LT |
1065 | clone->bi_io_vec->bv_offset = offset; |
1066 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1067 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1068 | |
bd2a49b8 | 1069 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1070 | } |
1071 | ||
1072 | /* | |
1073 | * Creates a bio that consists of range of complete bvecs. | |
1074 | */ | |
dba14160 MP |
1075 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1076 | sector_t sector, unsigned short idx, | |
e4c93811 | 1077 | unsigned short bv_count, unsigned len) |
1da177e4 | 1078 | { |
dba14160 | 1079 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1080 | unsigned trim = 0; |
1da177e4 | 1081 | |
9faf400f | 1082 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1083 | bio_setup_sector(clone, sector, len); |
1084 | bio_setup_bv(clone, idx, bv_count); | |
1085 | ||
1086 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1087 | trim = 1; | |
1088 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1089 | } |
1090 | ||
9015df24 | 1091 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1092 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1093 | unsigned target_bio_nr) |
f9ab94ce | 1094 | { |
dba14160 MP |
1095 | struct dm_target_io *tio; |
1096 | struct bio *clone; | |
1097 | ||
1098 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1099 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1100 | |
1101 | tio->io = ci->io; | |
1102 | tio->ti = ti; | |
f9ab94ce | 1103 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1104 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1105 | |
1106 | return tio; | |
1107 | } | |
1108 | ||
14fe594d AK |
1109 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1110 | struct dm_target *ti, | |
1111 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1112 | { |
55a62eef | 1113 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1114 | struct bio *clone = &tio->clone; |
9015df24 | 1115 | |
06a426ce MS |
1116 | /* |
1117 | * Discard requests require the bio's inline iovecs be initialized. | |
1118 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1119 | * and discard, so no need for concern about wasted bvec allocations. | |
1120 | */ | |
dba14160 | 1121 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1122 | if (len) |
1123 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1124 | |
bd2a49b8 | 1125 | __map_bio(tio); |
f9ab94ce MP |
1126 | } |
1127 | ||
14fe594d AK |
1128 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1129 | unsigned num_bios, sector_t len) | |
06a426ce | 1130 | { |
55a62eef | 1131 | unsigned target_bio_nr; |
06a426ce | 1132 | |
55a62eef | 1133 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1134 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1135 | } |
1136 | ||
14fe594d | 1137 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1138 | { |
06a426ce | 1139 | unsigned target_nr = 0; |
f9ab94ce MP |
1140 | struct dm_target *ti; |
1141 | ||
b372d360 | 1142 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1143 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1144 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1145 | |
f9ab94ce MP |
1146 | return 0; |
1147 | } | |
1148 | ||
e4c93811 AK |
1149 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1150 | sector_t sector, int nr_iovecs, | |
1151 | unsigned short idx, unsigned short bv_count, | |
1152 | unsigned offset, unsigned len, | |
1153 | unsigned split_bvec) | |
5ae89a87 | 1154 | { |
dba14160 | 1155 | struct bio *bio = ci->bio; |
5ae89a87 MS |
1156 | struct dm_target_io *tio; |
1157 | ||
e4c93811 AK |
1158 | tio = alloc_tio(ci, ti, nr_iovecs, 0); |
1159 | ||
1160 | if (split_bvec) | |
1161 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1162 | else | |
1163 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1164 | ||
bd2a49b8 | 1165 | __map_bio(tio); |
5ae89a87 MS |
1166 | } |
1167 | ||
55a62eef | 1168 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1169 | |
55a62eef | 1170 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1171 | { |
55a62eef | 1172 | return ti->num_discard_bios; |
23508a96 MS |
1173 | } |
1174 | ||
55a62eef | 1175 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1176 | { |
55a62eef | 1177 | return ti->num_write_same_bios; |
23508a96 MS |
1178 | } |
1179 | ||
1180 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1181 | ||
1182 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1183 | { | |
55a62eef | 1184 | return ti->split_discard_bios; |
23508a96 MS |
1185 | } |
1186 | ||
14fe594d AK |
1187 | static int __send_changing_extent_only(struct clone_info *ci, |
1188 | get_num_bios_fn get_num_bios, | |
1189 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1190 | { |
1191 | struct dm_target *ti; | |
a79245b3 | 1192 | sector_t len; |
55a62eef | 1193 | unsigned num_bios; |
5ae89a87 | 1194 | |
a79245b3 MS |
1195 | do { |
1196 | ti = dm_table_find_target(ci->map, ci->sector); | |
1197 | if (!dm_target_is_valid(ti)) | |
1198 | return -EIO; | |
5ae89a87 | 1199 | |
5ae89a87 | 1200 | /* |
23508a96 MS |
1201 | * Even though the device advertised support for this type of |
1202 | * request, that does not mean every target supports it, and | |
936688d7 | 1203 | * reconfiguration might also have changed that since the |
a79245b3 | 1204 | * check was performed. |
5ae89a87 | 1205 | */ |
55a62eef AK |
1206 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1207 | if (!num_bios) | |
a79245b3 | 1208 | return -EOPNOTSUPP; |
5ae89a87 | 1209 | |
23508a96 | 1210 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1211 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1212 | else | |
1213 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1214 | |
14fe594d | 1215 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1216 | |
1217 | ci->sector += len; | |
1218 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1219 | |
1220 | return 0; | |
1221 | } | |
1222 | ||
14fe594d | 1223 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1224 | { |
14fe594d AK |
1225 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1226 | is_split_required_for_discard); | |
23508a96 MS |
1227 | } |
1228 | ||
14fe594d | 1229 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1230 | { |
14fe594d | 1231 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1232 | } |
1233 | ||
e4c93811 AK |
1234 | /* |
1235 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1236 | */ | |
1237 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1238 | { | |
1239 | struct bio *bio = ci->bio; | |
1240 | sector_t bv_len, total_len = 0; | |
1241 | ||
1242 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1243 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1244 | ||
1245 | if (bv_len > max) | |
1246 | break; | |
1247 | ||
1248 | max -= bv_len; | |
1249 | total_len += bv_len; | |
1250 | } | |
1251 | ||
1252 | return total_len; | |
1253 | } | |
1254 | ||
1255 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1256 | struct dm_target *ti, sector_t max) | |
1257 | { | |
1258 | struct bio *bio = ci->bio; | |
1259 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1260 | sector_t remaining = to_sector(bv->bv_len); | |
1261 | unsigned offset = 0; | |
1262 | sector_t len; | |
1263 | ||
1264 | do { | |
1265 | if (offset) { | |
1266 | ti = dm_table_find_target(ci->map, ci->sector); | |
1267 | if (!dm_target_is_valid(ti)) | |
1268 | return -EIO; | |
1269 | ||
1270 | max = max_io_len(ci->sector, ti); | |
1271 | } | |
1272 | ||
1273 | len = min(remaining, max); | |
1274 | ||
1275 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1276 | bv->bv_offset + offset, len, 1); | |
1277 | ||
1278 | ci->sector += len; | |
1279 | ci->sector_count -= len; | |
1280 | offset += to_bytes(len); | |
1281 | } while (remaining -= len); | |
1282 | ||
1283 | ci->idx++; | |
1284 | ||
1285 | return 0; | |
1286 | } | |
1287 | ||
1288 | /* | |
1289 | * Select the correct strategy for processing a non-flush bio. | |
1290 | */ | |
14fe594d | 1291 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1292 | { |
dba14160 | 1293 | struct bio *bio = ci->bio; |
512875bd | 1294 | struct dm_target *ti; |
e4c93811 AK |
1295 | sector_t len, max; |
1296 | int idx; | |
1da177e4 | 1297 | |
5ae89a87 | 1298 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1299 | return __send_discard(ci); |
23508a96 | 1300 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1301 | return __send_write_same(ci); |
5ae89a87 | 1302 | |
512875bd JN |
1303 | ti = dm_table_find_target(ci->map, ci->sector); |
1304 | if (!dm_target_is_valid(ti)) | |
1305 | return -EIO; | |
1306 | ||
56a67df7 | 1307 | max = max_io_len(ci->sector, ti); |
512875bd | 1308 | |
e4c93811 AK |
1309 | /* |
1310 | * Optimise for the simple case where we can do all of | |
1311 | * the remaining io with a single clone. | |
1312 | */ | |
1da177e4 | 1313 | if (ci->sector_count <= max) { |
e4c93811 AK |
1314 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1315 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1316 | ci->sector_count, 0); | |
1317 | ci->sector_count = 0; | |
1318 | return 0; | |
1319 | } | |
1da177e4 | 1320 | |
e4c93811 AK |
1321 | /* |
1322 | * There are some bvecs that don't span targets. | |
1323 | * Do as many of these as possible. | |
1324 | */ | |
1325 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1326 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1327 | |
e4c93811 AK |
1328 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1329 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1330 | |
1331 | ci->sector += len; | |
1332 | ci->sector_count -= len; | |
e4c93811 | 1333 | ci->idx = idx; |
d2044a94 | 1334 | |
e4c93811 | 1335 | return 0; |
1da177e4 | 1336 | } |
512875bd | 1337 | |
e4c93811 AK |
1338 | /* |
1339 | * Handle a bvec that must be split between two or more targets. | |
1340 | */ | |
1341 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1342 | } |
1343 | ||
1344 | /* | |
14fe594d | 1345 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1346 | */ |
f0b9a450 | 1347 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1348 | { |
1349 | struct clone_info ci; | |
512875bd | 1350 | int error = 0; |
1da177e4 | 1351 | |
7c666411 | 1352 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1353 | if (unlikely(!ci.map)) { |
6a8736d1 | 1354 | bio_io_error(bio); |
f0b9a450 MP |
1355 | return; |
1356 | } | |
692d0eb9 | 1357 | |
1da177e4 | 1358 | ci.md = md; |
1da177e4 LT |
1359 | ci.io = alloc_io(md); |
1360 | ci.io->error = 0; | |
1361 | atomic_set(&ci.io->io_count, 1); | |
1362 | ci.io->bio = bio; | |
1363 | ci.io->md = md; | |
f88fb981 | 1364 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1365 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1366 | ci.idx = bio->bi_idx; |
1367 | ||
3eaf840e | 1368 | start_io_acct(ci.io); |
bd2a49b8 | 1369 | |
b372d360 MS |
1370 | if (bio->bi_rw & REQ_FLUSH) { |
1371 | ci.bio = &ci.md->flush_bio; | |
1372 | ci.sector_count = 0; | |
14fe594d | 1373 | error = __send_empty_flush(&ci); |
b372d360 MS |
1374 | /* dec_pending submits any data associated with flush */ |
1375 | } else { | |
6a8736d1 | 1376 | ci.bio = bio; |
d87f4c14 | 1377 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1378 | while (ci.sector_count && !error) |
14fe594d | 1379 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1380 | } |
1da177e4 LT |
1381 | |
1382 | /* drop the extra reference count */ | |
512875bd | 1383 | dec_pending(ci.io, error); |
1da177e4 LT |
1384 | dm_table_put(ci.map); |
1385 | } | |
1386 | /*----------------------------------------------------------------- | |
1387 | * CRUD END | |
1388 | *---------------------------------------------------------------*/ | |
1389 | ||
f6fccb12 MB |
1390 | static int dm_merge_bvec(struct request_queue *q, |
1391 | struct bvec_merge_data *bvm, | |
1392 | struct bio_vec *biovec) | |
1393 | { | |
1394 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1395 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1396 | struct dm_target *ti; |
1397 | sector_t max_sectors; | |
5037108a | 1398 | int max_size = 0; |
f6fccb12 MB |
1399 | |
1400 | if (unlikely(!map)) | |
5037108a | 1401 | goto out; |
f6fccb12 MB |
1402 | |
1403 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1404 | if (!dm_target_is_valid(ti)) |
1405 | goto out_table; | |
f6fccb12 MB |
1406 | |
1407 | /* | |
1408 | * Find maximum amount of I/O that won't need splitting | |
1409 | */ | |
56a67df7 | 1410 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1411 | (sector_t) BIO_MAX_SECTORS); |
1412 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1413 | if (max_size < 0) | |
1414 | max_size = 0; | |
1415 | ||
1416 | /* | |
1417 | * merge_bvec_fn() returns number of bytes | |
1418 | * it can accept at this offset | |
1419 | * max is precomputed maximal io size | |
1420 | */ | |
1421 | if (max_size && ti->type->merge) | |
1422 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1423 | /* |
1424 | * If the target doesn't support merge method and some of the devices | |
1425 | * provided their merge_bvec method (we know this by looking at | |
1426 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1427 | * entries. So always set max_size to 0, and the code below allows | |
1428 | * just one page. | |
1429 | */ | |
1430 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1431 | ||
1432 | max_size = 0; | |
f6fccb12 | 1433 | |
b01cd5ac | 1434 | out_table: |
5037108a MP |
1435 | dm_table_put(map); |
1436 | ||
1437 | out: | |
f6fccb12 MB |
1438 | /* |
1439 | * Always allow an entire first page | |
1440 | */ | |
1441 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1442 | max_size = biovec->bv_len; | |
1443 | ||
f6fccb12 MB |
1444 | return max_size; |
1445 | } | |
1446 | ||
1da177e4 LT |
1447 | /* |
1448 | * The request function that just remaps the bio built up by | |
1449 | * dm_merge_bvec. | |
1450 | */ | |
5a7bbad2 | 1451 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1452 | { |
12f03a49 | 1453 | int rw = bio_data_dir(bio); |
1da177e4 | 1454 | struct mapped_device *md = q->queuedata; |
c9959059 | 1455 | int cpu; |
1da177e4 | 1456 | |
2ca3310e | 1457 | down_read(&md->io_lock); |
1da177e4 | 1458 | |
074a7aca TH |
1459 | cpu = part_stat_lock(); |
1460 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1461 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1462 | part_stat_unlock(); | |
12f03a49 | 1463 | |
6a8736d1 TH |
1464 | /* if we're suspended, we have to queue this io for later */ |
1465 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1466 | up_read(&md->io_lock); |
1da177e4 | 1467 | |
6a8736d1 TH |
1468 | if (bio_rw(bio) != READA) |
1469 | queue_io(md, bio); | |
1470 | else | |
54d9a1b4 | 1471 | bio_io_error(bio); |
5a7bbad2 | 1472 | return; |
1da177e4 LT |
1473 | } |
1474 | ||
f0b9a450 | 1475 | __split_and_process_bio(md, bio); |
2ca3310e | 1476 | up_read(&md->io_lock); |
5a7bbad2 | 1477 | return; |
cec47e3d KU |
1478 | } |
1479 | ||
1480 | static int dm_request_based(struct mapped_device *md) | |
1481 | { | |
1482 | return blk_queue_stackable(md->queue); | |
1483 | } | |
1484 | ||
5a7bbad2 | 1485 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1486 | { |
1487 | struct mapped_device *md = q->queuedata; | |
1488 | ||
1489 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1490 | blk_queue_bio(q, bio); |
1491 | else | |
1492 | _dm_request(q, bio); | |
cec47e3d KU |
1493 | } |
1494 | ||
1495 | void dm_dispatch_request(struct request *rq) | |
1496 | { | |
1497 | int r; | |
1498 | ||
1499 | if (blk_queue_io_stat(rq->q)) | |
1500 | rq->cmd_flags |= REQ_IO_STAT; | |
1501 | ||
1502 | rq->start_time = jiffies; | |
1503 | r = blk_insert_cloned_request(rq->q, rq); | |
1504 | if (r) | |
1505 | dm_complete_request(rq, r); | |
1506 | } | |
1507 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1508 | ||
cec47e3d KU |
1509 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1510 | void *data) | |
1511 | { | |
1512 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1513 | struct dm_rq_clone_bio_info *info = |
1514 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1515 | |
1516 | info->orig = bio_orig; | |
1517 | info->tio = tio; | |
1518 | bio->bi_end_io = end_clone_bio; | |
1519 | bio->bi_private = info; | |
cec47e3d KU |
1520 | |
1521 | return 0; | |
1522 | } | |
1523 | ||
1524 | static int setup_clone(struct request *clone, struct request *rq, | |
1525 | struct dm_rq_target_io *tio) | |
1526 | { | |
d0bcb878 | 1527 | int r; |
cec47e3d | 1528 | |
29e4013d TH |
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; | |
cec47e3d | 1533 | |
29e4013d TH |
1534 | clone->cmd = rq->cmd; |
1535 | clone->cmd_len = rq->cmd_len; | |
1536 | clone->sense = rq->sense; | |
1537 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1538 | clone->end_io = end_clone_request; |
1539 | clone->end_io_data = tio; | |
1540 | ||
1541 | return 0; | |
1542 | } | |
1543 | ||
6facdaff KU |
1544 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1545 | gfp_t gfp_mask) | |
1546 | { | |
1547 | struct request *clone; | |
1548 | struct dm_rq_target_io *tio; | |
1549 | ||
1550 | tio = alloc_rq_tio(md, gfp_mask); | |
1551 | if (!tio) | |
1552 | return NULL; | |
1553 | ||
1554 | tio->md = md; | |
1555 | tio->ti = NULL; | |
1556 | tio->orig = rq; | |
1557 | tio->error = 0; | |
1558 | memset(&tio->info, 0, sizeof(tio->info)); | |
1559 | ||
1560 | clone = &tio->clone; | |
1561 | if (setup_clone(clone, rq, tio)) { | |
1562 | /* -ENOMEM */ | |
1563 | free_rq_tio(tio); | |
1564 | return NULL; | |
1565 | } | |
1566 | ||
1567 | return clone; | |
1568 | } | |
1569 | ||
cec47e3d KU |
1570 | /* |
1571 | * Called with the queue lock held. | |
1572 | */ | |
1573 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1574 | { | |
1575 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1576 | struct request *clone; |
1577 | ||
cec47e3d KU |
1578 | if (unlikely(rq->special)) { |
1579 | DMWARN("Already has something in rq->special."); | |
1580 | return BLKPREP_KILL; | |
1581 | } | |
1582 | ||
6facdaff KU |
1583 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1584 | if (!clone) | |
cec47e3d | 1585 | return BLKPREP_DEFER; |
cec47e3d KU |
1586 | |
1587 | rq->special = clone; | |
1588 | rq->cmd_flags |= REQ_DONTPREP; | |
1589 | ||
1590 | return BLKPREP_OK; | |
1591 | } | |
1592 | ||
9eef87da KU |
1593 | /* |
1594 | * Returns: | |
1595 | * 0 : the request has been processed (not requeued) | |
1596 | * !0 : the request has been requeued | |
1597 | */ | |
1598 | static int map_request(struct dm_target *ti, struct request *clone, | |
1599 | struct mapped_device *md) | |
cec47e3d | 1600 | { |
9eef87da | 1601 | int r, requeued = 0; |
cec47e3d KU |
1602 | struct dm_rq_target_io *tio = clone->end_io_data; |
1603 | ||
cec47e3d KU |
1604 | tio->ti = ti; |
1605 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1606 | switch (r) { | |
1607 | case DM_MAPIO_SUBMITTED: | |
1608 | /* The target has taken the I/O to submit by itself later */ | |
1609 | break; | |
1610 | case DM_MAPIO_REMAPPED: | |
1611 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1612 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1613 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1614 | dm_dispatch_request(clone); |
1615 | break; | |
1616 | case DM_MAPIO_REQUEUE: | |
1617 | /* The target wants to requeue the I/O */ | |
1618 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1619 | requeued = 1; |
cec47e3d KU |
1620 | break; |
1621 | default: | |
1622 | if (r > 0) { | |
1623 | DMWARN("unimplemented target map return value: %d", r); | |
1624 | BUG(); | |
1625 | } | |
1626 | ||
1627 | /* The target wants to complete the I/O */ | |
1628 | dm_kill_unmapped_request(clone, r); | |
1629 | break; | |
1630 | } | |
9eef87da KU |
1631 | |
1632 | return requeued; | |
cec47e3d KU |
1633 | } |
1634 | ||
ba1cbad9 MS |
1635 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1636 | { | |
1637 | struct request *clone; | |
1638 | ||
1639 | blk_start_request(orig); | |
1640 | clone = orig->special; | |
1641 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1642 | ||
1643 | /* | |
1644 | * Hold the md reference here for the in-flight I/O. | |
1645 | * We can't rely on the reference count by device opener, | |
1646 | * because the device may be closed during the request completion | |
1647 | * when all bios are completed. | |
1648 | * See the comment in rq_completed() too. | |
1649 | */ | |
1650 | dm_get(md); | |
1651 | ||
1652 | return clone; | |
1653 | } | |
1654 | ||
cec47e3d KU |
1655 | /* |
1656 | * q->request_fn for request-based dm. | |
1657 | * Called with the queue lock held. | |
1658 | */ | |
1659 | static void dm_request_fn(struct request_queue *q) | |
1660 | { | |
1661 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1662 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1663 | struct dm_target *ti; |
b4324fee | 1664 | struct request *rq, *clone; |
29e4013d | 1665 | sector_t pos; |
cec47e3d KU |
1666 | |
1667 | /* | |
b4324fee KU |
1668 | * For suspend, check blk_queue_stopped() and increment |
1669 | * ->pending within a single queue_lock not to increment the | |
1670 | * number of in-flight I/Os after the queue is stopped in | |
1671 | * dm_suspend(). | |
cec47e3d | 1672 | */ |
7eaceacc | 1673 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1674 | rq = blk_peek_request(q); |
1675 | if (!rq) | |
7eaceacc | 1676 | goto delay_and_out; |
cec47e3d | 1677 | |
29e4013d TH |
1678 | /* always use block 0 to find the target for flushes for now */ |
1679 | pos = 0; | |
1680 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1681 | pos = blk_rq_pos(rq); | |
1682 | ||
1683 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1684 | if (!dm_target_is_valid(ti)) { |
1685 | /* | |
1686 | * Must perform setup, that dm_done() requires, | |
1687 | * before calling dm_kill_unmapped_request | |
1688 | */ | |
1689 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1690 | clone = dm_start_request(md, rq); | |
1691 | dm_kill_unmapped_request(clone, -EIO); | |
1692 | continue; | |
1693 | } | |
d0bcb878 | 1694 | |
cec47e3d | 1695 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1696 | goto delay_and_out; |
cec47e3d | 1697 | |
ba1cbad9 | 1698 | clone = dm_start_request(md, rq); |
b4324fee | 1699 | |
cec47e3d | 1700 | spin_unlock(q->queue_lock); |
9eef87da KU |
1701 | if (map_request(ti, clone, md)) |
1702 | goto requeued; | |
1703 | ||
052189a2 KU |
1704 | BUG_ON(!irqs_disabled()); |
1705 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1706 | } |
1707 | ||
1708 | goto out; | |
1709 | ||
9eef87da | 1710 | requeued: |
052189a2 KU |
1711 | BUG_ON(!irqs_disabled()); |
1712 | spin_lock(q->queue_lock); | |
9eef87da | 1713 | |
7eaceacc JA |
1714 | delay_and_out: |
1715 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1716 | out: |
1717 | dm_table_put(map); | |
cec47e3d KU |
1718 | } |
1719 | ||
1720 | int dm_underlying_device_busy(struct request_queue *q) | |
1721 | { | |
1722 | return blk_lld_busy(q); | |
1723 | } | |
1724 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1725 | ||
1726 | static int dm_lld_busy(struct request_queue *q) | |
1727 | { | |
1728 | int r; | |
1729 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1730 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1731 | |
1732 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1733 | r = 1; | |
1734 | else | |
1735 | r = dm_table_any_busy_target(map); | |
1736 | ||
1737 | dm_table_put(map); | |
1738 | ||
1739 | return r; | |
1740 | } | |
1741 | ||
1da177e4 LT |
1742 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1743 | { | |
8a57dfc6 CS |
1744 | int r = bdi_bits; |
1745 | struct mapped_device *md = congested_data; | |
1746 | struct dm_table *map; | |
1da177e4 | 1747 | |
1eb787ec | 1748 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1749 | map = dm_get_live_table(md); |
8a57dfc6 | 1750 | if (map) { |
cec47e3d KU |
1751 | /* |
1752 | * Request-based dm cares about only own queue for | |
1753 | * the query about congestion status of request_queue | |
1754 | */ | |
1755 | if (dm_request_based(md)) | |
1756 | r = md->queue->backing_dev_info.state & | |
1757 | bdi_bits; | |
1758 | else | |
1759 | r = dm_table_any_congested(map, bdi_bits); | |
1760 | ||
8a57dfc6 CS |
1761 | dm_table_put(map); |
1762 | } | |
1763 | } | |
1764 | ||
1da177e4 LT |
1765 | return r; |
1766 | } | |
1767 | ||
1768 | /*----------------------------------------------------------------- | |
1769 | * An IDR is used to keep track of allocated minor numbers. | |
1770 | *---------------------------------------------------------------*/ | |
2b06cfff | 1771 | static void free_minor(int minor) |
1da177e4 | 1772 | { |
f32c10b0 | 1773 | spin_lock(&_minor_lock); |
1da177e4 | 1774 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1775 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1776 | } |
1777 | ||
1778 | /* | |
1779 | * See if the device with a specific minor # is free. | |
1780 | */ | |
cf13ab8e | 1781 | static int specific_minor(int minor) |
1da177e4 | 1782 | { |
c9d76be6 | 1783 | int r; |
1da177e4 LT |
1784 | |
1785 | if (minor >= (1 << MINORBITS)) | |
1786 | return -EINVAL; | |
1787 | ||
c9d76be6 | 1788 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1789 | spin_lock(&_minor_lock); |
1da177e4 | 1790 | |
c9d76be6 | 1791 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1792 | |
f32c10b0 | 1793 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1794 | idr_preload_end(); |
1795 | if (r < 0) | |
1796 | return r == -ENOSPC ? -EBUSY : r; | |
1797 | return 0; | |
1da177e4 LT |
1798 | } |
1799 | ||
cf13ab8e | 1800 | static int next_free_minor(int *minor) |
1da177e4 | 1801 | { |
c9d76be6 | 1802 | int r; |
62f75c2f | 1803 | |
c9d76be6 | 1804 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1805 | spin_lock(&_minor_lock); |
1da177e4 | 1806 | |
c9d76be6 | 1807 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1808 | |
f32c10b0 | 1809 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1810 | idr_preload_end(); |
1811 | if (r < 0) | |
1812 | return r; | |
1813 | *minor = r; | |
1814 | return 0; | |
1da177e4 LT |
1815 | } |
1816 | ||
83d5cde4 | 1817 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1818 | |
53d5914f MP |
1819 | static void dm_wq_work(struct work_struct *work); |
1820 | ||
4a0b4ddf MS |
1821 | static void dm_init_md_queue(struct mapped_device *md) |
1822 | { | |
1823 | /* | |
1824 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1825 | * devices. The type of this dm device has not been decided yet. | |
1826 | * The type is decided at the first table loading time. | |
1827 | * To prevent problematic device stacking, clear the queue flag | |
1828 | * for request stacking support until then. | |
1829 | * | |
1830 | * This queue is new, so no concurrency on the queue_flags. | |
1831 | */ | |
1832 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1833 | ||
1834 | md->queue->queuedata = md; | |
1835 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1836 | md->queue->backing_dev_info.congested_data = md; | |
1837 | blk_queue_make_request(md->queue, dm_request); | |
1838 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1839 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1840 | } | |
1841 | ||
1da177e4 LT |
1842 | /* |
1843 | * Allocate and initialise a blank device with a given minor. | |
1844 | */ | |
2b06cfff | 1845 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1846 | { |
1847 | int r; | |
cf13ab8e | 1848 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1849 | void *old_md; |
1da177e4 LT |
1850 | |
1851 | if (!md) { | |
1852 | DMWARN("unable to allocate device, out of memory."); | |
1853 | return NULL; | |
1854 | } | |
1855 | ||
10da4f79 | 1856 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1857 | goto bad_module_get; |
10da4f79 | 1858 | |
1da177e4 | 1859 | /* get a minor number for the dev */ |
2b06cfff | 1860 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1861 | r = next_free_minor(&minor); |
2b06cfff | 1862 | else |
cf13ab8e | 1863 | r = specific_minor(minor); |
1da177e4 | 1864 | if (r < 0) |
6ed7ade8 | 1865 | goto bad_minor; |
1da177e4 | 1866 | |
a5664dad | 1867 | md->type = DM_TYPE_NONE; |
2ca3310e | 1868 | init_rwsem(&md->io_lock); |
e61290a4 | 1869 | mutex_init(&md->suspend_lock); |
a5664dad | 1870 | mutex_init(&md->type_lock); |
022c2611 | 1871 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1872 | rwlock_init(&md->map_lock); |
1873 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1874 | atomic_set(&md->open_count, 0); |
1da177e4 | 1875 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1876 | atomic_set(&md->uevent_seq, 0); |
1877 | INIT_LIST_HEAD(&md->uevent_list); | |
1878 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1879 | |
4a0b4ddf | 1880 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1881 | if (!md->queue) |
6ed7ade8 | 1882 | goto bad_queue; |
1da177e4 | 1883 | |
4a0b4ddf | 1884 | dm_init_md_queue(md); |
9faf400f | 1885 | |
1da177e4 LT |
1886 | md->disk = alloc_disk(1); |
1887 | if (!md->disk) | |
6ed7ade8 | 1888 | goto bad_disk; |
1da177e4 | 1889 | |
316d315b NK |
1890 | atomic_set(&md->pending[0], 0); |
1891 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1892 | init_waitqueue_head(&md->wait); |
53d5914f | 1893 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1894 | init_waitqueue_head(&md->eventq); |
1895 | ||
1da177e4 LT |
1896 | md->disk->major = _major; |
1897 | md->disk->first_minor = minor; | |
1898 | md->disk->fops = &dm_blk_dops; | |
1899 | md->disk->queue = md->queue; | |
1900 | md->disk->private_data = md; | |
1901 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1902 | add_disk(md->disk); | |
7e51f257 | 1903 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1904 | |
9c4376de TH |
1905 | md->wq = alloc_workqueue("kdmflush", |
1906 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1907 | if (!md->wq) |
1908 | goto bad_thread; | |
1909 | ||
32a926da MP |
1910 | md->bdev = bdget_disk(md->disk, 0); |
1911 | if (!md->bdev) | |
1912 | goto bad_bdev; | |
1913 | ||
6a8736d1 TH |
1914 | bio_init(&md->flush_bio); |
1915 | md->flush_bio.bi_bdev = md->bdev; | |
1916 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1917 | ||
ba61fdd1 | 1918 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1919 | spin_lock(&_minor_lock); |
ba61fdd1 | 1920 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1921 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1922 | |
1923 | BUG_ON(old_md != MINOR_ALLOCED); | |
1924 | ||
1da177e4 LT |
1925 | return md; |
1926 | ||
32a926da MP |
1927 | bad_bdev: |
1928 | destroy_workqueue(md->wq); | |
304f3f6a | 1929 | bad_thread: |
03022c54 | 1930 | del_gendisk(md->disk); |
304f3f6a | 1931 | put_disk(md->disk); |
6ed7ade8 | 1932 | bad_disk: |
1312f40e | 1933 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1934 | bad_queue: |
1da177e4 | 1935 | free_minor(minor); |
6ed7ade8 | 1936 | bad_minor: |
10da4f79 | 1937 | module_put(THIS_MODULE); |
6ed7ade8 | 1938 | bad_module_get: |
1da177e4 LT |
1939 | kfree(md); |
1940 | return NULL; | |
1941 | } | |
1942 | ||
ae9da83f JN |
1943 | static void unlock_fs(struct mapped_device *md); |
1944 | ||
1da177e4 LT |
1945 | static void free_dev(struct mapped_device *md) |
1946 | { | |
f331c029 | 1947 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1948 | |
32a926da MP |
1949 | unlock_fs(md); |
1950 | bdput(md->bdev); | |
304f3f6a | 1951 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1952 | if (md->tio_pool) |
1953 | mempool_destroy(md->tio_pool); | |
1954 | if (md->io_pool) | |
1955 | mempool_destroy(md->io_pool); | |
1956 | if (md->bs) | |
1957 | bioset_free(md->bs); | |
9c47008d | 1958 | blk_integrity_unregister(md->disk); |
1da177e4 | 1959 | del_gendisk(md->disk); |
63d94e48 | 1960 | free_minor(minor); |
fba9f90e JM |
1961 | |
1962 | spin_lock(&_minor_lock); | |
1963 | md->disk->private_data = NULL; | |
1964 | spin_unlock(&_minor_lock); | |
1965 | ||
1da177e4 | 1966 | put_disk(md->disk); |
1312f40e | 1967 | blk_cleanup_queue(md->queue); |
10da4f79 | 1968 | module_put(THIS_MODULE); |
1da177e4 LT |
1969 | kfree(md); |
1970 | } | |
1971 | ||
e6ee8c0b KU |
1972 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1973 | { | |
c0820cf5 | 1974 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1975 | |
23e5083b | 1976 | if (md->bs) { |
16245bdc JN |
1977 | /* The md already has necessary mempools. */ |
1978 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
1979 | /* | |
1980 | * Reload bioset because front_pad may have changed | |
1981 | * because a different table was loaded. | |
1982 | */ | |
1983 | bioset_free(md->bs); | |
1984 | md->bs = p->bs; | |
1985 | p->bs = NULL; | |
1986 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
1987 | BUG_ON(!md->tio_pool); | |
1988 | /* | |
1989 | * There's no need to reload with request-based dm | |
1990 | * because the size of front_pad doesn't change. | |
1991 | * Note for future: If you are to reload bioset, | |
1992 | * prep-ed requests in the queue may refer | |
1993 | * to bio from the old bioset, so you must walk | |
1994 | * through the queue to unprep. | |
1995 | */ | |
1996 | } | |
e6ee8c0b | 1997 | goto out; |
c0820cf5 | 1998 | } |
e6ee8c0b | 1999 | |
e6ee8c0b KU |
2000 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); |
2001 | ||
2002 | md->io_pool = p->io_pool; | |
2003 | p->io_pool = NULL; | |
2004 | md->tio_pool = p->tio_pool; | |
2005 | p->tio_pool = NULL; | |
2006 | md->bs = p->bs; | |
2007 | p->bs = NULL; | |
2008 | ||
2009 | out: | |
2010 | /* mempool bind completed, now no need any mempools in the table */ | |
2011 | dm_table_free_md_mempools(t); | |
2012 | } | |
2013 | ||
1da177e4 LT |
2014 | /* |
2015 | * Bind a table to the device. | |
2016 | */ | |
2017 | static void event_callback(void *context) | |
2018 | { | |
7a8c3d3b MA |
2019 | unsigned long flags; |
2020 | LIST_HEAD(uevents); | |
1da177e4 LT |
2021 | struct mapped_device *md = (struct mapped_device *) context; |
2022 | ||
7a8c3d3b MA |
2023 | spin_lock_irqsave(&md->uevent_lock, flags); |
2024 | list_splice_init(&md->uevent_list, &uevents); | |
2025 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2026 | ||
ed9e1982 | 2027 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2028 | |
1da177e4 LT |
2029 | atomic_inc(&md->event_nr); |
2030 | wake_up(&md->eventq); | |
2031 | } | |
2032 | ||
c217649b MS |
2033 | /* |
2034 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2035 | */ | |
4e90188b | 2036 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2037 | { |
4e90188b | 2038 | set_capacity(md->disk, size); |
1da177e4 | 2039 | |
db8fef4f | 2040 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2041 | } |
2042 | ||
d5b9dd04 MP |
2043 | /* |
2044 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2045 | * | |
2046 | * If this function returns 0, then the device is either a non-dm | |
2047 | * device without a merge_bvec_fn, or it is a dm device that is | |
2048 | * able to split any bios it receives that are too big. | |
2049 | */ | |
2050 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2051 | { | |
2052 | struct mapped_device *dev_md; | |
2053 | ||
2054 | if (!q->merge_bvec_fn) | |
2055 | return 0; | |
2056 | ||
2057 | if (q->make_request_fn == dm_request) { | |
2058 | dev_md = q->queuedata; | |
2059 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2060 | return 0; | |
2061 | } | |
2062 | ||
2063 | return 1; | |
2064 | } | |
2065 | ||
2066 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2067 | struct dm_dev *dev, sector_t start, | |
2068 | sector_t len, void *data) | |
2069 | { | |
2070 | struct block_device *bdev = dev->bdev; | |
2071 | struct request_queue *q = bdev_get_queue(bdev); | |
2072 | ||
2073 | return dm_queue_merge_is_compulsory(q); | |
2074 | } | |
2075 | ||
2076 | /* | |
2077 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2078 | * on the properties of the underlying devices. | |
2079 | */ | |
2080 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2081 | { | |
2082 | unsigned i = 0; | |
2083 | struct dm_target *ti; | |
2084 | ||
2085 | while (i < dm_table_get_num_targets(table)) { | |
2086 | ti = dm_table_get_target(table, i++); | |
2087 | ||
2088 | if (ti->type->iterate_devices && | |
2089 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2090 | return 0; | |
2091 | } | |
2092 | ||
2093 | return 1; | |
2094 | } | |
2095 | ||
042d2a9b AK |
2096 | /* |
2097 | * Returns old map, which caller must destroy. | |
2098 | */ | |
2099 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2100 | struct queue_limits *limits) | |
1da177e4 | 2101 | { |
042d2a9b | 2102 | struct dm_table *old_map; |
165125e1 | 2103 | struct request_queue *q = md->queue; |
1da177e4 | 2104 | sector_t size; |
523d9297 | 2105 | unsigned long flags; |
d5b9dd04 | 2106 | int merge_is_optional; |
1da177e4 LT |
2107 | |
2108 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2109 | |
2110 | /* | |
2111 | * Wipe any geometry if the size of the table changed. | |
2112 | */ | |
2113 | if (size != get_capacity(md->disk)) | |
2114 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2115 | ||
32a926da | 2116 | __set_size(md, size); |
d5816876 | 2117 | |
2ca3310e AK |
2118 | dm_table_event_callback(t, event_callback, md); |
2119 | ||
e6ee8c0b KU |
2120 | /* |
2121 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2122 | * for request-based during suspension. So stop it to prevent | |
2123 | * I/O mapping before resume. | |
2124 | * This must be done before setting the queue restrictions, | |
2125 | * because request-based dm may be run just after the setting. | |
2126 | */ | |
2127 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2128 | stop_queue(q); | |
2129 | ||
2130 | __bind_mempools(md, t); | |
2131 | ||
d5b9dd04 MP |
2132 | merge_is_optional = dm_table_merge_is_optional(t); |
2133 | ||
523d9297 | 2134 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2135 | old_map = md->map; |
1da177e4 | 2136 | md->map = t; |
36a0456f AK |
2137 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2138 | ||
754c5fc7 | 2139 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2140 | if (merge_is_optional) |
2141 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2142 | else | |
2143 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2144 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2145 | |
042d2a9b | 2146 | return old_map; |
1da177e4 LT |
2147 | } |
2148 | ||
a7940155 AK |
2149 | /* |
2150 | * Returns unbound table for the caller to free. | |
2151 | */ | |
2152 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2153 | { |
2154 | struct dm_table *map = md->map; | |
523d9297 | 2155 | unsigned long flags; |
1da177e4 LT |
2156 | |
2157 | if (!map) | |
a7940155 | 2158 | return NULL; |
1da177e4 LT |
2159 | |
2160 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2161 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2162 | md->map = NULL; |
523d9297 | 2163 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2164 | |
2165 | return map; | |
1da177e4 LT |
2166 | } |
2167 | ||
2168 | /* | |
2169 | * Constructor for a new device. | |
2170 | */ | |
2b06cfff | 2171 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2172 | { |
2173 | struct mapped_device *md; | |
2174 | ||
2b06cfff | 2175 | md = alloc_dev(minor); |
1da177e4 LT |
2176 | if (!md) |
2177 | return -ENXIO; | |
2178 | ||
784aae73 MB |
2179 | dm_sysfs_init(md); |
2180 | ||
1da177e4 LT |
2181 | *result = md; |
2182 | return 0; | |
2183 | } | |
2184 | ||
a5664dad MS |
2185 | /* |
2186 | * Functions to manage md->type. | |
2187 | * All are required to hold md->type_lock. | |
2188 | */ | |
2189 | void dm_lock_md_type(struct mapped_device *md) | |
2190 | { | |
2191 | mutex_lock(&md->type_lock); | |
2192 | } | |
2193 | ||
2194 | void dm_unlock_md_type(struct mapped_device *md) | |
2195 | { | |
2196 | mutex_unlock(&md->type_lock); | |
2197 | } | |
2198 | ||
2199 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2200 | { | |
2201 | md->type = type; | |
2202 | } | |
2203 | ||
2204 | unsigned dm_get_md_type(struct mapped_device *md) | |
2205 | { | |
2206 | return md->type; | |
2207 | } | |
2208 | ||
36a0456f AK |
2209 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2210 | { | |
2211 | return md->immutable_target_type; | |
2212 | } | |
2213 | ||
4a0b4ddf MS |
2214 | /* |
2215 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2216 | */ | |
2217 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2218 | { | |
2219 | struct request_queue *q = NULL; | |
2220 | ||
2221 | if (md->queue->elevator) | |
2222 | return 1; | |
2223 | ||
2224 | /* Fully initialize the queue */ | |
2225 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2226 | if (!q) | |
2227 | return 0; | |
2228 | ||
2229 | md->queue = q; | |
4a0b4ddf MS |
2230 | dm_init_md_queue(md); |
2231 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2232 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2233 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2234 | |
2235 | elv_register_queue(md->queue); | |
2236 | ||
2237 | return 1; | |
2238 | } | |
2239 | ||
2240 | /* | |
2241 | * Setup the DM device's queue based on md's type | |
2242 | */ | |
2243 | int dm_setup_md_queue(struct mapped_device *md) | |
2244 | { | |
2245 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2246 | !dm_init_request_based_queue(md)) { | |
2247 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2248 | return -EINVAL; | |
2249 | } | |
2250 | ||
2251 | return 0; | |
2252 | } | |
2253 | ||
637842cf | 2254 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2255 | { |
2256 | struct mapped_device *md; | |
1da177e4 LT |
2257 | unsigned minor = MINOR(dev); |
2258 | ||
2259 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2260 | return NULL; | |
2261 | ||
f32c10b0 | 2262 | spin_lock(&_minor_lock); |
1da177e4 LT |
2263 | |
2264 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2265 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2266 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2267 | dm_deleting_md(md) || |
17b2f66f | 2268 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2269 | md = NULL; |
fba9f90e JM |
2270 | goto out; |
2271 | } | |
1da177e4 | 2272 | |
fba9f90e | 2273 | out: |
f32c10b0 | 2274 | spin_unlock(&_minor_lock); |
1da177e4 | 2275 | |
637842cf DT |
2276 | return md; |
2277 | } | |
2278 | ||
d229a958 DT |
2279 | struct mapped_device *dm_get_md(dev_t dev) |
2280 | { | |
2281 | struct mapped_device *md = dm_find_md(dev); | |
2282 | ||
2283 | if (md) | |
2284 | dm_get(md); | |
2285 | ||
2286 | return md; | |
2287 | } | |
3cf2e4ba | 2288 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2289 | |
9ade92a9 | 2290 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2291 | { |
9ade92a9 | 2292 | return md->interface_ptr; |
1da177e4 LT |
2293 | } |
2294 | ||
2295 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2296 | { | |
2297 | md->interface_ptr = ptr; | |
2298 | } | |
2299 | ||
2300 | void dm_get(struct mapped_device *md) | |
2301 | { | |
2302 | atomic_inc(&md->holders); | |
3f77316d | 2303 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2304 | } |
2305 | ||
72d94861 AK |
2306 | const char *dm_device_name(struct mapped_device *md) |
2307 | { | |
2308 | return md->name; | |
2309 | } | |
2310 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2311 | ||
3f77316d | 2312 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2313 | { |
1134e5ae | 2314 | struct dm_table *map; |
1da177e4 | 2315 | |
3f77316d | 2316 | might_sleep(); |
fba9f90e | 2317 | |
3f77316d KU |
2318 | spin_lock(&_minor_lock); |
2319 | map = dm_get_live_table(md); | |
2320 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2321 | set_bit(DMF_FREEING, &md->flags); | |
2322 | spin_unlock(&_minor_lock); | |
2323 | ||
2324 | if (!dm_suspended_md(md)) { | |
2325 | dm_table_presuspend_targets(map); | |
2326 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2327 | } |
3f77316d KU |
2328 | |
2329 | /* | |
2330 | * Rare, but there may be I/O requests still going to complete, | |
2331 | * for example. Wait for all references to disappear. | |
2332 | * No one should increment the reference count of the mapped_device, | |
2333 | * after the mapped_device state becomes DMF_FREEING. | |
2334 | */ | |
2335 | if (wait) | |
2336 | while (atomic_read(&md->holders)) | |
2337 | msleep(1); | |
2338 | else if (atomic_read(&md->holders)) | |
2339 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2340 | dm_device_name(md), atomic_read(&md->holders)); | |
2341 | ||
2342 | dm_sysfs_exit(md); | |
2343 | dm_table_put(map); | |
2344 | dm_table_destroy(__unbind(md)); | |
2345 | free_dev(md); | |
2346 | } | |
2347 | ||
2348 | void dm_destroy(struct mapped_device *md) | |
2349 | { | |
2350 | __dm_destroy(md, true); | |
2351 | } | |
2352 | ||
2353 | void dm_destroy_immediate(struct mapped_device *md) | |
2354 | { | |
2355 | __dm_destroy(md, false); | |
2356 | } | |
2357 | ||
2358 | void dm_put(struct mapped_device *md) | |
2359 | { | |
2360 | atomic_dec(&md->holders); | |
1da177e4 | 2361 | } |
79eb885c | 2362 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2363 | |
401600df | 2364 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2365 | { |
2366 | int r = 0; | |
b44ebeb0 MP |
2367 | DECLARE_WAITQUEUE(wait, current); |
2368 | ||
b44ebeb0 | 2369 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2370 | |
2371 | while (1) { | |
401600df | 2372 | set_current_state(interruptible); |
46125c1c | 2373 | |
b4324fee | 2374 | if (!md_in_flight(md)) |
46125c1c MB |
2375 | break; |
2376 | ||
401600df MP |
2377 | if (interruptible == TASK_INTERRUPTIBLE && |
2378 | signal_pending(current)) { | |
46125c1c MB |
2379 | r = -EINTR; |
2380 | break; | |
2381 | } | |
2382 | ||
2383 | io_schedule(); | |
2384 | } | |
2385 | set_current_state(TASK_RUNNING); | |
2386 | ||
b44ebeb0 MP |
2387 | remove_wait_queue(&md->wait, &wait); |
2388 | ||
46125c1c MB |
2389 | return r; |
2390 | } | |
2391 | ||
1da177e4 LT |
2392 | /* |
2393 | * Process the deferred bios | |
2394 | */ | |
ef208587 | 2395 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2396 | { |
ef208587 MP |
2397 | struct mapped_device *md = container_of(work, struct mapped_device, |
2398 | work); | |
6d6f10df | 2399 | struct bio *c; |
1da177e4 | 2400 | |
6a8736d1 | 2401 | down_read(&md->io_lock); |
ef208587 | 2402 | |
3b00b203 | 2403 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2404 | spin_lock_irq(&md->deferred_lock); |
2405 | c = bio_list_pop(&md->deferred); | |
2406 | spin_unlock_irq(&md->deferred_lock); | |
2407 | ||
6a8736d1 | 2408 | if (!c) |
df12ee99 | 2409 | break; |
022c2611 | 2410 | |
6a8736d1 | 2411 | up_read(&md->io_lock); |
3b00b203 | 2412 | |
e6ee8c0b KU |
2413 | if (dm_request_based(md)) |
2414 | generic_make_request(c); | |
6a8736d1 TH |
2415 | else |
2416 | __split_and_process_bio(md, c); | |
3b00b203 | 2417 | |
6a8736d1 | 2418 | down_read(&md->io_lock); |
022c2611 | 2419 | } |
73d410c0 | 2420 | |
6a8736d1 | 2421 | up_read(&md->io_lock); |
1da177e4 LT |
2422 | } |
2423 | ||
9a1fb464 | 2424 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2425 | { |
3b00b203 MP |
2426 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2427 | smp_mb__after_clear_bit(); | |
53d5914f | 2428 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2429 | } |
2430 | ||
1da177e4 | 2431 | /* |
042d2a9b | 2432 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2433 | */ |
042d2a9b | 2434 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2435 | { |
87eb5b21 | 2436 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2437 | struct queue_limits limits; |
042d2a9b | 2438 | int r; |
1da177e4 | 2439 | |
e61290a4 | 2440 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2441 | |
2442 | /* device must be suspended */ | |
4f186f8b | 2443 | if (!dm_suspended_md(md)) |
93c534ae | 2444 | goto out; |
1da177e4 | 2445 | |
3ae70656 MS |
2446 | /* |
2447 | * If the new table has no data devices, retain the existing limits. | |
2448 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2449 | * then new I/O is queued based on these limits, and then some paths | |
2450 | * reappear. | |
2451 | */ | |
2452 | if (dm_table_has_no_data_devices(table)) { | |
2453 | live_map = dm_get_live_table(md); | |
2454 | if (live_map) | |
2455 | limits = md->queue->limits; | |
2456 | dm_table_put(live_map); | |
2457 | } | |
2458 | ||
87eb5b21 MC |
2459 | if (!live_map) { |
2460 | r = dm_calculate_queue_limits(table, &limits); | |
2461 | if (r) { | |
2462 | map = ERR_PTR(r); | |
2463 | goto out; | |
2464 | } | |
042d2a9b | 2465 | } |
754c5fc7 | 2466 | |
042d2a9b | 2467 | map = __bind(md, table, &limits); |
1da177e4 | 2468 | |
93c534ae | 2469 | out: |
e61290a4 | 2470 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2471 | return map; |
1da177e4 LT |
2472 | } |
2473 | ||
2474 | /* | |
2475 | * Functions to lock and unlock any filesystem running on the | |
2476 | * device. | |
2477 | */ | |
2ca3310e | 2478 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2479 | { |
e39e2e95 | 2480 | int r; |
1da177e4 LT |
2481 | |
2482 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2483 | |
db8fef4f | 2484 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2485 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2486 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2487 | md->frozen_sb = NULL; |
2488 | return r; | |
dfbe03f6 AK |
2489 | } |
2490 | ||
aa8d7c2f AK |
2491 | set_bit(DMF_FROZEN, &md->flags); |
2492 | ||
1da177e4 LT |
2493 | return 0; |
2494 | } | |
2495 | ||
2ca3310e | 2496 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2497 | { |
aa8d7c2f AK |
2498 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2499 | return; | |
2500 | ||
db8fef4f | 2501 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2502 | md->frozen_sb = NULL; |
aa8d7c2f | 2503 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2504 | } |
2505 | ||
2506 | /* | |
2507 | * We need to be able to change a mapping table under a mounted | |
2508 | * filesystem. For example we might want to move some data in | |
2509 | * the background. Before the table can be swapped with | |
2510 | * dm_bind_table, dm_suspend must be called to flush any in | |
2511 | * flight bios and ensure that any further io gets deferred. | |
2512 | */ | |
cec47e3d KU |
2513 | /* |
2514 | * Suspend mechanism in request-based dm. | |
2515 | * | |
9f518b27 KU |
2516 | * 1. Flush all I/Os by lock_fs() if needed. |
2517 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2518 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2519 | * |
9f518b27 | 2520 | * To abort suspend, start the request_queue. |
cec47e3d | 2521 | */ |
a3d77d35 | 2522 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2523 | { |
2ca3310e | 2524 | struct dm_table *map = NULL; |
46125c1c | 2525 | int r = 0; |
a3d77d35 | 2526 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2527 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2528 | |
e61290a4 | 2529 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2530 | |
4f186f8b | 2531 | if (dm_suspended_md(md)) { |
73d410c0 | 2532 | r = -EINVAL; |
d287483d | 2533 | goto out_unlock; |
73d410c0 | 2534 | } |
1da177e4 | 2535 | |
7c666411 | 2536 | map = dm_get_live_table(md); |
1da177e4 | 2537 | |
2e93ccc1 KU |
2538 | /* |
2539 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2540 | * This flag is cleared before dm_suspend returns. | |
2541 | */ | |
2542 | if (noflush) | |
2543 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2544 | ||
cf222b37 AK |
2545 | /* This does not get reverted if there's an error later. */ |
2546 | dm_table_presuspend_targets(map); | |
2547 | ||
32a926da | 2548 | /* |
9f518b27 KU |
2549 | * Flush I/O to the device. |
2550 | * Any I/O submitted after lock_fs() may not be flushed. | |
2551 | * noflush takes precedence over do_lockfs. | |
2552 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2553 | */ |
2554 | if (!noflush && do_lockfs) { | |
2555 | r = lock_fs(md); | |
2556 | if (r) | |
f431d966 | 2557 | goto out; |
aa8d7c2f | 2558 | } |
1da177e4 LT |
2559 | |
2560 | /* | |
3b00b203 MP |
2561 | * Here we must make sure that no processes are submitting requests |
2562 | * to target drivers i.e. no one may be executing | |
2563 | * __split_and_process_bio. This is called from dm_request and | |
2564 | * dm_wq_work. | |
2565 | * | |
2566 | * To get all processes out of __split_and_process_bio in dm_request, | |
2567 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2568 | * __split_and_process_bio from dm_request and quiesce the thread |
2569 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2570 | * flush_workqueue(md->wq). | |
1da177e4 | 2571 | */ |
2ca3310e | 2572 | down_write(&md->io_lock); |
1eb787ec | 2573 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2574 | up_write(&md->io_lock); |
1da177e4 | 2575 | |
d0bcb878 | 2576 | /* |
29e4013d TH |
2577 | * Stop md->queue before flushing md->wq in case request-based |
2578 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2579 | */ |
cec47e3d | 2580 | if (dm_request_based(md)) |
9f518b27 | 2581 | stop_queue(md->queue); |
cec47e3d | 2582 | |
d0bcb878 KU |
2583 | flush_workqueue(md->wq); |
2584 | ||
1da177e4 | 2585 | /* |
3b00b203 MP |
2586 | * At this point no more requests are entering target request routines. |
2587 | * We call dm_wait_for_completion to wait for all existing requests | |
2588 | * to finish. | |
1da177e4 | 2589 | */ |
401600df | 2590 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2591 | |
2ca3310e | 2592 | down_write(&md->io_lock); |
6d6f10df | 2593 | if (noflush) |
022c2611 | 2594 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2595 | up_write(&md->io_lock); |
2e93ccc1 | 2596 | |
1da177e4 | 2597 | /* were we interrupted ? */ |
46125c1c | 2598 | if (r < 0) { |
9a1fb464 | 2599 | dm_queue_flush(md); |
73d410c0 | 2600 | |
cec47e3d | 2601 | if (dm_request_based(md)) |
9f518b27 | 2602 | start_queue(md->queue); |
cec47e3d | 2603 | |
2ca3310e | 2604 | unlock_fs(md); |
2e93ccc1 | 2605 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2606 | } |
1da177e4 | 2607 | |
3b00b203 MP |
2608 | /* |
2609 | * If dm_wait_for_completion returned 0, the device is completely | |
2610 | * quiescent now. There is no request-processing activity. All new | |
2611 | * requests are being added to md->deferred list. | |
2612 | */ | |
2613 | ||
2ca3310e | 2614 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2615 | |
4d4471cb KU |
2616 | dm_table_postsuspend_targets(map); |
2617 | ||
2ca3310e AK |
2618 | out: |
2619 | dm_table_put(map); | |
d287483d AK |
2620 | |
2621 | out_unlock: | |
e61290a4 | 2622 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2623 | return r; |
1da177e4 LT |
2624 | } |
2625 | ||
2626 | int dm_resume(struct mapped_device *md) | |
2627 | { | |
cf222b37 | 2628 | int r = -EINVAL; |
cf222b37 | 2629 | struct dm_table *map = NULL; |
1da177e4 | 2630 | |
e61290a4 | 2631 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2632 | if (!dm_suspended_md(md)) |
cf222b37 | 2633 | goto out; |
cf222b37 | 2634 | |
7c666411 | 2635 | map = dm_get_live_table(md); |
2ca3310e | 2636 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2637 | goto out; |
1da177e4 | 2638 | |
8757b776 MB |
2639 | r = dm_table_resume_targets(map); |
2640 | if (r) | |
2641 | goto out; | |
2ca3310e | 2642 | |
9a1fb464 | 2643 | dm_queue_flush(md); |
2ca3310e | 2644 | |
cec47e3d KU |
2645 | /* |
2646 | * Flushing deferred I/Os must be done after targets are resumed | |
2647 | * so that mapping of targets can work correctly. | |
2648 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2649 | */ | |
2650 | if (dm_request_based(md)) | |
2651 | start_queue(md->queue); | |
2652 | ||
2ca3310e AK |
2653 | unlock_fs(md); |
2654 | ||
2655 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2656 | ||
cf222b37 AK |
2657 | r = 0; |
2658 | out: | |
2659 | dm_table_put(map); | |
e61290a4 | 2660 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2661 | |
cf222b37 | 2662 | return r; |
1da177e4 LT |
2663 | } |
2664 | ||
2665 | /*----------------------------------------------------------------- | |
2666 | * Event notification. | |
2667 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2668 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2669 | unsigned cookie) |
69267a30 | 2670 | { |
60935eb2 MB |
2671 | char udev_cookie[DM_COOKIE_LENGTH]; |
2672 | char *envp[] = { udev_cookie, NULL }; | |
2673 | ||
2674 | if (!cookie) | |
3abf85b5 | 2675 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2676 | else { |
2677 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2678 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2679 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2680 | action, envp); | |
60935eb2 | 2681 | } |
69267a30 AK |
2682 | } |
2683 | ||
7a8c3d3b MA |
2684 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2685 | { | |
2686 | return atomic_add_return(1, &md->uevent_seq); | |
2687 | } | |
2688 | ||
1da177e4 LT |
2689 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2690 | { | |
2691 | return atomic_read(&md->event_nr); | |
2692 | } | |
2693 | ||
2694 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2695 | { | |
2696 | return wait_event_interruptible(md->eventq, | |
2697 | (event_nr != atomic_read(&md->event_nr))); | |
2698 | } | |
2699 | ||
7a8c3d3b MA |
2700 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2701 | { | |
2702 | unsigned long flags; | |
2703 | ||
2704 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2705 | list_add(elist, &md->uevent_list); | |
2706 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2707 | } | |
2708 | ||
1da177e4 LT |
2709 | /* |
2710 | * The gendisk is only valid as long as you have a reference | |
2711 | * count on 'md'. | |
2712 | */ | |
2713 | struct gendisk *dm_disk(struct mapped_device *md) | |
2714 | { | |
2715 | return md->disk; | |
2716 | } | |
2717 | ||
784aae73 MB |
2718 | struct kobject *dm_kobject(struct mapped_device *md) |
2719 | { | |
2720 | return &md->kobj; | |
2721 | } | |
2722 | ||
2723 | /* | |
2724 | * struct mapped_device should not be exported outside of dm.c | |
2725 | * so use this check to verify that kobj is part of md structure | |
2726 | */ | |
2727 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2728 | { | |
2729 | struct mapped_device *md; | |
2730 | ||
2731 | md = container_of(kobj, struct mapped_device, kobj); | |
2732 | if (&md->kobj != kobj) | |
2733 | return NULL; | |
2734 | ||
4d89b7b4 | 2735 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2736 | dm_deleting_md(md)) |
4d89b7b4 MB |
2737 | return NULL; |
2738 | ||
784aae73 MB |
2739 | dm_get(md); |
2740 | return md; | |
2741 | } | |
2742 | ||
4f186f8b | 2743 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2744 | { |
2745 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2746 | } | |
2747 | ||
64dbce58 KU |
2748 | int dm_suspended(struct dm_target *ti) |
2749 | { | |
ecdb2e25 | 2750 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2751 | } |
2752 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2753 | ||
2e93ccc1 KU |
2754 | int dm_noflush_suspending(struct dm_target *ti) |
2755 | { | |
ecdb2e25 | 2756 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2757 | } |
2758 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2759 | ||
c0820cf5 | 2760 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b KU |
2761 | { |
2762 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2763 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2764 | |
2765 | if (!pools) | |
2766 | return NULL; | |
2767 | ||
c0820cf5 MP |
2768 | per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io)); |
2769 | ||
23e5083b JN |
2770 | pools->io_pool = NULL; |
2771 | if (type == DM_TYPE_BIO_BASED) { | |
2772 | pools->io_pool = mempool_create_slab_pool(MIN_IOS, _io_cache); | |
2773 | if (!pools->io_pool) | |
2774 | goto free_pools_and_out; | |
2775 | } | |
e6ee8c0b | 2776 | |
dba14160 MP |
2777 | pools->tio_pool = NULL; |
2778 | if (type == DM_TYPE_REQUEST_BASED) { | |
2779 | pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2780 | if (!pools->tio_pool) | |
2781 | goto free_io_pool_and_out; | |
2782 | } | |
e6ee8c0b | 2783 | |
94818742 | 2784 | pools->bs = (type == DM_TYPE_BIO_BASED) ? |
dba14160 | 2785 | bioset_create(pool_size, |
c0820cf5 | 2786 | per_bio_data_size + offsetof(struct dm_target_io, clone)) : |
94818742 KO |
2787 | bioset_create(pool_size, |
2788 | offsetof(struct dm_rq_clone_bio_info, clone)); | |
e6ee8c0b KU |
2789 | if (!pools->bs) |
2790 | goto free_tio_pool_and_out; | |
2791 | ||
a91a2785 MP |
2792 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2793 | goto free_bioset_and_out; | |
2794 | ||
e6ee8c0b KU |
2795 | return pools; |
2796 | ||
a91a2785 MP |
2797 | free_bioset_and_out: |
2798 | bioset_free(pools->bs); | |
2799 | ||
e6ee8c0b | 2800 | free_tio_pool_and_out: |
dba14160 MP |
2801 | if (pools->tio_pool) |
2802 | mempool_destroy(pools->tio_pool); | |
e6ee8c0b KU |
2803 | |
2804 | free_io_pool_and_out: | |
23e5083b JN |
2805 | if (pools->io_pool) |
2806 | mempool_destroy(pools->io_pool); | |
e6ee8c0b KU |
2807 | |
2808 | free_pools_and_out: | |
2809 | kfree(pools); | |
2810 | ||
2811 | return NULL; | |
2812 | } | |
2813 | ||
2814 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2815 | { | |
2816 | if (!pools) | |
2817 | return; | |
2818 | ||
2819 | if (pools->io_pool) | |
2820 | mempool_destroy(pools->io_pool); | |
2821 | ||
2822 | if (pools->tio_pool) | |
2823 | mempool_destroy(pools->tio_pool); | |
2824 | ||
2825 | if (pools->bs) | |
2826 | bioset_free(pools->bs); | |
2827 | ||
2828 | kfree(pools); | |
2829 | } | |
2830 | ||
83d5cde4 | 2831 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2832 | .open = dm_blk_open, |
2833 | .release = dm_blk_close, | |
aa129a22 | 2834 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2835 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2836 | .owner = THIS_MODULE |
2837 | }; | |
2838 | ||
2839 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2840 | ||
2841 | /* | |
2842 | * module hooks | |
2843 | */ | |
2844 | module_init(dm_init); | |
2845 | module_exit(dm_exit); | |
2846 | ||
2847 | module_param(major, uint, 0); | |
2848 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2849 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2850 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2851 | MODULE_LICENSE("GPL"); |