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); |
2c140a24 MP |
52 | |
53 | static void do_deferred_remove(struct work_struct *w); | |
54 | ||
55 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
56 | ||
1da177e4 | 57 | /* |
8fbf26ad | 58 | * For bio-based dm. |
1da177e4 LT |
59 | * One of these is allocated per bio. |
60 | */ | |
61 | struct dm_io { | |
62 | struct mapped_device *md; | |
63 | int error; | |
1da177e4 | 64 | atomic_t io_count; |
6ae2fa67 | 65 | struct bio *bio; |
3eaf840e | 66 | unsigned long start_time; |
f88fb981 | 67 | spinlock_t endio_lock; |
fd2ed4d2 | 68 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
69 | }; |
70 | ||
8fbf26ad KU |
71 | /* |
72 | * For request-based dm. | |
73 | * One of these is allocated per request. | |
74 | */ | |
75 | struct dm_rq_target_io { | |
76 | struct mapped_device *md; | |
77 | struct dm_target *ti; | |
78 | struct request *orig, clone; | |
79 | int error; | |
80 | union map_info info; | |
81 | }; | |
82 | ||
83 | /* | |
94818742 KO |
84 | * For request-based dm - the bio clones we allocate are embedded in these |
85 | * structs. | |
86 | * | |
87 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
88 | * the bioset is created - this means the bio has to come at the end of the | |
89 | * struct. | |
8fbf26ad KU |
90 | */ |
91 | struct dm_rq_clone_bio_info { | |
92 | struct bio *orig; | |
cec47e3d | 93 | struct dm_rq_target_io *tio; |
94818742 | 94 | struct bio clone; |
8fbf26ad KU |
95 | }; |
96 | ||
1da177e4 LT |
97 | union map_info *dm_get_mapinfo(struct bio *bio) |
98 | { | |
17b2f66f | 99 | if (bio && bio->bi_private) |
028867ac | 100 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 101 | return NULL; |
1da177e4 LT |
102 | } |
103 | ||
cec47e3d KU |
104 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
105 | { | |
106 | if (rq && rq->end_io_data) | |
107 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
108 | return NULL; | |
109 | } | |
110 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
111 | ||
ba61fdd1 JM |
112 | #define MINOR_ALLOCED ((void *)-1) |
113 | ||
1da177e4 LT |
114 | /* |
115 | * Bits for the md->flags field. | |
116 | */ | |
1eb787ec | 117 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 118 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 119 | #define DMF_FROZEN 2 |
fba9f90e | 120 | #define DMF_FREEING 3 |
5c6bd75d | 121 | #define DMF_DELETING 4 |
2e93ccc1 | 122 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 123 | #define DMF_MERGE_IS_OPTIONAL 6 |
2c140a24 | 124 | #define DMF_DEFERRED_REMOVE 7 |
1da177e4 | 125 | |
83d5e5b0 MP |
126 | /* |
127 | * A dummy definition to make RCU happy. | |
128 | * struct dm_table should never be dereferenced in this file. | |
129 | */ | |
130 | struct dm_table { | |
131 | int undefined__; | |
132 | }; | |
133 | ||
304f3f6a MB |
134 | /* |
135 | * Work processed by per-device workqueue. | |
136 | */ | |
1da177e4 | 137 | struct mapped_device { |
83d5e5b0 | 138 | struct srcu_struct io_barrier; |
e61290a4 | 139 | struct mutex suspend_lock; |
1da177e4 | 140 | atomic_t holders; |
5c6bd75d | 141 | atomic_t open_count; |
1da177e4 | 142 | |
2a7faeb1 MP |
143 | /* |
144 | * The current mapping. | |
145 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
146 | * dereference. | |
147 | */ | |
148 | struct dm_table *map; | |
149 | ||
1da177e4 LT |
150 | unsigned long flags; |
151 | ||
165125e1 | 152 | struct request_queue *queue; |
a5664dad | 153 | unsigned type; |
4a0b4ddf | 154 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
155 | struct mutex type_lock; |
156 | ||
36a0456f AK |
157 | struct target_type *immutable_target_type; |
158 | ||
1da177e4 | 159 | struct gendisk *disk; |
7e51f257 | 160 | char name[16]; |
1da177e4 LT |
161 | |
162 | void *interface_ptr; | |
163 | ||
164 | /* | |
165 | * A list of ios that arrived while we were suspended. | |
166 | */ | |
316d315b | 167 | atomic_t pending[2]; |
1da177e4 | 168 | wait_queue_head_t wait; |
53d5914f | 169 | struct work_struct work; |
74859364 | 170 | struct bio_list deferred; |
022c2611 | 171 | spinlock_t deferred_lock; |
1da177e4 | 172 | |
af7e466a | 173 | /* |
29e4013d | 174 | * Processing queue (flush) |
304f3f6a MB |
175 | */ |
176 | struct workqueue_struct *wq; | |
177 | ||
1da177e4 LT |
178 | /* |
179 | * io objects are allocated from here. | |
180 | */ | |
181 | mempool_t *io_pool; | |
1da177e4 | 182 | |
9faf400f SB |
183 | struct bio_set *bs; |
184 | ||
1da177e4 LT |
185 | /* |
186 | * Event handling. | |
187 | */ | |
188 | atomic_t event_nr; | |
189 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
190 | atomic_t uevent_seq; |
191 | struct list_head uevent_list; | |
192 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
193 | |
194 | /* | |
195 | * freeze/thaw support require holding onto a super block | |
196 | */ | |
197 | struct super_block *frozen_sb; | |
db8fef4f | 198 | struct block_device *bdev; |
3ac51e74 DW |
199 | |
200 | /* forced geometry settings */ | |
201 | struct hd_geometry geometry; | |
784aae73 MB |
202 | |
203 | /* sysfs handle */ | |
204 | struct kobject kobj; | |
52b1fd5a | 205 | |
d87f4c14 TH |
206 | /* zero-length flush that will be cloned and submitted to targets */ |
207 | struct bio flush_bio; | |
fd2ed4d2 MP |
208 | |
209 | struct dm_stats stats; | |
1da177e4 LT |
210 | }; |
211 | ||
e6ee8c0b KU |
212 | /* |
213 | * For mempools pre-allocation at the table loading time. | |
214 | */ | |
215 | struct dm_md_mempools { | |
216 | mempool_t *io_pool; | |
e6ee8c0b KU |
217 | struct bio_set *bs; |
218 | }; | |
219 | ||
6cfa5857 MS |
220 | #define RESERVED_BIO_BASED_IOS 16 |
221 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 222 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 223 | static struct kmem_cache *_io_cache; |
8fbf26ad | 224 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 225 | |
e8603136 MS |
226 | /* |
227 | * Bio-based DM's mempools' reserved IOs set by the user. | |
228 | */ | |
229 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
230 | ||
f4790826 MS |
231 | /* |
232 | * Request-based DM's mempools' reserved IOs set by the user. | |
233 | */ | |
234 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
235 | ||
236 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
237 | unsigned def, unsigned max) | |
238 | { | |
239 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
240 | unsigned modified_ios = 0; | |
241 | ||
242 | if (!ios) | |
243 | modified_ios = def; | |
244 | else if (ios > max) | |
245 | modified_ios = max; | |
246 | ||
247 | if (modified_ios) { | |
248 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
249 | ios = modified_ios; | |
250 | } | |
251 | ||
252 | return ios; | |
253 | } | |
254 | ||
e8603136 MS |
255 | unsigned dm_get_reserved_bio_based_ios(void) |
256 | { | |
257 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
258 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
259 | } | |
260 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
261 | ||
f4790826 MS |
262 | unsigned dm_get_reserved_rq_based_ios(void) |
263 | { | |
264 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
265 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
266 | } | |
267 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
268 | ||
1da177e4 LT |
269 | static int __init local_init(void) |
270 | { | |
51157b4a | 271 | int r = -ENOMEM; |
1da177e4 | 272 | |
1da177e4 | 273 | /* allocate a slab for the dm_ios */ |
028867ac | 274 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 275 | if (!_io_cache) |
51157b4a | 276 | return r; |
1da177e4 | 277 | |
8fbf26ad KU |
278 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
279 | if (!_rq_tio_cache) | |
dba14160 | 280 | goto out_free_io_cache; |
8fbf26ad | 281 | |
51e5b2bd | 282 | r = dm_uevent_init(); |
51157b4a | 283 | if (r) |
23e5083b | 284 | goto out_free_rq_tio_cache; |
51e5b2bd | 285 | |
1da177e4 LT |
286 | _major = major; |
287 | r = register_blkdev(_major, _name); | |
51157b4a KU |
288 | if (r < 0) |
289 | goto out_uevent_exit; | |
1da177e4 LT |
290 | |
291 | if (!_major) | |
292 | _major = r; | |
293 | ||
294 | return 0; | |
51157b4a KU |
295 | |
296 | out_uevent_exit: | |
297 | dm_uevent_exit(); | |
8fbf26ad KU |
298 | out_free_rq_tio_cache: |
299 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
300 | out_free_io_cache: |
301 | kmem_cache_destroy(_io_cache); | |
302 | ||
303 | return r; | |
1da177e4 LT |
304 | } |
305 | ||
306 | static void local_exit(void) | |
307 | { | |
2c140a24 MP |
308 | flush_scheduled_work(); |
309 | ||
8fbf26ad | 310 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 311 | kmem_cache_destroy(_io_cache); |
00d59405 | 312 | unregister_blkdev(_major, _name); |
51e5b2bd | 313 | dm_uevent_exit(); |
1da177e4 LT |
314 | |
315 | _major = 0; | |
316 | ||
317 | DMINFO("cleaned up"); | |
318 | } | |
319 | ||
b9249e55 | 320 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
321 | local_init, |
322 | dm_target_init, | |
323 | dm_linear_init, | |
324 | dm_stripe_init, | |
952b3557 | 325 | dm_io_init, |
945fa4d2 | 326 | dm_kcopyd_init, |
1da177e4 | 327 | dm_interface_init, |
fd2ed4d2 | 328 | dm_statistics_init, |
1da177e4 LT |
329 | }; |
330 | ||
b9249e55 | 331 | static void (*_exits[])(void) = { |
1da177e4 LT |
332 | local_exit, |
333 | dm_target_exit, | |
334 | dm_linear_exit, | |
335 | dm_stripe_exit, | |
952b3557 | 336 | dm_io_exit, |
945fa4d2 | 337 | dm_kcopyd_exit, |
1da177e4 | 338 | dm_interface_exit, |
fd2ed4d2 | 339 | dm_statistics_exit, |
1da177e4 LT |
340 | }; |
341 | ||
342 | static int __init dm_init(void) | |
343 | { | |
344 | const int count = ARRAY_SIZE(_inits); | |
345 | ||
346 | int r, i; | |
347 | ||
348 | for (i = 0; i < count; i++) { | |
349 | r = _inits[i](); | |
350 | if (r) | |
351 | goto bad; | |
352 | } | |
353 | ||
354 | return 0; | |
355 | ||
356 | bad: | |
357 | while (i--) | |
358 | _exits[i](); | |
359 | ||
360 | return r; | |
361 | } | |
362 | ||
363 | static void __exit dm_exit(void) | |
364 | { | |
365 | int i = ARRAY_SIZE(_exits); | |
366 | ||
367 | while (i--) | |
368 | _exits[i](); | |
d15b774c AK |
369 | |
370 | /* | |
371 | * Should be empty by this point. | |
372 | */ | |
d15b774c | 373 | idr_destroy(&_minor_idr); |
1da177e4 LT |
374 | } |
375 | ||
376 | /* | |
377 | * Block device functions | |
378 | */ | |
432a212c MA |
379 | int dm_deleting_md(struct mapped_device *md) |
380 | { | |
381 | return test_bit(DMF_DELETING, &md->flags); | |
382 | } | |
383 | ||
fe5f9f2c | 384 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
385 | { |
386 | struct mapped_device *md; | |
387 | ||
fba9f90e JM |
388 | spin_lock(&_minor_lock); |
389 | ||
fe5f9f2c | 390 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
391 | if (!md) |
392 | goto out; | |
393 | ||
5c6bd75d | 394 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 395 | dm_deleting_md(md)) { |
fba9f90e JM |
396 | md = NULL; |
397 | goto out; | |
398 | } | |
399 | ||
1da177e4 | 400 | dm_get(md); |
5c6bd75d | 401 | atomic_inc(&md->open_count); |
fba9f90e JM |
402 | |
403 | out: | |
404 | spin_unlock(&_minor_lock); | |
405 | ||
406 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
407 | } |
408 | ||
db2a144b | 409 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 410 | { |
fe5f9f2c | 411 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 412 | |
4a1aeb98 MB |
413 | spin_lock(&_minor_lock); |
414 | ||
2c140a24 MP |
415 | if (atomic_dec_and_test(&md->open_count) && |
416 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
417 | schedule_work(&deferred_remove_work); | |
418 | ||
1da177e4 | 419 | dm_put(md); |
4a1aeb98 MB |
420 | |
421 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
422 | } |
423 | ||
5c6bd75d AK |
424 | int dm_open_count(struct mapped_device *md) |
425 | { | |
426 | return atomic_read(&md->open_count); | |
427 | } | |
428 | ||
429 | /* | |
430 | * Guarantees nothing is using the device before it's deleted. | |
431 | */ | |
2c140a24 | 432 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
433 | { |
434 | int r = 0; | |
435 | ||
436 | spin_lock(&_minor_lock); | |
437 | ||
2c140a24 | 438 | if (dm_open_count(md)) { |
5c6bd75d | 439 | r = -EBUSY; |
2c140a24 MP |
440 | if (mark_deferred) |
441 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
442 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
443 | r = -EEXIST; | |
5c6bd75d AK |
444 | else |
445 | set_bit(DMF_DELETING, &md->flags); | |
446 | ||
447 | spin_unlock(&_minor_lock); | |
448 | ||
449 | return r; | |
450 | } | |
451 | ||
2c140a24 MP |
452 | int dm_cancel_deferred_remove(struct mapped_device *md) |
453 | { | |
454 | int r = 0; | |
455 | ||
456 | spin_lock(&_minor_lock); | |
457 | ||
458 | if (test_bit(DMF_DELETING, &md->flags)) | |
459 | r = -EBUSY; | |
460 | else | |
461 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
462 | ||
463 | spin_unlock(&_minor_lock); | |
464 | ||
465 | return r; | |
466 | } | |
467 | ||
468 | static void do_deferred_remove(struct work_struct *w) | |
469 | { | |
470 | dm_deferred_remove(); | |
471 | } | |
472 | ||
fd2ed4d2 MP |
473 | sector_t dm_get_size(struct mapped_device *md) |
474 | { | |
475 | return get_capacity(md->disk); | |
476 | } | |
477 | ||
478 | struct dm_stats *dm_get_stats(struct mapped_device *md) | |
479 | { | |
480 | return &md->stats; | |
481 | } | |
482 | ||
3ac51e74 DW |
483 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
484 | { | |
485 | struct mapped_device *md = bdev->bd_disk->private_data; | |
486 | ||
487 | return dm_get_geometry(md, geo); | |
488 | } | |
489 | ||
fe5f9f2c | 490 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
491 | unsigned int cmd, unsigned long arg) |
492 | { | |
fe5f9f2c | 493 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 494 | int srcu_idx; |
6c182cd8 | 495 | struct dm_table *map; |
aa129a22 MB |
496 | struct dm_target *tgt; |
497 | int r = -ENOTTY; | |
498 | ||
6c182cd8 | 499 | retry: |
83d5e5b0 MP |
500 | map = dm_get_live_table(md, &srcu_idx); |
501 | ||
aa129a22 MB |
502 | if (!map || !dm_table_get_size(map)) |
503 | goto out; | |
504 | ||
505 | /* We only support devices that have a single target */ | |
506 | if (dm_table_get_num_targets(map) != 1) | |
507 | goto out; | |
508 | ||
509 | tgt = dm_table_get_target(map, 0); | |
510 | ||
4f186f8b | 511 | if (dm_suspended_md(md)) { |
aa129a22 MB |
512 | r = -EAGAIN; |
513 | goto out; | |
514 | } | |
515 | ||
516 | if (tgt->type->ioctl) | |
647b3d00 | 517 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
518 | |
519 | out: | |
83d5e5b0 | 520 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 521 | |
6c182cd8 HR |
522 | if (r == -ENOTCONN) { |
523 | msleep(10); | |
524 | goto retry; | |
525 | } | |
526 | ||
aa129a22 MB |
527 | return r; |
528 | } | |
529 | ||
028867ac | 530 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
531 | { |
532 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
533 | } | |
534 | ||
028867ac | 535 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
536 | { |
537 | mempool_free(io, md->io_pool); | |
538 | } | |
539 | ||
028867ac | 540 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 541 | { |
dba14160 | 542 | bio_put(&tio->clone); |
1da177e4 LT |
543 | } |
544 | ||
08885643 KU |
545 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
546 | gfp_t gfp_mask) | |
cec47e3d | 547 | { |
5f015204 | 548 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
549 | } |
550 | ||
551 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
552 | { | |
5f015204 | 553 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
554 | } |
555 | ||
90abb8c4 KU |
556 | static int md_in_flight(struct mapped_device *md) |
557 | { | |
558 | return atomic_read(&md->pending[READ]) + | |
559 | atomic_read(&md->pending[WRITE]); | |
560 | } | |
561 | ||
3eaf840e JNN |
562 | static void start_io_acct(struct dm_io *io) |
563 | { | |
564 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 565 | struct bio *bio = io->bio; |
c9959059 | 566 | int cpu; |
fd2ed4d2 | 567 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
568 | |
569 | io->start_time = jiffies; | |
570 | ||
074a7aca TH |
571 | cpu = part_stat_lock(); |
572 | part_round_stats(cpu, &dm_disk(md)->part0); | |
573 | part_stat_unlock(); | |
1e9bb880 SL |
574 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
575 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
576 | |
577 | if (unlikely(dm_stats_used(&md->stats))) | |
578 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector, | |
579 | bio_sectors(bio), false, 0, &io->stats_aux); | |
3eaf840e JNN |
580 | } |
581 | ||
d221d2e7 | 582 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
583 | { |
584 | struct mapped_device *md = io->md; | |
585 | struct bio *bio = io->bio; | |
586 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 587 | int pending, cpu; |
3eaf840e JNN |
588 | int rw = bio_data_dir(bio); |
589 | ||
074a7aca TH |
590 | cpu = part_stat_lock(); |
591 | part_round_stats(cpu, &dm_disk(md)->part0); | |
592 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
593 | part_stat_unlock(); | |
3eaf840e | 594 | |
fd2ed4d2 MP |
595 | if (unlikely(dm_stats_used(&md->stats))) |
596 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_sector, | |
597 | bio_sectors(bio), true, duration, &io->stats_aux); | |
598 | ||
af7e466a MP |
599 | /* |
600 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 601 | * a flush. |
af7e466a | 602 | */ |
1e9bb880 SL |
603 | pending = atomic_dec_return(&md->pending[rw]); |
604 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 605 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 606 | |
d221d2e7 MP |
607 | /* nudge anyone waiting on suspend queue */ |
608 | if (!pending) | |
609 | wake_up(&md->wait); | |
3eaf840e JNN |
610 | } |
611 | ||
1da177e4 LT |
612 | /* |
613 | * Add the bio to the list of deferred io. | |
614 | */ | |
92c63902 | 615 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 616 | { |
05447420 | 617 | unsigned long flags; |
1da177e4 | 618 | |
05447420 | 619 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 620 | bio_list_add(&md->deferred, bio); |
05447420 | 621 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 622 | queue_work(md->wq, &md->work); |
1da177e4 LT |
623 | } |
624 | ||
625 | /* | |
626 | * Everyone (including functions in this file), should use this | |
627 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 628 | * dm_put_live_table() when finished. |
1da177e4 | 629 | */ |
83d5e5b0 | 630 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 631 | { |
83d5e5b0 MP |
632 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
633 | ||
634 | return srcu_dereference(md->map, &md->io_barrier); | |
635 | } | |
1da177e4 | 636 | |
83d5e5b0 MP |
637 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
638 | { | |
639 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
640 | } | |
641 | ||
642 | void dm_sync_table(struct mapped_device *md) | |
643 | { | |
644 | synchronize_srcu(&md->io_barrier); | |
645 | synchronize_rcu_expedited(); | |
646 | } | |
647 | ||
648 | /* | |
649 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
650 | * The caller must not block between these two functions. | |
651 | */ | |
652 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
653 | { | |
654 | rcu_read_lock(); | |
655 | return rcu_dereference(md->map); | |
656 | } | |
1da177e4 | 657 | |
83d5e5b0 MP |
658 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
659 | { | |
660 | rcu_read_unlock(); | |
1da177e4 LT |
661 | } |
662 | ||
3ac51e74 DW |
663 | /* |
664 | * Get the geometry associated with a dm device | |
665 | */ | |
666 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
667 | { | |
668 | *geo = md->geometry; | |
669 | ||
670 | return 0; | |
671 | } | |
672 | ||
673 | /* | |
674 | * Set the geometry of a device. | |
675 | */ | |
676 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
677 | { | |
678 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
679 | ||
680 | if (geo->start > sz) { | |
681 | DMWARN("Start sector is beyond the geometry limits."); | |
682 | return -EINVAL; | |
683 | } | |
684 | ||
685 | md->geometry = *geo; | |
686 | ||
687 | return 0; | |
688 | } | |
689 | ||
1da177e4 LT |
690 | /*----------------------------------------------------------------- |
691 | * CRUD START: | |
692 | * A more elegant soln is in the works that uses the queue | |
693 | * merge fn, unfortunately there are a couple of changes to | |
694 | * the block layer that I want to make for this. So in the | |
695 | * interests of getting something for people to use I give | |
696 | * you this clearly demarcated crap. | |
697 | *---------------------------------------------------------------*/ | |
698 | ||
2e93ccc1 KU |
699 | static int __noflush_suspending(struct mapped_device *md) |
700 | { | |
701 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
702 | } | |
703 | ||
1da177e4 LT |
704 | /* |
705 | * Decrements the number of outstanding ios that a bio has been | |
706 | * cloned into, completing the original io if necc. | |
707 | */ | |
858119e1 | 708 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 709 | { |
2e93ccc1 | 710 | unsigned long flags; |
b35f8caa MB |
711 | int io_error; |
712 | struct bio *bio; | |
713 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
714 | |
715 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
716 | if (unlikely(error)) { |
717 | spin_lock_irqsave(&io->endio_lock, flags); | |
718 | if (!(io->error > 0 && __noflush_suspending(md))) | |
719 | io->error = error; | |
720 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
721 | } | |
1da177e4 LT |
722 | |
723 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
724 | if (io->error == DM_ENDIO_REQUEUE) { |
725 | /* | |
726 | * Target requested pushing back the I/O. | |
2e93ccc1 | 727 | */ |
022c2611 | 728 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
729 | if (__noflush_suspending(md)) |
730 | bio_list_add_head(&md->deferred, io->bio); | |
731 | else | |
2e93ccc1 KU |
732 | /* noflush suspend was interrupted. */ |
733 | io->error = -EIO; | |
022c2611 | 734 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
735 | } |
736 | ||
b35f8caa MB |
737 | io_error = io->error; |
738 | bio = io->bio; | |
6a8736d1 TH |
739 | end_io_acct(io); |
740 | free_io(md, io); | |
741 | ||
742 | if (io_error == DM_ENDIO_REQUEUE) | |
743 | return; | |
2e93ccc1 | 744 | |
b372d360 | 745 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 746 | /* |
6a8736d1 TH |
747 | * Preflush done for flush with data, reissue |
748 | * without REQ_FLUSH. | |
af7e466a | 749 | */ |
6a8736d1 TH |
750 | bio->bi_rw &= ~REQ_FLUSH; |
751 | queue_io(md, bio); | |
af7e466a | 752 | } else { |
b372d360 | 753 | /* done with normal IO or empty flush */ |
0a82a8d1 | 754 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 755 | bio_endio(bio, io_error); |
b35f8caa | 756 | } |
1da177e4 LT |
757 | } |
758 | } | |
759 | ||
6712ecf8 | 760 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
761 | { |
762 | int r = 0; | |
028867ac | 763 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 764 | struct dm_io *io = tio->io; |
9faf400f | 765 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
766 | dm_endio_fn endio = tio->ti->type->end_io; |
767 | ||
1da177e4 LT |
768 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
769 | error = -EIO; | |
770 | ||
771 | if (endio) { | |
7de3ee57 | 772 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
773 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
774 | /* | |
775 | * error and requeue request are handled | |
776 | * in dec_pending(). | |
777 | */ | |
1da177e4 | 778 | error = r; |
45cbcd79 KU |
779 | else if (r == DM_ENDIO_INCOMPLETE) |
780 | /* The target will handle the io */ | |
6712ecf8 | 781 | return; |
45cbcd79 KU |
782 | else if (r) { |
783 | DMWARN("unimplemented target endio return value: %d", r); | |
784 | BUG(); | |
785 | } | |
1da177e4 LT |
786 | } |
787 | ||
9faf400f | 788 | free_tio(md, tio); |
b35f8caa | 789 | dec_pending(io, error); |
1da177e4 LT |
790 | } |
791 | ||
cec47e3d KU |
792 | /* |
793 | * Partial completion handling for request-based dm | |
794 | */ | |
795 | static void end_clone_bio(struct bio *clone, int error) | |
796 | { | |
797 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
798 | struct dm_rq_target_io *tio = info->tio; | |
799 | struct bio *bio = info->orig; | |
800 | unsigned int nr_bytes = info->orig->bi_size; | |
801 | ||
802 | bio_put(clone); | |
803 | ||
804 | if (tio->error) | |
805 | /* | |
806 | * An error has already been detected on the request. | |
807 | * Once error occurred, just let clone->end_io() handle | |
808 | * the remainder. | |
809 | */ | |
810 | return; | |
811 | else if (error) { | |
812 | /* | |
813 | * Don't notice the error to the upper layer yet. | |
814 | * The error handling decision is made by the target driver, | |
815 | * when the request is completed. | |
816 | */ | |
817 | tio->error = error; | |
818 | return; | |
819 | } | |
820 | ||
821 | /* | |
822 | * I/O for the bio successfully completed. | |
823 | * Notice the data completion to the upper layer. | |
824 | */ | |
825 | ||
826 | /* | |
827 | * bios are processed from the head of the list. | |
828 | * So the completing bio should always be rq->bio. | |
829 | * If it's not, something wrong is happening. | |
830 | */ | |
831 | if (tio->orig->bio != bio) | |
832 | DMERR("bio completion is going in the middle of the request"); | |
833 | ||
834 | /* | |
835 | * Update the original request. | |
836 | * Do not use blk_end_request() here, because it may complete | |
837 | * the original request before the clone, and break the ordering. | |
838 | */ | |
839 | blk_update_request(tio->orig, 0, nr_bytes); | |
840 | } | |
841 | ||
842 | /* | |
843 | * Don't touch any member of the md after calling this function because | |
844 | * the md may be freed in dm_put() at the end of this function. | |
845 | * Or do dm_get() before calling this function and dm_put() later. | |
846 | */ | |
b4324fee | 847 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 848 | { |
b4324fee | 849 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
850 | |
851 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 852 | if (!md_in_flight(md)) |
cec47e3d KU |
853 | wake_up(&md->wait); |
854 | ||
a8c32a5c JA |
855 | /* |
856 | * Run this off this callpath, as drivers could invoke end_io while | |
857 | * inside their request_fn (and holding the queue lock). Calling | |
858 | * back into ->request_fn() could deadlock attempting to grab the | |
859 | * queue lock again. | |
860 | */ | |
cec47e3d | 861 | if (run_queue) |
a8c32a5c | 862 | blk_run_queue_async(md->queue); |
cec47e3d KU |
863 | |
864 | /* | |
865 | * dm_put() must be at the end of this function. See the comment above | |
866 | */ | |
867 | dm_put(md); | |
868 | } | |
869 | ||
a77e28c7 KU |
870 | static void free_rq_clone(struct request *clone) |
871 | { | |
872 | struct dm_rq_target_io *tio = clone->end_io_data; | |
873 | ||
874 | blk_rq_unprep_clone(clone); | |
875 | free_rq_tio(tio); | |
876 | } | |
877 | ||
980691e5 KU |
878 | /* |
879 | * Complete the clone and the original request. | |
880 | * Must be called without queue lock. | |
881 | */ | |
882 | static void dm_end_request(struct request *clone, int error) | |
883 | { | |
884 | int rw = rq_data_dir(clone); | |
885 | struct dm_rq_target_io *tio = clone->end_io_data; | |
886 | struct mapped_device *md = tio->md; | |
887 | struct request *rq = tio->orig; | |
888 | ||
29e4013d | 889 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
890 | rq->errors = clone->errors; |
891 | rq->resid_len = clone->resid_len; | |
892 | ||
893 | if (rq->sense) | |
894 | /* | |
895 | * We are using the sense buffer of the original | |
896 | * request. | |
897 | * So setting the length of the sense data is enough. | |
898 | */ | |
899 | rq->sense_len = clone->sense_len; | |
900 | } | |
901 | ||
902 | free_rq_clone(clone); | |
29e4013d TH |
903 | blk_end_request_all(rq, error); |
904 | rq_completed(md, rw, true); | |
980691e5 KU |
905 | } |
906 | ||
cec47e3d KU |
907 | static void dm_unprep_request(struct request *rq) |
908 | { | |
909 | struct request *clone = rq->special; | |
cec47e3d KU |
910 | |
911 | rq->special = NULL; | |
912 | rq->cmd_flags &= ~REQ_DONTPREP; | |
913 | ||
a77e28c7 | 914 | free_rq_clone(clone); |
cec47e3d KU |
915 | } |
916 | ||
917 | /* | |
918 | * Requeue the original request of a clone. | |
919 | */ | |
920 | void dm_requeue_unmapped_request(struct request *clone) | |
921 | { | |
b4324fee | 922 | int rw = rq_data_dir(clone); |
cec47e3d KU |
923 | struct dm_rq_target_io *tio = clone->end_io_data; |
924 | struct mapped_device *md = tio->md; | |
925 | struct request *rq = tio->orig; | |
926 | struct request_queue *q = rq->q; | |
927 | unsigned long flags; | |
928 | ||
929 | dm_unprep_request(rq); | |
930 | ||
931 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
932 | blk_requeue_request(q, rq); |
933 | spin_unlock_irqrestore(q->queue_lock, flags); | |
934 | ||
b4324fee | 935 | rq_completed(md, rw, 0); |
cec47e3d KU |
936 | } |
937 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
938 | ||
939 | static void __stop_queue(struct request_queue *q) | |
940 | { | |
941 | blk_stop_queue(q); | |
942 | } | |
943 | ||
944 | static void stop_queue(struct request_queue *q) | |
945 | { | |
946 | unsigned long flags; | |
947 | ||
948 | spin_lock_irqsave(q->queue_lock, flags); | |
949 | __stop_queue(q); | |
950 | spin_unlock_irqrestore(q->queue_lock, flags); | |
951 | } | |
952 | ||
953 | static void __start_queue(struct request_queue *q) | |
954 | { | |
955 | if (blk_queue_stopped(q)) | |
956 | blk_start_queue(q); | |
957 | } | |
958 | ||
959 | static void start_queue(struct request_queue *q) | |
960 | { | |
961 | unsigned long flags; | |
962 | ||
963 | spin_lock_irqsave(q->queue_lock, flags); | |
964 | __start_queue(q); | |
965 | spin_unlock_irqrestore(q->queue_lock, flags); | |
966 | } | |
967 | ||
11a68244 | 968 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 969 | { |
11a68244 | 970 | int r = error; |
cec47e3d | 971 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 972 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 973 | |
ba1cbad9 MS |
974 | if (tio->ti) { |
975 | rq_end_io = tio->ti->type->rq_end_io; | |
976 | ||
977 | if (mapped && rq_end_io) | |
978 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
979 | } | |
cec47e3d | 980 | |
11a68244 | 981 | if (r <= 0) |
cec47e3d | 982 | /* The target wants to complete the I/O */ |
11a68244 KU |
983 | dm_end_request(clone, r); |
984 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
985 | /* The target will handle the I/O */ |
986 | return; | |
11a68244 | 987 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
988 | /* The target wants to requeue the I/O */ |
989 | dm_requeue_unmapped_request(clone); | |
990 | else { | |
11a68244 | 991 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
992 | BUG(); |
993 | } | |
994 | } | |
995 | ||
11a68244 KU |
996 | /* |
997 | * Request completion handler for request-based dm | |
998 | */ | |
999 | static void dm_softirq_done(struct request *rq) | |
1000 | { | |
1001 | bool mapped = true; | |
1002 | struct request *clone = rq->completion_data; | |
1003 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1004 | ||
1005 | if (rq->cmd_flags & REQ_FAILED) | |
1006 | mapped = false; | |
1007 | ||
1008 | dm_done(clone, tio->error, mapped); | |
1009 | } | |
1010 | ||
cec47e3d KU |
1011 | /* |
1012 | * Complete the clone and the original request with the error status | |
1013 | * through softirq context. | |
1014 | */ | |
1015 | static void dm_complete_request(struct request *clone, int error) | |
1016 | { | |
1017 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1018 | struct request *rq = tio->orig; | |
1019 | ||
1020 | tio->error = error; | |
1021 | rq->completion_data = clone; | |
1022 | blk_complete_request(rq); | |
1023 | } | |
1024 | ||
1025 | /* | |
1026 | * Complete the not-mapped clone and the original request with the error status | |
1027 | * through softirq context. | |
1028 | * Target's rq_end_io() function isn't called. | |
1029 | * This may be used when the target's map_rq() function fails. | |
1030 | */ | |
1031 | void dm_kill_unmapped_request(struct request *clone, int error) | |
1032 | { | |
1033 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1034 | struct request *rq = tio->orig; | |
1035 | ||
1036 | rq->cmd_flags |= REQ_FAILED; | |
1037 | dm_complete_request(clone, error); | |
1038 | } | |
1039 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1040 | ||
1041 | /* | |
1042 | * Called with the queue lock held | |
1043 | */ | |
1044 | static void end_clone_request(struct request *clone, int error) | |
1045 | { | |
1046 | /* | |
1047 | * For just cleaning up the information of the queue in which | |
1048 | * the clone was dispatched. | |
1049 | * The clone is *NOT* freed actually here because it is alloced from | |
1050 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1051 | */ | |
1052 | __blk_put_request(clone->q, clone); | |
1053 | ||
1054 | /* | |
1055 | * Actual request completion is done in a softirq context which doesn't | |
1056 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1057 | * - another request may be submitted by the upper level driver | |
1058 | * of the stacking during the completion | |
1059 | * - the submission which requires queue lock may be done | |
1060 | * against this queue | |
1061 | */ | |
1062 | dm_complete_request(clone, error); | |
1063 | } | |
1064 | ||
56a67df7 MS |
1065 | /* |
1066 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1067 | * target boundary. | |
1068 | */ | |
1069 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1070 | { | |
1071 | sector_t target_offset = dm_target_offset(ti, sector); | |
1072 | ||
1073 | return ti->len - target_offset; | |
1074 | } | |
1075 | ||
1076 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1077 | { |
56a67df7 | 1078 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1079 | sector_t offset, max_len; |
1da177e4 LT |
1080 | |
1081 | /* | |
542f9038 | 1082 | * Does the target need to split even further? |
1da177e4 | 1083 | */ |
542f9038 MS |
1084 | if (ti->max_io_len) { |
1085 | offset = dm_target_offset(ti, sector); | |
1086 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1087 | max_len = sector_div(offset, ti->max_io_len); | |
1088 | else | |
1089 | max_len = offset & (ti->max_io_len - 1); | |
1090 | max_len = ti->max_io_len - max_len; | |
1091 | ||
1092 | if (len > max_len) | |
1093 | len = max_len; | |
1da177e4 LT |
1094 | } |
1095 | ||
1096 | return len; | |
1097 | } | |
1098 | ||
542f9038 MS |
1099 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1100 | { | |
1101 | if (len > UINT_MAX) { | |
1102 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1103 | (unsigned long long)len, UINT_MAX); | |
1104 | ti->error = "Maximum size of target IO is too large"; | |
1105 | return -EINVAL; | |
1106 | } | |
1107 | ||
1108 | ti->max_io_len = (uint32_t) len; | |
1109 | ||
1110 | return 0; | |
1111 | } | |
1112 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1113 | ||
bd2a49b8 | 1114 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1115 | { |
1116 | int r; | |
2056a782 | 1117 | sector_t sector; |
9faf400f | 1118 | struct mapped_device *md; |
dba14160 | 1119 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1120 | struct dm_target *ti = tio->ti; |
1da177e4 | 1121 | |
1da177e4 LT |
1122 | clone->bi_end_io = clone_endio; |
1123 | clone->bi_private = tio; | |
1124 | ||
1125 | /* | |
1126 | * Map the clone. If r == 0 we don't need to do | |
1127 | * anything, the target has assumed ownership of | |
1128 | * this io. | |
1129 | */ | |
1130 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1131 | sector = clone->bi_sector; |
7de3ee57 | 1132 | r = ti->type->map(ti, clone); |
45cbcd79 | 1133 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1134 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1135 | |
d07335e5 MS |
1136 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1137 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1138 | |
1da177e4 | 1139 | generic_make_request(clone); |
2e93ccc1 KU |
1140 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1141 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1142 | md = tio->io->md; |
1143 | dec_pending(tio->io, r); | |
9faf400f | 1144 | free_tio(md, tio); |
45cbcd79 KU |
1145 | } else if (r) { |
1146 | DMWARN("unimplemented target map return value: %d", r); | |
1147 | BUG(); | |
1da177e4 LT |
1148 | } |
1149 | } | |
1150 | ||
1151 | struct clone_info { | |
1152 | struct mapped_device *md; | |
1153 | struct dm_table *map; | |
1154 | struct bio *bio; | |
1155 | struct dm_io *io; | |
1156 | sector_t sector; | |
1157 | sector_t sector_count; | |
1158 | unsigned short idx; | |
1159 | }; | |
1160 | ||
bd2a49b8 AK |
1161 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1162 | { | |
1163 | bio->bi_sector = sector; | |
1164 | bio->bi_size = to_bytes(len); | |
1165 | } | |
1166 | ||
1167 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1168 | { | |
1169 | bio->bi_idx = idx; | |
1170 | bio->bi_vcnt = idx + bv_count; | |
1171 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1172 | } | |
1173 | ||
1174 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1175 | unsigned short idx, unsigned len, unsigned offset, | |
1176 | unsigned trim) | |
1177 | { | |
1178 | if (!bio_integrity(bio)) | |
1179 | return; | |
1180 | ||
1181 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1182 | ||
1183 | if (trim) | |
1184 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1185 | } | |
1186 | ||
1da177e4 | 1187 | /* |
d87f4c14 | 1188 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1189 | */ |
14fe594d AK |
1190 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1191 | sector_t sector, unsigned short idx, | |
1192 | unsigned offset, unsigned len) | |
1da177e4 | 1193 | { |
dba14160 | 1194 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1195 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1196 | ||
1da177e4 LT |
1197 | *clone->bi_io_vec = *bv; |
1198 | ||
bd2a49b8 AK |
1199 | bio_setup_sector(clone, sector, len); |
1200 | ||
1da177e4 | 1201 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1202 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1203 | clone->bi_vcnt = 1; |
1da177e4 LT |
1204 | clone->bi_io_vec->bv_offset = offset; |
1205 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1206 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1207 | |
bd2a49b8 | 1208 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1209 | } |
1210 | ||
1211 | /* | |
1212 | * Creates a bio that consists of range of complete bvecs. | |
1213 | */ | |
dba14160 MP |
1214 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1215 | sector_t sector, unsigned short idx, | |
e4c93811 | 1216 | unsigned short bv_count, unsigned len) |
1da177e4 | 1217 | { |
dba14160 | 1218 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1219 | unsigned trim = 0; |
1da177e4 | 1220 | |
9faf400f | 1221 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1222 | bio_setup_sector(clone, sector, len); |
1223 | bio_setup_bv(clone, idx, bv_count); | |
1224 | ||
1225 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1226 | trim = 1; | |
1227 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1228 | } |
1229 | ||
9015df24 | 1230 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1231 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1232 | unsigned target_bio_nr) |
f9ab94ce | 1233 | { |
dba14160 MP |
1234 | struct dm_target_io *tio; |
1235 | struct bio *clone; | |
1236 | ||
1237 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1238 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1239 | |
1240 | tio->io = ci->io; | |
1241 | tio->ti = ti; | |
f9ab94ce | 1242 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1243 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1244 | |
1245 | return tio; | |
1246 | } | |
1247 | ||
14fe594d AK |
1248 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1249 | struct dm_target *ti, | |
1250 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1251 | { |
55a62eef | 1252 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1253 | struct bio *clone = &tio->clone; |
9015df24 | 1254 | |
06a426ce MS |
1255 | /* |
1256 | * Discard requests require the bio's inline iovecs be initialized. | |
1257 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1258 | * and discard, so no need for concern about wasted bvec allocations. | |
1259 | */ | |
dba14160 | 1260 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1261 | if (len) |
1262 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1263 | |
bd2a49b8 | 1264 | __map_bio(tio); |
f9ab94ce MP |
1265 | } |
1266 | ||
14fe594d AK |
1267 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1268 | unsigned num_bios, sector_t len) | |
06a426ce | 1269 | { |
55a62eef | 1270 | unsigned target_bio_nr; |
06a426ce | 1271 | |
55a62eef | 1272 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1273 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1274 | } |
1275 | ||
14fe594d | 1276 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1277 | { |
06a426ce | 1278 | unsigned target_nr = 0; |
f9ab94ce MP |
1279 | struct dm_target *ti; |
1280 | ||
b372d360 | 1281 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1282 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1283 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1284 | |
f9ab94ce MP |
1285 | return 0; |
1286 | } | |
1287 | ||
e4c93811 AK |
1288 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1289 | sector_t sector, int nr_iovecs, | |
1290 | unsigned short idx, unsigned short bv_count, | |
1291 | unsigned offset, unsigned len, | |
1292 | unsigned split_bvec) | |
5ae89a87 | 1293 | { |
dba14160 | 1294 | struct bio *bio = ci->bio; |
5ae89a87 | 1295 | struct dm_target_io *tio; |
b0d8ed4d AK |
1296 | unsigned target_bio_nr; |
1297 | unsigned num_target_bios = 1; | |
5ae89a87 | 1298 | |
b0d8ed4d AK |
1299 | /* |
1300 | * Does the target want to receive duplicate copies of the bio? | |
1301 | */ | |
1302 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1303 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1304 | |
b0d8ed4d AK |
1305 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
1306 | tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr); | |
1307 | if (split_bvec) | |
1308 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1309 | else | |
1310 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1311 | __map_bio(tio); | |
1312 | } | |
5ae89a87 MS |
1313 | } |
1314 | ||
55a62eef | 1315 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1316 | |
55a62eef | 1317 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1318 | { |
55a62eef | 1319 | return ti->num_discard_bios; |
23508a96 MS |
1320 | } |
1321 | ||
55a62eef | 1322 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1323 | { |
55a62eef | 1324 | return ti->num_write_same_bios; |
23508a96 MS |
1325 | } |
1326 | ||
1327 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1328 | ||
1329 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1330 | { | |
55a62eef | 1331 | return ti->split_discard_bios; |
23508a96 MS |
1332 | } |
1333 | ||
14fe594d AK |
1334 | static int __send_changing_extent_only(struct clone_info *ci, |
1335 | get_num_bios_fn get_num_bios, | |
1336 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1337 | { |
1338 | struct dm_target *ti; | |
a79245b3 | 1339 | sector_t len; |
55a62eef | 1340 | unsigned num_bios; |
5ae89a87 | 1341 | |
a79245b3 MS |
1342 | do { |
1343 | ti = dm_table_find_target(ci->map, ci->sector); | |
1344 | if (!dm_target_is_valid(ti)) | |
1345 | return -EIO; | |
5ae89a87 | 1346 | |
5ae89a87 | 1347 | /* |
23508a96 MS |
1348 | * Even though the device advertised support for this type of |
1349 | * request, that does not mean every target supports it, and | |
936688d7 | 1350 | * reconfiguration might also have changed that since the |
a79245b3 | 1351 | * check was performed. |
5ae89a87 | 1352 | */ |
55a62eef AK |
1353 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1354 | if (!num_bios) | |
a79245b3 | 1355 | return -EOPNOTSUPP; |
5ae89a87 | 1356 | |
23508a96 | 1357 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1358 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1359 | else | |
1360 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1361 | |
14fe594d | 1362 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1363 | |
1364 | ci->sector += len; | |
1365 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1366 | |
1367 | return 0; | |
1368 | } | |
1369 | ||
14fe594d | 1370 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1371 | { |
14fe594d AK |
1372 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1373 | is_split_required_for_discard); | |
23508a96 MS |
1374 | } |
1375 | ||
14fe594d | 1376 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1377 | { |
14fe594d | 1378 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1379 | } |
1380 | ||
e4c93811 AK |
1381 | /* |
1382 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1383 | */ | |
1384 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1385 | { | |
1386 | struct bio *bio = ci->bio; | |
1387 | sector_t bv_len, total_len = 0; | |
1388 | ||
1389 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1390 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1391 | ||
1392 | if (bv_len > max) | |
1393 | break; | |
1394 | ||
1395 | max -= bv_len; | |
1396 | total_len += bv_len; | |
1397 | } | |
1398 | ||
1399 | return total_len; | |
1400 | } | |
1401 | ||
1402 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1403 | struct dm_target *ti, sector_t max) | |
1404 | { | |
1405 | struct bio *bio = ci->bio; | |
1406 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1407 | sector_t remaining = to_sector(bv->bv_len); | |
1408 | unsigned offset = 0; | |
1409 | sector_t len; | |
1410 | ||
1411 | do { | |
1412 | if (offset) { | |
1413 | ti = dm_table_find_target(ci->map, ci->sector); | |
1414 | if (!dm_target_is_valid(ti)) | |
1415 | return -EIO; | |
1416 | ||
1417 | max = max_io_len(ci->sector, ti); | |
1418 | } | |
1419 | ||
1420 | len = min(remaining, max); | |
1421 | ||
1422 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1423 | bv->bv_offset + offset, len, 1); | |
1424 | ||
1425 | ci->sector += len; | |
1426 | ci->sector_count -= len; | |
1427 | offset += to_bytes(len); | |
1428 | } while (remaining -= len); | |
1429 | ||
1430 | ci->idx++; | |
1431 | ||
1432 | return 0; | |
1433 | } | |
1434 | ||
1435 | /* | |
1436 | * Select the correct strategy for processing a non-flush bio. | |
1437 | */ | |
14fe594d | 1438 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1439 | { |
dba14160 | 1440 | struct bio *bio = ci->bio; |
512875bd | 1441 | struct dm_target *ti; |
e4c93811 AK |
1442 | sector_t len, max; |
1443 | int idx; | |
1da177e4 | 1444 | |
5ae89a87 | 1445 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1446 | return __send_discard(ci); |
23508a96 | 1447 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1448 | return __send_write_same(ci); |
5ae89a87 | 1449 | |
512875bd JN |
1450 | ti = dm_table_find_target(ci->map, ci->sector); |
1451 | if (!dm_target_is_valid(ti)) | |
1452 | return -EIO; | |
1453 | ||
56a67df7 | 1454 | max = max_io_len(ci->sector, ti); |
512875bd | 1455 | |
e4c93811 AK |
1456 | /* |
1457 | * Optimise for the simple case where we can do all of | |
1458 | * the remaining io with a single clone. | |
1459 | */ | |
1da177e4 | 1460 | if (ci->sector_count <= max) { |
e4c93811 AK |
1461 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1462 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1463 | ci->sector_count, 0); | |
1464 | ci->sector_count = 0; | |
1465 | return 0; | |
1466 | } | |
1da177e4 | 1467 | |
e4c93811 AK |
1468 | /* |
1469 | * There are some bvecs that don't span targets. | |
1470 | * Do as many of these as possible. | |
1471 | */ | |
1472 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1473 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1474 | |
e4c93811 AK |
1475 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1476 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1477 | |
1478 | ci->sector += len; | |
1479 | ci->sector_count -= len; | |
e4c93811 | 1480 | ci->idx = idx; |
d2044a94 | 1481 | |
e4c93811 | 1482 | return 0; |
1da177e4 | 1483 | } |
512875bd | 1484 | |
e4c93811 AK |
1485 | /* |
1486 | * Handle a bvec that must be split between two or more targets. | |
1487 | */ | |
1488 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1489 | } |
1490 | ||
1491 | /* | |
14fe594d | 1492 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1493 | */ |
83d5e5b0 MP |
1494 | static void __split_and_process_bio(struct mapped_device *md, |
1495 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1496 | { |
1497 | struct clone_info ci; | |
512875bd | 1498 | int error = 0; |
1da177e4 | 1499 | |
83d5e5b0 | 1500 | if (unlikely(!map)) { |
6a8736d1 | 1501 | bio_io_error(bio); |
f0b9a450 MP |
1502 | return; |
1503 | } | |
692d0eb9 | 1504 | |
83d5e5b0 | 1505 | ci.map = map; |
1da177e4 | 1506 | ci.md = md; |
1da177e4 LT |
1507 | ci.io = alloc_io(md); |
1508 | ci.io->error = 0; | |
1509 | atomic_set(&ci.io->io_count, 1); | |
1510 | ci.io->bio = bio; | |
1511 | ci.io->md = md; | |
f88fb981 | 1512 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1513 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1514 | ci.idx = bio->bi_idx; |
1515 | ||
3eaf840e | 1516 | start_io_acct(ci.io); |
bd2a49b8 | 1517 | |
b372d360 MS |
1518 | if (bio->bi_rw & REQ_FLUSH) { |
1519 | ci.bio = &ci.md->flush_bio; | |
1520 | ci.sector_count = 0; | |
14fe594d | 1521 | error = __send_empty_flush(&ci); |
b372d360 MS |
1522 | /* dec_pending submits any data associated with flush */ |
1523 | } else { | |
6a8736d1 | 1524 | ci.bio = bio; |
d87f4c14 | 1525 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1526 | while (ci.sector_count && !error) |
14fe594d | 1527 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1528 | } |
1da177e4 LT |
1529 | |
1530 | /* drop the extra reference count */ | |
512875bd | 1531 | dec_pending(ci.io, error); |
1da177e4 LT |
1532 | } |
1533 | /*----------------------------------------------------------------- | |
1534 | * CRUD END | |
1535 | *---------------------------------------------------------------*/ | |
1536 | ||
f6fccb12 MB |
1537 | static int dm_merge_bvec(struct request_queue *q, |
1538 | struct bvec_merge_data *bvm, | |
1539 | struct bio_vec *biovec) | |
1540 | { | |
1541 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1542 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1543 | struct dm_target *ti; |
1544 | sector_t max_sectors; | |
5037108a | 1545 | int max_size = 0; |
f6fccb12 MB |
1546 | |
1547 | if (unlikely(!map)) | |
5037108a | 1548 | goto out; |
f6fccb12 MB |
1549 | |
1550 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1551 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1552 | goto out; |
f6fccb12 MB |
1553 | |
1554 | /* | |
1555 | * Find maximum amount of I/O that won't need splitting | |
1556 | */ | |
56a67df7 | 1557 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1558 | (sector_t) BIO_MAX_SECTORS); |
1559 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1560 | if (max_size < 0) | |
1561 | max_size = 0; | |
1562 | ||
1563 | /* | |
1564 | * merge_bvec_fn() returns number of bytes | |
1565 | * it can accept at this offset | |
1566 | * max is precomputed maximal io size | |
1567 | */ | |
1568 | if (max_size && ti->type->merge) | |
1569 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1570 | /* |
1571 | * If the target doesn't support merge method and some of the devices | |
1572 | * provided their merge_bvec method (we know this by looking at | |
1573 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1574 | * entries. So always set max_size to 0, and the code below allows | |
1575 | * just one page. | |
1576 | */ | |
1577 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1578 | ||
1579 | max_size = 0; | |
f6fccb12 | 1580 | |
5037108a | 1581 | out: |
83d5e5b0 | 1582 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1583 | /* |
1584 | * Always allow an entire first page | |
1585 | */ | |
1586 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1587 | max_size = biovec->bv_len; | |
1588 | ||
f6fccb12 MB |
1589 | return max_size; |
1590 | } | |
1591 | ||
1da177e4 LT |
1592 | /* |
1593 | * The request function that just remaps the bio built up by | |
1594 | * dm_merge_bvec. | |
1595 | */ | |
5a7bbad2 | 1596 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1597 | { |
12f03a49 | 1598 | int rw = bio_data_dir(bio); |
1da177e4 | 1599 | struct mapped_device *md = q->queuedata; |
c9959059 | 1600 | int cpu; |
83d5e5b0 MP |
1601 | int srcu_idx; |
1602 | struct dm_table *map; | |
1da177e4 | 1603 | |
83d5e5b0 | 1604 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1605 | |
074a7aca TH |
1606 | cpu = part_stat_lock(); |
1607 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1608 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1609 | part_stat_unlock(); | |
12f03a49 | 1610 | |
6a8736d1 TH |
1611 | /* if we're suspended, we have to queue this io for later */ |
1612 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1613 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1614 | |
6a8736d1 TH |
1615 | if (bio_rw(bio) != READA) |
1616 | queue_io(md, bio); | |
1617 | else | |
54d9a1b4 | 1618 | bio_io_error(bio); |
5a7bbad2 | 1619 | return; |
1da177e4 LT |
1620 | } |
1621 | ||
83d5e5b0 MP |
1622 | __split_and_process_bio(md, map, bio); |
1623 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1624 | return; |
cec47e3d KU |
1625 | } |
1626 | ||
fd2ed4d2 | 1627 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1628 | { |
1629 | return blk_queue_stackable(md->queue); | |
1630 | } | |
1631 | ||
5a7bbad2 | 1632 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1633 | { |
1634 | struct mapped_device *md = q->queuedata; | |
1635 | ||
1636 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1637 | blk_queue_bio(q, bio); |
1638 | else | |
1639 | _dm_request(q, bio); | |
cec47e3d KU |
1640 | } |
1641 | ||
1642 | void dm_dispatch_request(struct request *rq) | |
1643 | { | |
1644 | int r; | |
1645 | ||
1646 | if (blk_queue_io_stat(rq->q)) | |
1647 | rq->cmd_flags |= REQ_IO_STAT; | |
1648 | ||
1649 | rq->start_time = jiffies; | |
1650 | r = blk_insert_cloned_request(rq->q, rq); | |
1651 | if (r) | |
1652 | dm_complete_request(rq, r); | |
1653 | } | |
1654 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1655 | ||
cec47e3d KU |
1656 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1657 | void *data) | |
1658 | { | |
1659 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1660 | struct dm_rq_clone_bio_info *info = |
1661 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1662 | |
1663 | info->orig = bio_orig; | |
1664 | info->tio = tio; | |
1665 | bio->bi_end_io = end_clone_bio; | |
1666 | bio->bi_private = info; | |
cec47e3d KU |
1667 | |
1668 | return 0; | |
1669 | } | |
1670 | ||
1671 | static int setup_clone(struct request *clone, struct request *rq, | |
1672 | struct dm_rq_target_io *tio) | |
1673 | { | |
d0bcb878 | 1674 | int r; |
cec47e3d | 1675 | |
29e4013d TH |
1676 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1677 | dm_rq_bio_constructor, tio); | |
1678 | if (r) | |
1679 | return r; | |
cec47e3d | 1680 | |
29e4013d TH |
1681 | clone->cmd = rq->cmd; |
1682 | clone->cmd_len = rq->cmd_len; | |
1683 | clone->sense = rq->sense; | |
1684 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1685 | clone->end_io = end_clone_request; |
1686 | clone->end_io_data = tio; | |
1687 | ||
1688 | return 0; | |
1689 | } | |
1690 | ||
6facdaff KU |
1691 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1692 | gfp_t gfp_mask) | |
1693 | { | |
1694 | struct request *clone; | |
1695 | struct dm_rq_target_io *tio; | |
1696 | ||
1697 | tio = alloc_rq_tio(md, gfp_mask); | |
1698 | if (!tio) | |
1699 | return NULL; | |
1700 | ||
1701 | tio->md = md; | |
1702 | tio->ti = NULL; | |
1703 | tio->orig = rq; | |
1704 | tio->error = 0; | |
1705 | memset(&tio->info, 0, sizeof(tio->info)); | |
1706 | ||
1707 | clone = &tio->clone; | |
1708 | if (setup_clone(clone, rq, tio)) { | |
1709 | /* -ENOMEM */ | |
1710 | free_rq_tio(tio); | |
1711 | return NULL; | |
1712 | } | |
1713 | ||
1714 | return clone; | |
1715 | } | |
1716 | ||
cec47e3d KU |
1717 | /* |
1718 | * Called with the queue lock held. | |
1719 | */ | |
1720 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1721 | { | |
1722 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1723 | struct request *clone; |
1724 | ||
cec47e3d KU |
1725 | if (unlikely(rq->special)) { |
1726 | DMWARN("Already has something in rq->special."); | |
1727 | return BLKPREP_KILL; | |
1728 | } | |
1729 | ||
6facdaff KU |
1730 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1731 | if (!clone) | |
cec47e3d | 1732 | return BLKPREP_DEFER; |
cec47e3d KU |
1733 | |
1734 | rq->special = clone; | |
1735 | rq->cmd_flags |= REQ_DONTPREP; | |
1736 | ||
1737 | return BLKPREP_OK; | |
1738 | } | |
1739 | ||
9eef87da KU |
1740 | /* |
1741 | * Returns: | |
1742 | * 0 : the request has been processed (not requeued) | |
1743 | * !0 : the request has been requeued | |
1744 | */ | |
1745 | static int map_request(struct dm_target *ti, struct request *clone, | |
1746 | struct mapped_device *md) | |
cec47e3d | 1747 | { |
9eef87da | 1748 | int r, requeued = 0; |
cec47e3d KU |
1749 | struct dm_rq_target_io *tio = clone->end_io_data; |
1750 | ||
cec47e3d KU |
1751 | tio->ti = ti; |
1752 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1753 | switch (r) { | |
1754 | case DM_MAPIO_SUBMITTED: | |
1755 | /* The target has taken the I/O to submit by itself later */ | |
1756 | break; | |
1757 | case DM_MAPIO_REMAPPED: | |
1758 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1759 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1760 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1761 | dm_dispatch_request(clone); |
1762 | break; | |
1763 | case DM_MAPIO_REQUEUE: | |
1764 | /* The target wants to requeue the I/O */ | |
1765 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1766 | requeued = 1; |
cec47e3d KU |
1767 | break; |
1768 | default: | |
1769 | if (r > 0) { | |
1770 | DMWARN("unimplemented target map return value: %d", r); | |
1771 | BUG(); | |
1772 | } | |
1773 | ||
1774 | /* The target wants to complete the I/O */ | |
1775 | dm_kill_unmapped_request(clone, r); | |
1776 | break; | |
1777 | } | |
9eef87da KU |
1778 | |
1779 | return requeued; | |
cec47e3d KU |
1780 | } |
1781 | ||
ba1cbad9 MS |
1782 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1783 | { | |
1784 | struct request *clone; | |
1785 | ||
1786 | blk_start_request(orig); | |
1787 | clone = orig->special; | |
1788 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1789 | ||
1790 | /* | |
1791 | * Hold the md reference here for the in-flight I/O. | |
1792 | * We can't rely on the reference count by device opener, | |
1793 | * because the device may be closed during the request completion | |
1794 | * when all bios are completed. | |
1795 | * See the comment in rq_completed() too. | |
1796 | */ | |
1797 | dm_get(md); | |
1798 | ||
1799 | return clone; | |
1800 | } | |
1801 | ||
cec47e3d KU |
1802 | /* |
1803 | * q->request_fn for request-based dm. | |
1804 | * Called with the queue lock held. | |
1805 | */ | |
1806 | static void dm_request_fn(struct request_queue *q) | |
1807 | { | |
1808 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1809 | int srcu_idx; |
1810 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1811 | struct dm_target *ti; |
b4324fee | 1812 | struct request *rq, *clone; |
29e4013d | 1813 | sector_t pos; |
cec47e3d KU |
1814 | |
1815 | /* | |
b4324fee KU |
1816 | * For suspend, check blk_queue_stopped() and increment |
1817 | * ->pending within a single queue_lock not to increment the | |
1818 | * number of in-flight I/Os after the queue is stopped in | |
1819 | * dm_suspend(). | |
cec47e3d | 1820 | */ |
7eaceacc | 1821 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1822 | rq = blk_peek_request(q); |
1823 | if (!rq) | |
7eaceacc | 1824 | goto delay_and_out; |
cec47e3d | 1825 | |
29e4013d TH |
1826 | /* always use block 0 to find the target for flushes for now */ |
1827 | pos = 0; | |
1828 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1829 | pos = blk_rq_pos(rq); | |
1830 | ||
1831 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1832 | if (!dm_target_is_valid(ti)) { |
1833 | /* | |
1834 | * Must perform setup, that dm_done() requires, | |
1835 | * before calling dm_kill_unmapped_request | |
1836 | */ | |
1837 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1838 | clone = dm_start_request(md, rq); | |
1839 | dm_kill_unmapped_request(clone, -EIO); | |
1840 | continue; | |
1841 | } | |
d0bcb878 | 1842 | |
cec47e3d | 1843 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1844 | goto delay_and_out; |
cec47e3d | 1845 | |
ba1cbad9 | 1846 | clone = dm_start_request(md, rq); |
b4324fee | 1847 | |
cec47e3d | 1848 | spin_unlock(q->queue_lock); |
9eef87da KU |
1849 | if (map_request(ti, clone, md)) |
1850 | goto requeued; | |
1851 | ||
052189a2 KU |
1852 | BUG_ON(!irqs_disabled()); |
1853 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1854 | } |
1855 | ||
1856 | goto out; | |
1857 | ||
9eef87da | 1858 | requeued: |
052189a2 KU |
1859 | BUG_ON(!irqs_disabled()); |
1860 | spin_lock(q->queue_lock); | |
9eef87da | 1861 | |
7eaceacc JA |
1862 | delay_and_out: |
1863 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1864 | out: |
83d5e5b0 | 1865 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1866 | } |
1867 | ||
1868 | int dm_underlying_device_busy(struct request_queue *q) | |
1869 | { | |
1870 | return blk_lld_busy(q); | |
1871 | } | |
1872 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1873 | ||
1874 | static int dm_lld_busy(struct request_queue *q) | |
1875 | { | |
1876 | int r; | |
1877 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1878 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1879 | |
1880 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1881 | r = 1; | |
1882 | else | |
1883 | r = dm_table_any_busy_target(map); | |
1884 | ||
83d5e5b0 | 1885 | dm_put_live_table_fast(md); |
cec47e3d KU |
1886 | |
1887 | return r; | |
1888 | } | |
1889 | ||
1da177e4 LT |
1890 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1891 | { | |
8a57dfc6 CS |
1892 | int r = bdi_bits; |
1893 | struct mapped_device *md = congested_data; | |
1894 | struct dm_table *map; | |
1da177e4 | 1895 | |
1eb787ec | 1896 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1897 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1898 | if (map) { |
cec47e3d KU |
1899 | /* |
1900 | * Request-based dm cares about only own queue for | |
1901 | * the query about congestion status of request_queue | |
1902 | */ | |
1903 | if (dm_request_based(md)) | |
1904 | r = md->queue->backing_dev_info.state & | |
1905 | bdi_bits; | |
1906 | else | |
1907 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1908 | } |
83d5e5b0 | 1909 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1910 | } |
1911 | ||
1da177e4 LT |
1912 | return r; |
1913 | } | |
1914 | ||
1915 | /*----------------------------------------------------------------- | |
1916 | * An IDR is used to keep track of allocated minor numbers. | |
1917 | *---------------------------------------------------------------*/ | |
2b06cfff | 1918 | static void free_minor(int minor) |
1da177e4 | 1919 | { |
f32c10b0 | 1920 | spin_lock(&_minor_lock); |
1da177e4 | 1921 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1922 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1923 | } |
1924 | ||
1925 | /* | |
1926 | * See if the device with a specific minor # is free. | |
1927 | */ | |
cf13ab8e | 1928 | static int specific_minor(int minor) |
1da177e4 | 1929 | { |
c9d76be6 | 1930 | int r; |
1da177e4 LT |
1931 | |
1932 | if (minor >= (1 << MINORBITS)) | |
1933 | return -EINVAL; | |
1934 | ||
c9d76be6 | 1935 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1936 | spin_lock(&_minor_lock); |
1da177e4 | 1937 | |
c9d76be6 | 1938 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1939 | |
f32c10b0 | 1940 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1941 | idr_preload_end(); |
1942 | if (r < 0) | |
1943 | return r == -ENOSPC ? -EBUSY : r; | |
1944 | return 0; | |
1da177e4 LT |
1945 | } |
1946 | ||
cf13ab8e | 1947 | static int next_free_minor(int *minor) |
1da177e4 | 1948 | { |
c9d76be6 | 1949 | int r; |
62f75c2f | 1950 | |
c9d76be6 | 1951 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1952 | spin_lock(&_minor_lock); |
1da177e4 | 1953 | |
c9d76be6 | 1954 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1955 | |
f32c10b0 | 1956 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1957 | idr_preload_end(); |
1958 | if (r < 0) | |
1959 | return r; | |
1960 | *minor = r; | |
1961 | return 0; | |
1da177e4 LT |
1962 | } |
1963 | ||
83d5cde4 | 1964 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1965 | |
53d5914f MP |
1966 | static void dm_wq_work(struct work_struct *work); |
1967 | ||
4a0b4ddf MS |
1968 | static void dm_init_md_queue(struct mapped_device *md) |
1969 | { | |
1970 | /* | |
1971 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1972 | * devices. The type of this dm device has not been decided yet. | |
1973 | * The type is decided at the first table loading time. | |
1974 | * To prevent problematic device stacking, clear the queue flag | |
1975 | * for request stacking support until then. | |
1976 | * | |
1977 | * This queue is new, so no concurrency on the queue_flags. | |
1978 | */ | |
1979 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1980 | ||
1981 | md->queue->queuedata = md; | |
1982 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1983 | md->queue->backing_dev_info.congested_data = md; | |
1984 | blk_queue_make_request(md->queue, dm_request); | |
1985 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1986 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1987 | } | |
1988 | ||
1da177e4 LT |
1989 | /* |
1990 | * Allocate and initialise a blank device with a given minor. | |
1991 | */ | |
2b06cfff | 1992 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1993 | { |
1994 | int r; | |
cf13ab8e | 1995 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1996 | void *old_md; |
1da177e4 LT |
1997 | |
1998 | if (!md) { | |
1999 | DMWARN("unable to allocate device, out of memory."); | |
2000 | return NULL; | |
2001 | } | |
2002 | ||
10da4f79 | 2003 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2004 | goto bad_module_get; |
10da4f79 | 2005 | |
1da177e4 | 2006 | /* get a minor number for the dev */ |
2b06cfff | 2007 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2008 | r = next_free_minor(&minor); |
2b06cfff | 2009 | else |
cf13ab8e | 2010 | r = specific_minor(minor); |
1da177e4 | 2011 | if (r < 0) |
6ed7ade8 | 2012 | goto bad_minor; |
1da177e4 | 2013 | |
83d5e5b0 MP |
2014 | r = init_srcu_struct(&md->io_barrier); |
2015 | if (r < 0) | |
2016 | goto bad_io_barrier; | |
2017 | ||
a5664dad | 2018 | md->type = DM_TYPE_NONE; |
e61290a4 | 2019 | mutex_init(&md->suspend_lock); |
a5664dad | 2020 | mutex_init(&md->type_lock); |
022c2611 | 2021 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2022 | atomic_set(&md->holders, 1); |
5c6bd75d | 2023 | atomic_set(&md->open_count, 0); |
1da177e4 | 2024 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2025 | atomic_set(&md->uevent_seq, 0); |
2026 | INIT_LIST_HEAD(&md->uevent_list); | |
2027 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 2028 | |
4a0b4ddf | 2029 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2030 | if (!md->queue) |
6ed7ade8 | 2031 | goto bad_queue; |
1da177e4 | 2032 | |
4a0b4ddf | 2033 | dm_init_md_queue(md); |
9faf400f | 2034 | |
1da177e4 LT |
2035 | md->disk = alloc_disk(1); |
2036 | if (!md->disk) | |
6ed7ade8 | 2037 | goto bad_disk; |
1da177e4 | 2038 | |
316d315b NK |
2039 | atomic_set(&md->pending[0], 0); |
2040 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2041 | init_waitqueue_head(&md->wait); |
53d5914f | 2042 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
2043 | init_waitqueue_head(&md->eventq); |
2044 | ||
1da177e4 LT |
2045 | md->disk->major = _major; |
2046 | md->disk->first_minor = minor; | |
2047 | md->disk->fops = &dm_blk_dops; | |
2048 | md->disk->queue = md->queue; | |
2049 | md->disk->private_data = md; | |
2050 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2051 | add_disk(md->disk); | |
7e51f257 | 2052 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2053 | |
670368a8 | 2054 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2055 | if (!md->wq) |
2056 | goto bad_thread; | |
2057 | ||
32a926da MP |
2058 | md->bdev = bdget_disk(md->disk, 0); |
2059 | if (!md->bdev) | |
2060 | goto bad_bdev; | |
2061 | ||
6a8736d1 TH |
2062 | bio_init(&md->flush_bio); |
2063 | md->flush_bio.bi_bdev = md->bdev; | |
2064 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2065 | ||
fd2ed4d2 MP |
2066 | dm_stats_init(&md->stats); |
2067 | ||
ba61fdd1 | 2068 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2069 | spin_lock(&_minor_lock); |
ba61fdd1 | 2070 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2071 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2072 | |
2073 | BUG_ON(old_md != MINOR_ALLOCED); | |
2074 | ||
1da177e4 LT |
2075 | return md; |
2076 | ||
32a926da MP |
2077 | bad_bdev: |
2078 | destroy_workqueue(md->wq); | |
304f3f6a | 2079 | bad_thread: |
03022c54 | 2080 | del_gendisk(md->disk); |
304f3f6a | 2081 | put_disk(md->disk); |
6ed7ade8 | 2082 | bad_disk: |
1312f40e | 2083 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2084 | bad_queue: |
83d5e5b0 MP |
2085 | cleanup_srcu_struct(&md->io_barrier); |
2086 | bad_io_barrier: | |
1da177e4 | 2087 | free_minor(minor); |
6ed7ade8 | 2088 | bad_minor: |
10da4f79 | 2089 | module_put(THIS_MODULE); |
6ed7ade8 | 2090 | bad_module_get: |
1da177e4 LT |
2091 | kfree(md); |
2092 | return NULL; | |
2093 | } | |
2094 | ||
ae9da83f JN |
2095 | static void unlock_fs(struct mapped_device *md); |
2096 | ||
1da177e4 LT |
2097 | static void free_dev(struct mapped_device *md) |
2098 | { | |
f331c029 | 2099 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2100 | |
32a926da MP |
2101 | unlock_fs(md); |
2102 | bdput(md->bdev); | |
304f3f6a | 2103 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
2104 | if (md->io_pool) |
2105 | mempool_destroy(md->io_pool); | |
2106 | if (md->bs) | |
2107 | bioset_free(md->bs); | |
9c47008d | 2108 | blk_integrity_unregister(md->disk); |
1da177e4 | 2109 | del_gendisk(md->disk); |
83d5e5b0 | 2110 | cleanup_srcu_struct(&md->io_barrier); |
63d94e48 | 2111 | free_minor(minor); |
fba9f90e JM |
2112 | |
2113 | spin_lock(&_minor_lock); | |
2114 | md->disk->private_data = NULL; | |
2115 | spin_unlock(&_minor_lock); | |
2116 | ||
1da177e4 | 2117 | put_disk(md->disk); |
1312f40e | 2118 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2119 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2120 | module_put(THIS_MODULE); |
1da177e4 LT |
2121 | kfree(md); |
2122 | } | |
2123 | ||
e6ee8c0b KU |
2124 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2125 | { | |
c0820cf5 | 2126 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2127 | |
5f015204 | 2128 | if (md->io_pool && md->bs) { |
16245bdc JN |
2129 | /* The md already has necessary mempools. */ |
2130 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2131 | /* | |
2132 | * Reload bioset because front_pad may have changed | |
2133 | * because a different table was loaded. | |
2134 | */ | |
2135 | bioset_free(md->bs); | |
2136 | md->bs = p->bs; | |
2137 | p->bs = NULL; | |
2138 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2139 | /* |
2140 | * There's no need to reload with request-based dm | |
2141 | * because the size of front_pad doesn't change. | |
2142 | * Note for future: If you are to reload bioset, | |
2143 | * prep-ed requests in the queue may refer | |
2144 | * to bio from the old bioset, so you must walk | |
2145 | * through the queue to unprep. | |
2146 | */ | |
2147 | } | |
e6ee8c0b | 2148 | goto out; |
c0820cf5 | 2149 | } |
e6ee8c0b | 2150 | |
5f015204 | 2151 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2152 | |
2153 | md->io_pool = p->io_pool; | |
2154 | p->io_pool = NULL; | |
e6ee8c0b KU |
2155 | md->bs = p->bs; |
2156 | p->bs = NULL; | |
2157 | ||
2158 | out: | |
2159 | /* mempool bind completed, now no need any mempools in the table */ | |
2160 | dm_table_free_md_mempools(t); | |
2161 | } | |
2162 | ||
1da177e4 LT |
2163 | /* |
2164 | * Bind a table to the device. | |
2165 | */ | |
2166 | static void event_callback(void *context) | |
2167 | { | |
7a8c3d3b MA |
2168 | unsigned long flags; |
2169 | LIST_HEAD(uevents); | |
1da177e4 LT |
2170 | struct mapped_device *md = (struct mapped_device *) context; |
2171 | ||
7a8c3d3b MA |
2172 | spin_lock_irqsave(&md->uevent_lock, flags); |
2173 | list_splice_init(&md->uevent_list, &uevents); | |
2174 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2175 | ||
ed9e1982 | 2176 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2177 | |
1da177e4 LT |
2178 | atomic_inc(&md->event_nr); |
2179 | wake_up(&md->eventq); | |
2180 | } | |
2181 | ||
c217649b MS |
2182 | /* |
2183 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2184 | */ | |
4e90188b | 2185 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2186 | { |
4e90188b | 2187 | set_capacity(md->disk, size); |
1da177e4 | 2188 | |
db8fef4f | 2189 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2190 | } |
2191 | ||
d5b9dd04 MP |
2192 | /* |
2193 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2194 | * | |
2195 | * If this function returns 0, then the device is either a non-dm | |
2196 | * device without a merge_bvec_fn, or it is a dm device that is | |
2197 | * able to split any bios it receives that are too big. | |
2198 | */ | |
2199 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2200 | { | |
2201 | struct mapped_device *dev_md; | |
2202 | ||
2203 | if (!q->merge_bvec_fn) | |
2204 | return 0; | |
2205 | ||
2206 | if (q->make_request_fn == dm_request) { | |
2207 | dev_md = q->queuedata; | |
2208 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2209 | return 0; | |
2210 | } | |
2211 | ||
2212 | return 1; | |
2213 | } | |
2214 | ||
2215 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2216 | struct dm_dev *dev, sector_t start, | |
2217 | sector_t len, void *data) | |
2218 | { | |
2219 | struct block_device *bdev = dev->bdev; | |
2220 | struct request_queue *q = bdev_get_queue(bdev); | |
2221 | ||
2222 | return dm_queue_merge_is_compulsory(q); | |
2223 | } | |
2224 | ||
2225 | /* | |
2226 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2227 | * on the properties of the underlying devices. | |
2228 | */ | |
2229 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2230 | { | |
2231 | unsigned i = 0; | |
2232 | struct dm_target *ti; | |
2233 | ||
2234 | while (i < dm_table_get_num_targets(table)) { | |
2235 | ti = dm_table_get_target(table, i++); | |
2236 | ||
2237 | if (ti->type->iterate_devices && | |
2238 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2239 | return 0; | |
2240 | } | |
2241 | ||
2242 | return 1; | |
2243 | } | |
2244 | ||
042d2a9b AK |
2245 | /* |
2246 | * Returns old map, which caller must destroy. | |
2247 | */ | |
2248 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2249 | struct queue_limits *limits) | |
1da177e4 | 2250 | { |
042d2a9b | 2251 | struct dm_table *old_map; |
165125e1 | 2252 | struct request_queue *q = md->queue; |
1da177e4 | 2253 | sector_t size; |
d5b9dd04 | 2254 | int merge_is_optional; |
1da177e4 LT |
2255 | |
2256 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2257 | |
2258 | /* | |
2259 | * Wipe any geometry if the size of the table changed. | |
2260 | */ | |
fd2ed4d2 | 2261 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2262 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2263 | ||
32a926da | 2264 | __set_size(md, size); |
d5816876 | 2265 | |
2ca3310e AK |
2266 | dm_table_event_callback(t, event_callback, md); |
2267 | ||
e6ee8c0b KU |
2268 | /* |
2269 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2270 | * for request-based during suspension. So stop it to prevent | |
2271 | * I/O mapping before resume. | |
2272 | * This must be done before setting the queue restrictions, | |
2273 | * because request-based dm may be run just after the setting. | |
2274 | */ | |
2275 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2276 | stop_queue(q); | |
2277 | ||
2278 | __bind_mempools(md, t); | |
2279 | ||
d5b9dd04 MP |
2280 | merge_is_optional = dm_table_merge_is_optional(t); |
2281 | ||
042d2a9b | 2282 | old_map = md->map; |
83d5e5b0 | 2283 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2284 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2285 | ||
754c5fc7 | 2286 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2287 | if (merge_is_optional) |
2288 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2289 | else | |
2290 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
83d5e5b0 | 2291 | dm_sync_table(md); |
1da177e4 | 2292 | |
042d2a9b | 2293 | return old_map; |
1da177e4 LT |
2294 | } |
2295 | ||
a7940155 AK |
2296 | /* |
2297 | * Returns unbound table for the caller to free. | |
2298 | */ | |
2299 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2300 | { |
2301 | struct dm_table *map = md->map; | |
2302 | ||
2303 | if (!map) | |
a7940155 | 2304 | return NULL; |
1da177e4 LT |
2305 | |
2306 | dm_table_event_callback(map, NULL, NULL); | |
83d5e5b0 MP |
2307 | rcu_assign_pointer(md->map, NULL); |
2308 | dm_sync_table(md); | |
a7940155 AK |
2309 | |
2310 | return map; | |
1da177e4 LT |
2311 | } |
2312 | ||
2313 | /* | |
2314 | * Constructor for a new device. | |
2315 | */ | |
2b06cfff | 2316 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2317 | { |
2318 | struct mapped_device *md; | |
2319 | ||
2b06cfff | 2320 | md = alloc_dev(minor); |
1da177e4 LT |
2321 | if (!md) |
2322 | return -ENXIO; | |
2323 | ||
784aae73 MB |
2324 | dm_sysfs_init(md); |
2325 | ||
1da177e4 LT |
2326 | *result = md; |
2327 | return 0; | |
2328 | } | |
2329 | ||
a5664dad MS |
2330 | /* |
2331 | * Functions to manage md->type. | |
2332 | * All are required to hold md->type_lock. | |
2333 | */ | |
2334 | void dm_lock_md_type(struct mapped_device *md) | |
2335 | { | |
2336 | mutex_lock(&md->type_lock); | |
2337 | } | |
2338 | ||
2339 | void dm_unlock_md_type(struct mapped_device *md) | |
2340 | { | |
2341 | mutex_unlock(&md->type_lock); | |
2342 | } | |
2343 | ||
2344 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2345 | { | |
00c4fc3b | 2346 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2347 | md->type = type; |
2348 | } | |
2349 | ||
2350 | unsigned dm_get_md_type(struct mapped_device *md) | |
2351 | { | |
00c4fc3b | 2352 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2353 | return md->type; |
2354 | } | |
2355 | ||
36a0456f AK |
2356 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2357 | { | |
2358 | return md->immutable_target_type; | |
2359 | } | |
2360 | ||
f84cb8a4 MS |
2361 | /* |
2362 | * The queue_limits are only valid as long as you have a reference | |
2363 | * count on 'md'. | |
2364 | */ | |
2365 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2366 | { | |
2367 | BUG_ON(!atomic_read(&md->holders)); | |
2368 | return &md->queue->limits; | |
2369 | } | |
2370 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2371 | ||
4a0b4ddf MS |
2372 | /* |
2373 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2374 | */ | |
2375 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2376 | { | |
2377 | struct request_queue *q = NULL; | |
2378 | ||
2379 | if (md->queue->elevator) | |
2380 | return 1; | |
2381 | ||
2382 | /* Fully initialize the queue */ | |
2383 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2384 | if (!q) | |
2385 | return 0; | |
2386 | ||
2387 | md->queue = q; | |
4a0b4ddf MS |
2388 | dm_init_md_queue(md); |
2389 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2390 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2391 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2392 | |
2393 | elv_register_queue(md->queue); | |
2394 | ||
2395 | return 1; | |
2396 | } | |
2397 | ||
2398 | /* | |
2399 | * Setup the DM device's queue based on md's type | |
2400 | */ | |
2401 | int dm_setup_md_queue(struct mapped_device *md) | |
2402 | { | |
2403 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2404 | !dm_init_request_based_queue(md)) { | |
2405 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2406 | return -EINVAL; | |
2407 | } | |
2408 | ||
2409 | return 0; | |
2410 | } | |
2411 | ||
637842cf | 2412 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2413 | { |
2414 | struct mapped_device *md; | |
1da177e4 LT |
2415 | unsigned minor = MINOR(dev); |
2416 | ||
2417 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2418 | return NULL; | |
2419 | ||
f32c10b0 | 2420 | spin_lock(&_minor_lock); |
1da177e4 LT |
2421 | |
2422 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2423 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2424 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2425 | dm_deleting_md(md) || |
17b2f66f | 2426 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2427 | md = NULL; |
fba9f90e JM |
2428 | goto out; |
2429 | } | |
1da177e4 | 2430 | |
fba9f90e | 2431 | out: |
f32c10b0 | 2432 | spin_unlock(&_minor_lock); |
1da177e4 | 2433 | |
637842cf DT |
2434 | return md; |
2435 | } | |
2436 | ||
d229a958 DT |
2437 | struct mapped_device *dm_get_md(dev_t dev) |
2438 | { | |
2439 | struct mapped_device *md = dm_find_md(dev); | |
2440 | ||
2441 | if (md) | |
2442 | dm_get(md); | |
2443 | ||
2444 | return md; | |
2445 | } | |
3cf2e4ba | 2446 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2447 | |
9ade92a9 | 2448 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2449 | { |
9ade92a9 | 2450 | return md->interface_ptr; |
1da177e4 LT |
2451 | } |
2452 | ||
2453 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2454 | { | |
2455 | md->interface_ptr = ptr; | |
2456 | } | |
2457 | ||
2458 | void dm_get(struct mapped_device *md) | |
2459 | { | |
2460 | atomic_inc(&md->holders); | |
3f77316d | 2461 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2462 | } |
2463 | ||
72d94861 AK |
2464 | const char *dm_device_name(struct mapped_device *md) |
2465 | { | |
2466 | return md->name; | |
2467 | } | |
2468 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2469 | ||
3f77316d | 2470 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2471 | { |
1134e5ae | 2472 | struct dm_table *map; |
83d5e5b0 | 2473 | int srcu_idx; |
1da177e4 | 2474 | |
3f77316d | 2475 | might_sleep(); |
fba9f90e | 2476 | |
3f77316d | 2477 | spin_lock(&_minor_lock); |
83d5e5b0 | 2478 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2479 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2480 | set_bit(DMF_FREEING, &md->flags); | |
2481 | spin_unlock(&_minor_lock); | |
2482 | ||
2483 | if (!dm_suspended_md(md)) { | |
2484 | dm_table_presuspend_targets(map); | |
2485 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2486 | } |
3f77316d | 2487 | |
83d5e5b0 MP |
2488 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2489 | dm_put_live_table(md, srcu_idx); | |
2490 | ||
3f77316d KU |
2491 | /* |
2492 | * Rare, but there may be I/O requests still going to complete, | |
2493 | * for example. Wait for all references to disappear. | |
2494 | * No one should increment the reference count of the mapped_device, | |
2495 | * after the mapped_device state becomes DMF_FREEING. | |
2496 | */ | |
2497 | if (wait) | |
2498 | while (atomic_read(&md->holders)) | |
2499 | msleep(1); | |
2500 | else if (atomic_read(&md->holders)) | |
2501 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2502 | dm_device_name(md), atomic_read(&md->holders)); | |
2503 | ||
2504 | dm_sysfs_exit(md); | |
3f77316d KU |
2505 | dm_table_destroy(__unbind(md)); |
2506 | free_dev(md); | |
2507 | } | |
2508 | ||
2509 | void dm_destroy(struct mapped_device *md) | |
2510 | { | |
2511 | __dm_destroy(md, true); | |
2512 | } | |
2513 | ||
2514 | void dm_destroy_immediate(struct mapped_device *md) | |
2515 | { | |
2516 | __dm_destroy(md, false); | |
2517 | } | |
2518 | ||
2519 | void dm_put(struct mapped_device *md) | |
2520 | { | |
2521 | atomic_dec(&md->holders); | |
1da177e4 | 2522 | } |
79eb885c | 2523 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2524 | |
401600df | 2525 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2526 | { |
2527 | int r = 0; | |
b44ebeb0 MP |
2528 | DECLARE_WAITQUEUE(wait, current); |
2529 | ||
b44ebeb0 | 2530 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2531 | |
2532 | while (1) { | |
401600df | 2533 | set_current_state(interruptible); |
46125c1c | 2534 | |
b4324fee | 2535 | if (!md_in_flight(md)) |
46125c1c MB |
2536 | break; |
2537 | ||
401600df MP |
2538 | if (interruptible == TASK_INTERRUPTIBLE && |
2539 | signal_pending(current)) { | |
46125c1c MB |
2540 | r = -EINTR; |
2541 | break; | |
2542 | } | |
2543 | ||
2544 | io_schedule(); | |
2545 | } | |
2546 | set_current_state(TASK_RUNNING); | |
2547 | ||
b44ebeb0 MP |
2548 | remove_wait_queue(&md->wait, &wait); |
2549 | ||
46125c1c MB |
2550 | return r; |
2551 | } | |
2552 | ||
1da177e4 LT |
2553 | /* |
2554 | * Process the deferred bios | |
2555 | */ | |
ef208587 | 2556 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2557 | { |
ef208587 MP |
2558 | struct mapped_device *md = container_of(work, struct mapped_device, |
2559 | work); | |
6d6f10df | 2560 | struct bio *c; |
83d5e5b0 MP |
2561 | int srcu_idx; |
2562 | struct dm_table *map; | |
1da177e4 | 2563 | |
83d5e5b0 | 2564 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2565 | |
3b00b203 | 2566 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2567 | spin_lock_irq(&md->deferred_lock); |
2568 | c = bio_list_pop(&md->deferred); | |
2569 | spin_unlock_irq(&md->deferred_lock); | |
2570 | ||
6a8736d1 | 2571 | if (!c) |
df12ee99 | 2572 | break; |
022c2611 | 2573 | |
e6ee8c0b KU |
2574 | if (dm_request_based(md)) |
2575 | generic_make_request(c); | |
6a8736d1 | 2576 | else |
83d5e5b0 | 2577 | __split_and_process_bio(md, map, c); |
022c2611 | 2578 | } |
73d410c0 | 2579 | |
83d5e5b0 | 2580 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2581 | } |
2582 | ||
9a1fb464 | 2583 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2584 | { |
3b00b203 MP |
2585 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2586 | smp_mb__after_clear_bit(); | |
53d5914f | 2587 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2588 | } |
2589 | ||
1da177e4 | 2590 | /* |
042d2a9b | 2591 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2592 | */ |
042d2a9b | 2593 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2594 | { |
87eb5b21 | 2595 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2596 | struct queue_limits limits; |
042d2a9b | 2597 | int r; |
1da177e4 | 2598 | |
e61290a4 | 2599 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2600 | |
2601 | /* device must be suspended */ | |
4f186f8b | 2602 | if (!dm_suspended_md(md)) |
93c534ae | 2603 | goto out; |
1da177e4 | 2604 | |
3ae70656 MS |
2605 | /* |
2606 | * If the new table has no data devices, retain the existing limits. | |
2607 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2608 | * then new I/O is queued based on these limits, and then some paths | |
2609 | * reappear. | |
2610 | */ | |
2611 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2612 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2613 | if (live_map) |
2614 | limits = md->queue->limits; | |
83d5e5b0 | 2615 | dm_put_live_table_fast(md); |
3ae70656 MS |
2616 | } |
2617 | ||
87eb5b21 MC |
2618 | if (!live_map) { |
2619 | r = dm_calculate_queue_limits(table, &limits); | |
2620 | if (r) { | |
2621 | map = ERR_PTR(r); | |
2622 | goto out; | |
2623 | } | |
042d2a9b | 2624 | } |
754c5fc7 | 2625 | |
042d2a9b | 2626 | map = __bind(md, table, &limits); |
1da177e4 | 2627 | |
93c534ae | 2628 | out: |
e61290a4 | 2629 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2630 | return map; |
1da177e4 LT |
2631 | } |
2632 | ||
2633 | /* | |
2634 | * Functions to lock and unlock any filesystem running on the | |
2635 | * device. | |
2636 | */ | |
2ca3310e | 2637 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2638 | { |
e39e2e95 | 2639 | int r; |
1da177e4 LT |
2640 | |
2641 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2642 | |
db8fef4f | 2643 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2644 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2645 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2646 | md->frozen_sb = NULL; |
2647 | return r; | |
dfbe03f6 AK |
2648 | } |
2649 | ||
aa8d7c2f AK |
2650 | set_bit(DMF_FROZEN, &md->flags); |
2651 | ||
1da177e4 LT |
2652 | return 0; |
2653 | } | |
2654 | ||
2ca3310e | 2655 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2656 | { |
aa8d7c2f AK |
2657 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2658 | return; | |
2659 | ||
db8fef4f | 2660 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2661 | md->frozen_sb = NULL; |
aa8d7c2f | 2662 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2663 | } |
2664 | ||
2665 | /* | |
2666 | * We need to be able to change a mapping table under a mounted | |
2667 | * filesystem. For example we might want to move some data in | |
2668 | * the background. Before the table can be swapped with | |
2669 | * dm_bind_table, dm_suspend must be called to flush any in | |
2670 | * flight bios and ensure that any further io gets deferred. | |
2671 | */ | |
cec47e3d KU |
2672 | /* |
2673 | * Suspend mechanism in request-based dm. | |
2674 | * | |
9f518b27 KU |
2675 | * 1. Flush all I/Os by lock_fs() if needed. |
2676 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2677 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2678 | * |
9f518b27 | 2679 | * To abort suspend, start the request_queue. |
cec47e3d | 2680 | */ |
a3d77d35 | 2681 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2682 | { |
2ca3310e | 2683 | struct dm_table *map = NULL; |
46125c1c | 2684 | int r = 0; |
a3d77d35 | 2685 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2686 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2687 | |
e61290a4 | 2688 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2689 | |
4f186f8b | 2690 | if (dm_suspended_md(md)) { |
73d410c0 | 2691 | r = -EINVAL; |
d287483d | 2692 | goto out_unlock; |
73d410c0 | 2693 | } |
1da177e4 | 2694 | |
83d5e5b0 | 2695 | map = md->map; |
1da177e4 | 2696 | |
2e93ccc1 KU |
2697 | /* |
2698 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2699 | * This flag is cleared before dm_suspend returns. | |
2700 | */ | |
2701 | if (noflush) | |
2702 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2703 | ||
cf222b37 AK |
2704 | /* This does not get reverted if there's an error later. */ |
2705 | dm_table_presuspend_targets(map); | |
2706 | ||
32a926da | 2707 | /* |
9f518b27 KU |
2708 | * Flush I/O to the device. |
2709 | * Any I/O submitted after lock_fs() may not be flushed. | |
2710 | * noflush takes precedence over do_lockfs. | |
2711 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2712 | */ |
2713 | if (!noflush && do_lockfs) { | |
2714 | r = lock_fs(md); | |
2715 | if (r) | |
83d5e5b0 | 2716 | goto out_unlock; |
aa8d7c2f | 2717 | } |
1da177e4 LT |
2718 | |
2719 | /* | |
3b00b203 MP |
2720 | * Here we must make sure that no processes are submitting requests |
2721 | * to target drivers i.e. no one may be executing | |
2722 | * __split_and_process_bio. This is called from dm_request and | |
2723 | * dm_wq_work. | |
2724 | * | |
2725 | * To get all processes out of __split_and_process_bio in dm_request, | |
2726 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2727 | * __split_and_process_bio from dm_request and quiesce the thread |
2728 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2729 | * flush_workqueue(md->wq). | |
1da177e4 | 2730 | */ |
1eb787ec | 2731 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
83d5e5b0 | 2732 | synchronize_srcu(&md->io_barrier); |
1da177e4 | 2733 | |
d0bcb878 | 2734 | /* |
29e4013d TH |
2735 | * Stop md->queue before flushing md->wq in case request-based |
2736 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2737 | */ |
cec47e3d | 2738 | if (dm_request_based(md)) |
9f518b27 | 2739 | stop_queue(md->queue); |
cec47e3d | 2740 | |
d0bcb878 KU |
2741 | flush_workqueue(md->wq); |
2742 | ||
1da177e4 | 2743 | /* |
3b00b203 MP |
2744 | * At this point no more requests are entering target request routines. |
2745 | * We call dm_wait_for_completion to wait for all existing requests | |
2746 | * to finish. | |
1da177e4 | 2747 | */ |
401600df | 2748 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2749 | |
6d6f10df | 2750 | if (noflush) |
022c2611 | 2751 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
83d5e5b0 | 2752 | synchronize_srcu(&md->io_barrier); |
2e93ccc1 | 2753 | |
1da177e4 | 2754 | /* were we interrupted ? */ |
46125c1c | 2755 | if (r < 0) { |
9a1fb464 | 2756 | dm_queue_flush(md); |
73d410c0 | 2757 | |
cec47e3d | 2758 | if (dm_request_based(md)) |
9f518b27 | 2759 | start_queue(md->queue); |
cec47e3d | 2760 | |
2ca3310e | 2761 | unlock_fs(md); |
83d5e5b0 | 2762 | goto out_unlock; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2763 | } |
1da177e4 | 2764 | |
3b00b203 MP |
2765 | /* |
2766 | * If dm_wait_for_completion returned 0, the device is completely | |
2767 | * quiescent now. There is no request-processing activity. All new | |
2768 | * requests are being added to md->deferred list. | |
2769 | */ | |
2770 | ||
2ca3310e | 2771 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2772 | |
4d4471cb KU |
2773 | dm_table_postsuspend_targets(map); |
2774 | ||
d287483d | 2775 | out_unlock: |
e61290a4 | 2776 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2777 | return r; |
1da177e4 LT |
2778 | } |
2779 | ||
2780 | int dm_resume(struct mapped_device *md) | |
2781 | { | |
cf222b37 | 2782 | int r = -EINVAL; |
cf222b37 | 2783 | struct dm_table *map = NULL; |
1da177e4 | 2784 | |
e61290a4 | 2785 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2786 | if (!dm_suspended_md(md)) |
cf222b37 | 2787 | goto out; |
cf222b37 | 2788 | |
83d5e5b0 | 2789 | map = md->map; |
2ca3310e | 2790 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2791 | goto out; |
1da177e4 | 2792 | |
8757b776 MB |
2793 | r = dm_table_resume_targets(map); |
2794 | if (r) | |
2795 | goto out; | |
2ca3310e | 2796 | |
9a1fb464 | 2797 | dm_queue_flush(md); |
2ca3310e | 2798 | |
cec47e3d KU |
2799 | /* |
2800 | * Flushing deferred I/Os must be done after targets are resumed | |
2801 | * so that mapping of targets can work correctly. | |
2802 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2803 | */ | |
2804 | if (dm_request_based(md)) | |
2805 | start_queue(md->queue); | |
2806 | ||
2ca3310e AK |
2807 | unlock_fs(md); |
2808 | ||
2809 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2810 | ||
cf222b37 AK |
2811 | r = 0; |
2812 | out: | |
e61290a4 | 2813 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2814 | |
cf222b37 | 2815 | return r; |
1da177e4 LT |
2816 | } |
2817 | ||
fd2ed4d2 MP |
2818 | /* |
2819 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2820 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2821 | * It may be used only from the kernel. | |
2822 | * | |
2823 | * Internal suspend holds md->suspend_lock, which prevents interaction with | |
2824 | * userspace-driven suspend. | |
2825 | */ | |
2826 | ||
2827 | void dm_internal_suspend(struct mapped_device *md) | |
2828 | { | |
2829 | mutex_lock(&md->suspend_lock); | |
2830 | if (dm_suspended_md(md)) | |
2831 | return; | |
2832 | ||
2833 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2834 | synchronize_srcu(&md->io_barrier); | |
2835 | flush_workqueue(md->wq); | |
2836 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2837 | } | |
2838 | ||
2839 | void dm_internal_resume(struct mapped_device *md) | |
2840 | { | |
2841 | if (dm_suspended_md(md)) | |
2842 | goto done; | |
2843 | ||
2844 | dm_queue_flush(md); | |
2845 | ||
2846 | done: | |
2847 | mutex_unlock(&md->suspend_lock); | |
2848 | } | |
2849 | ||
1da177e4 LT |
2850 | /*----------------------------------------------------------------- |
2851 | * Event notification. | |
2852 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2853 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2854 | unsigned cookie) |
69267a30 | 2855 | { |
60935eb2 MB |
2856 | char udev_cookie[DM_COOKIE_LENGTH]; |
2857 | char *envp[] = { udev_cookie, NULL }; | |
2858 | ||
2859 | if (!cookie) | |
3abf85b5 | 2860 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2861 | else { |
2862 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2863 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2864 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2865 | action, envp); | |
60935eb2 | 2866 | } |
69267a30 AK |
2867 | } |
2868 | ||
7a8c3d3b MA |
2869 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2870 | { | |
2871 | return atomic_add_return(1, &md->uevent_seq); | |
2872 | } | |
2873 | ||
1da177e4 LT |
2874 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2875 | { | |
2876 | return atomic_read(&md->event_nr); | |
2877 | } | |
2878 | ||
2879 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2880 | { | |
2881 | return wait_event_interruptible(md->eventq, | |
2882 | (event_nr != atomic_read(&md->event_nr))); | |
2883 | } | |
2884 | ||
7a8c3d3b MA |
2885 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2886 | { | |
2887 | unsigned long flags; | |
2888 | ||
2889 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2890 | list_add(elist, &md->uevent_list); | |
2891 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2892 | } | |
2893 | ||
1da177e4 LT |
2894 | /* |
2895 | * The gendisk is only valid as long as you have a reference | |
2896 | * count on 'md'. | |
2897 | */ | |
2898 | struct gendisk *dm_disk(struct mapped_device *md) | |
2899 | { | |
2900 | return md->disk; | |
2901 | } | |
2902 | ||
784aae73 MB |
2903 | struct kobject *dm_kobject(struct mapped_device *md) |
2904 | { | |
2905 | return &md->kobj; | |
2906 | } | |
2907 | ||
2908 | /* | |
2909 | * struct mapped_device should not be exported outside of dm.c | |
2910 | * so use this check to verify that kobj is part of md structure | |
2911 | */ | |
2912 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2913 | { | |
2914 | struct mapped_device *md; | |
2915 | ||
2916 | md = container_of(kobj, struct mapped_device, kobj); | |
2917 | if (&md->kobj != kobj) | |
2918 | return NULL; | |
2919 | ||
4d89b7b4 | 2920 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2921 | dm_deleting_md(md)) |
4d89b7b4 MB |
2922 | return NULL; |
2923 | ||
784aae73 MB |
2924 | dm_get(md); |
2925 | return md; | |
2926 | } | |
2927 | ||
4f186f8b | 2928 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2929 | { |
2930 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2931 | } | |
2932 | ||
2c140a24 MP |
2933 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2934 | { | |
2935 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2936 | } | |
2937 | ||
64dbce58 KU |
2938 | int dm_suspended(struct dm_target *ti) |
2939 | { | |
ecdb2e25 | 2940 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2941 | } |
2942 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2943 | ||
2e93ccc1 KU |
2944 | int dm_noflush_suspending(struct dm_target *ti) |
2945 | { | |
ecdb2e25 | 2946 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2947 | } |
2948 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2949 | ||
c0820cf5 | 2950 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 2951 | { |
5f015204 JN |
2952 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
2953 | struct kmem_cache *cachep; | |
2954 | unsigned int pool_size; | |
2955 | unsigned int front_pad; | |
e6ee8c0b KU |
2956 | |
2957 | if (!pools) | |
2958 | return NULL; | |
2959 | ||
23e5083b | 2960 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 | 2961 | cachep = _io_cache; |
e8603136 | 2962 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 JN |
2963 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
2964 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
2965 | cachep = _rq_tio_cache; | |
f4790826 | 2966 | pool_size = dm_get_reserved_rq_based_ios(); |
5f015204 JN |
2967 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
2968 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
2969 | WARN_ON(per_bio_data_size != 0); | |
2970 | } else | |
2971 | goto out; | |
e6ee8c0b | 2972 | |
6cfa5857 | 2973 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
2974 | if (!pools->io_pool) |
2975 | goto out; | |
e6ee8c0b | 2976 | |
5f015204 | 2977 | pools->bs = bioset_create(pool_size, front_pad); |
e6ee8c0b | 2978 | if (!pools->bs) |
5f015204 | 2979 | goto out; |
e6ee8c0b | 2980 | |
a91a2785 | 2981 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2982 | goto out; |
a91a2785 | 2983 | |
e6ee8c0b KU |
2984 | return pools; |
2985 | ||
5f015204 JN |
2986 | out: |
2987 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
2988 | |
2989 | return NULL; | |
2990 | } | |
2991 | ||
2992 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2993 | { | |
2994 | if (!pools) | |
2995 | return; | |
2996 | ||
2997 | if (pools->io_pool) | |
2998 | mempool_destroy(pools->io_pool); | |
2999 | ||
e6ee8c0b KU |
3000 | if (pools->bs) |
3001 | bioset_free(pools->bs); | |
3002 | ||
3003 | kfree(pools); | |
3004 | } | |
3005 | ||
83d5cde4 | 3006 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3007 | .open = dm_blk_open, |
3008 | .release = dm_blk_close, | |
aa129a22 | 3009 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3010 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3011 | .owner = THIS_MODULE |
3012 | }; | |
3013 | ||
3014 | EXPORT_SYMBOL(dm_get_mapinfo); | |
3015 | ||
3016 | /* | |
3017 | * module hooks | |
3018 | */ | |
3019 | module_init(dm_init); | |
3020 | module_exit(dm_exit); | |
3021 | ||
3022 | module_param(major, uint, 0); | |
3023 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3024 | |
e8603136 MS |
3025 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3026 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3027 | ||
f4790826 MS |
3028 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3029 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3030 | ||
1da177e4 LT |
3031 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3032 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3033 | MODULE_LICENSE("GPL"); |