Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
1da177e4 LT |
63 | }; |
64 | ||
65 | /* | |
8fbf26ad | 66 | * For bio-based dm. |
1da177e4 LT |
67 | * One of these is allocated per target within a bio. Hopefully |
68 | * this will be simplified out one day. | |
69 | */ | |
028867ac | 70 | struct dm_target_io { |
1da177e4 LT |
71 | struct dm_io *io; |
72 | struct dm_target *ti; | |
73 | union map_info info; | |
dba14160 | 74 | struct bio clone; |
1da177e4 LT |
75 | }; |
76 | ||
8fbf26ad KU |
77 | /* |
78 | * For request-based dm. | |
79 | * One of these is allocated per request. | |
80 | */ | |
81 | struct dm_rq_target_io { | |
82 | struct mapped_device *md; | |
83 | struct dm_target *ti; | |
84 | struct request *orig, clone; | |
85 | int error; | |
86 | union map_info info; | |
87 | }; | |
88 | ||
89 | /* | |
94818742 KO |
90 | * For request-based dm - the bio clones we allocate are embedded in these |
91 | * structs. | |
92 | * | |
93 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
94 | * the bioset is created - this means the bio has to come at the end of the | |
95 | * struct. | |
8fbf26ad KU |
96 | */ |
97 | struct dm_rq_clone_bio_info { | |
98 | struct bio *orig; | |
cec47e3d | 99 | struct dm_rq_target_io *tio; |
94818742 | 100 | struct bio clone; |
8fbf26ad KU |
101 | }; |
102 | ||
1da177e4 LT |
103 | union map_info *dm_get_mapinfo(struct bio *bio) |
104 | { | |
17b2f66f | 105 | if (bio && bio->bi_private) |
028867ac | 106 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 107 | return NULL; |
1da177e4 LT |
108 | } |
109 | ||
cec47e3d KU |
110 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
111 | { | |
112 | if (rq && rq->end_io_data) | |
113 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
114 | return NULL; | |
115 | } | |
116 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
117 | ||
ba61fdd1 JM |
118 | #define MINOR_ALLOCED ((void *)-1) |
119 | ||
1da177e4 LT |
120 | /* |
121 | * Bits for the md->flags field. | |
122 | */ | |
1eb787ec | 123 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 124 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 125 | #define DMF_FROZEN 2 |
fba9f90e | 126 | #define DMF_FREEING 3 |
5c6bd75d | 127 | #define DMF_DELETING 4 |
2e93ccc1 | 128 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 129 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 130 | |
304f3f6a MB |
131 | /* |
132 | * Work processed by per-device workqueue. | |
133 | */ | |
1da177e4 | 134 | struct mapped_device { |
2ca3310e | 135 | struct rw_semaphore io_lock; |
e61290a4 | 136 | struct mutex suspend_lock; |
1da177e4 LT |
137 | rwlock_t map_lock; |
138 | atomic_t holders; | |
5c6bd75d | 139 | atomic_t open_count; |
1da177e4 LT |
140 | |
141 | unsigned long flags; | |
142 | ||
165125e1 | 143 | struct request_queue *queue; |
a5664dad | 144 | unsigned type; |
4a0b4ddf | 145 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
146 | struct mutex type_lock; |
147 | ||
36a0456f AK |
148 | struct target_type *immutable_target_type; |
149 | ||
1da177e4 | 150 | struct gendisk *disk; |
7e51f257 | 151 | char name[16]; |
1da177e4 LT |
152 | |
153 | void *interface_ptr; | |
154 | ||
155 | /* | |
156 | * A list of ios that arrived while we were suspended. | |
157 | */ | |
316d315b | 158 | atomic_t pending[2]; |
1da177e4 | 159 | wait_queue_head_t wait; |
53d5914f | 160 | struct work_struct work; |
74859364 | 161 | struct bio_list deferred; |
022c2611 | 162 | spinlock_t deferred_lock; |
1da177e4 | 163 | |
af7e466a | 164 | /* |
29e4013d | 165 | * Processing queue (flush) |
304f3f6a MB |
166 | */ |
167 | struct workqueue_struct *wq; | |
168 | ||
1da177e4 LT |
169 | /* |
170 | * The current mapping. | |
171 | */ | |
172 | struct dm_table *map; | |
173 | ||
174 | /* | |
175 | * io objects are allocated from here. | |
176 | */ | |
177 | mempool_t *io_pool; | |
178 | mempool_t *tio_pool; | |
179 | ||
9faf400f SB |
180 | struct bio_set *bs; |
181 | ||
1da177e4 LT |
182 | /* |
183 | * Event handling. | |
184 | */ | |
185 | atomic_t event_nr; | |
186 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
187 | atomic_t uevent_seq; |
188 | struct list_head uevent_list; | |
189 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
190 | |
191 | /* | |
192 | * freeze/thaw support require holding onto a super block | |
193 | */ | |
194 | struct super_block *frozen_sb; | |
db8fef4f | 195 | struct block_device *bdev; |
3ac51e74 DW |
196 | |
197 | /* forced geometry settings */ | |
198 | struct hd_geometry geometry; | |
784aae73 MB |
199 | |
200 | /* sysfs handle */ | |
201 | struct kobject kobj; | |
52b1fd5a | 202 | |
d87f4c14 TH |
203 | /* zero-length flush that will be cloned and submitted to targets */ |
204 | struct bio flush_bio; | |
1da177e4 LT |
205 | }; |
206 | ||
e6ee8c0b KU |
207 | /* |
208 | * For mempools pre-allocation at the table loading time. | |
209 | */ | |
210 | struct dm_md_mempools { | |
211 | mempool_t *io_pool; | |
212 | mempool_t *tio_pool; | |
213 | struct bio_set *bs; | |
214 | }; | |
215 | ||
1da177e4 | 216 | #define MIN_IOS 256 |
e18b890b | 217 | static struct kmem_cache *_io_cache; |
8fbf26ad | 218 | static struct kmem_cache *_rq_tio_cache; |
94818742 KO |
219 | |
220 | /* | |
221 | * Unused now, and needs to be deleted. But since io_pool is overloaded and it's | |
222 | * still used for _io_cache, I'm leaving this for a later cleanup | |
223 | */ | |
8fbf26ad | 224 | static struct kmem_cache *_rq_bio_info_cache; |
1da177e4 | 225 | |
1da177e4 LT |
226 | static int __init local_init(void) |
227 | { | |
51157b4a | 228 | int r = -ENOMEM; |
1da177e4 | 229 | |
1da177e4 | 230 | /* allocate a slab for the dm_ios */ |
028867ac | 231 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 232 | if (!_io_cache) |
51157b4a | 233 | return r; |
1da177e4 | 234 | |
8fbf26ad KU |
235 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
236 | if (!_rq_tio_cache) | |
dba14160 | 237 | goto out_free_io_cache; |
8fbf26ad KU |
238 | |
239 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
240 | if (!_rq_bio_info_cache) | |
241 | goto out_free_rq_tio_cache; | |
242 | ||
51e5b2bd | 243 | r = dm_uevent_init(); |
51157b4a | 244 | if (r) |
8fbf26ad | 245 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 246 | |
1da177e4 LT |
247 | _major = major; |
248 | r = register_blkdev(_major, _name); | |
51157b4a KU |
249 | if (r < 0) |
250 | goto out_uevent_exit; | |
1da177e4 LT |
251 | |
252 | if (!_major) | |
253 | _major = r; | |
254 | ||
255 | return 0; | |
51157b4a KU |
256 | |
257 | out_uevent_exit: | |
258 | dm_uevent_exit(); | |
8fbf26ad KU |
259 | out_free_rq_bio_info_cache: |
260 | kmem_cache_destroy(_rq_bio_info_cache); | |
261 | out_free_rq_tio_cache: | |
262 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
263 | out_free_io_cache: |
264 | kmem_cache_destroy(_io_cache); | |
265 | ||
266 | return r; | |
1da177e4 LT |
267 | } |
268 | ||
269 | static void local_exit(void) | |
270 | { | |
8fbf26ad KU |
271 | kmem_cache_destroy(_rq_bio_info_cache); |
272 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 | 273 | kmem_cache_destroy(_io_cache); |
00d59405 | 274 | unregister_blkdev(_major, _name); |
51e5b2bd | 275 | dm_uevent_exit(); |
1da177e4 LT |
276 | |
277 | _major = 0; | |
278 | ||
279 | DMINFO("cleaned up"); | |
280 | } | |
281 | ||
b9249e55 | 282 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
283 | local_init, |
284 | dm_target_init, | |
285 | dm_linear_init, | |
286 | dm_stripe_init, | |
952b3557 | 287 | dm_io_init, |
945fa4d2 | 288 | dm_kcopyd_init, |
1da177e4 LT |
289 | dm_interface_init, |
290 | }; | |
291 | ||
b9249e55 | 292 | static void (*_exits[])(void) = { |
1da177e4 LT |
293 | local_exit, |
294 | dm_target_exit, | |
295 | dm_linear_exit, | |
296 | dm_stripe_exit, | |
952b3557 | 297 | dm_io_exit, |
945fa4d2 | 298 | dm_kcopyd_exit, |
1da177e4 LT |
299 | dm_interface_exit, |
300 | }; | |
301 | ||
302 | static int __init dm_init(void) | |
303 | { | |
304 | const int count = ARRAY_SIZE(_inits); | |
305 | ||
306 | int r, i; | |
307 | ||
308 | for (i = 0; i < count; i++) { | |
309 | r = _inits[i](); | |
310 | if (r) | |
311 | goto bad; | |
312 | } | |
313 | ||
314 | return 0; | |
315 | ||
316 | bad: | |
317 | while (i--) | |
318 | _exits[i](); | |
319 | ||
320 | return r; | |
321 | } | |
322 | ||
323 | static void __exit dm_exit(void) | |
324 | { | |
325 | int i = ARRAY_SIZE(_exits); | |
326 | ||
327 | while (i--) | |
328 | _exits[i](); | |
d15b774c AK |
329 | |
330 | /* | |
331 | * Should be empty by this point. | |
332 | */ | |
333 | idr_remove_all(&_minor_idr); | |
334 | idr_destroy(&_minor_idr); | |
1da177e4 LT |
335 | } |
336 | ||
337 | /* | |
338 | * Block device functions | |
339 | */ | |
432a212c MA |
340 | int dm_deleting_md(struct mapped_device *md) |
341 | { | |
342 | return test_bit(DMF_DELETING, &md->flags); | |
343 | } | |
344 | ||
fe5f9f2c | 345 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
346 | { |
347 | struct mapped_device *md; | |
348 | ||
fba9f90e JM |
349 | spin_lock(&_minor_lock); |
350 | ||
fe5f9f2c | 351 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
352 | if (!md) |
353 | goto out; | |
354 | ||
5c6bd75d | 355 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 356 | dm_deleting_md(md)) { |
fba9f90e JM |
357 | md = NULL; |
358 | goto out; | |
359 | } | |
360 | ||
1da177e4 | 361 | dm_get(md); |
5c6bd75d | 362 | atomic_inc(&md->open_count); |
fba9f90e JM |
363 | |
364 | out: | |
365 | spin_unlock(&_minor_lock); | |
366 | ||
367 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
368 | } |
369 | ||
fe5f9f2c | 370 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 371 | { |
fe5f9f2c | 372 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 373 | |
4a1aeb98 MB |
374 | spin_lock(&_minor_lock); |
375 | ||
5c6bd75d | 376 | atomic_dec(&md->open_count); |
1da177e4 | 377 | dm_put(md); |
4a1aeb98 MB |
378 | |
379 | spin_unlock(&_minor_lock); | |
6e9624b8 | 380 | |
1da177e4 LT |
381 | return 0; |
382 | } | |
383 | ||
5c6bd75d AK |
384 | int dm_open_count(struct mapped_device *md) |
385 | { | |
386 | return atomic_read(&md->open_count); | |
387 | } | |
388 | ||
389 | /* | |
390 | * Guarantees nothing is using the device before it's deleted. | |
391 | */ | |
392 | int dm_lock_for_deletion(struct mapped_device *md) | |
393 | { | |
394 | int r = 0; | |
395 | ||
396 | spin_lock(&_minor_lock); | |
397 | ||
398 | if (dm_open_count(md)) | |
399 | r = -EBUSY; | |
400 | else | |
401 | set_bit(DMF_DELETING, &md->flags); | |
402 | ||
403 | spin_unlock(&_minor_lock); | |
404 | ||
405 | return r; | |
406 | } | |
407 | ||
3ac51e74 DW |
408 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
409 | { | |
410 | struct mapped_device *md = bdev->bd_disk->private_data; | |
411 | ||
412 | return dm_get_geometry(md, geo); | |
413 | } | |
414 | ||
fe5f9f2c | 415 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
416 | unsigned int cmd, unsigned long arg) |
417 | { | |
fe5f9f2c | 418 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 419 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
420 | struct dm_target *tgt; |
421 | int r = -ENOTTY; | |
422 | ||
aa129a22 MB |
423 | if (!map || !dm_table_get_size(map)) |
424 | goto out; | |
425 | ||
426 | /* We only support devices that have a single target */ | |
427 | if (dm_table_get_num_targets(map) != 1) | |
428 | goto out; | |
429 | ||
430 | tgt = dm_table_get_target(map, 0); | |
431 | ||
4f186f8b | 432 | if (dm_suspended_md(md)) { |
aa129a22 MB |
433 | r = -EAGAIN; |
434 | goto out; | |
435 | } | |
436 | ||
437 | if (tgt->type->ioctl) | |
647b3d00 | 438 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
439 | |
440 | out: | |
441 | dm_table_put(map); | |
442 | ||
aa129a22 MB |
443 | return r; |
444 | } | |
445 | ||
028867ac | 446 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
447 | { |
448 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
449 | } | |
450 | ||
028867ac | 451 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
452 | { |
453 | mempool_free(io, md->io_pool); | |
454 | } | |
455 | ||
028867ac | 456 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 457 | { |
dba14160 | 458 | bio_put(&tio->clone); |
1da177e4 LT |
459 | } |
460 | ||
08885643 KU |
461 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
462 | gfp_t gfp_mask) | |
cec47e3d | 463 | { |
08885643 | 464 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
465 | } |
466 | ||
467 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
468 | { | |
469 | mempool_free(tio, tio->md->tio_pool); | |
470 | } | |
471 | ||
90abb8c4 KU |
472 | static int md_in_flight(struct mapped_device *md) |
473 | { | |
474 | return atomic_read(&md->pending[READ]) + | |
475 | atomic_read(&md->pending[WRITE]); | |
476 | } | |
477 | ||
3eaf840e JNN |
478 | static void start_io_acct(struct dm_io *io) |
479 | { | |
480 | struct mapped_device *md = io->md; | |
c9959059 | 481 | int cpu; |
316d315b | 482 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
483 | |
484 | io->start_time = jiffies; | |
485 | ||
074a7aca TH |
486 | cpu = part_stat_lock(); |
487 | part_round_stats(cpu, &dm_disk(md)->part0); | |
488 | part_stat_unlock(); | |
1e9bb880 SL |
489 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
490 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
491 | } |
492 | ||
d221d2e7 | 493 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
494 | { |
495 | struct mapped_device *md = io->md; | |
496 | struct bio *bio = io->bio; | |
497 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 498 | int pending, cpu; |
3eaf840e JNN |
499 | int rw = bio_data_dir(bio); |
500 | ||
074a7aca TH |
501 | cpu = part_stat_lock(); |
502 | part_round_stats(cpu, &dm_disk(md)->part0); | |
503 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
504 | part_stat_unlock(); | |
3eaf840e | 505 | |
af7e466a MP |
506 | /* |
507 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 508 | * a flush. |
af7e466a | 509 | */ |
1e9bb880 SL |
510 | pending = atomic_dec_return(&md->pending[rw]); |
511 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 512 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 513 | |
d221d2e7 MP |
514 | /* nudge anyone waiting on suspend queue */ |
515 | if (!pending) | |
516 | wake_up(&md->wait); | |
3eaf840e JNN |
517 | } |
518 | ||
1da177e4 LT |
519 | /* |
520 | * Add the bio to the list of deferred io. | |
521 | */ | |
92c63902 | 522 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 523 | { |
05447420 | 524 | unsigned long flags; |
1da177e4 | 525 | |
05447420 | 526 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 527 | bio_list_add(&md->deferred, bio); |
05447420 | 528 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 529 | queue_work(md->wq, &md->work); |
1da177e4 LT |
530 | } |
531 | ||
532 | /* | |
533 | * Everyone (including functions in this file), should use this | |
534 | * function to access the md->map field, and make sure they call | |
535 | * dm_table_put() when finished. | |
536 | */ | |
7c666411 | 537 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
538 | { |
539 | struct dm_table *t; | |
523d9297 | 540 | unsigned long flags; |
1da177e4 | 541 | |
523d9297 | 542 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
543 | t = md->map; |
544 | if (t) | |
545 | dm_table_get(t); | |
523d9297 | 546 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
547 | |
548 | return t; | |
549 | } | |
550 | ||
3ac51e74 DW |
551 | /* |
552 | * Get the geometry associated with a dm device | |
553 | */ | |
554 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
555 | { | |
556 | *geo = md->geometry; | |
557 | ||
558 | return 0; | |
559 | } | |
560 | ||
561 | /* | |
562 | * Set the geometry of a device. | |
563 | */ | |
564 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
565 | { | |
566 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
567 | ||
568 | if (geo->start > sz) { | |
569 | DMWARN("Start sector is beyond the geometry limits."); | |
570 | return -EINVAL; | |
571 | } | |
572 | ||
573 | md->geometry = *geo; | |
574 | ||
575 | return 0; | |
576 | } | |
577 | ||
1da177e4 LT |
578 | /*----------------------------------------------------------------- |
579 | * CRUD START: | |
580 | * A more elegant soln is in the works that uses the queue | |
581 | * merge fn, unfortunately there are a couple of changes to | |
582 | * the block layer that I want to make for this. So in the | |
583 | * interests of getting something for people to use I give | |
584 | * you this clearly demarcated crap. | |
585 | *---------------------------------------------------------------*/ | |
586 | ||
2e93ccc1 KU |
587 | static int __noflush_suspending(struct mapped_device *md) |
588 | { | |
589 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
590 | } | |
591 | ||
1da177e4 LT |
592 | /* |
593 | * Decrements the number of outstanding ios that a bio has been | |
594 | * cloned into, completing the original io if necc. | |
595 | */ | |
858119e1 | 596 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 597 | { |
2e93ccc1 | 598 | unsigned long flags; |
b35f8caa MB |
599 | int io_error; |
600 | struct bio *bio; | |
601 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
602 | |
603 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
604 | if (unlikely(error)) { |
605 | spin_lock_irqsave(&io->endio_lock, flags); | |
606 | if (!(io->error > 0 && __noflush_suspending(md))) | |
607 | io->error = error; | |
608 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
609 | } | |
1da177e4 LT |
610 | |
611 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
612 | if (io->error == DM_ENDIO_REQUEUE) { |
613 | /* | |
614 | * Target requested pushing back the I/O. | |
2e93ccc1 | 615 | */ |
022c2611 | 616 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
617 | if (__noflush_suspending(md)) |
618 | bio_list_add_head(&md->deferred, io->bio); | |
619 | else | |
2e93ccc1 KU |
620 | /* noflush suspend was interrupted. */ |
621 | io->error = -EIO; | |
022c2611 | 622 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
623 | } |
624 | ||
b35f8caa MB |
625 | io_error = io->error; |
626 | bio = io->bio; | |
6a8736d1 TH |
627 | end_io_acct(io); |
628 | free_io(md, io); | |
629 | ||
630 | if (io_error == DM_ENDIO_REQUEUE) | |
631 | return; | |
2e93ccc1 | 632 | |
b372d360 | 633 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 634 | /* |
6a8736d1 TH |
635 | * Preflush done for flush with data, reissue |
636 | * without REQ_FLUSH. | |
af7e466a | 637 | */ |
6a8736d1 TH |
638 | bio->bi_rw &= ~REQ_FLUSH; |
639 | queue_io(md, bio); | |
af7e466a | 640 | } else { |
b372d360 | 641 | /* done with normal IO or empty flush */ |
b7908c10 | 642 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 643 | bio_endio(bio, io_error); |
b35f8caa | 644 | } |
1da177e4 LT |
645 | } |
646 | } | |
647 | ||
6712ecf8 | 648 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
649 | { |
650 | int r = 0; | |
028867ac | 651 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 652 | struct dm_io *io = tio->io; |
9faf400f | 653 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
654 | dm_endio_fn endio = tio->ti->type->end_io; |
655 | ||
1da177e4 LT |
656 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
657 | error = -EIO; | |
658 | ||
659 | if (endio) { | |
660 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
661 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
662 | /* | |
663 | * error and requeue request are handled | |
664 | * in dec_pending(). | |
665 | */ | |
1da177e4 | 666 | error = r; |
45cbcd79 KU |
667 | else if (r == DM_ENDIO_INCOMPLETE) |
668 | /* The target will handle the io */ | |
6712ecf8 | 669 | return; |
45cbcd79 KU |
670 | else if (r) { |
671 | DMWARN("unimplemented target endio return value: %d", r); | |
672 | BUG(); | |
673 | } | |
1da177e4 LT |
674 | } |
675 | ||
9faf400f | 676 | free_tio(md, tio); |
b35f8caa | 677 | dec_pending(io, error); |
1da177e4 LT |
678 | } |
679 | ||
cec47e3d KU |
680 | /* |
681 | * Partial completion handling for request-based dm | |
682 | */ | |
683 | static void end_clone_bio(struct bio *clone, int error) | |
684 | { | |
685 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
686 | struct dm_rq_target_io *tio = info->tio; | |
687 | struct bio *bio = info->orig; | |
688 | unsigned int nr_bytes = info->orig->bi_size; | |
689 | ||
690 | bio_put(clone); | |
691 | ||
692 | if (tio->error) | |
693 | /* | |
694 | * An error has already been detected on the request. | |
695 | * Once error occurred, just let clone->end_io() handle | |
696 | * the remainder. | |
697 | */ | |
698 | return; | |
699 | else if (error) { | |
700 | /* | |
701 | * Don't notice the error to the upper layer yet. | |
702 | * The error handling decision is made by the target driver, | |
703 | * when the request is completed. | |
704 | */ | |
705 | tio->error = error; | |
706 | return; | |
707 | } | |
708 | ||
709 | /* | |
710 | * I/O for the bio successfully completed. | |
711 | * Notice the data completion to the upper layer. | |
712 | */ | |
713 | ||
714 | /* | |
715 | * bios are processed from the head of the list. | |
716 | * So the completing bio should always be rq->bio. | |
717 | * If it's not, something wrong is happening. | |
718 | */ | |
719 | if (tio->orig->bio != bio) | |
720 | DMERR("bio completion is going in the middle of the request"); | |
721 | ||
722 | /* | |
723 | * Update the original request. | |
724 | * Do not use blk_end_request() here, because it may complete | |
725 | * the original request before the clone, and break the ordering. | |
726 | */ | |
727 | blk_update_request(tio->orig, 0, nr_bytes); | |
728 | } | |
729 | ||
730 | /* | |
731 | * Don't touch any member of the md after calling this function because | |
732 | * the md may be freed in dm_put() at the end of this function. | |
733 | * Or do dm_get() before calling this function and dm_put() later. | |
734 | */ | |
b4324fee | 735 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 736 | { |
b4324fee | 737 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
738 | |
739 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 740 | if (!md_in_flight(md)) |
cec47e3d KU |
741 | wake_up(&md->wait); |
742 | ||
743 | if (run_queue) | |
b4324fee | 744 | blk_run_queue(md->queue); |
cec47e3d KU |
745 | |
746 | /* | |
747 | * dm_put() must be at the end of this function. See the comment above | |
748 | */ | |
749 | dm_put(md); | |
750 | } | |
751 | ||
a77e28c7 KU |
752 | static void free_rq_clone(struct request *clone) |
753 | { | |
754 | struct dm_rq_target_io *tio = clone->end_io_data; | |
755 | ||
756 | blk_rq_unprep_clone(clone); | |
757 | free_rq_tio(tio); | |
758 | } | |
759 | ||
980691e5 KU |
760 | /* |
761 | * Complete the clone and the original request. | |
762 | * Must be called without queue lock. | |
763 | */ | |
764 | static void dm_end_request(struct request *clone, int error) | |
765 | { | |
766 | int rw = rq_data_dir(clone); | |
767 | struct dm_rq_target_io *tio = clone->end_io_data; | |
768 | struct mapped_device *md = tio->md; | |
769 | struct request *rq = tio->orig; | |
770 | ||
29e4013d | 771 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
772 | rq->errors = clone->errors; |
773 | rq->resid_len = clone->resid_len; | |
774 | ||
775 | if (rq->sense) | |
776 | /* | |
777 | * We are using the sense buffer of the original | |
778 | * request. | |
779 | * So setting the length of the sense data is enough. | |
780 | */ | |
781 | rq->sense_len = clone->sense_len; | |
782 | } | |
783 | ||
784 | free_rq_clone(clone); | |
29e4013d TH |
785 | blk_end_request_all(rq, error); |
786 | rq_completed(md, rw, true); | |
980691e5 KU |
787 | } |
788 | ||
cec47e3d KU |
789 | static void dm_unprep_request(struct request *rq) |
790 | { | |
791 | struct request *clone = rq->special; | |
cec47e3d KU |
792 | |
793 | rq->special = NULL; | |
794 | rq->cmd_flags &= ~REQ_DONTPREP; | |
795 | ||
a77e28c7 | 796 | free_rq_clone(clone); |
cec47e3d KU |
797 | } |
798 | ||
799 | /* | |
800 | * Requeue the original request of a clone. | |
801 | */ | |
802 | void dm_requeue_unmapped_request(struct request *clone) | |
803 | { | |
b4324fee | 804 | int rw = rq_data_dir(clone); |
cec47e3d KU |
805 | struct dm_rq_target_io *tio = clone->end_io_data; |
806 | struct mapped_device *md = tio->md; | |
807 | struct request *rq = tio->orig; | |
808 | struct request_queue *q = rq->q; | |
809 | unsigned long flags; | |
810 | ||
811 | dm_unprep_request(rq); | |
812 | ||
813 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
814 | blk_requeue_request(q, rq); |
815 | spin_unlock_irqrestore(q->queue_lock, flags); | |
816 | ||
b4324fee | 817 | rq_completed(md, rw, 0); |
cec47e3d KU |
818 | } |
819 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
820 | ||
821 | static void __stop_queue(struct request_queue *q) | |
822 | { | |
823 | blk_stop_queue(q); | |
824 | } | |
825 | ||
826 | static void stop_queue(struct request_queue *q) | |
827 | { | |
828 | unsigned long flags; | |
829 | ||
830 | spin_lock_irqsave(q->queue_lock, flags); | |
831 | __stop_queue(q); | |
832 | spin_unlock_irqrestore(q->queue_lock, flags); | |
833 | } | |
834 | ||
835 | static void __start_queue(struct request_queue *q) | |
836 | { | |
837 | if (blk_queue_stopped(q)) | |
838 | blk_start_queue(q); | |
839 | } | |
840 | ||
841 | static void start_queue(struct request_queue *q) | |
842 | { | |
843 | unsigned long flags; | |
844 | ||
845 | spin_lock_irqsave(q->queue_lock, flags); | |
846 | __start_queue(q); | |
847 | spin_unlock_irqrestore(q->queue_lock, flags); | |
848 | } | |
849 | ||
11a68244 | 850 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 851 | { |
11a68244 | 852 | int r = error; |
cec47e3d | 853 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 854 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 855 | |
ba1cbad9 MS |
856 | if (tio->ti) { |
857 | rq_end_io = tio->ti->type->rq_end_io; | |
858 | ||
859 | if (mapped && rq_end_io) | |
860 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
861 | } | |
cec47e3d | 862 | |
11a68244 | 863 | if (r <= 0) |
cec47e3d | 864 | /* The target wants to complete the I/O */ |
11a68244 KU |
865 | dm_end_request(clone, r); |
866 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
867 | /* The target will handle the I/O */ |
868 | return; | |
11a68244 | 869 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
870 | /* The target wants to requeue the I/O */ |
871 | dm_requeue_unmapped_request(clone); | |
872 | else { | |
11a68244 | 873 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
874 | BUG(); |
875 | } | |
876 | } | |
877 | ||
11a68244 KU |
878 | /* |
879 | * Request completion handler for request-based dm | |
880 | */ | |
881 | static void dm_softirq_done(struct request *rq) | |
882 | { | |
883 | bool mapped = true; | |
884 | struct request *clone = rq->completion_data; | |
885 | struct dm_rq_target_io *tio = clone->end_io_data; | |
886 | ||
887 | if (rq->cmd_flags & REQ_FAILED) | |
888 | mapped = false; | |
889 | ||
890 | dm_done(clone, tio->error, mapped); | |
891 | } | |
892 | ||
cec47e3d KU |
893 | /* |
894 | * Complete the clone and the original request with the error status | |
895 | * through softirq context. | |
896 | */ | |
897 | static void dm_complete_request(struct request *clone, int error) | |
898 | { | |
899 | struct dm_rq_target_io *tio = clone->end_io_data; | |
900 | struct request *rq = tio->orig; | |
901 | ||
902 | tio->error = error; | |
903 | rq->completion_data = clone; | |
904 | blk_complete_request(rq); | |
905 | } | |
906 | ||
907 | /* | |
908 | * Complete the not-mapped clone and the original request with the error status | |
909 | * through softirq context. | |
910 | * Target's rq_end_io() function isn't called. | |
911 | * This may be used when the target's map_rq() function fails. | |
912 | */ | |
913 | void dm_kill_unmapped_request(struct request *clone, int error) | |
914 | { | |
915 | struct dm_rq_target_io *tio = clone->end_io_data; | |
916 | struct request *rq = tio->orig; | |
917 | ||
918 | rq->cmd_flags |= REQ_FAILED; | |
919 | dm_complete_request(clone, error); | |
920 | } | |
921 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
922 | ||
923 | /* | |
924 | * Called with the queue lock held | |
925 | */ | |
926 | static void end_clone_request(struct request *clone, int error) | |
927 | { | |
928 | /* | |
929 | * For just cleaning up the information of the queue in which | |
930 | * the clone was dispatched. | |
931 | * The clone is *NOT* freed actually here because it is alloced from | |
932 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
933 | */ | |
934 | __blk_put_request(clone->q, clone); | |
935 | ||
936 | /* | |
937 | * Actual request completion is done in a softirq context which doesn't | |
938 | * hold the queue lock. Otherwise, deadlock could occur because: | |
939 | * - another request may be submitted by the upper level driver | |
940 | * of the stacking during the completion | |
941 | * - the submission which requires queue lock may be done | |
942 | * against this queue | |
943 | */ | |
944 | dm_complete_request(clone, error); | |
945 | } | |
946 | ||
56a67df7 MS |
947 | /* |
948 | * Return maximum size of I/O possible at the supplied sector up to the current | |
949 | * target boundary. | |
950 | */ | |
951 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
952 | { | |
953 | sector_t target_offset = dm_target_offset(ti, sector); | |
954 | ||
955 | return ti->len - target_offset; | |
956 | } | |
957 | ||
958 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 959 | { |
56a67df7 | 960 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 961 | sector_t offset, max_len; |
1da177e4 LT |
962 | |
963 | /* | |
542f9038 | 964 | * Does the target need to split even further? |
1da177e4 | 965 | */ |
542f9038 MS |
966 | if (ti->max_io_len) { |
967 | offset = dm_target_offset(ti, sector); | |
968 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
969 | max_len = sector_div(offset, ti->max_io_len); | |
970 | else | |
971 | max_len = offset & (ti->max_io_len - 1); | |
972 | max_len = ti->max_io_len - max_len; | |
973 | ||
974 | if (len > max_len) | |
975 | len = max_len; | |
1da177e4 LT |
976 | } |
977 | ||
978 | return len; | |
979 | } | |
980 | ||
542f9038 MS |
981 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
982 | { | |
983 | if (len > UINT_MAX) { | |
984 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
985 | (unsigned long long)len, UINT_MAX); | |
986 | ti->error = "Maximum size of target IO is too large"; | |
987 | return -EINVAL; | |
988 | } | |
989 | ||
990 | ti->max_io_len = (uint32_t) len; | |
991 | ||
992 | return 0; | |
993 | } | |
994 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
995 | ||
dba14160 | 996 | static void __map_bio(struct dm_target *ti, struct dm_target_io *tio) |
1da177e4 LT |
997 | { |
998 | int r; | |
2056a782 | 999 | sector_t sector; |
9faf400f | 1000 | struct mapped_device *md; |
dba14160 | 1001 | struct bio *clone = &tio->clone; |
1da177e4 | 1002 | |
1da177e4 LT |
1003 | clone->bi_end_io = clone_endio; |
1004 | clone->bi_private = tio; | |
1005 | ||
1006 | /* | |
1007 | * Map the clone. If r == 0 we don't need to do | |
1008 | * anything, the target has assumed ownership of | |
1009 | * this io. | |
1010 | */ | |
1011 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1012 | sector = clone->bi_sector; |
1da177e4 | 1013 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 1014 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1015 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1016 | |
d07335e5 MS |
1017 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1018 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1019 | |
1da177e4 | 1020 | generic_make_request(clone); |
2e93ccc1 KU |
1021 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1022 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1023 | md = tio->io->md; |
1024 | dec_pending(tio->io, r); | |
9faf400f | 1025 | free_tio(md, tio); |
45cbcd79 KU |
1026 | } else if (r) { |
1027 | DMWARN("unimplemented target map return value: %d", r); | |
1028 | BUG(); | |
1da177e4 LT |
1029 | } |
1030 | } | |
1031 | ||
1032 | struct clone_info { | |
1033 | struct mapped_device *md; | |
1034 | struct dm_table *map; | |
1035 | struct bio *bio; | |
1036 | struct dm_io *io; | |
1037 | sector_t sector; | |
1038 | sector_t sector_count; | |
1039 | unsigned short idx; | |
1040 | }; | |
1041 | ||
1042 | /* | |
d87f4c14 | 1043 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1044 | */ |
dba14160 MP |
1045 | static void split_bvec(struct dm_target_io *tio, struct bio *bio, |
1046 | sector_t sector, unsigned short idx, unsigned int offset, | |
1047 | unsigned int len, struct bio_set *bs) | |
1da177e4 | 1048 | { |
dba14160 | 1049 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1050 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1051 | ||
1da177e4 LT |
1052 | *clone->bi_io_vec = *bv; |
1053 | ||
1054 | clone->bi_sector = sector; | |
1055 | clone->bi_bdev = bio->bi_bdev; | |
d87f4c14 | 1056 | clone->bi_rw = bio->bi_rw; |
1da177e4 LT |
1057 | clone->bi_vcnt = 1; |
1058 | clone->bi_size = to_bytes(len); | |
1059 | clone->bi_io_vec->bv_offset = offset; | |
1060 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1061 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1062 | |
9c47008d | 1063 | if (bio_integrity(bio)) { |
1e2a410f | 1064 | bio_integrity_clone(clone, bio, GFP_NOIO); |
9c47008d MP |
1065 | bio_integrity_trim(clone, |
1066 | bio_sector_offset(bio, idx, offset), len); | |
1067 | } | |
1da177e4 LT |
1068 | } |
1069 | ||
1070 | /* | |
1071 | * Creates a bio that consists of range of complete bvecs. | |
1072 | */ | |
dba14160 MP |
1073 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1074 | sector_t sector, unsigned short idx, | |
1075 | unsigned short bv_count, unsigned int len, | |
1076 | struct bio_set *bs) | |
1da177e4 | 1077 | { |
dba14160 | 1078 | struct bio *clone = &tio->clone; |
1da177e4 | 1079 | |
9faf400f | 1080 | __bio_clone(clone, bio); |
1da177e4 LT |
1081 | clone->bi_sector = sector; |
1082 | clone->bi_idx = idx; | |
1083 | clone->bi_vcnt = idx + bv_count; | |
1084 | clone->bi_size = to_bytes(len); | |
1085 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1086 | ||
9c47008d | 1087 | if (bio_integrity(bio)) { |
1e2a410f | 1088 | bio_integrity_clone(clone, bio, GFP_NOIO); |
9c47008d MP |
1089 | |
1090 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1091 | bio_integrity_trim(clone, | |
1092 | bio_sector_offset(bio, idx, 0), len); | |
1093 | } | |
1da177e4 LT |
1094 | } |
1095 | ||
9015df24 | 1096 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
dba14160 | 1097 | struct dm_target *ti, int nr_iovecs) |
f9ab94ce | 1098 | { |
dba14160 MP |
1099 | struct dm_target_io *tio; |
1100 | struct bio *clone; | |
1101 | ||
1102 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1103 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1104 | |
1105 | tio->io = ci->io; | |
1106 | tio->ti = ti; | |
f9ab94ce | 1107 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1108 | |
1109 | return tio; | |
1110 | } | |
1111 | ||
06a426ce | 1112 | static void __issue_target_request(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1113 | unsigned request_nr, sector_t len) |
9015df24 | 1114 | { |
dba14160 MP |
1115 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs); |
1116 | struct bio *clone = &tio->clone; | |
9015df24 | 1117 | |
57cba5d3 | 1118 | tio->info.target_request_nr = request_nr; |
f9ab94ce | 1119 | |
06a426ce MS |
1120 | /* |
1121 | * Discard requests require the bio's inline iovecs be initialized. | |
1122 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1123 | * and discard, so no need for concern about wasted bvec allocations. | |
1124 | */ | |
bf800ef1 | 1125 | |
dba14160 | 1126 | __bio_clone(clone, ci->bio); |
a79245b3 MS |
1127 | if (len) { |
1128 | clone->bi_sector = ci->sector; | |
1129 | clone->bi_size = to_bytes(len); | |
1130 | } | |
f9ab94ce | 1131 | |
dba14160 | 1132 | __map_bio(ti, tio); |
f9ab94ce MP |
1133 | } |
1134 | ||
06a426ce | 1135 | static void __issue_target_requests(struct clone_info *ci, struct dm_target *ti, |
a79245b3 | 1136 | unsigned num_requests, sector_t len) |
06a426ce MS |
1137 | { |
1138 | unsigned request_nr; | |
1139 | ||
1140 | for (request_nr = 0; request_nr < num_requests; request_nr++) | |
a79245b3 | 1141 | __issue_target_request(ci, ti, request_nr, len); |
06a426ce MS |
1142 | } |
1143 | ||
b372d360 | 1144 | static int __clone_and_map_empty_flush(struct clone_info *ci) |
f9ab94ce | 1145 | { |
06a426ce | 1146 | unsigned target_nr = 0; |
f9ab94ce MP |
1147 | struct dm_target *ti; |
1148 | ||
b372d360 | 1149 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1150 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
a79245b3 | 1151 | __issue_target_requests(ci, ti, ti->num_flush_requests, 0); |
f9ab94ce | 1152 | |
f9ab94ce MP |
1153 | return 0; |
1154 | } | |
1155 | ||
5ae89a87 MS |
1156 | /* |
1157 | * Perform all io with a single clone. | |
1158 | */ | |
1159 | static void __clone_and_map_simple(struct clone_info *ci, struct dm_target *ti) | |
1160 | { | |
dba14160 | 1161 | struct bio *bio = ci->bio; |
5ae89a87 MS |
1162 | struct dm_target_io *tio; |
1163 | ||
dba14160 MP |
1164 | tio = alloc_tio(ci, ti, bio->bi_max_vecs); |
1165 | clone_bio(tio, bio, ci->sector, ci->idx, bio->bi_vcnt - ci->idx, | |
1166 | ci->sector_count, ci->md->bs); | |
1167 | __map_bio(ti, tio); | |
5ae89a87 MS |
1168 | ci->sector_count = 0; |
1169 | } | |
1170 | ||
1171 | static int __clone_and_map_discard(struct clone_info *ci) | |
1172 | { | |
1173 | struct dm_target *ti; | |
a79245b3 | 1174 | sector_t len; |
5ae89a87 | 1175 | |
a79245b3 MS |
1176 | do { |
1177 | ti = dm_table_find_target(ci->map, ci->sector); | |
1178 | if (!dm_target_is_valid(ti)) | |
1179 | return -EIO; | |
5ae89a87 | 1180 | |
5ae89a87 | 1181 | /* |
a79245b3 | 1182 | * Even though the device advertised discard support, |
936688d7 MS |
1183 | * that does not mean every target supports it, and |
1184 | * reconfiguration might also have changed that since the | |
a79245b3 | 1185 | * check was performed. |
5ae89a87 | 1186 | */ |
a79245b3 MS |
1187 | if (!ti->num_discard_requests) |
1188 | return -EOPNOTSUPP; | |
5ae89a87 | 1189 | |
7acf0277 MP |
1190 | if (!ti->split_discard_requests) |
1191 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); | |
1192 | else | |
1193 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1194 | |
a79245b3 MS |
1195 | __issue_target_requests(ci, ti, ti->num_discard_requests, len); |
1196 | ||
1197 | ci->sector += len; | |
1198 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1199 | |
1200 | return 0; | |
1201 | } | |
1202 | ||
512875bd | 1203 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 | 1204 | { |
dba14160 | 1205 | struct bio *bio = ci->bio; |
512875bd JN |
1206 | struct dm_target *ti; |
1207 | sector_t len = 0, max; | |
028867ac | 1208 | struct dm_target_io *tio; |
1da177e4 | 1209 | |
5ae89a87 MS |
1210 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
1211 | return __clone_and_map_discard(ci); | |
1212 | ||
512875bd JN |
1213 | ti = dm_table_find_target(ci->map, ci->sector); |
1214 | if (!dm_target_is_valid(ti)) | |
1215 | return -EIO; | |
1216 | ||
56a67df7 | 1217 | max = max_io_len(ci->sector, ti); |
512875bd | 1218 | |
1da177e4 LT |
1219 | if (ci->sector_count <= max) { |
1220 | /* | |
1221 | * Optimise for the simple case where we can do all of | |
1222 | * the remaining io with a single clone. | |
1223 | */ | |
5ae89a87 | 1224 | __clone_and_map_simple(ci, ti); |
1da177e4 LT |
1225 | |
1226 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1227 | /* | |
1228 | * There are some bvecs that don't span targets. | |
1229 | * Do as many of these as possible. | |
1230 | */ | |
1231 | int i; | |
1232 | sector_t remaining = max; | |
1233 | sector_t bv_len; | |
1234 | ||
1235 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1236 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1237 | ||
1238 | if (bv_len > remaining) | |
1239 | break; | |
1240 | ||
1241 | remaining -= bv_len; | |
1242 | len += bv_len; | |
1243 | } | |
1244 | ||
dba14160 MP |
1245 | tio = alloc_tio(ci, ti, bio->bi_max_vecs); |
1246 | clone_bio(tio, bio, ci->sector, ci->idx, i - ci->idx, len, | |
1247 | ci->md->bs); | |
1248 | __map_bio(ti, tio); | |
1da177e4 LT |
1249 | |
1250 | ci->sector += len; | |
1251 | ci->sector_count -= len; | |
1252 | ci->idx = i; | |
1253 | ||
1254 | } else { | |
1255 | /* | |
d2044a94 | 1256 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1257 | */ |
1258 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1259 | sector_t remaining = to_sector(bv->bv_len); |
1260 | unsigned int offset = 0; | |
1da177e4 | 1261 | |
d2044a94 AK |
1262 | do { |
1263 | if (offset) { | |
1264 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1265 | if (!dm_target_is_valid(ti)) |
1266 | return -EIO; | |
1267 | ||
56a67df7 | 1268 | max = max_io_len(ci->sector, ti); |
d2044a94 AK |
1269 | } |
1270 | ||
1271 | len = min(remaining, max); | |
1272 | ||
dba14160 MP |
1273 | tio = alloc_tio(ci, ti, 1); |
1274 | split_bvec(tio, bio, ci->sector, ci->idx, | |
1275 | bv->bv_offset + offset, len, ci->md->bs); | |
d2044a94 | 1276 | |
dba14160 | 1277 | __map_bio(ti, tio); |
d2044a94 AK |
1278 | |
1279 | ci->sector += len; | |
1280 | ci->sector_count -= len; | |
1281 | offset += to_bytes(len); | |
1282 | } while (remaining -= len); | |
1da177e4 | 1283 | |
1da177e4 LT |
1284 | ci->idx++; |
1285 | } | |
512875bd JN |
1286 | |
1287 | return 0; | |
1da177e4 LT |
1288 | } |
1289 | ||
1290 | /* | |
8a53c28d | 1291 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1292 | */ |
f0b9a450 | 1293 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1294 | { |
1295 | struct clone_info ci; | |
512875bd | 1296 | int error = 0; |
1da177e4 | 1297 | |
7c666411 | 1298 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1299 | if (unlikely(!ci.map)) { |
6a8736d1 | 1300 | bio_io_error(bio); |
f0b9a450 MP |
1301 | return; |
1302 | } | |
692d0eb9 | 1303 | |
1da177e4 | 1304 | ci.md = md; |
1da177e4 LT |
1305 | ci.io = alloc_io(md); |
1306 | ci.io->error = 0; | |
1307 | atomic_set(&ci.io->io_count, 1); | |
1308 | ci.io->bio = bio; | |
1309 | ci.io->md = md; | |
f88fb981 | 1310 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1311 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1312 | ci.idx = bio->bi_idx; |
1313 | ||
3eaf840e | 1314 | start_io_acct(ci.io); |
b372d360 MS |
1315 | if (bio->bi_rw & REQ_FLUSH) { |
1316 | ci.bio = &ci.md->flush_bio; | |
1317 | ci.sector_count = 0; | |
1318 | error = __clone_and_map_empty_flush(&ci); | |
1319 | /* dec_pending submits any data associated with flush */ | |
1320 | } else { | |
6a8736d1 | 1321 | ci.bio = bio; |
d87f4c14 | 1322 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1323 | while (ci.sector_count && !error) |
d87f4c14 | 1324 | error = __clone_and_map(&ci); |
d87f4c14 | 1325 | } |
1da177e4 LT |
1326 | |
1327 | /* drop the extra reference count */ | |
512875bd | 1328 | dec_pending(ci.io, error); |
1da177e4 LT |
1329 | dm_table_put(ci.map); |
1330 | } | |
1331 | /*----------------------------------------------------------------- | |
1332 | * CRUD END | |
1333 | *---------------------------------------------------------------*/ | |
1334 | ||
f6fccb12 MB |
1335 | static int dm_merge_bvec(struct request_queue *q, |
1336 | struct bvec_merge_data *bvm, | |
1337 | struct bio_vec *biovec) | |
1338 | { | |
1339 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1340 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1341 | struct dm_target *ti; |
1342 | sector_t max_sectors; | |
5037108a | 1343 | int max_size = 0; |
f6fccb12 MB |
1344 | |
1345 | if (unlikely(!map)) | |
5037108a | 1346 | goto out; |
f6fccb12 MB |
1347 | |
1348 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1349 | if (!dm_target_is_valid(ti)) |
1350 | goto out_table; | |
f6fccb12 MB |
1351 | |
1352 | /* | |
1353 | * Find maximum amount of I/O that won't need splitting | |
1354 | */ | |
56a67df7 | 1355 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1356 | (sector_t) BIO_MAX_SECTORS); |
1357 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1358 | if (max_size < 0) | |
1359 | max_size = 0; | |
1360 | ||
1361 | /* | |
1362 | * merge_bvec_fn() returns number of bytes | |
1363 | * it can accept at this offset | |
1364 | * max is precomputed maximal io size | |
1365 | */ | |
1366 | if (max_size && ti->type->merge) | |
1367 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1368 | /* |
1369 | * If the target doesn't support merge method and some of the devices | |
1370 | * provided their merge_bvec method (we know this by looking at | |
1371 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1372 | * entries. So always set max_size to 0, and the code below allows | |
1373 | * just one page. | |
1374 | */ | |
1375 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1376 | ||
1377 | max_size = 0; | |
f6fccb12 | 1378 | |
b01cd5ac | 1379 | out_table: |
5037108a MP |
1380 | dm_table_put(map); |
1381 | ||
1382 | out: | |
f6fccb12 MB |
1383 | /* |
1384 | * Always allow an entire first page | |
1385 | */ | |
1386 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1387 | max_size = biovec->bv_len; | |
1388 | ||
f6fccb12 MB |
1389 | return max_size; |
1390 | } | |
1391 | ||
1da177e4 LT |
1392 | /* |
1393 | * The request function that just remaps the bio built up by | |
1394 | * dm_merge_bvec. | |
1395 | */ | |
5a7bbad2 | 1396 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1397 | { |
12f03a49 | 1398 | int rw = bio_data_dir(bio); |
1da177e4 | 1399 | struct mapped_device *md = q->queuedata; |
c9959059 | 1400 | int cpu; |
1da177e4 | 1401 | |
2ca3310e | 1402 | down_read(&md->io_lock); |
1da177e4 | 1403 | |
074a7aca TH |
1404 | cpu = part_stat_lock(); |
1405 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1406 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1407 | part_stat_unlock(); | |
12f03a49 | 1408 | |
6a8736d1 TH |
1409 | /* if we're suspended, we have to queue this io for later */ |
1410 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1411 | up_read(&md->io_lock); |
1da177e4 | 1412 | |
6a8736d1 TH |
1413 | if (bio_rw(bio) != READA) |
1414 | queue_io(md, bio); | |
1415 | else | |
54d9a1b4 | 1416 | bio_io_error(bio); |
5a7bbad2 | 1417 | return; |
1da177e4 LT |
1418 | } |
1419 | ||
f0b9a450 | 1420 | __split_and_process_bio(md, bio); |
2ca3310e | 1421 | up_read(&md->io_lock); |
5a7bbad2 | 1422 | return; |
cec47e3d KU |
1423 | } |
1424 | ||
1425 | static int dm_request_based(struct mapped_device *md) | |
1426 | { | |
1427 | return blk_queue_stackable(md->queue); | |
1428 | } | |
1429 | ||
5a7bbad2 | 1430 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1431 | { |
1432 | struct mapped_device *md = q->queuedata; | |
1433 | ||
1434 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1435 | blk_queue_bio(q, bio); |
1436 | else | |
1437 | _dm_request(q, bio); | |
cec47e3d KU |
1438 | } |
1439 | ||
1440 | void dm_dispatch_request(struct request *rq) | |
1441 | { | |
1442 | int r; | |
1443 | ||
1444 | if (blk_queue_io_stat(rq->q)) | |
1445 | rq->cmd_flags |= REQ_IO_STAT; | |
1446 | ||
1447 | rq->start_time = jiffies; | |
1448 | r = blk_insert_cloned_request(rq->q, rq); | |
1449 | if (r) | |
1450 | dm_complete_request(rq, r); | |
1451 | } | |
1452 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1453 | ||
cec47e3d KU |
1454 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1455 | void *data) | |
1456 | { | |
1457 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1458 | struct dm_rq_clone_bio_info *info = |
1459 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1460 | |
1461 | info->orig = bio_orig; | |
1462 | info->tio = tio; | |
1463 | bio->bi_end_io = end_clone_bio; | |
1464 | bio->bi_private = info; | |
cec47e3d KU |
1465 | |
1466 | return 0; | |
1467 | } | |
1468 | ||
1469 | static int setup_clone(struct request *clone, struct request *rq, | |
1470 | struct dm_rq_target_io *tio) | |
1471 | { | |
d0bcb878 | 1472 | int r; |
cec47e3d | 1473 | |
29e4013d TH |
1474 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1475 | dm_rq_bio_constructor, tio); | |
1476 | if (r) | |
1477 | return r; | |
cec47e3d | 1478 | |
29e4013d TH |
1479 | clone->cmd = rq->cmd; |
1480 | clone->cmd_len = rq->cmd_len; | |
1481 | clone->sense = rq->sense; | |
1482 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1483 | clone->end_io = end_clone_request; |
1484 | clone->end_io_data = tio; | |
1485 | ||
1486 | return 0; | |
1487 | } | |
1488 | ||
6facdaff KU |
1489 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1490 | gfp_t gfp_mask) | |
1491 | { | |
1492 | struct request *clone; | |
1493 | struct dm_rq_target_io *tio; | |
1494 | ||
1495 | tio = alloc_rq_tio(md, gfp_mask); | |
1496 | if (!tio) | |
1497 | return NULL; | |
1498 | ||
1499 | tio->md = md; | |
1500 | tio->ti = NULL; | |
1501 | tio->orig = rq; | |
1502 | tio->error = 0; | |
1503 | memset(&tio->info, 0, sizeof(tio->info)); | |
1504 | ||
1505 | clone = &tio->clone; | |
1506 | if (setup_clone(clone, rq, tio)) { | |
1507 | /* -ENOMEM */ | |
1508 | free_rq_tio(tio); | |
1509 | return NULL; | |
1510 | } | |
1511 | ||
1512 | return clone; | |
1513 | } | |
1514 | ||
cec47e3d KU |
1515 | /* |
1516 | * Called with the queue lock held. | |
1517 | */ | |
1518 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1519 | { | |
1520 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1521 | struct request *clone; |
1522 | ||
cec47e3d KU |
1523 | if (unlikely(rq->special)) { |
1524 | DMWARN("Already has something in rq->special."); | |
1525 | return BLKPREP_KILL; | |
1526 | } | |
1527 | ||
6facdaff KU |
1528 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1529 | if (!clone) | |
cec47e3d | 1530 | return BLKPREP_DEFER; |
cec47e3d KU |
1531 | |
1532 | rq->special = clone; | |
1533 | rq->cmd_flags |= REQ_DONTPREP; | |
1534 | ||
1535 | return BLKPREP_OK; | |
1536 | } | |
1537 | ||
9eef87da KU |
1538 | /* |
1539 | * Returns: | |
1540 | * 0 : the request has been processed (not requeued) | |
1541 | * !0 : the request has been requeued | |
1542 | */ | |
1543 | static int map_request(struct dm_target *ti, struct request *clone, | |
1544 | struct mapped_device *md) | |
cec47e3d | 1545 | { |
9eef87da | 1546 | int r, requeued = 0; |
cec47e3d KU |
1547 | struct dm_rq_target_io *tio = clone->end_io_data; |
1548 | ||
cec47e3d KU |
1549 | tio->ti = ti; |
1550 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1551 | switch (r) { | |
1552 | case DM_MAPIO_SUBMITTED: | |
1553 | /* The target has taken the I/O to submit by itself later */ | |
1554 | break; | |
1555 | case DM_MAPIO_REMAPPED: | |
1556 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1557 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1558 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1559 | dm_dispatch_request(clone); |
1560 | break; | |
1561 | case DM_MAPIO_REQUEUE: | |
1562 | /* The target wants to requeue the I/O */ | |
1563 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1564 | requeued = 1; |
cec47e3d KU |
1565 | break; |
1566 | default: | |
1567 | if (r > 0) { | |
1568 | DMWARN("unimplemented target map return value: %d", r); | |
1569 | BUG(); | |
1570 | } | |
1571 | ||
1572 | /* The target wants to complete the I/O */ | |
1573 | dm_kill_unmapped_request(clone, r); | |
1574 | break; | |
1575 | } | |
9eef87da KU |
1576 | |
1577 | return requeued; | |
cec47e3d KU |
1578 | } |
1579 | ||
ba1cbad9 MS |
1580 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1581 | { | |
1582 | struct request *clone; | |
1583 | ||
1584 | blk_start_request(orig); | |
1585 | clone = orig->special; | |
1586 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1587 | ||
1588 | /* | |
1589 | * Hold the md reference here for the in-flight I/O. | |
1590 | * We can't rely on the reference count by device opener, | |
1591 | * because the device may be closed during the request completion | |
1592 | * when all bios are completed. | |
1593 | * See the comment in rq_completed() too. | |
1594 | */ | |
1595 | dm_get(md); | |
1596 | ||
1597 | return clone; | |
1598 | } | |
1599 | ||
cec47e3d KU |
1600 | /* |
1601 | * q->request_fn for request-based dm. | |
1602 | * Called with the queue lock held. | |
1603 | */ | |
1604 | static void dm_request_fn(struct request_queue *q) | |
1605 | { | |
1606 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1607 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1608 | struct dm_target *ti; |
b4324fee | 1609 | struct request *rq, *clone; |
29e4013d | 1610 | sector_t pos; |
cec47e3d KU |
1611 | |
1612 | /* | |
b4324fee KU |
1613 | * For suspend, check blk_queue_stopped() and increment |
1614 | * ->pending within a single queue_lock not to increment the | |
1615 | * number of in-flight I/Os after the queue is stopped in | |
1616 | * dm_suspend(). | |
cec47e3d | 1617 | */ |
7eaceacc | 1618 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1619 | rq = blk_peek_request(q); |
1620 | if (!rq) | |
7eaceacc | 1621 | goto delay_and_out; |
cec47e3d | 1622 | |
29e4013d TH |
1623 | /* always use block 0 to find the target for flushes for now */ |
1624 | pos = 0; | |
1625 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1626 | pos = blk_rq_pos(rq); | |
1627 | ||
1628 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1629 | if (!dm_target_is_valid(ti)) { |
1630 | /* | |
1631 | * Must perform setup, that dm_done() requires, | |
1632 | * before calling dm_kill_unmapped_request | |
1633 | */ | |
1634 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1635 | clone = dm_start_request(md, rq); | |
1636 | dm_kill_unmapped_request(clone, -EIO); | |
1637 | continue; | |
1638 | } | |
d0bcb878 | 1639 | |
cec47e3d | 1640 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1641 | goto delay_and_out; |
cec47e3d | 1642 | |
ba1cbad9 | 1643 | clone = dm_start_request(md, rq); |
b4324fee | 1644 | |
cec47e3d | 1645 | spin_unlock(q->queue_lock); |
9eef87da KU |
1646 | if (map_request(ti, clone, md)) |
1647 | goto requeued; | |
1648 | ||
052189a2 KU |
1649 | BUG_ON(!irqs_disabled()); |
1650 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1651 | } |
1652 | ||
1653 | goto out; | |
1654 | ||
9eef87da | 1655 | requeued: |
052189a2 KU |
1656 | BUG_ON(!irqs_disabled()); |
1657 | spin_lock(q->queue_lock); | |
9eef87da | 1658 | |
7eaceacc JA |
1659 | delay_and_out: |
1660 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1661 | out: |
1662 | dm_table_put(map); | |
cec47e3d KU |
1663 | } |
1664 | ||
1665 | int dm_underlying_device_busy(struct request_queue *q) | |
1666 | { | |
1667 | return blk_lld_busy(q); | |
1668 | } | |
1669 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1670 | ||
1671 | static int dm_lld_busy(struct request_queue *q) | |
1672 | { | |
1673 | int r; | |
1674 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1675 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1676 | |
1677 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1678 | r = 1; | |
1679 | else | |
1680 | r = dm_table_any_busy_target(map); | |
1681 | ||
1682 | dm_table_put(map); | |
1683 | ||
1684 | return r; | |
1685 | } | |
1686 | ||
1da177e4 LT |
1687 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1688 | { | |
8a57dfc6 CS |
1689 | int r = bdi_bits; |
1690 | struct mapped_device *md = congested_data; | |
1691 | struct dm_table *map; | |
1da177e4 | 1692 | |
1eb787ec | 1693 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1694 | map = dm_get_live_table(md); |
8a57dfc6 | 1695 | if (map) { |
cec47e3d KU |
1696 | /* |
1697 | * Request-based dm cares about only own queue for | |
1698 | * the query about congestion status of request_queue | |
1699 | */ | |
1700 | if (dm_request_based(md)) | |
1701 | r = md->queue->backing_dev_info.state & | |
1702 | bdi_bits; | |
1703 | else | |
1704 | r = dm_table_any_congested(map, bdi_bits); | |
1705 | ||
8a57dfc6 CS |
1706 | dm_table_put(map); |
1707 | } | |
1708 | } | |
1709 | ||
1da177e4 LT |
1710 | return r; |
1711 | } | |
1712 | ||
1713 | /*----------------------------------------------------------------- | |
1714 | * An IDR is used to keep track of allocated minor numbers. | |
1715 | *---------------------------------------------------------------*/ | |
2b06cfff | 1716 | static void free_minor(int minor) |
1da177e4 | 1717 | { |
f32c10b0 | 1718 | spin_lock(&_minor_lock); |
1da177e4 | 1719 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1720 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1721 | } |
1722 | ||
1723 | /* | |
1724 | * See if the device with a specific minor # is free. | |
1725 | */ | |
cf13ab8e | 1726 | static int specific_minor(int minor) |
1da177e4 LT |
1727 | { |
1728 | int r, m; | |
1729 | ||
1730 | if (minor >= (1 << MINORBITS)) | |
1731 | return -EINVAL; | |
1732 | ||
62f75c2f JM |
1733 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1734 | if (!r) | |
1735 | return -ENOMEM; | |
1736 | ||
f32c10b0 | 1737 | spin_lock(&_minor_lock); |
1da177e4 LT |
1738 | |
1739 | if (idr_find(&_minor_idr, minor)) { | |
1740 | r = -EBUSY; | |
1741 | goto out; | |
1742 | } | |
1743 | ||
ba61fdd1 | 1744 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1745 | if (r) |
1da177e4 | 1746 | goto out; |
1da177e4 LT |
1747 | |
1748 | if (m != minor) { | |
1749 | idr_remove(&_minor_idr, m); | |
1750 | r = -EBUSY; | |
1751 | goto out; | |
1752 | } | |
1753 | ||
1754 | out: | |
f32c10b0 | 1755 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1756 | return r; |
1757 | } | |
1758 | ||
cf13ab8e | 1759 | static int next_free_minor(int *minor) |
1da177e4 | 1760 | { |
2b06cfff | 1761 | int r, m; |
1da177e4 | 1762 | |
1da177e4 | 1763 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1764 | if (!r) |
1765 | return -ENOMEM; | |
1766 | ||
f32c10b0 | 1767 | spin_lock(&_minor_lock); |
1da177e4 | 1768 | |
ba61fdd1 | 1769 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1770 | if (r) |
1da177e4 | 1771 | goto out; |
1da177e4 LT |
1772 | |
1773 | if (m >= (1 << MINORBITS)) { | |
1774 | idr_remove(&_minor_idr, m); | |
1775 | r = -ENOSPC; | |
1776 | goto out; | |
1777 | } | |
1778 | ||
1779 | *minor = m; | |
1780 | ||
1781 | out: | |
f32c10b0 | 1782 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1783 | return r; |
1784 | } | |
1785 | ||
83d5cde4 | 1786 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1787 | |
53d5914f MP |
1788 | static void dm_wq_work(struct work_struct *work); |
1789 | ||
4a0b4ddf MS |
1790 | static void dm_init_md_queue(struct mapped_device *md) |
1791 | { | |
1792 | /* | |
1793 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1794 | * devices. The type of this dm device has not been decided yet. | |
1795 | * The type is decided at the first table loading time. | |
1796 | * To prevent problematic device stacking, clear the queue flag | |
1797 | * for request stacking support until then. | |
1798 | * | |
1799 | * This queue is new, so no concurrency on the queue_flags. | |
1800 | */ | |
1801 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1802 | ||
1803 | md->queue->queuedata = md; | |
1804 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1805 | md->queue->backing_dev_info.congested_data = md; | |
1806 | blk_queue_make_request(md->queue, dm_request); | |
1807 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1808 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1809 | } | |
1810 | ||
1da177e4 LT |
1811 | /* |
1812 | * Allocate and initialise a blank device with a given minor. | |
1813 | */ | |
2b06cfff | 1814 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1815 | { |
1816 | int r; | |
cf13ab8e | 1817 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1818 | void *old_md; |
1da177e4 LT |
1819 | |
1820 | if (!md) { | |
1821 | DMWARN("unable to allocate device, out of memory."); | |
1822 | return NULL; | |
1823 | } | |
1824 | ||
10da4f79 | 1825 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1826 | goto bad_module_get; |
10da4f79 | 1827 | |
1da177e4 | 1828 | /* get a minor number for the dev */ |
2b06cfff | 1829 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1830 | r = next_free_minor(&minor); |
2b06cfff | 1831 | else |
cf13ab8e | 1832 | r = specific_minor(minor); |
1da177e4 | 1833 | if (r < 0) |
6ed7ade8 | 1834 | goto bad_minor; |
1da177e4 | 1835 | |
a5664dad | 1836 | md->type = DM_TYPE_NONE; |
2ca3310e | 1837 | init_rwsem(&md->io_lock); |
e61290a4 | 1838 | mutex_init(&md->suspend_lock); |
a5664dad | 1839 | mutex_init(&md->type_lock); |
022c2611 | 1840 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1841 | rwlock_init(&md->map_lock); |
1842 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1843 | atomic_set(&md->open_count, 0); |
1da177e4 | 1844 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1845 | atomic_set(&md->uevent_seq, 0); |
1846 | INIT_LIST_HEAD(&md->uevent_list); | |
1847 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1848 | |
4a0b4ddf | 1849 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1850 | if (!md->queue) |
6ed7ade8 | 1851 | goto bad_queue; |
1da177e4 | 1852 | |
4a0b4ddf | 1853 | dm_init_md_queue(md); |
9faf400f | 1854 | |
1da177e4 LT |
1855 | md->disk = alloc_disk(1); |
1856 | if (!md->disk) | |
6ed7ade8 | 1857 | goto bad_disk; |
1da177e4 | 1858 | |
316d315b NK |
1859 | atomic_set(&md->pending[0], 0); |
1860 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1861 | init_waitqueue_head(&md->wait); |
53d5914f | 1862 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1863 | init_waitqueue_head(&md->eventq); |
1864 | ||
1da177e4 LT |
1865 | md->disk->major = _major; |
1866 | md->disk->first_minor = minor; | |
1867 | md->disk->fops = &dm_blk_dops; | |
1868 | md->disk->queue = md->queue; | |
1869 | md->disk->private_data = md; | |
1870 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1871 | add_disk(md->disk); | |
7e51f257 | 1872 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1873 | |
9c4376de TH |
1874 | md->wq = alloc_workqueue("kdmflush", |
1875 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1876 | if (!md->wq) |
1877 | goto bad_thread; | |
1878 | ||
32a926da MP |
1879 | md->bdev = bdget_disk(md->disk, 0); |
1880 | if (!md->bdev) | |
1881 | goto bad_bdev; | |
1882 | ||
6a8736d1 TH |
1883 | bio_init(&md->flush_bio); |
1884 | md->flush_bio.bi_bdev = md->bdev; | |
1885 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1886 | ||
ba61fdd1 | 1887 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1888 | spin_lock(&_minor_lock); |
ba61fdd1 | 1889 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1890 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1891 | |
1892 | BUG_ON(old_md != MINOR_ALLOCED); | |
1893 | ||
1da177e4 LT |
1894 | return md; |
1895 | ||
32a926da MP |
1896 | bad_bdev: |
1897 | destroy_workqueue(md->wq); | |
304f3f6a | 1898 | bad_thread: |
03022c54 | 1899 | del_gendisk(md->disk); |
304f3f6a | 1900 | put_disk(md->disk); |
6ed7ade8 | 1901 | bad_disk: |
1312f40e | 1902 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1903 | bad_queue: |
1da177e4 | 1904 | free_minor(minor); |
6ed7ade8 | 1905 | bad_minor: |
10da4f79 | 1906 | module_put(THIS_MODULE); |
6ed7ade8 | 1907 | bad_module_get: |
1da177e4 LT |
1908 | kfree(md); |
1909 | return NULL; | |
1910 | } | |
1911 | ||
ae9da83f JN |
1912 | static void unlock_fs(struct mapped_device *md); |
1913 | ||
1da177e4 LT |
1914 | static void free_dev(struct mapped_device *md) |
1915 | { | |
f331c029 | 1916 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1917 | |
32a926da MP |
1918 | unlock_fs(md); |
1919 | bdput(md->bdev); | |
304f3f6a | 1920 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1921 | if (md->tio_pool) |
1922 | mempool_destroy(md->tio_pool); | |
1923 | if (md->io_pool) | |
1924 | mempool_destroy(md->io_pool); | |
1925 | if (md->bs) | |
1926 | bioset_free(md->bs); | |
9c47008d | 1927 | blk_integrity_unregister(md->disk); |
1da177e4 | 1928 | del_gendisk(md->disk); |
63d94e48 | 1929 | free_minor(minor); |
fba9f90e JM |
1930 | |
1931 | spin_lock(&_minor_lock); | |
1932 | md->disk->private_data = NULL; | |
1933 | spin_unlock(&_minor_lock); | |
1934 | ||
1da177e4 | 1935 | put_disk(md->disk); |
1312f40e | 1936 | blk_cleanup_queue(md->queue); |
10da4f79 | 1937 | module_put(THIS_MODULE); |
1da177e4 LT |
1938 | kfree(md); |
1939 | } | |
1940 | ||
e6ee8c0b KU |
1941 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1942 | { | |
1943 | struct dm_md_mempools *p; | |
1944 | ||
dba14160 | 1945 | if (md->io_pool && (md->tio_pool || dm_table_get_type(t) == DM_TYPE_BIO_BASED) && md->bs) |
e6ee8c0b KU |
1946 | /* the md already has necessary mempools */ |
1947 | goto out; | |
1948 | ||
1949 | p = dm_table_get_md_mempools(t); | |
1950 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1951 | ||
1952 | md->io_pool = p->io_pool; | |
1953 | p->io_pool = NULL; | |
1954 | md->tio_pool = p->tio_pool; | |
1955 | p->tio_pool = NULL; | |
1956 | md->bs = p->bs; | |
1957 | p->bs = NULL; | |
1958 | ||
1959 | out: | |
1960 | /* mempool bind completed, now no need any mempools in the table */ | |
1961 | dm_table_free_md_mempools(t); | |
1962 | } | |
1963 | ||
1da177e4 LT |
1964 | /* |
1965 | * Bind a table to the device. | |
1966 | */ | |
1967 | static void event_callback(void *context) | |
1968 | { | |
7a8c3d3b MA |
1969 | unsigned long flags; |
1970 | LIST_HEAD(uevents); | |
1da177e4 LT |
1971 | struct mapped_device *md = (struct mapped_device *) context; |
1972 | ||
7a8c3d3b MA |
1973 | spin_lock_irqsave(&md->uevent_lock, flags); |
1974 | list_splice_init(&md->uevent_list, &uevents); | |
1975 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1976 | ||
ed9e1982 | 1977 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1978 | |
1da177e4 LT |
1979 | atomic_inc(&md->event_nr); |
1980 | wake_up(&md->eventq); | |
1981 | } | |
1982 | ||
c217649b MS |
1983 | /* |
1984 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
1985 | */ | |
4e90188b | 1986 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1987 | { |
4e90188b | 1988 | set_capacity(md->disk, size); |
1da177e4 | 1989 | |
db8fef4f | 1990 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
1991 | } |
1992 | ||
d5b9dd04 MP |
1993 | /* |
1994 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
1995 | * | |
1996 | * If this function returns 0, then the device is either a non-dm | |
1997 | * device without a merge_bvec_fn, or it is a dm device that is | |
1998 | * able to split any bios it receives that are too big. | |
1999 | */ | |
2000 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2001 | { | |
2002 | struct mapped_device *dev_md; | |
2003 | ||
2004 | if (!q->merge_bvec_fn) | |
2005 | return 0; | |
2006 | ||
2007 | if (q->make_request_fn == dm_request) { | |
2008 | dev_md = q->queuedata; | |
2009 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2010 | return 0; | |
2011 | } | |
2012 | ||
2013 | return 1; | |
2014 | } | |
2015 | ||
2016 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2017 | struct dm_dev *dev, sector_t start, | |
2018 | sector_t len, void *data) | |
2019 | { | |
2020 | struct block_device *bdev = dev->bdev; | |
2021 | struct request_queue *q = bdev_get_queue(bdev); | |
2022 | ||
2023 | return dm_queue_merge_is_compulsory(q); | |
2024 | } | |
2025 | ||
2026 | /* | |
2027 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2028 | * on the properties of the underlying devices. | |
2029 | */ | |
2030 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2031 | { | |
2032 | unsigned i = 0; | |
2033 | struct dm_target *ti; | |
2034 | ||
2035 | while (i < dm_table_get_num_targets(table)) { | |
2036 | ti = dm_table_get_target(table, i++); | |
2037 | ||
2038 | if (ti->type->iterate_devices && | |
2039 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2040 | return 0; | |
2041 | } | |
2042 | ||
2043 | return 1; | |
2044 | } | |
2045 | ||
042d2a9b AK |
2046 | /* |
2047 | * Returns old map, which caller must destroy. | |
2048 | */ | |
2049 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2050 | struct queue_limits *limits) | |
1da177e4 | 2051 | { |
042d2a9b | 2052 | struct dm_table *old_map; |
165125e1 | 2053 | struct request_queue *q = md->queue; |
1da177e4 | 2054 | sector_t size; |
523d9297 | 2055 | unsigned long flags; |
d5b9dd04 | 2056 | int merge_is_optional; |
1da177e4 LT |
2057 | |
2058 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2059 | |
2060 | /* | |
2061 | * Wipe any geometry if the size of the table changed. | |
2062 | */ | |
2063 | if (size != get_capacity(md->disk)) | |
2064 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2065 | ||
32a926da | 2066 | __set_size(md, size); |
d5816876 | 2067 | |
2ca3310e AK |
2068 | dm_table_event_callback(t, event_callback, md); |
2069 | ||
e6ee8c0b KU |
2070 | /* |
2071 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2072 | * for request-based during suspension. So stop it to prevent | |
2073 | * I/O mapping before resume. | |
2074 | * This must be done before setting the queue restrictions, | |
2075 | * because request-based dm may be run just after the setting. | |
2076 | */ | |
2077 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2078 | stop_queue(q); | |
2079 | ||
2080 | __bind_mempools(md, t); | |
2081 | ||
d5b9dd04 MP |
2082 | merge_is_optional = dm_table_merge_is_optional(t); |
2083 | ||
523d9297 | 2084 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2085 | old_map = md->map; |
1da177e4 | 2086 | md->map = t; |
36a0456f AK |
2087 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2088 | ||
754c5fc7 | 2089 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2090 | if (merge_is_optional) |
2091 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2092 | else | |
2093 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2094 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2095 | |
042d2a9b | 2096 | return old_map; |
1da177e4 LT |
2097 | } |
2098 | ||
a7940155 AK |
2099 | /* |
2100 | * Returns unbound table for the caller to free. | |
2101 | */ | |
2102 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2103 | { |
2104 | struct dm_table *map = md->map; | |
523d9297 | 2105 | unsigned long flags; |
1da177e4 LT |
2106 | |
2107 | if (!map) | |
a7940155 | 2108 | return NULL; |
1da177e4 LT |
2109 | |
2110 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2111 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2112 | md->map = NULL; |
523d9297 | 2113 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2114 | |
2115 | return map; | |
1da177e4 LT |
2116 | } |
2117 | ||
2118 | /* | |
2119 | * Constructor for a new device. | |
2120 | */ | |
2b06cfff | 2121 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2122 | { |
2123 | struct mapped_device *md; | |
2124 | ||
2b06cfff | 2125 | md = alloc_dev(minor); |
1da177e4 LT |
2126 | if (!md) |
2127 | return -ENXIO; | |
2128 | ||
784aae73 MB |
2129 | dm_sysfs_init(md); |
2130 | ||
1da177e4 LT |
2131 | *result = md; |
2132 | return 0; | |
2133 | } | |
2134 | ||
a5664dad MS |
2135 | /* |
2136 | * Functions to manage md->type. | |
2137 | * All are required to hold md->type_lock. | |
2138 | */ | |
2139 | void dm_lock_md_type(struct mapped_device *md) | |
2140 | { | |
2141 | mutex_lock(&md->type_lock); | |
2142 | } | |
2143 | ||
2144 | void dm_unlock_md_type(struct mapped_device *md) | |
2145 | { | |
2146 | mutex_unlock(&md->type_lock); | |
2147 | } | |
2148 | ||
2149 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2150 | { | |
2151 | md->type = type; | |
2152 | } | |
2153 | ||
2154 | unsigned dm_get_md_type(struct mapped_device *md) | |
2155 | { | |
2156 | return md->type; | |
2157 | } | |
2158 | ||
36a0456f AK |
2159 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2160 | { | |
2161 | return md->immutable_target_type; | |
2162 | } | |
2163 | ||
4a0b4ddf MS |
2164 | /* |
2165 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2166 | */ | |
2167 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2168 | { | |
2169 | struct request_queue *q = NULL; | |
2170 | ||
2171 | if (md->queue->elevator) | |
2172 | return 1; | |
2173 | ||
2174 | /* Fully initialize the queue */ | |
2175 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2176 | if (!q) | |
2177 | return 0; | |
2178 | ||
2179 | md->queue = q; | |
4a0b4ddf MS |
2180 | dm_init_md_queue(md); |
2181 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2182 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2183 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2184 | |
2185 | elv_register_queue(md->queue); | |
2186 | ||
2187 | return 1; | |
2188 | } | |
2189 | ||
2190 | /* | |
2191 | * Setup the DM device's queue based on md's type | |
2192 | */ | |
2193 | int dm_setup_md_queue(struct mapped_device *md) | |
2194 | { | |
2195 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2196 | !dm_init_request_based_queue(md)) { | |
2197 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2198 | return -EINVAL; | |
2199 | } | |
2200 | ||
2201 | return 0; | |
2202 | } | |
2203 | ||
637842cf | 2204 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2205 | { |
2206 | struct mapped_device *md; | |
1da177e4 LT |
2207 | unsigned minor = MINOR(dev); |
2208 | ||
2209 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2210 | return NULL; | |
2211 | ||
f32c10b0 | 2212 | spin_lock(&_minor_lock); |
1da177e4 LT |
2213 | |
2214 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2215 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2216 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2217 | dm_deleting_md(md) || |
17b2f66f | 2218 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2219 | md = NULL; |
fba9f90e JM |
2220 | goto out; |
2221 | } | |
1da177e4 | 2222 | |
fba9f90e | 2223 | out: |
f32c10b0 | 2224 | spin_unlock(&_minor_lock); |
1da177e4 | 2225 | |
637842cf DT |
2226 | return md; |
2227 | } | |
2228 | ||
d229a958 DT |
2229 | struct mapped_device *dm_get_md(dev_t dev) |
2230 | { | |
2231 | struct mapped_device *md = dm_find_md(dev); | |
2232 | ||
2233 | if (md) | |
2234 | dm_get(md); | |
2235 | ||
2236 | return md; | |
2237 | } | |
3cf2e4ba | 2238 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2239 | |
9ade92a9 | 2240 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2241 | { |
9ade92a9 | 2242 | return md->interface_ptr; |
1da177e4 LT |
2243 | } |
2244 | ||
2245 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2246 | { | |
2247 | md->interface_ptr = ptr; | |
2248 | } | |
2249 | ||
2250 | void dm_get(struct mapped_device *md) | |
2251 | { | |
2252 | atomic_inc(&md->holders); | |
3f77316d | 2253 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2254 | } |
2255 | ||
72d94861 AK |
2256 | const char *dm_device_name(struct mapped_device *md) |
2257 | { | |
2258 | return md->name; | |
2259 | } | |
2260 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2261 | ||
3f77316d | 2262 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2263 | { |
1134e5ae | 2264 | struct dm_table *map; |
1da177e4 | 2265 | |
3f77316d | 2266 | might_sleep(); |
fba9f90e | 2267 | |
3f77316d KU |
2268 | spin_lock(&_minor_lock); |
2269 | map = dm_get_live_table(md); | |
2270 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2271 | set_bit(DMF_FREEING, &md->flags); | |
2272 | spin_unlock(&_minor_lock); | |
2273 | ||
2274 | if (!dm_suspended_md(md)) { | |
2275 | dm_table_presuspend_targets(map); | |
2276 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2277 | } |
3f77316d KU |
2278 | |
2279 | /* | |
2280 | * Rare, but there may be I/O requests still going to complete, | |
2281 | * for example. Wait for all references to disappear. | |
2282 | * No one should increment the reference count of the mapped_device, | |
2283 | * after the mapped_device state becomes DMF_FREEING. | |
2284 | */ | |
2285 | if (wait) | |
2286 | while (atomic_read(&md->holders)) | |
2287 | msleep(1); | |
2288 | else if (atomic_read(&md->holders)) | |
2289 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2290 | dm_device_name(md), atomic_read(&md->holders)); | |
2291 | ||
2292 | dm_sysfs_exit(md); | |
2293 | dm_table_put(map); | |
2294 | dm_table_destroy(__unbind(md)); | |
2295 | free_dev(md); | |
2296 | } | |
2297 | ||
2298 | void dm_destroy(struct mapped_device *md) | |
2299 | { | |
2300 | __dm_destroy(md, true); | |
2301 | } | |
2302 | ||
2303 | void dm_destroy_immediate(struct mapped_device *md) | |
2304 | { | |
2305 | __dm_destroy(md, false); | |
2306 | } | |
2307 | ||
2308 | void dm_put(struct mapped_device *md) | |
2309 | { | |
2310 | atomic_dec(&md->holders); | |
1da177e4 | 2311 | } |
79eb885c | 2312 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2313 | |
401600df | 2314 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2315 | { |
2316 | int r = 0; | |
b44ebeb0 MP |
2317 | DECLARE_WAITQUEUE(wait, current); |
2318 | ||
b44ebeb0 | 2319 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2320 | |
2321 | while (1) { | |
401600df | 2322 | set_current_state(interruptible); |
46125c1c | 2323 | |
b4324fee | 2324 | if (!md_in_flight(md)) |
46125c1c MB |
2325 | break; |
2326 | ||
401600df MP |
2327 | if (interruptible == TASK_INTERRUPTIBLE && |
2328 | signal_pending(current)) { | |
46125c1c MB |
2329 | r = -EINTR; |
2330 | break; | |
2331 | } | |
2332 | ||
2333 | io_schedule(); | |
2334 | } | |
2335 | set_current_state(TASK_RUNNING); | |
2336 | ||
b44ebeb0 MP |
2337 | remove_wait_queue(&md->wait, &wait); |
2338 | ||
46125c1c MB |
2339 | return r; |
2340 | } | |
2341 | ||
1da177e4 LT |
2342 | /* |
2343 | * Process the deferred bios | |
2344 | */ | |
ef208587 | 2345 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2346 | { |
ef208587 MP |
2347 | struct mapped_device *md = container_of(work, struct mapped_device, |
2348 | work); | |
6d6f10df | 2349 | struct bio *c; |
1da177e4 | 2350 | |
6a8736d1 | 2351 | down_read(&md->io_lock); |
ef208587 | 2352 | |
3b00b203 | 2353 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2354 | spin_lock_irq(&md->deferred_lock); |
2355 | c = bio_list_pop(&md->deferred); | |
2356 | spin_unlock_irq(&md->deferred_lock); | |
2357 | ||
6a8736d1 | 2358 | if (!c) |
df12ee99 | 2359 | break; |
022c2611 | 2360 | |
6a8736d1 | 2361 | up_read(&md->io_lock); |
3b00b203 | 2362 | |
e6ee8c0b KU |
2363 | if (dm_request_based(md)) |
2364 | generic_make_request(c); | |
6a8736d1 TH |
2365 | else |
2366 | __split_and_process_bio(md, c); | |
3b00b203 | 2367 | |
6a8736d1 | 2368 | down_read(&md->io_lock); |
022c2611 | 2369 | } |
73d410c0 | 2370 | |
6a8736d1 | 2371 | up_read(&md->io_lock); |
1da177e4 LT |
2372 | } |
2373 | ||
9a1fb464 | 2374 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2375 | { |
3b00b203 MP |
2376 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2377 | smp_mb__after_clear_bit(); | |
53d5914f | 2378 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2379 | } |
2380 | ||
1da177e4 | 2381 | /* |
042d2a9b | 2382 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2383 | */ |
042d2a9b | 2384 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2385 | { |
3ae70656 | 2386 | struct dm_table *live_map, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2387 | struct queue_limits limits; |
042d2a9b | 2388 | int r; |
1da177e4 | 2389 | |
e61290a4 | 2390 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2391 | |
2392 | /* device must be suspended */ | |
4f186f8b | 2393 | if (!dm_suspended_md(md)) |
93c534ae | 2394 | goto out; |
1da177e4 | 2395 | |
3ae70656 MS |
2396 | /* |
2397 | * If the new table has no data devices, retain the existing limits. | |
2398 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2399 | * then new I/O is queued based on these limits, and then some paths | |
2400 | * reappear. | |
2401 | */ | |
2402 | if (dm_table_has_no_data_devices(table)) { | |
2403 | live_map = dm_get_live_table(md); | |
2404 | if (live_map) | |
2405 | limits = md->queue->limits; | |
2406 | dm_table_put(live_map); | |
2407 | } | |
2408 | ||
754c5fc7 | 2409 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2410 | if (r) { |
2411 | map = ERR_PTR(r); | |
754c5fc7 | 2412 | goto out; |
042d2a9b | 2413 | } |
754c5fc7 | 2414 | |
042d2a9b | 2415 | map = __bind(md, table, &limits); |
1da177e4 | 2416 | |
93c534ae | 2417 | out: |
e61290a4 | 2418 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2419 | return map; |
1da177e4 LT |
2420 | } |
2421 | ||
2422 | /* | |
2423 | * Functions to lock and unlock any filesystem running on the | |
2424 | * device. | |
2425 | */ | |
2ca3310e | 2426 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2427 | { |
e39e2e95 | 2428 | int r; |
1da177e4 LT |
2429 | |
2430 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2431 | |
db8fef4f | 2432 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2433 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2434 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2435 | md->frozen_sb = NULL; |
2436 | return r; | |
dfbe03f6 AK |
2437 | } |
2438 | ||
aa8d7c2f AK |
2439 | set_bit(DMF_FROZEN, &md->flags); |
2440 | ||
1da177e4 LT |
2441 | return 0; |
2442 | } | |
2443 | ||
2ca3310e | 2444 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2445 | { |
aa8d7c2f AK |
2446 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2447 | return; | |
2448 | ||
db8fef4f | 2449 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2450 | md->frozen_sb = NULL; |
aa8d7c2f | 2451 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2452 | } |
2453 | ||
2454 | /* | |
2455 | * We need to be able to change a mapping table under a mounted | |
2456 | * filesystem. For example we might want to move some data in | |
2457 | * the background. Before the table can be swapped with | |
2458 | * dm_bind_table, dm_suspend must be called to flush any in | |
2459 | * flight bios and ensure that any further io gets deferred. | |
2460 | */ | |
cec47e3d KU |
2461 | /* |
2462 | * Suspend mechanism in request-based dm. | |
2463 | * | |
9f518b27 KU |
2464 | * 1. Flush all I/Os by lock_fs() if needed. |
2465 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2466 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2467 | * |
9f518b27 | 2468 | * To abort suspend, start the request_queue. |
cec47e3d | 2469 | */ |
a3d77d35 | 2470 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2471 | { |
2ca3310e | 2472 | struct dm_table *map = NULL; |
46125c1c | 2473 | int r = 0; |
a3d77d35 | 2474 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2475 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2476 | |
e61290a4 | 2477 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2478 | |
4f186f8b | 2479 | if (dm_suspended_md(md)) { |
73d410c0 | 2480 | r = -EINVAL; |
d287483d | 2481 | goto out_unlock; |
73d410c0 | 2482 | } |
1da177e4 | 2483 | |
7c666411 | 2484 | map = dm_get_live_table(md); |
1da177e4 | 2485 | |
2e93ccc1 KU |
2486 | /* |
2487 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2488 | * This flag is cleared before dm_suspend returns. | |
2489 | */ | |
2490 | if (noflush) | |
2491 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2492 | ||
cf222b37 AK |
2493 | /* This does not get reverted if there's an error later. */ |
2494 | dm_table_presuspend_targets(map); | |
2495 | ||
32a926da | 2496 | /* |
9f518b27 KU |
2497 | * Flush I/O to the device. |
2498 | * Any I/O submitted after lock_fs() may not be flushed. | |
2499 | * noflush takes precedence over do_lockfs. | |
2500 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2501 | */ |
2502 | if (!noflush && do_lockfs) { | |
2503 | r = lock_fs(md); | |
2504 | if (r) | |
f431d966 | 2505 | goto out; |
aa8d7c2f | 2506 | } |
1da177e4 LT |
2507 | |
2508 | /* | |
3b00b203 MP |
2509 | * Here we must make sure that no processes are submitting requests |
2510 | * to target drivers i.e. no one may be executing | |
2511 | * __split_and_process_bio. This is called from dm_request and | |
2512 | * dm_wq_work. | |
2513 | * | |
2514 | * To get all processes out of __split_and_process_bio in dm_request, | |
2515 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2516 | * __split_and_process_bio from dm_request and quiesce the thread |
2517 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2518 | * flush_workqueue(md->wq). | |
1da177e4 | 2519 | */ |
2ca3310e | 2520 | down_write(&md->io_lock); |
1eb787ec | 2521 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2522 | up_write(&md->io_lock); |
1da177e4 | 2523 | |
d0bcb878 | 2524 | /* |
29e4013d TH |
2525 | * Stop md->queue before flushing md->wq in case request-based |
2526 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2527 | */ |
cec47e3d | 2528 | if (dm_request_based(md)) |
9f518b27 | 2529 | stop_queue(md->queue); |
cec47e3d | 2530 | |
d0bcb878 KU |
2531 | flush_workqueue(md->wq); |
2532 | ||
1da177e4 | 2533 | /* |
3b00b203 MP |
2534 | * At this point no more requests are entering target request routines. |
2535 | * We call dm_wait_for_completion to wait for all existing requests | |
2536 | * to finish. | |
1da177e4 | 2537 | */ |
401600df | 2538 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2539 | |
2ca3310e | 2540 | down_write(&md->io_lock); |
6d6f10df | 2541 | if (noflush) |
022c2611 | 2542 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2543 | up_write(&md->io_lock); |
2e93ccc1 | 2544 | |
1da177e4 | 2545 | /* were we interrupted ? */ |
46125c1c | 2546 | if (r < 0) { |
9a1fb464 | 2547 | dm_queue_flush(md); |
73d410c0 | 2548 | |
cec47e3d | 2549 | if (dm_request_based(md)) |
9f518b27 | 2550 | start_queue(md->queue); |
cec47e3d | 2551 | |
2ca3310e | 2552 | unlock_fs(md); |
2e93ccc1 | 2553 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2554 | } |
1da177e4 | 2555 | |
3b00b203 MP |
2556 | /* |
2557 | * If dm_wait_for_completion returned 0, the device is completely | |
2558 | * quiescent now. There is no request-processing activity. All new | |
2559 | * requests are being added to md->deferred list. | |
2560 | */ | |
2561 | ||
2ca3310e | 2562 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2563 | |
4d4471cb KU |
2564 | dm_table_postsuspend_targets(map); |
2565 | ||
2ca3310e AK |
2566 | out: |
2567 | dm_table_put(map); | |
d287483d AK |
2568 | |
2569 | out_unlock: | |
e61290a4 | 2570 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2571 | return r; |
1da177e4 LT |
2572 | } |
2573 | ||
2574 | int dm_resume(struct mapped_device *md) | |
2575 | { | |
cf222b37 | 2576 | int r = -EINVAL; |
cf222b37 | 2577 | struct dm_table *map = NULL; |
1da177e4 | 2578 | |
e61290a4 | 2579 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2580 | if (!dm_suspended_md(md)) |
cf222b37 | 2581 | goto out; |
cf222b37 | 2582 | |
7c666411 | 2583 | map = dm_get_live_table(md); |
2ca3310e | 2584 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2585 | goto out; |
1da177e4 | 2586 | |
8757b776 MB |
2587 | r = dm_table_resume_targets(map); |
2588 | if (r) | |
2589 | goto out; | |
2ca3310e | 2590 | |
9a1fb464 | 2591 | dm_queue_flush(md); |
2ca3310e | 2592 | |
cec47e3d KU |
2593 | /* |
2594 | * Flushing deferred I/Os must be done after targets are resumed | |
2595 | * so that mapping of targets can work correctly. | |
2596 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2597 | */ | |
2598 | if (dm_request_based(md)) | |
2599 | start_queue(md->queue); | |
2600 | ||
2ca3310e AK |
2601 | unlock_fs(md); |
2602 | ||
2603 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2604 | ||
cf222b37 AK |
2605 | r = 0; |
2606 | out: | |
2607 | dm_table_put(map); | |
e61290a4 | 2608 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2609 | |
cf222b37 | 2610 | return r; |
1da177e4 LT |
2611 | } |
2612 | ||
2613 | /*----------------------------------------------------------------- | |
2614 | * Event notification. | |
2615 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2616 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2617 | unsigned cookie) |
69267a30 | 2618 | { |
60935eb2 MB |
2619 | char udev_cookie[DM_COOKIE_LENGTH]; |
2620 | char *envp[] = { udev_cookie, NULL }; | |
2621 | ||
2622 | if (!cookie) | |
3abf85b5 | 2623 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2624 | else { |
2625 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2626 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2627 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2628 | action, envp); | |
60935eb2 | 2629 | } |
69267a30 AK |
2630 | } |
2631 | ||
7a8c3d3b MA |
2632 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2633 | { | |
2634 | return atomic_add_return(1, &md->uevent_seq); | |
2635 | } | |
2636 | ||
1da177e4 LT |
2637 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2638 | { | |
2639 | return atomic_read(&md->event_nr); | |
2640 | } | |
2641 | ||
2642 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2643 | { | |
2644 | return wait_event_interruptible(md->eventq, | |
2645 | (event_nr != atomic_read(&md->event_nr))); | |
2646 | } | |
2647 | ||
7a8c3d3b MA |
2648 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2649 | { | |
2650 | unsigned long flags; | |
2651 | ||
2652 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2653 | list_add(elist, &md->uevent_list); | |
2654 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2655 | } | |
2656 | ||
1da177e4 LT |
2657 | /* |
2658 | * The gendisk is only valid as long as you have a reference | |
2659 | * count on 'md'. | |
2660 | */ | |
2661 | struct gendisk *dm_disk(struct mapped_device *md) | |
2662 | { | |
2663 | return md->disk; | |
2664 | } | |
2665 | ||
784aae73 MB |
2666 | struct kobject *dm_kobject(struct mapped_device *md) |
2667 | { | |
2668 | return &md->kobj; | |
2669 | } | |
2670 | ||
2671 | /* | |
2672 | * struct mapped_device should not be exported outside of dm.c | |
2673 | * so use this check to verify that kobj is part of md structure | |
2674 | */ | |
2675 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2676 | { | |
2677 | struct mapped_device *md; | |
2678 | ||
2679 | md = container_of(kobj, struct mapped_device, kobj); | |
2680 | if (&md->kobj != kobj) | |
2681 | return NULL; | |
2682 | ||
4d89b7b4 | 2683 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2684 | dm_deleting_md(md)) |
4d89b7b4 MB |
2685 | return NULL; |
2686 | ||
784aae73 MB |
2687 | dm_get(md); |
2688 | return md; | |
2689 | } | |
2690 | ||
4f186f8b | 2691 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2692 | { |
2693 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2694 | } | |
2695 | ||
64dbce58 KU |
2696 | int dm_suspended(struct dm_target *ti) |
2697 | { | |
ecdb2e25 | 2698 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2699 | } |
2700 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2701 | ||
2e93ccc1 KU |
2702 | int dm_noflush_suspending(struct dm_target *ti) |
2703 | { | |
ecdb2e25 | 2704 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2705 | } |
2706 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2707 | ||
a91a2785 | 2708 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity) |
e6ee8c0b KU |
2709 | { |
2710 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2711 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2712 | |
2713 | if (!pools) | |
2714 | return NULL; | |
2715 | ||
2716 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2717 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2718 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2719 | if (!pools->io_pool) | |
2720 | goto free_pools_and_out; | |
2721 | ||
dba14160 MP |
2722 | pools->tio_pool = NULL; |
2723 | if (type == DM_TYPE_REQUEST_BASED) { | |
2724 | pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2725 | if (!pools->tio_pool) | |
2726 | goto free_io_pool_and_out; | |
2727 | } | |
e6ee8c0b | 2728 | |
94818742 | 2729 | pools->bs = (type == DM_TYPE_BIO_BASED) ? |
dba14160 MP |
2730 | bioset_create(pool_size, |
2731 | offsetof(struct dm_target_io, clone)) : | |
94818742 KO |
2732 | bioset_create(pool_size, |
2733 | offsetof(struct dm_rq_clone_bio_info, clone)); | |
e6ee8c0b KU |
2734 | if (!pools->bs) |
2735 | goto free_tio_pool_and_out; | |
2736 | ||
a91a2785 MP |
2737 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2738 | goto free_bioset_and_out; | |
2739 | ||
e6ee8c0b KU |
2740 | return pools; |
2741 | ||
a91a2785 MP |
2742 | free_bioset_and_out: |
2743 | bioset_free(pools->bs); | |
2744 | ||
e6ee8c0b | 2745 | free_tio_pool_and_out: |
dba14160 MP |
2746 | if (pools->tio_pool) |
2747 | mempool_destroy(pools->tio_pool); | |
e6ee8c0b KU |
2748 | |
2749 | free_io_pool_and_out: | |
2750 | mempool_destroy(pools->io_pool); | |
2751 | ||
2752 | free_pools_and_out: | |
2753 | kfree(pools); | |
2754 | ||
2755 | return NULL; | |
2756 | } | |
2757 | ||
2758 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2759 | { | |
2760 | if (!pools) | |
2761 | return; | |
2762 | ||
2763 | if (pools->io_pool) | |
2764 | mempool_destroy(pools->io_pool); | |
2765 | ||
2766 | if (pools->tio_pool) | |
2767 | mempool_destroy(pools->tio_pool); | |
2768 | ||
2769 | if (pools->bs) | |
2770 | bioset_free(pools->bs); | |
2771 | ||
2772 | kfree(pools); | |
2773 | } | |
2774 | ||
83d5cde4 | 2775 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2776 | .open = dm_blk_open, |
2777 | .release = dm_blk_close, | |
aa129a22 | 2778 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2779 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2780 | .owner = THIS_MODULE |
2781 | }; | |
2782 | ||
2783 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2784 | ||
2785 | /* | |
2786 | * module hooks | |
2787 | */ | |
2788 | module_init(dm_init); | |
2789 | module_exit(dm_exit); | |
2790 | ||
2791 | module_param(major, uint, 0); | |
2792 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2793 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2794 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2795 | MODULE_LICENSE("GPL"); |