Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
1da177e4 LT |
63 | }; |
64 | ||
8fbf26ad KU |
65 | /* |
66 | * For request-based dm. | |
67 | * One of these is allocated per request. | |
68 | */ | |
69 | struct dm_rq_target_io { | |
70 | struct mapped_device *md; | |
71 | struct dm_target *ti; | |
72 | struct request *orig, clone; | |
73 | int error; | |
74 | union map_info info; | |
75 | }; | |
76 | ||
77 | /* | |
94818742 KO |
78 | * For request-based dm - the bio clones we allocate are embedded in these |
79 | * structs. | |
80 | * | |
81 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
82 | * the bioset is created - this means the bio has to come at the end of the | |
83 | * struct. | |
8fbf26ad KU |
84 | */ |
85 | struct dm_rq_clone_bio_info { | |
86 | struct bio *orig; | |
cec47e3d | 87 | struct dm_rq_target_io *tio; |
94818742 | 88 | struct bio clone; |
8fbf26ad KU |
89 | }; |
90 | ||
1da177e4 LT |
91 | union map_info *dm_get_mapinfo(struct bio *bio) |
92 | { | |
17b2f66f | 93 | if (bio && bio->bi_private) |
028867ac | 94 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 95 | return NULL; |
1da177e4 LT |
96 | } |
97 | ||
cec47e3d KU |
98 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
99 | { | |
100 | if (rq && rq->end_io_data) | |
101 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
102 | return NULL; | |
103 | } | |
104 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
105 | ||
ba61fdd1 JM |
106 | #define MINOR_ALLOCED ((void *)-1) |
107 | ||
1da177e4 LT |
108 | /* |
109 | * Bits for the md->flags field. | |
110 | */ | |
1eb787ec | 111 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 112 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 113 | #define DMF_FROZEN 2 |
fba9f90e | 114 | #define DMF_FREEING 3 |
5c6bd75d | 115 | #define DMF_DELETING 4 |
2e93ccc1 | 116 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 117 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 118 | |
304f3f6a MB |
119 | /* |
120 | * Work processed by per-device workqueue. | |
121 | */ | |
1da177e4 | 122 | struct mapped_device { |
2ca3310e | 123 | struct rw_semaphore io_lock; |
e61290a4 | 124 | struct mutex suspend_lock; |
1da177e4 LT |
125 | rwlock_t map_lock; |
126 | atomic_t holders; | |
5c6bd75d | 127 | atomic_t open_count; |
1da177e4 LT |
128 | |
129 | unsigned long flags; | |
130 | ||
165125e1 | 131 | struct request_queue *queue; |
a5664dad | 132 | unsigned type; |
4a0b4ddf | 133 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
134 | struct mutex type_lock; |
135 | ||
36a0456f AK |
136 | struct target_type *immutable_target_type; |
137 | ||
1da177e4 | 138 | struct gendisk *disk; |
7e51f257 | 139 | char name[16]; |
1da177e4 LT |
140 | |
141 | void *interface_ptr; | |
142 | ||
143 | /* | |
144 | * A list of ios that arrived while we were suspended. | |
145 | */ | |
316d315b | 146 | atomic_t pending[2]; |
1da177e4 | 147 | wait_queue_head_t wait; |
53d5914f | 148 | struct work_struct work; |
74859364 | 149 | struct bio_list deferred; |
022c2611 | 150 | spinlock_t deferred_lock; |
1da177e4 | 151 | |
af7e466a | 152 | /* |
29e4013d | 153 | * Processing queue (flush) |
304f3f6a MB |
154 | */ |
155 | struct workqueue_struct *wq; | |
156 | ||
1da177e4 LT |
157 | /* |
158 | * The current mapping. | |
159 | */ | |
160 | struct dm_table *map; | |
161 | ||
162 | /* | |
163 | * io objects are allocated from here. | |
164 | */ | |
165 | mempool_t *io_pool; | |
166 | mempool_t *tio_pool; | |
167 | ||
9faf400f SB |
168 | struct bio_set *bs; |
169 | ||
1da177e4 LT |
170 | /* |
171 | * Event handling. | |
172 | */ | |
173 | atomic_t event_nr; | |
174 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
175 | atomic_t uevent_seq; |
176 | struct list_head uevent_list; | |
177 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
178 | |
179 | /* | |
180 | * freeze/thaw support require holding onto a super block | |
181 | */ | |
182 | struct super_block *frozen_sb; | |
db8fef4f | 183 | struct block_device *bdev; |
3ac51e74 DW |
184 | |
185 | /* forced geometry settings */ | |
186 | struct hd_geometry geometry; | |
784aae73 MB |
187 | |
188 | /* sysfs handle */ | |
189 | struct kobject kobj; | |
52b1fd5a | 190 | |
d87f4c14 TH |
191 | /* zero-length flush that will be cloned and submitted to targets */ |
192 | struct bio flush_bio; | |
1da177e4 LT |
193 | }; |
194 | ||
e6ee8c0b KU |
195 | /* |
196 | * For mempools pre-allocation at the table loading time. | |
197 | */ | |
198 | struct dm_md_mempools { | |
199 | mempool_t *io_pool; | |
200 | mempool_t *tio_pool; | |
201 | struct bio_set *bs; | |
202 | }; | |
203 | ||
1da177e4 | 204 | #define MIN_IOS 256 |
e18b890b | 205 | static struct kmem_cache *_io_cache; |
8fbf26ad | 206 | static struct kmem_cache *_rq_tio_cache; |
94818742 KO |
207 | |
208 | /* | |
209 | * Unused now, and needs to be deleted. But since io_pool is overloaded and it's | |
210 | * still used for _io_cache, I'm leaving this for a later cleanup | |
211 | */ | |
8fbf26ad | 212 | static struct kmem_cache *_rq_bio_info_cache; |
1da177e4 | 213 | |
1da177e4 LT |
214 | static int __init local_init(void) |
215 | { | |
51157b4a | 216 | int r = -ENOMEM; |
1da177e4 | 217 | |
1da177e4 | 218 | /* allocate a slab for the dm_ios */ |
028867ac | 219 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 220 | if (!_io_cache) |
51157b4a | 221 | return r; |
1da177e4 | 222 | |
8fbf26ad KU |
223 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
224 | if (!_rq_tio_cache) | |
dba14160 | 225 | goto out_free_io_cache; |
8fbf26ad KU |
226 | |
227 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
228 | if (!_rq_bio_info_cache) | |
229 | goto out_free_rq_tio_cache; | |
230 | ||
51e5b2bd | 231 | r = dm_uevent_init(); |
51157b4a | 232 | if (r) |
8fbf26ad | 233 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 234 | |
1da177e4 LT |
235 | _major = major; |
236 | r = register_blkdev(_major, _name); | |
51157b4a KU |
237 | if (r < 0) |
238 | goto out_uevent_exit; | |
1da177e4 LT |
239 | |
240 | if (!_major) | |
241 | _major = r; | |
242 | ||
243 | return 0; | |
51157b4a KU |
244 | |
245 | out_uevent_exit: | |
246 | dm_uevent_exit(); | |
8fbf26ad KU |
247 | out_free_rq_bio_info_cache: |
248 | kmem_cache_destroy(_rq_bio_info_cache); | |
249 | out_free_rq_tio_cache: | |
250 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
251 | out_free_io_cache: |
252 | kmem_cache_destroy(_io_cache); | |
253 | ||
254 | return r; | |
1da177e4 LT |
255 | } |
256 | ||
257 | static void local_exit(void) | |
258 | { | |
8fbf26ad KU |
259 | kmem_cache_destroy(_rq_bio_info_cache); |
260 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 | 261 | kmem_cache_destroy(_io_cache); |
00d59405 | 262 | unregister_blkdev(_major, _name); |
51e5b2bd | 263 | dm_uevent_exit(); |
1da177e4 LT |
264 | |
265 | _major = 0; | |
266 | ||
267 | DMINFO("cleaned up"); | |
268 | } | |
269 | ||
b9249e55 | 270 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
271 | local_init, |
272 | dm_target_init, | |
273 | dm_linear_init, | |
274 | dm_stripe_init, | |
952b3557 | 275 | dm_io_init, |
945fa4d2 | 276 | dm_kcopyd_init, |
1da177e4 LT |
277 | dm_interface_init, |
278 | }; | |
279 | ||
b9249e55 | 280 | static void (*_exits[])(void) = { |
1da177e4 LT |
281 | local_exit, |
282 | dm_target_exit, | |
283 | dm_linear_exit, | |
284 | dm_stripe_exit, | |
952b3557 | 285 | dm_io_exit, |
945fa4d2 | 286 | dm_kcopyd_exit, |
1da177e4 LT |
287 | dm_interface_exit, |
288 | }; | |
289 | ||
290 | static int __init dm_init(void) | |
291 | { | |
292 | const int count = ARRAY_SIZE(_inits); | |
293 | ||
294 | int r, i; | |
295 | ||
296 | for (i = 0; i < count; i++) { | |
297 | r = _inits[i](); | |
298 | if (r) | |
299 | goto bad; | |
300 | } | |
301 | ||
302 | return 0; | |
303 | ||
304 | bad: | |
305 | while (i--) | |
306 | _exits[i](); | |
307 | ||
308 | return r; | |
309 | } | |
310 | ||
311 | static void __exit dm_exit(void) | |
312 | { | |
313 | int i = ARRAY_SIZE(_exits); | |
314 | ||
315 | while (i--) | |
316 | _exits[i](); | |
d15b774c AK |
317 | |
318 | /* | |
319 | * Should be empty by this point. | |
320 | */ | |
d15b774c | 321 | idr_destroy(&_minor_idr); |
1da177e4 LT |
322 | } |
323 | ||
324 | /* | |
325 | * Block device functions | |
326 | */ | |
432a212c MA |
327 | int dm_deleting_md(struct mapped_device *md) |
328 | { | |
329 | return test_bit(DMF_DELETING, &md->flags); | |
330 | } | |
331 | ||
fe5f9f2c | 332 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
333 | { |
334 | struct mapped_device *md; | |
335 | ||
fba9f90e JM |
336 | spin_lock(&_minor_lock); |
337 | ||
fe5f9f2c | 338 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
339 | if (!md) |
340 | goto out; | |
341 | ||
5c6bd75d | 342 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 343 | dm_deleting_md(md)) { |
fba9f90e JM |
344 | md = NULL; |
345 | goto out; | |
346 | } | |
347 | ||
1da177e4 | 348 | dm_get(md); |
5c6bd75d | 349 | atomic_inc(&md->open_count); |
fba9f90e JM |
350 | |
351 | out: | |
352 | spin_unlock(&_minor_lock); | |
353 | ||
354 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
355 | } |
356 | ||
fe5f9f2c | 357 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 358 | { |
fe5f9f2c | 359 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 360 | |
4a1aeb98 MB |
361 | spin_lock(&_minor_lock); |
362 | ||
5c6bd75d | 363 | atomic_dec(&md->open_count); |
1da177e4 | 364 | dm_put(md); |
4a1aeb98 MB |
365 | |
366 | spin_unlock(&_minor_lock); | |
6e9624b8 | 367 | |
1da177e4 LT |
368 | return 0; |
369 | } | |
370 | ||
5c6bd75d AK |
371 | int dm_open_count(struct mapped_device *md) |
372 | { | |
373 | return atomic_read(&md->open_count); | |
374 | } | |
375 | ||
376 | /* | |
377 | * Guarantees nothing is using the device before it's deleted. | |
378 | */ | |
379 | int dm_lock_for_deletion(struct mapped_device *md) | |
380 | { | |
381 | int r = 0; | |
382 | ||
383 | spin_lock(&_minor_lock); | |
384 | ||
385 | if (dm_open_count(md)) | |
386 | r = -EBUSY; | |
387 | else | |
388 | set_bit(DMF_DELETING, &md->flags); | |
389 | ||
390 | spin_unlock(&_minor_lock); | |
391 | ||
392 | return r; | |
393 | } | |
394 | ||
3ac51e74 DW |
395 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
396 | { | |
397 | struct mapped_device *md = bdev->bd_disk->private_data; | |
398 | ||
399 | return dm_get_geometry(md, geo); | |
400 | } | |
401 | ||
fe5f9f2c | 402 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
403 | unsigned int cmd, unsigned long arg) |
404 | { | |
fe5f9f2c | 405 | struct mapped_device *md = bdev->bd_disk->private_data; |
7c666411 | 406 | struct dm_table *map = dm_get_live_table(md); |
aa129a22 MB |
407 | struct dm_target *tgt; |
408 | int r = -ENOTTY; | |
409 | ||
aa129a22 MB |
410 | if (!map || !dm_table_get_size(map)) |
411 | goto out; | |
412 | ||
413 | /* We only support devices that have a single target */ | |
414 | if (dm_table_get_num_targets(map) != 1) | |
415 | goto out; | |
416 | ||
417 | tgt = dm_table_get_target(map, 0); | |
418 | ||
4f186f8b | 419 | if (dm_suspended_md(md)) { |
aa129a22 MB |
420 | r = -EAGAIN; |
421 | goto out; | |
422 | } | |
423 | ||
424 | if (tgt->type->ioctl) | |
647b3d00 | 425 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
426 | |
427 | out: | |
428 | dm_table_put(map); | |
429 | ||
aa129a22 MB |
430 | return r; |
431 | } | |
432 | ||
028867ac | 433 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
434 | { |
435 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
436 | } | |
437 | ||
028867ac | 438 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
439 | { |
440 | mempool_free(io, md->io_pool); | |
441 | } | |
442 | ||
028867ac | 443 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 444 | { |
dba14160 | 445 | bio_put(&tio->clone); |
1da177e4 LT |
446 | } |
447 | ||
08885643 KU |
448 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
449 | gfp_t gfp_mask) | |
cec47e3d | 450 | { |
08885643 | 451 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
452 | } |
453 | ||
454 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
455 | { | |
456 | mempool_free(tio, tio->md->tio_pool); | |
457 | } | |
458 | ||
90abb8c4 KU |
459 | static int md_in_flight(struct mapped_device *md) |
460 | { | |
461 | return atomic_read(&md->pending[READ]) + | |
462 | atomic_read(&md->pending[WRITE]); | |
463 | } | |
464 | ||
3eaf840e JNN |
465 | static void start_io_acct(struct dm_io *io) |
466 | { | |
467 | struct mapped_device *md = io->md; | |
c9959059 | 468 | int cpu; |
316d315b | 469 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
470 | |
471 | io->start_time = jiffies; | |
472 | ||
074a7aca TH |
473 | cpu = part_stat_lock(); |
474 | part_round_stats(cpu, &dm_disk(md)->part0); | |
475 | part_stat_unlock(); | |
1e9bb880 SL |
476 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
477 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
478 | } |
479 | ||
d221d2e7 | 480 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
481 | { |
482 | struct mapped_device *md = io->md; | |
483 | struct bio *bio = io->bio; | |
484 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 485 | int pending, cpu; |
3eaf840e JNN |
486 | int rw = bio_data_dir(bio); |
487 | ||
074a7aca TH |
488 | cpu = part_stat_lock(); |
489 | part_round_stats(cpu, &dm_disk(md)->part0); | |
490 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
491 | part_stat_unlock(); | |
3eaf840e | 492 | |
af7e466a MP |
493 | /* |
494 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 495 | * a flush. |
af7e466a | 496 | */ |
1e9bb880 SL |
497 | pending = atomic_dec_return(&md->pending[rw]); |
498 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 499 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 500 | |
d221d2e7 MP |
501 | /* nudge anyone waiting on suspend queue */ |
502 | if (!pending) | |
503 | wake_up(&md->wait); | |
3eaf840e JNN |
504 | } |
505 | ||
1da177e4 LT |
506 | /* |
507 | * Add the bio to the list of deferred io. | |
508 | */ | |
92c63902 | 509 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 510 | { |
05447420 | 511 | unsigned long flags; |
1da177e4 | 512 | |
05447420 | 513 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 514 | bio_list_add(&md->deferred, bio); |
05447420 | 515 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 516 | queue_work(md->wq, &md->work); |
1da177e4 LT |
517 | } |
518 | ||
519 | /* | |
520 | * Everyone (including functions in this file), should use this | |
521 | * function to access the md->map field, and make sure they call | |
522 | * dm_table_put() when finished. | |
523 | */ | |
7c666411 | 524 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
525 | { |
526 | struct dm_table *t; | |
523d9297 | 527 | unsigned long flags; |
1da177e4 | 528 | |
523d9297 | 529 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
530 | t = md->map; |
531 | if (t) | |
532 | dm_table_get(t); | |
523d9297 | 533 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
534 | |
535 | return t; | |
536 | } | |
537 | ||
3ac51e74 DW |
538 | /* |
539 | * Get the geometry associated with a dm device | |
540 | */ | |
541 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
542 | { | |
543 | *geo = md->geometry; | |
544 | ||
545 | return 0; | |
546 | } | |
547 | ||
548 | /* | |
549 | * Set the geometry of a device. | |
550 | */ | |
551 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
552 | { | |
553 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
554 | ||
555 | if (geo->start > sz) { | |
556 | DMWARN("Start sector is beyond the geometry limits."); | |
557 | return -EINVAL; | |
558 | } | |
559 | ||
560 | md->geometry = *geo; | |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
1da177e4 LT |
565 | /*----------------------------------------------------------------- |
566 | * CRUD START: | |
567 | * A more elegant soln is in the works that uses the queue | |
568 | * merge fn, unfortunately there are a couple of changes to | |
569 | * the block layer that I want to make for this. So in the | |
570 | * interests of getting something for people to use I give | |
571 | * you this clearly demarcated crap. | |
572 | *---------------------------------------------------------------*/ | |
573 | ||
2e93ccc1 KU |
574 | static int __noflush_suspending(struct mapped_device *md) |
575 | { | |
576 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
577 | } | |
578 | ||
1da177e4 LT |
579 | /* |
580 | * Decrements the number of outstanding ios that a bio has been | |
581 | * cloned into, completing the original io if necc. | |
582 | */ | |
858119e1 | 583 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 584 | { |
2e93ccc1 | 585 | unsigned long flags; |
b35f8caa MB |
586 | int io_error; |
587 | struct bio *bio; | |
588 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
589 | |
590 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
591 | if (unlikely(error)) { |
592 | spin_lock_irqsave(&io->endio_lock, flags); | |
593 | if (!(io->error > 0 && __noflush_suspending(md))) | |
594 | io->error = error; | |
595 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
596 | } | |
1da177e4 LT |
597 | |
598 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
599 | if (io->error == DM_ENDIO_REQUEUE) { |
600 | /* | |
601 | * Target requested pushing back the I/O. | |
2e93ccc1 | 602 | */ |
022c2611 | 603 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
604 | if (__noflush_suspending(md)) |
605 | bio_list_add_head(&md->deferred, io->bio); | |
606 | else | |
2e93ccc1 KU |
607 | /* noflush suspend was interrupted. */ |
608 | io->error = -EIO; | |
022c2611 | 609 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
610 | } |
611 | ||
b35f8caa MB |
612 | io_error = io->error; |
613 | bio = io->bio; | |
6a8736d1 TH |
614 | end_io_acct(io); |
615 | free_io(md, io); | |
616 | ||
617 | if (io_error == DM_ENDIO_REQUEUE) | |
618 | return; | |
2e93ccc1 | 619 | |
b372d360 | 620 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 621 | /* |
6a8736d1 TH |
622 | * Preflush done for flush with data, reissue |
623 | * without REQ_FLUSH. | |
af7e466a | 624 | */ |
6a8736d1 TH |
625 | bio->bi_rw &= ~REQ_FLUSH; |
626 | queue_io(md, bio); | |
af7e466a | 627 | } else { |
b372d360 | 628 | /* done with normal IO or empty flush */ |
b372d360 | 629 | bio_endio(bio, io_error); |
b35f8caa | 630 | } |
1da177e4 LT |
631 | } |
632 | } | |
633 | ||
6712ecf8 | 634 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
635 | { |
636 | int r = 0; | |
028867ac | 637 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 638 | struct dm_io *io = tio->io; |
9faf400f | 639 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
640 | dm_endio_fn endio = tio->ti->type->end_io; |
641 | ||
1da177e4 LT |
642 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
643 | error = -EIO; | |
644 | ||
645 | if (endio) { | |
7de3ee57 | 646 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
647 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
648 | /* | |
649 | * error and requeue request are handled | |
650 | * in dec_pending(). | |
651 | */ | |
1da177e4 | 652 | error = r; |
45cbcd79 KU |
653 | else if (r == DM_ENDIO_INCOMPLETE) |
654 | /* The target will handle the io */ | |
6712ecf8 | 655 | return; |
45cbcd79 KU |
656 | else if (r) { |
657 | DMWARN("unimplemented target endio return value: %d", r); | |
658 | BUG(); | |
659 | } | |
1da177e4 LT |
660 | } |
661 | ||
9faf400f | 662 | free_tio(md, tio); |
b35f8caa | 663 | dec_pending(io, error); |
1da177e4 LT |
664 | } |
665 | ||
cec47e3d KU |
666 | /* |
667 | * Partial completion handling for request-based dm | |
668 | */ | |
669 | static void end_clone_bio(struct bio *clone, int error) | |
670 | { | |
671 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
672 | struct dm_rq_target_io *tio = info->tio; | |
673 | struct bio *bio = info->orig; | |
674 | unsigned int nr_bytes = info->orig->bi_size; | |
675 | ||
676 | bio_put(clone); | |
677 | ||
678 | if (tio->error) | |
679 | /* | |
680 | * An error has already been detected on the request. | |
681 | * Once error occurred, just let clone->end_io() handle | |
682 | * the remainder. | |
683 | */ | |
684 | return; | |
685 | else if (error) { | |
686 | /* | |
687 | * Don't notice the error to the upper layer yet. | |
688 | * The error handling decision is made by the target driver, | |
689 | * when the request is completed. | |
690 | */ | |
691 | tio->error = error; | |
692 | return; | |
693 | } | |
694 | ||
695 | /* | |
696 | * I/O for the bio successfully completed. | |
697 | * Notice the data completion to the upper layer. | |
698 | */ | |
699 | ||
700 | /* | |
701 | * bios are processed from the head of the list. | |
702 | * So the completing bio should always be rq->bio. | |
703 | * If it's not, something wrong is happening. | |
704 | */ | |
705 | if (tio->orig->bio != bio) | |
706 | DMERR("bio completion is going in the middle of the request"); | |
707 | ||
708 | /* | |
709 | * Update the original request. | |
710 | * Do not use blk_end_request() here, because it may complete | |
711 | * the original request before the clone, and break the ordering. | |
712 | */ | |
713 | blk_update_request(tio->orig, 0, nr_bytes); | |
714 | } | |
715 | ||
716 | /* | |
717 | * Don't touch any member of the md after calling this function because | |
718 | * the md may be freed in dm_put() at the end of this function. | |
719 | * Or do dm_get() before calling this function and dm_put() later. | |
720 | */ | |
b4324fee | 721 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 722 | { |
b4324fee | 723 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
724 | |
725 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 726 | if (!md_in_flight(md)) |
cec47e3d KU |
727 | wake_up(&md->wait); |
728 | ||
a8c32a5c JA |
729 | /* |
730 | * Run this off this callpath, as drivers could invoke end_io while | |
731 | * inside their request_fn (and holding the queue lock). Calling | |
732 | * back into ->request_fn() could deadlock attempting to grab the | |
733 | * queue lock again. | |
734 | */ | |
cec47e3d | 735 | if (run_queue) |
a8c32a5c | 736 | blk_run_queue_async(md->queue); |
cec47e3d KU |
737 | |
738 | /* | |
739 | * dm_put() must be at the end of this function. See the comment above | |
740 | */ | |
741 | dm_put(md); | |
742 | } | |
743 | ||
a77e28c7 KU |
744 | static void free_rq_clone(struct request *clone) |
745 | { | |
746 | struct dm_rq_target_io *tio = clone->end_io_data; | |
747 | ||
748 | blk_rq_unprep_clone(clone); | |
749 | free_rq_tio(tio); | |
750 | } | |
751 | ||
980691e5 KU |
752 | /* |
753 | * Complete the clone and the original request. | |
754 | * Must be called without queue lock. | |
755 | */ | |
756 | static void dm_end_request(struct request *clone, int error) | |
757 | { | |
758 | int rw = rq_data_dir(clone); | |
759 | struct dm_rq_target_io *tio = clone->end_io_data; | |
760 | struct mapped_device *md = tio->md; | |
761 | struct request *rq = tio->orig; | |
762 | ||
29e4013d | 763 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
764 | rq->errors = clone->errors; |
765 | rq->resid_len = clone->resid_len; | |
766 | ||
767 | if (rq->sense) | |
768 | /* | |
769 | * We are using the sense buffer of the original | |
770 | * request. | |
771 | * So setting the length of the sense data is enough. | |
772 | */ | |
773 | rq->sense_len = clone->sense_len; | |
774 | } | |
775 | ||
776 | free_rq_clone(clone); | |
29e4013d TH |
777 | blk_end_request_all(rq, error); |
778 | rq_completed(md, rw, true); | |
980691e5 KU |
779 | } |
780 | ||
cec47e3d KU |
781 | static void dm_unprep_request(struct request *rq) |
782 | { | |
783 | struct request *clone = rq->special; | |
cec47e3d KU |
784 | |
785 | rq->special = NULL; | |
786 | rq->cmd_flags &= ~REQ_DONTPREP; | |
787 | ||
a77e28c7 | 788 | free_rq_clone(clone); |
cec47e3d KU |
789 | } |
790 | ||
791 | /* | |
792 | * Requeue the original request of a clone. | |
793 | */ | |
794 | void dm_requeue_unmapped_request(struct request *clone) | |
795 | { | |
b4324fee | 796 | int rw = rq_data_dir(clone); |
cec47e3d KU |
797 | struct dm_rq_target_io *tio = clone->end_io_data; |
798 | struct mapped_device *md = tio->md; | |
799 | struct request *rq = tio->orig; | |
800 | struct request_queue *q = rq->q; | |
801 | unsigned long flags; | |
802 | ||
803 | dm_unprep_request(rq); | |
804 | ||
805 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
806 | blk_requeue_request(q, rq); |
807 | spin_unlock_irqrestore(q->queue_lock, flags); | |
808 | ||
b4324fee | 809 | rq_completed(md, rw, 0); |
cec47e3d KU |
810 | } |
811 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
812 | ||
813 | static void __stop_queue(struct request_queue *q) | |
814 | { | |
815 | blk_stop_queue(q); | |
816 | } | |
817 | ||
818 | static void stop_queue(struct request_queue *q) | |
819 | { | |
820 | unsigned long flags; | |
821 | ||
822 | spin_lock_irqsave(q->queue_lock, flags); | |
823 | __stop_queue(q); | |
824 | spin_unlock_irqrestore(q->queue_lock, flags); | |
825 | } | |
826 | ||
827 | static void __start_queue(struct request_queue *q) | |
828 | { | |
829 | if (blk_queue_stopped(q)) | |
830 | blk_start_queue(q); | |
831 | } | |
832 | ||
833 | static void start_queue(struct request_queue *q) | |
834 | { | |
835 | unsigned long flags; | |
836 | ||
837 | spin_lock_irqsave(q->queue_lock, flags); | |
838 | __start_queue(q); | |
839 | spin_unlock_irqrestore(q->queue_lock, flags); | |
840 | } | |
841 | ||
11a68244 | 842 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 843 | { |
11a68244 | 844 | int r = error; |
cec47e3d | 845 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 846 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 847 | |
ba1cbad9 MS |
848 | if (tio->ti) { |
849 | rq_end_io = tio->ti->type->rq_end_io; | |
850 | ||
851 | if (mapped && rq_end_io) | |
852 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
853 | } | |
cec47e3d | 854 | |
11a68244 | 855 | if (r <= 0) |
cec47e3d | 856 | /* The target wants to complete the I/O */ |
11a68244 KU |
857 | dm_end_request(clone, r); |
858 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
859 | /* The target will handle the I/O */ |
860 | return; | |
11a68244 | 861 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
862 | /* The target wants to requeue the I/O */ |
863 | dm_requeue_unmapped_request(clone); | |
864 | else { | |
11a68244 | 865 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
866 | BUG(); |
867 | } | |
868 | } | |
869 | ||
11a68244 KU |
870 | /* |
871 | * Request completion handler for request-based dm | |
872 | */ | |
873 | static void dm_softirq_done(struct request *rq) | |
874 | { | |
875 | bool mapped = true; | |
876 | struct request *clone = rq->completion_data; | |
877 | struct dm_rq_target_io *tio = clone->end_io_data; | |
878 | ||
879 | if (rq->cmd_flags & REQ_FAILED) | |
880 | mapped = false; | |
881 | ||
882 | dm_done(clone, tio->error, mapped); | |
883 | } | |
884 | ||
cec47e3d KU |
885 | /* |
886 | * Complete the clone and the original request with the error status | |
887 | * through softirq context. | |
888 | */ | |
889 | static void dm_complete_request(struct request *clone, int error) | |
890 | { | |
891 | struct dm_rq_target_io *tio = clone->end_io_data; | |
892 | struct request *rq = tio->orig; | |
893 | ||
894 | tio->error = error; | |
895 | rq->completion_data = clone; | |
896 | blk_complete_request(rq); | |
897 | } | |
898 | ||
899 | /* | |
900 | * Complete the not-mapped clone and the original request with the error status | |
901 | * through softirq context. | |
902 | * Target's rq_end_io() function isn't called. | |
903 | * This may be used when the target's map_rq() function fails. | |
904 | */ | |
905 | void dm_kill_unmapped_request(struct request *clone, int error) | |
906 | { | |
907 | struct dm_rq_target_io *tio = clone->end_io_data; | |
908 | struct request *rq = tio->orig; | |
909 | ||
910 | rq->cmd_flags |= REQ_FAILED; | |
911 | dm_complete_request(clone, error); | |
912 | } | |
913 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
914 | ||
915 | /* | |
916 | * Called with the queue lock held | |
917 | */ | |
918 | static void end_clone_request(struct request *clone, int error) | |
919 | { | |
920 | /* | |
921 | * For just cleaning up the information of the queue in which | |
922 | * the clone was dispatched. | |
923 | * The clone is *NOT* freed actually here because it is alloced from | |
924 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
925 | */ | |
926 | __blk_put_request(clone->q, clone); | |
927 | ||
928 | /* | |
929 | * Actual request completion is done in a softirq context which doesn't | |
930 | * hold the queue lock. Otherwise, deadlock could occur because: | |
931 | * - another request may be submitted by the upper level driver | |
932 | * of the stacking during the completion | |
933 | * - the submission which requires queue lock may be done | |
934 | * against this queue | |
935 | */ | |
936 | dm_complete_request(clone, error); | |
937 | } | |
938 | ||
56a67df7 MS |
939 | /* |
940 | * Return maximum size of I/O possible at the supplied sector up to the current | |
941 | * target boundary. | |
942 | */ | |
943 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
944 | { | |
945 | sector_t target_offset = dm_target_offset(ti, sector); | |
946 | ||
947 | return ti->len - target_offset; | |
948 | } | |
949 | ||
950 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 951 | { |
56a67df7 | 952 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 953 | sector_t offset, max_len; |
1da177e4 LT |
954 | |
955 | /* | |
542f9038 | 956 | * Does the target need to split even further? |
1da177e4 | 957 | */ |
542f9038 MS |
958 | if (ti->max_io_len) { |
959 | offset = dm_target_offset(ti, sector); | |
960 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
961 | max_len = sector_div(offset, ti->max_io_len); | |
962 | else | |
963 | max_len = offset & (ti->max_io_len - 1); | |
964 | max_len = ti->max_io_len - max_len; | |
965 | ||
966 | if (len > max_len) | |
967 | len = max_len; | |
1da177e4 LT |
968 | } |
969 | ||
970 | return len; | |
971 | } | |
972 | ||
542f9038 MS |
973 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
974 | { | |
975 | if (len > UINT_MAX) { | |
976 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
977 | (unsigned long long)len, UINT_MAX); | |
978 | ti->error = "Maximum size of target IO is too large"; | |
979 | return -EINVAL; | |
980 | } | |
981 | ||
982 | ti->max_io_len = (uint32_t) len; | |
983 | ||
984 | return 0; | |
985 | } | |
986 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
987 | ||
bd2a49b8 | 988 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
989 | { |
990 | int r; | |
2056a782 | 991 | sector_t sector; |
9faf400f | 992 | struct mapped_device *md; |
dba14160 | 993 | struct bio *clone = &tio->clone; |
bd2a49b8 | 994 | struct dm_target *ti = tio->ti; |
1da177e4 | 995 | |
1da177e4 LT |
996 | clone->bi_end_io = clone_endio; |
997 | clone->bi_private = tio; | |
998 | ||
999 | /* | |
1000 | * Map the clone. If r == 0 we don't need to do | |
1001 | * anything, the target has assumed ownership of | |
1002 | * this io. | |
1003 | */ | |
1004 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1005 | sector = clone->bi_sector; |
7de3ee57 | 1006 | r = ti->type->map(ti, clone); |
45cbcd79 | 1007 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1008 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1009 | |
d07335e5 MS |
1010 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1011 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1012 | |
1da177e4 | 1013 | generic_make_request(clone); |
2e93ccc1 KU |
1014 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1015 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1016 | md = tio->io->md; |
1017 | dec_pending(tio->io, r); | |
9faf400f | 1018 | free_tio(md, tio); |
45cbcd79 KU |
1019 | } else if (r) { |
1020 | DMWARN("unimplemented target map return value: %d", r); | |
1021 | BUG(); | |
1da177e4 LT |
1022 | } |
1023 | } | |
1024 | ||
1025 | struct clone_info { | |
1026 | struct mapped_device *md; | |
1027 | struct dm_table *map; | |
1028 | struct bio *bio; | |
1029 | struct dm_io *io; | |
1030 | sector_t sector; | |
1031 | sector_t sector_count; | |
1032 | unsigned short idx; | |
1033 | }; | |
1034 | ||
bd2a49b8 AK |
1035 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1036 | { | |
1037 | bio->bi_sector = sector; | |
1038 | bio->bi_size = to_bytes(len); | |
1039 | } | |
1040 | ||
1041 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1042 | { | |
1043 | bio->bi_idx = idx; | |
1044 | bio->bi_vcnt = idx + bv_count; | |
1045 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1046 | } | |
1047 | ||
1048 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1049 | unsigned short idx, unsigned len, unsigned offset, | |
1050 | unsigned trim) | |
1051 | { | |
1052 | if (!bio_integrity(bio)) | |
1053 | return; | |
1054 | ||
1055 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1056 | ||
1057 | if (trim) | |
1058 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1059 | } | |
1060 | ||
1da177e4 | 1061 | /* |
d87f4c14 | 1062 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1063 | */ |
14fe594d AK |
1064 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1065 | sector_t sector, unsigned short idx, | |
1066 | unsigned offset, unsigned len) | |
1da177e4 | 1067 | { |
dba14160 | 1068 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1069 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1070 | ||
1da177e4 LT |
1071 | *clone->bi_io_vec = *bv; |
1072 | ||
bd2a49b8 AK |
1073 | bio_setup_sector(clone, sector, len); |
1074 | ||
1da177e4 | 1075 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1076 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1077 | clone->bi_vcnt = 1; |
1da177e4 LT |
1078 | clone->bi_io_vec->bv_offset = offset; |
1079 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1080 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1081 | |
bd2a49b8 | 1082 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1083 | } |
1084 | ||
1085 | /* | |
1086 | * Creates a bio that consists of range of complete bvecs. | |
1087 | */ | |
dba14160 MP |
1088 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1089 | sector_t sector, unsigned short idx, | |
e4c93811 | 1090 | unsigned short bv_count, unsigned len) |
1da177e4 | 1091 | { |
dba14160 | 1092 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1093 | unsigned trim = 0; |
1da177e4 | 1094 | |
9faf400f | 1095 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1096 | bio_setup_sector(clone, sector, len); |
1097 | bio_setup_bv(clone, idx, bv_count); | |
1098 | ||
1099 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1100 | trim = 1; | |
1101 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1102 | } |
1103 | ||
9015df24 | 1104 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1105 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1106 | unsigned target_bio_nr) |
f9ab94ce | 1107 | { |
dba14160 MP |
1108 | struct dm_target_io *tio; |
1109 | struct bio *clone; | |
1110 | ||
1111 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1112 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1113 | |
1114 | tio->io = ci->io; | |
1115 | tio->ti = ti; | |
f9ab94ce | 1116 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1117 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1118 | |
1119 | return tio; | |
1120 | } | |
1121 | ||
14fe594d AK |
1122 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1123 | struct dm_target *ti, | |
1124 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1125 | { |
55a62eef | 1126 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1127 | struct bio *clone = &tio->clone; |
9015df24 | 1128 | |
06a426ce MS |
1129 | /* |
1130 | * Discard requests require the bio's inline iovecs be initialized. | |
1131 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1132 | * and discard, so no need for concern about wasted bvec allocations. | |
1133 | */ | |
dba14160 | 1134 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1135 | if (len) |
1136 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1137 | |
bd2a49b8 | 1138 | __map_bio(tio); |
f9ab94ce MP |
1139 | } |
1140 | ||
14fe594d AK |
1141 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1142 | unsigned num_bios, sector_t len) | |
06a426ce | 1143 | { |
55a62eef | 1144 | unsigned target_bio_nr; |
06a426ce | 1145 | |
55a62eef | 1146 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1147 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1148 | } |
1149 | ||
14fe594d | 1150 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1151 | { |
06a426ce | 1152 | unsigned target_nr = 0; |
f9ab94ce MP |
1153 | struct dm_target *ti; |
1154 | ||
b372d360 | 1155 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1156 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1157 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1158 | |
f9ab94ce MP |
1159 | return 0; |
1160 | } | |
1161 | ||
e4c93811 AK |
1162 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1163 | sector_t sector, int nr_iovecs, | |
1164 | unsigned short idx, unsigned short bv_count, | |
1165 | unsigned offset, unsigned len, | |
1166 | unsigned split_bvec) | |
5ae89a87 | 1167 | { |
dba14160 | 1168 | struct bio *bio = ci->bio; |
5ae89a87 MS |
1169 | struct dm_target_io *tio; |
1170 | ||
e4c93811 AK |
1171 | tio = alloc_tio(ci, ti, nr_iovecs, 0); |
1172 | ||
1173 | if (split_bvec) | |
1174 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1175 | else | |
1176 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1177 | ||
bd2a49b8 | 1178 | __map_bio(tio); |
5ae89a87 MS |
1179 | } |
1180 | ||
55a62eef | 1181 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1182 | |
55a62eef | 1183 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1184 | { |
55a62eef | 1185 | return ti->num_discard_bios; |
23508a96 MS |
1186 | } |
1187 | ||
55a62eef | 1188 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1189 | { |
55a62eef | 1190 | return ti->num_write_same_bios; |
23508a96 MS |
1191 | } |
1192 | ||
1193 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1194 | ||
1195 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1196 | { | |
55a62eef | 1197 | return ti->split_discard_bios; |
23508a96 MS |
1198 | } |
1199 | ||
14fe594d AK |
1200 | static int __send_changing_extent_only(struct clone_info *ci, |
1201 | get_num_bios_fn get_num_bios, | |
1202 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1203 | { |
1204 | struct dm_target *ti; | |
a79245b3 | 1205 | sector_t len; |
55a62eef | 1206 | unsigned num_bios; |
5ae89a87 | 1207 | |
a79245b3 MS |
1208 | do { |
1209 | ti = dm_table_find_target(ci->map, ci->sector); | |
1210 | if (!dm_target_is_valid(ti)) | |
1211 | return -EIO; | |
5ae89a87 | 1212 | |
5ae89a87 | 1213 | /* |
23508a96 MS |
1214 | * Even though the device advertised support for this type of |
1215 | * request, that does not mean every target supports it, and | |
936688d7 | 1216 | * reconfiguration might also have changed that since the |
a79245b3 | 1217 | * check was performed. |
5ae89a87 | 1218 | */ |
55a62eef AK |
1219 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1220 | if (!num_bios) | |
a79245b3 | 1221 | return -EOPNOTSUPP; |
5ae89a87 | 1222 | |
23508a96 | 1223 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1224 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1225 | else | |
1226 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1227 | |
14fe594d | 1228 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1229 | |
1230 | ci->sector += len; | |
1231 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1232 | |
1233 | return 0; | |
1234 | } | |
1235 | ||
14fe594d | 1236 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1237 | { |
14fe594d AK |
1238 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1239 | is_split_required_for_discard); | |
23508a96 MS |
1240 | } |
1241 | ||
14fe594d | 1242 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1243 | { |
14fe594d | 1244 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1245 | } |
1246 | ||
e4c93811 AK |
1247 | /* |
1248 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1249 | */ | |
1250 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1251 | { | |
1252 | struct bio *bio = ci->bio; | |
1253 | sector_t bv_len, total_len = 0; | |
1254 | ||
1255 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1256 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1257 | ||
1258 | if (bv_len > max) | |
1259 | break; | |
1260 | ||
1261 | max -= bv_len; | |
1262 | total_len += bv_len; | |
1263 | } | |
1264 | ||
1265 | return total_len; | |
1266 | } | |
1267 | ||
1268 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1269 | struct dm_target *ti, sector_t max) | |
1270 | { | |
1271 | struct bio *bio = ci->bio; | |
1272 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1273 | sector_t remaining = to_sector(bv->bv_len); | |
1274 | unsigned offset = 0; | |
1275 | sector_t len; | |
1276 | ||
1277 | do { | |
1278 | if (offset) { | |
1279 | ti = dm_table_find_target(ci->map, ci->sector); | |
1280 | if (!dm_target_is_valid(ti)) | |
1281 | return -EIO; | |
1282 | ||
1283 | max = max_io_len(ci->sector, ti); | |
1284 | } | |
1285 | ||
1286 | len = min(remaining, max); | |
1287 | ||
1288 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1289 | bv->bv_offset + offset, len, 1); | |
1290 | ||
1291 | ci->sector += len; | |
1292 | ci->sector_count -= len; | |
1293 | offset += to_bytes(len); | |
1294 | } while (remaining -= len); | |
1295 | ||
1296 | ci->idx++; | |
1297 | ||
1298 | return 0; | |
1299 | } | |
1300 | ||
1301 | /* | |
1302 | * Select the correct strategy for processing a non-flush bio. | |
1303 | */ | |
14fe594d | 1304 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1305 | { |
dba14160 | 1306 | struct bio *bio = ci->bio; |
512875bd | 1307 | struct dm_target *ti; |
e4c93811 AK |
1308 | sector_t len, max; |
1309 | int idx; | |
1da177e4 | 1310 | |
5ae89a87 | 1311 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1312 | return __send_discard(ci); |
23508a96 | 1313 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1314 | return __send_write_same(ci); |
5ae89a87 | 1315 | |
512875bd JN |
1316 | ti = dm_table_find_target(ci->map, ci->sector); |
1317 | if (!dm_target_is_valid(ti)) | |
1318 | return -EIO; | |
1319 | ||
56a67df7 | 1320 | max = max_io_len(ci->sector, ti); |
512875bd | 1321 | |
e4c93811 AK |
1322 | /* |
1323 | * Optimise for the simple case where we can do all of | |
1324 | * the remaining io with a single clone. | |
1325 | */ | |
1da177e4 | 1326 | if (ci->sector_count <= max) { |
e4c93811 AK |
1327 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1328 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1329 | ci->sector_count, 0); | |
1330 | ci->sector_count = 0; | |
1331 | return 0; | |
1332 | } | |
1da177e4 | 1333 | |
e4c93811 AK |
1334 | /* |
1335 | * There are some bvecs that don't span targets. | |
1336 | * Do as many of these as possible. | |
1337 | */ | |
1338 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1339 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1340 | |
e4c93811 AK |
1341 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1342 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1343 | |
1344 | ci->sector += len; | |
1345 | ci->sector_count -= len; | |
e4c93811 | 1346 | ci->idx = idx; |
d2044a94 | 1347 | |
e4c93811 | 1348 | return 0; |
1da177e4 | 1349 | } |
512875bd | 1350 | |
e4c93811 AK |
1351 | /* |
1352 | * Handle a bvec that must be split between two or more targets. | |
1353 | */ | |
1354 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1355 | } |
1356 | ||
1357 | /* | |
14fe594d | 1358 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1359 | */ |
f0b9a450 | 1360 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1361 | { |
1362 | struct clone_info ci; | |
512875bd | 1363 | int error = 0; |
1da177e4 | 1364 | |
7c666411 | 1365 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1366 | if (unlikely(!ci.map)) { |
6a8736d1 | 1367 | bio_io_error(bio); |
f0b9a450 MP |
1368 | return; |
1369 | } | |
692d0eb9 | 1370 | |
1da177e4 | 1371 | ci.md = md; |
1da177e4 LT |
1372 | ci.io = alloc_io(md); |
1373 | ci.io->error = 0; | |
1374 | atomic_set(&ci.io->io_count, 1); | |
1375 | ci.io->bio = bio; | |
1376 | ci.io->md = md; | |
f88fb981 | 1377 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1378 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1379 | ci.idx = bio->bi_idx; |
1380 | ||
3eaf840e | 1381 | start_io_acct(ci.io); |
bd2a49b8 | 1382 | |
b372d360 MS |
1383 | if (bio->bi_rw & REQ_FLUSH) { |
1384 | ci.bio = &ci.md->flush_bio; | |
1385 | ci.sector_count = 0; | |
14fe594d | 1386 | error = __send_empty_flush(&ci); |
b372d360 MS |
1387 | /* dec_pending submits any data associated with flush */ |
1388 | } else { | |
6a8736d1 | 1389 | ci.bio = bio; |
d87f4c14 | 1390 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1391 | while (ci.sector_count && !error) |
14fe594d | 1392 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1393 | } |
1da177e4 LT |
1394 | |
1395 | /* drop the extra reference count */ | |
512875bd | 1396 | dec_pending(ci.io, error); |
1da177e4 LT |
1397 | dm_table_put(ci.map); |
1398 | } | |
1399 | /*----------------------------------------------------------------- | |
1400 | * CRUD END | |
1401 | *---------------------------------------------------------------*/ | |
1402 | ||
f6fccb12 MB |
1403 | static int dm_merge_bvec(struct request_queue *q, |
1404 | struct bvec_merge_data *bvm, | |
1405 | struct bio_vec *biovec) | |
1406 | { | |
1407 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1408 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1409 | struct dm_target *ti; |
1410 | sector_t max_sectors; | |
5037108a | 1411 | int max_size = 0; |
f6fccb12 MB |
1412 | |
1413 | if (unlikely(!map)) | |
5037108a | 1414 | goto out; |
f6fccb12 MB |
1415 | |
1416 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1417 | if (!dm_target_is_valid(ti)) |
1418 | goto out_table; | |
f6fccb12 MB |
1419 | |
1420 | /* | |
1421 | * Find maximum amount of I/O that won't need splitting | |
1422 | */ | |
56a67df7 | 1423 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1424 | (sector_t) BIO_MAX_SECTORS); |
1425 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1426 | if (max_size < 0) | |
1427 | max_size = 0; | |
1428 | ||
1429 | /* | |
1430 | * merge_bvec_fn() returns number of bytes | |
1431 | * it can accept at this offset | |
1432 | * max is precomputed maximal io size | |
1433 | */ | |
1434 | if (max_size && ti->type->merge) | |
1435 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1436 | /* |
1437 | * If the target doesn't support merge method and some of the devices | |
1438 | * provided their merge_bvec method (we know this by looking at | |
1439 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1440 | * entries. So always set max_size to 0, and the code below allows | |
1441 | * just one page. | |
1442 | */ | |
1443 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1444 | ||
1445 | max_size = 0; | |
f6fccb12 | 1446 | |
b01cd5ac | 1447 | out_table: |
5037108a MP |
1448 | dm_table_put(map); |
1449 | ||
1450 | out: | |
f6fccb12 MB |
1451 | /* |
1452 | * Always allow an entire first page | |
1453 | */ | |
1454 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1455 | max_size = biovec->bv_len; | |
1456 | ||
f6fccb12 MB |
1457 | return max_size; |
1458 | } | |
1459 | ||
1da177e4 LT |
1460 | /* |
1461 | * The request function that just remaps the bio built up by | |
1462 | * dm_merge_bvec. | |
1463 | */ | |
5a7bbad2 | 1464 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1465 | { |
12f03a49 | 1466 | int rw = bio_data_dir(bio); |
1da177e4 | 1467 | struct mapped_device *md = q->queuedata; |
c9959059 | 1468 | int cpu; |
1da177e4 | 1469 | |
2ca3310e | 1470 | down_read(&md->io_lock); |
1da177e4 | 1471 | |
074a7aca TH |
1472 | cpu = part_stat_lock(); |
1473 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1474 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1475 | part_stat_unlock(); | |
12f03a49 | 1476 | |
6a8736d1 TH |
1477 | /* if we're suspended, we have to queue this io for later */ |
1478 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1479 | up_read(&md->io_lock); |
1da177e4 | 1480 | |
6a8736d1 TH |
1481 | if (bio_rw(bio) != READA) |
1482 | queue_io(md, bio); | |
1483 | else | |
54d9a1b4 | 1484 | bio_io_error(bio); |
5a7bbad2 | 1485 | return; |
1da177e4 LT |
1486 | } |
1487 | ||
f0b9a450 | 1488 | __split_and_process_bio(md, bio); |
2ca3310e | 1489 | up_read(&md->io_lock); |
5a7bbad2 | 1490 | return; |
cec47e3d KU |
1491 | } |
1492 | ||
1493 | static int dm_request_based(struct mapped_device *md) | |
1494 | { | |
1495 | return blk_queue_stackable(md->queue); | |
1496 | } | |
1497 | ||
5a7bbad2 | 1498 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1499 | { |
1500 | struct mapped_device *md = q->queuedata; | |
1501 | ||
1502 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1503 | blk_queue_bio(q, bio); |
1504 | else | |
1505 | _dm_request(q, bio); | |
cec47e3d KU |
1506 | } |
1507 | ||
1508 | void dm_dispatch_request(struct request *rq) | |
1509 | { | |
1510 | int r; | |
1511 | ||
1512 | if (blk_queue_io_stat(rq->q)) | |
1513 | rq->cmd_flags |= REQ_IO_STAT; | |
1514 | ||
1515 | rq->start_time = jiffies; | |
1516 | r = blk_insert_cloned_request(rq->q, rq); | |
1517 | if (r) | |
1518 | dm_complete_request(rq, r); | |
1519 | } | |
1520 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1521 | ||
cec47e3d KU |
1522 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1523 | void *data) | |
1524 | { | |
1525 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1526 | struct dm_rq_clone_bio_info *info = |
1527 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1528 | |
1529 | info->orig = bio_orig; | |
1530 | info->tio = tio; | |
1531 | bio->bi_end_io = end_clone_bio; | |
1532 | bio->bi_private = info; | |
cec47e3d KU |
1533 | |
1534 | return 0; | |
1535 | } | |
1536 | ||
1537 | static int setup_clone(struct request *clone, struct request *rq, | |
1538 | struct dm_rq_target_io *tio) | |
1539 | { | |
d0bcb878 | 1540 | int r; |
cec47e3d | 1541 | |
29e4013d TH |
1542 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1543 | dm_rq_bio_constructor, tio); | |
1544 | if (r) | |
1545 | return r; | |
cec47e3d | 1546 | |
29e4013d TH |
1547 | clone->cmd = rq->cmd; |
1548 | clone->cmd_len = rq->cmd_len; | |
1549 | clone->sense = rq->sense; | |
1550 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1551 | clone->end_io = end_clone_request; |
1552 | clone->end_io_data = tio; | |
1553 | ||
1554 | return 0; | |
1555 | } | |
1556 | ||
6facdaff KU |
1557 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1558 | gfp_t gfp_mask) | |
1559 | { | |
1560 | struct request *clone; | |
1561 | struct dm_rq_target_io *tio; | |
1562 | ||
1563 | tio = alloc_rq_tio(md, gfp_mask); | |
1564 | if (!tio) | |
1565 | return NULL; | |
1566 | ||
1567 | tio->md = md; | |
1568 | tio->ti = NULL; | |
1569 | tio->orig = rq; | |
1570 | tio->error = 0; | |
1571 | memset(&tio->info, 0, sizeof(tio->info)); | |
1572 | ||
1573 | clone = &tio->clone; | |
1574 | if (setup_clone(clone, rq, tio)) { | |
1575 | /* -ENOMEM */ | |
1576 | free_rq_tio(tio); | |
1577 | return NULL; | |
1578 | } | |
1579 | ||
1580 | return clone; | |
1581 | } | |
1582 | ||
cec47e3d KU |
1583 | /* |
1584 | * Called with the queue lock held. | |
1585 | */ | |
1586 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1587 | { | |
1588 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1589 | struct request *clone; |
1590 | ||
cec47e3d KU |
1591 | if (unlikely(rq->special)) { |
1592 | DMWARN("Already has something in rq->special."); | |
1593 | return BLKPREP_KILL; | |
1594 | } | |
1595 | ||
6facdaff KU |
1596 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1597 | if (!clone) | |
cec47e3d | 1598 | return BLKPREP_DEFER; |
cec47e3d KU |
1599 | |
1600 | rq->special = clone; | |
1601 | rq->cmd_flags |= REQ_DONTPREP; | |
1602 | ||
1603 | return BLKPREP_OK; | |
1604 | } | |
1605 | ||
9eef87da KU |
1606 | /* |
1607 | * Returns: | |
1608 | * 0 : the request has been processed (not requeued) | |
1609 | * !0 : the request has been requeued | |
1610 | */ | |
1611 | static int map_request(struct dm_target *ti, struct request *clone, | |
1612 | struct mapped_device *md) | |
cec47e3d | 1613 | { |
9eef87da | 1614 | int r, requeued = 0; |
cec47e3d KU |
1615 | struct dm_rq_target_io *tio = clone->end_io_data; |
1616 | ||
cec47e3d KU |
1617 | tio->ti = ti; |
1618 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1619 | switch (r) { | |
1620 | case DM_MAPIO_SUBMITTED: | |
1621 | /* The target has taken the I/O to submit by itself later */ | |
1622 | break; | |
1623 | case DM_MAPIO_REMAPPED: | |
1624 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1625 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1626 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1627 | dm_dispatch_request(clone); |
1628 | break; | |
1629 | case DM_MAPIO_REQUEUE: | |
1630 | /* The target wants to requeue the I/O */ | |
1631 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1632 | requeued = 1; |
cec47e3d KU |
1633 | break; |
1634 | default: | |
1635 | if (r > 0) { | |
1636 | DMWARN("unimplemented target map return value: %d", r); | |
1637 | BUG(); | |
1638 | } | |
1639 | ||
1640 | /* The target wants to complete the I/O */ | |
1641 | dm_kill_unmapped_request(clone, r); | |
1642 | break; | |
1643 | } | |
9eef87da KU |
1644 | |
1645 | return requeued; | |
cec47e3d KU |
1646 | } |
1647 | ||
ba1cbad9 MS |
1648 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1649 | { | |
1650 | struct request *clone; | |
1651 | ||
1652 | blk_start_request(orig); | |
1653 | clone = orig->special; | |
1654 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1655 | ||
1656 | /* | |
1657 | * Hold the md reference here for the in-flight I/O. | |
1658 | * We can't rely on the reference count by device opener, | |
1659 | * because the device may be closed during the request completion | |
1660 | * when all bios are completed. | |
1661 | * See the comment in rq_completed() too. | |
1662 | */ | |
1663 | dm_get(md); | |
1664 | ||
1665 | return clone; | |
1666 | } | |
1667 | ||
cec47e3d KU |
1668 | /* |
1669 | * q->request_fn for request-based dm. | |
1670 | * Called with the queue lock held. | |
1671 | */ | |
1672 | static void dm_request_fn(struct request_queue *q) | |
1673 | { | |
1674 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1675 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1676 | struct dm_target *ti; |
b4324fee | 1677 | struct request *rq, *clone; |
29e4013d | 1678 | sector_t pos; |
cec47e3d KU |
1679 | |
1680 | /* | |
b4324fee KU |
1681 | * For suspend, check blk_queue_stopped() and increment |
1682 | * ->pending within a single queue_lock not to increment the | |
1683 | * number of in-flight I/Os after the queue is stopped in | |
1684 | * dm_suspend(). | |
cec47e3d | 1685 | */ |
7eaceacc | 1686 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1687 | rq = blk_peek_request(q); |
1688 | if (!rq) | |
7eaceacc | 1689 | goto delay_and_out; |
cec47e3d | 1690 | |
29e4013d TH |
1691 | /* always use block 0 to find the target for flushes for now */ |
1692 | pos = 0; | |
1693 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1694 | pos = blk_rq_pos(rq); | |
1695 | ||
1696 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1697 | if (!dm_target_is_valid(ti)) { |
1698 | /* | |
1699 | * Must perform setup, that dm_done() requires, | |
1700 | * before calling dm_kill_unmapped_request | |
1701 | */ | |
1702 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1703 | clone = dm_start_request(md, rq); | |
1704 | dm_kill_unmapped_request(clone, -EIO); | |
1705 | continue; | |
1706 | } | |
d0bcb878 | 1707 | |
cec47e3d | 1708 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1709 | goto delay_and_out; |
cec47e3d | 1710 | |
ba1cbad9 | 1711 | clone = dm_start_request(md, rq); |
b4324fee | 1712 | |
cec47e3d | 1713 | spin_unlock(q->queue_lock); |
9eef87da KU |
1714 | if (map_request(ti, clone, md)) |
1715 | goto requeued; | |
1716 | ||
052189a2 KU |
1717 | BUG_ON(!irqs_disabled()); |
1718 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1719 | } |
1720 | ||
1721 | goto out; | |
1722 | ||
9eef87da | 1723 | requeued: |
052189a2 KU |
1724 | BUG_ON(!irqs_disabled()); |
1725 | spin_lock(q->queue_lock); | |
9eef87da | 1726 | |
7eaceacc JA |
1727 | delay_and_out: |
1728 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1729 | out: |
1730 | dm_table_put(map); | |
cec47e3d KU |
1731 | } |
1732 | ||
1733 | int dm_underlying_device_busy(struct request_queue *q) | |
1734 | { | |
1735 | return blk_lld_busy(q); | |
1736 | } | |
1737 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1738 | ||
1739 | static int dm_lld_busy(struct request_queue *q) | |
1740 | { | |
1741 | int r; | |
1742 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1743 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1744 | |
1745 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1746 | r = 1; | |
1747 | else | |
1748 | r = dm_table_any_busy_target(map); | |
1749 | ||
1750 | dm_table_put(map); | |
1751 | ||
1752 | return r; | |
1753 | } | |
1754 | ||
1da177e4 LT |
1755 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1756 | { | |
8a57dfc6 CS |
1757 | int r = bdi_bits; |
1758 | struct mapped_device *md = congested_data; | |
1759 | struct dm_table *map; | |
1da177e4 | 1760 | |
1eb787ec | 1761 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1762 | map = dm_get_live_table(md); |
8a57dfc6 | 1763 | if (map) { |
cec47e3d KU |
1764 | /* |
1765 | * Request-based dm cares about only own queue for | |
1766 | * the query about congestion status of request_queue | |
1767 | */ | |
1768 | if (dm_request_based(md)) | |
1769 | r = md->queue->backing_dev_info.state & | |
1770 | bdi_bits; | |
1771 | else | |
1772 | r = dm_table_any_congested(map, bdi_bits); | |
1773 | ||
8a57dfc6 CS |
1774 | dm_table_put(map); |
1775 | } | |
1776 | } | |
1777 | ||
1da177e4 LT |
1778 | return r; |
1779 | } | |
1780 | ||
1781 | /*----------------------------------------------------------------- | |
1782 | * An IDR is used to keep track of allocated minor numbers. | |
1783 | *---------------------------------------------------------------*/ | |
2b06cfff | 1784 | static void free_minor(int minor) |
1da177e4 | 1785 | { |
f32c10b0 | 1786 | spin_lock(&_minor_lock); |
1da177e4 | 1787 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1788 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1789 | } |
1790 | ||
1791 | /* | |
1792 | * See if the device with a specific minor # is free. | |
1793 | */ | |
cf13ab8e | 1794 | static int specific_minor(int minor) |
1da177e4 | 1795 | { |
c9d76be6 | 1796 | int r; |
1da177e4 LT |
1797 | |
1798 | if (minor >= (1 << MINORBITS)) | |
1799 | return -EINVAL; | |
1800 | ||
c9d76be6 | 1801 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1802 | spin_lock(&_minor_lock); |
1da177e4 | 1803 | |
c9d76be6 | 1804 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1805 | |
f32c10b0 | 1806 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1807 | idr_preload_end(); |
1808 | if (r < 0) | |
1809 | return r == -ENOSPC ? -EBUSY : r; | |
1810 | return 0; | |
1da177e4 LT |
1811 | } |
1812 | ||
cf13ab8e | 1813 | static int next_free_minor(int *minor) |
1da177e4 | 1814 | { |
c9d76be6 | 1815 | int r; |
62f75c2f | 1816 | |
c9d76be6 | 1817 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1818 | spin_lock(&_minor_lock); |
1da177e4 | 1819 | |
c9d76be6 | 1820 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1821 | |
f32c10b0 | 1822 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1823 | idr_preload_end(); |
1824 | if (r < 0) | |
1825 | return r; | |
1826 | *minor = r; | |
1827 | return 0; | |
1da177e4 LT |
1828 | } |
1829 | ||
83d5cde4 | 1830 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1831 | |
53d5914f MP |
1832 | static void dm_wq_work(struct work_struct *work); |
1833 | ||
4a0b4ddf MS |
1834 | static void dm_init_md_queue(struct mapped_device *md) |
1835 | { | |
1836 | /* | |
1837 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1838 | * devices. The type of this dm device has not been decided yet. | |
1839 | * The type is decided at the first table loading time. | |
1840 | * To prevent problematic device stacking, clear the queue flag | |
1841 | * for request stacking support until then. | |
1842 | * | |
1843 | * This queue is new, so no concurrency on the queue_flags. | |
1844 | */ | |
1845 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1846 | ||
1847 | md->queue->queuedata = md; | |
1848 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1849 | md->queue->backing_dev_info.congested_data = md; | |
1850 | blk_queue_make_request(md->queue, dm_request); | |
1851 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1852 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1853 | } | |
1854 | ||
1da177e4 LT |
1855 | /* |
1856 | * Allocate and initialise a blank device with a given minor. | |
1857 | */ | |
2b06cfff | 1858 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1859 | { |
1860 | int r; | |
cf13ab8e | 1861 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1862 | void *old_md; |
1da177e4 LT |
1863 | |
1864 | if (!md) { | |
1865 | DMWARN("unable to allocate device, out of memory."); | |
1866 | return NULL; | |
1867 | } | |
1868 | ||
10da4f79 | 1869 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1870 | goto bad_module_get; |
10da4f79 | 1871 | |
1da177e4 | 1872 | /* get a minor number for the dev */ |
2b06cfff | 1873 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1874 | r = next_free_minor(&minor); |
2b06cfff | 1875 | else |
cf13ab8e | 1876 | r = specific_minor(minor); |
1da177e4 | 1877 | if (r < 0) |
6ed7ade8 | 1878 | goto bad_minor; |
1da177e4 | 1879 | |
a5664dad | 1880 | md->type = DM_TYPE_NONE; |
2ca3310e | 1881 | init_rwsem(&md->io_lock); |
e61290a4 | 1882 | mutex_init(&md->suspend_lock); |
a5664dad | 1883 | mutex_init(&md->type_lock); |
022c2611 | 1884 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1885 | rwlock_init(&md->map_lock); |
1886 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1887 | atomic_set(&md->open_count, 0); |
1da177e4 | 1888 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1889 | atomic_set(&md->uevent_seq, 0); |
1890 | INIT_LIST_HEAD(&md->uevent_list); | |
1891 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1892 | |
4a0b4ddf | 1893 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1894 | if (!md->queue) |
6ed7ade8 | 1895 | goto bad_queue; |
1da177e4 | 1896 | |
4a0b4ddf | 1897 | dm_init_md_queue(md); |
9faf400f | 1898 | |
1da177e4 LT |
1899 | md->disk = alloc_disk(1); |
1900 | if (!md->disk) | |
6ed7ade8 | 1901 | goto bad_disk; |
1da177e4 | 1902 | |
316d315b NK |
1903 | atomic_set(&md->pending[0], 0); |
1904 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1905 | init_waitqueue_head(&md->wait); |
53d5914f | 1906 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1907 | init_waitqueue_head(&md->eventq); |
1908 | ||
1da177e4 LT |
1909 | md->disk->major = _major; |
1910 | md->disk->first_minor = minor; | |
1911 | md->disk->fops = &dm_blk_dops; | |
1912 | md->disk->queue = md->queue; | |
1913 | md->disk->private_data = md; | |
1914 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1915 | add_disk(md->disk); | |
7e51f257 | 1916 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1917 | |
9c4376de TH |
1918 | md->wq = alloc_workqueue("kdmflush", |
1919 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1920 | if (!md->wq) |
1921 | goto bad_thread; | |
1922 | ||
32a926da MP |
1923 | md->bdev = bdget_disk(md->disk, 0); |
1924 | if (!md->bdev) | |
1925 | goto bad_bdev; | |
1926 | ||
6a8736d1 TH |
1927 | bio_init(&md->flush_bio); |
1928 | md->flush_bio.bi_bdev = md->bdev; | |
1929 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1930 | ||
ba61fdd1 | 1931 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1932 | spin_lock(&_minor_lock); |
ba61fdd1 | 1933 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1934 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1935 | |
1936 | BUG_ON(old_md != MINOR_ALLOCED); | |
1937 | ||
1da177e4 LT |
1938 | return md; |
1939 | ||
32a926da MP |
1940 | bad_bdev: |
1941 | destroy_workqueue(md->wq); | |
304f3f6a | 1942 | bad_thread: |
03022c54 | 1943 | del_gendisk(md->disk); |
304f3f6a | 1944 | put_disk(md->disk); |
6ed7ade8 | 1945 | bad_disk: |
1312f40e | 1946 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1947 | bad_queue: |
1da177e4 | 1948 | free_minor(minor); |
6ed7ade8 | 1949 | bad_minor: |
10da4f79 | 1950 | module_put(THIS_MODULE); |
6ed7ade8 | 1951 | bad_module_get: |
1da177e4 LT |
1952 | kfree(md); |
1953 | return NULL; | |
1954 | } | |
1955 | ||
ae9da83f JN |
1956 | static void unlock_fs(struct mapped_device *md); |
1957 | ||
1da177e4 LT |
1958 | static void free_dev(struct mapped_device *md) |
1959 | { | |
f331c029 | 1960 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1961 | |
32a926da MP |
1962 | unlock_fs(md); |
1963 | bdput(md->bdev); | |
304f3f6a | 1964 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1965 | if (md->tio_pool) |
1966 | mempool_destroy(md->tio_pool); | |
1967 | if (md->io_pool) | |
1968 | mempool_destroy(md->io_pool); | |
1969 | if (md->bs) | |
1970 | bioset_free(md->bs); | |
9c47008d | 1971 | blk_integrity_unregister(md->disk); |
1da177e4 | 1972 | del_gendisk(md->disk); |
63d94e48 | 1973 | free_minor(minor); |
fba9f90e JM |
1974 | |
1975 | spin_lock(&_minor_lock); | |
1976 | md->disk->private_data = NULL; | |
1977 | spin_unlock(&_minor_lock); | |
1978 | ||
1da177e4 | 1979 | put_disk(md->disk); |
1312f40e | 1980 | blk_cleanup_queue(md->queue); |
10da4f79 | 1981 | module_put(THIS_MODULE); |
1da177e4 LT |
1982 | kfree(md); |
1983 | } | |
1984 | ||
e6ee8c0b KU |
1985 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1986 | { | |
c0820cf5 | 1987 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1988 | |
16245bdc JN |
1989 | if (md->io_pool && md->bs) { |
1990 | /* The md already has necessary mempools. */ | |
1991 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
1992 | /* | |
1993 | * Reload bioset because front_pad may have changed | |
1994 | * because a different table was loaded. | |
1995 | */ | |
1996 | bioset_free(md->bs); | |
1997 | md->bs = p->bs; | |
1998 | p->bs = NULL; | |
1999 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
2000 | BUG_ON(!md->tio_pool); | |
2001 | /* | |
2002 | * There's no need to reload with request-based dm | |
2003 | * because the size of front_pad doesn't change. | |
2004 | * Note for future: If you are to reload bioset, | |
2005 | * prep-ed requests in the queue may refer | |
2006 | * to bio from the old bioset, so you must walk | |
2007 | * through the queue to unprep. | |
2008 | */ | |
2009 | } | |
e6ee8c0b | 2010 | goto out; |
c0820cf5 | 2011 | } |
e6ee8c0b | 2012 | |
e6ee8c0b KU |
2013 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); |
2014 | ||
2015 | md->io_pool = p->io_pool; | |
2016 | p->io_pool = NULL; | |
2017 | md->tio_pool = p->tio_pool; | |
2018 | p->tio_pool = NULL; | |
2019 | md->bs = p->bs; | |
2020 | p->bs = NULL; | |
2021 | ||
2022 | out: | |
2023 | /* mempool bind completed, now no need any mempools in the table */ | |
2024 | dm_table_free_md_mempools(t); | |
2025 | } | |
2026 | ||
1da177e4 LT |
2027 | /* |
2028 | * Bind a table to the device. | |
2029 | */ | |
2030 | static void event_callback(void *context) | |
2031 | { | |
7a8c3d3b MA |
2032 | unsigned long flags; |
2033 | LIST_HEAD(uevents); | |
1da177e4 LT |
2034 | struct mapped_device *md = (struct mapped_device *) context; |
2035 | ||
7a8c3d3b MA |
2036 | spin_lock_irqsave(&md->uevent_lock, flags); |
2037 | list_splice_init(&md->uevent_list, &uevents); | |
2038 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2039 | ||
ed9e1982 | 2040 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2041 | |
1da177e4 LT |
2042 | atomic_inc(&md->event_nr); |
2043 | wake_up(&md->eventq); | |
2044 | } | |
2045 | ||
c217649b MS |
2046 | /* |
2047 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2048 | */ | |
4e90188b | 2049 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2050 | { |
4e90188b | 2051 | set_capacity(md->disk, size); |
1da177e4 | 2052 | |
db8fef4f | 2053 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2054 | } |
2055 | ||
d5b9dd04 MP |
2056 | /* |
2057 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2058 | * | |
2059 | * If this function returns 0, then the device is either a non-dm | |
2060 | * device without a merge_bvec_fn, or it is a dm device that is | |
2061 | * able to split any bios it receives that are too big. | |
2062 | */ | |
2063 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2064 | { | |
2065 | struct mapped_device *dev_md; | |
2066 | ||
2067 | if (!q->merge_bvec_fn) | |
2068 | return 0; | |
2069 | ||
2070 | if (q->make_request_fn == dm_request) { | |
2071 | dev_md = q->queuedata; | |
2072 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2073 | return 0; | |
2074 | } | |
2075 | ||
2076 | return 1; | |
2077 | } | |
2078 | ||
2079 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2080 | struct dm_dev *dev, sector_t start, | |
2081 | sector_t len, void *data) | |
2082 | { | |
2083 | struct block_device *bdev = dev->bdev; | |
2084 | struct request_queue *q = bdev_get_queue(bdev); | |
2085 | ||
2086 | return dm_queue_merge_is_compulsory(q); | |
2087 | } | |
2088 | ||
2089 | /* | |
2090 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2091 | * on the properties of the underlying devices. | |
2092 | */ | |
2093 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2094 | { | |
2095 | unsigned i = 0; | |
2096 | struct dm_target *ti; | |
2097 | ||
2098 | while (i < dm_table_get_num_targets(table)) { | |
2099 | ti = dm_table_get_target(table, i++); | |
2100 | ||
2101 | if (ti->type->iterate_devices && | |
2102 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2103 | return 0; | |
2104 | } | |
2105 | ||
2106 | return 1; | |
2107 | } | |
2108 | ||
042d2a9b AK |
2109 | /* |
2110 | * Returns old map, which caller must destroy. | |
2111 | */ | |
2112 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2113 | struct queue_limits *limits) | |
1da177e4 | 2114 | { |
042d2a9b | 2115 | struct dm_table *old_map; |
165125e1 | 2116 | struct request_queue *q = md->queue; |
1da177e4 | 2117 | sector_t size; |
523d9297 | 2118 | unsigned long flags; |
d5b9dd04 | 2119 | int merge_is_optional; |
1da177e4 LT |
2120 | |
2121 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2122 | |
2123 | /* | |
2124 | * Wipe any geometry if the size of the table changed. | |
2125 | */ | |
2126 | if (size != get_capacity(md->disk)) | |
2127 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2128 | ||
32a926da | 2129 | __set_size(md, size); |
d5816876 | 2130 | |
2ca3310e AK |
2131 | dm_table_event_callback(t, event_callback, md); |
2132 | ||
e6ee8c0b KU |
2133 | /* |
2134 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2135 | * for request-based during suspension. So stop it to prevent | |
2136 | * I/O mapping before resume. | |
2137 | * This must be done before setting the queue restrictions, | |
2138 | * because request-based dm may be run just after the setting. | |
2139 | */ | |
2140 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2141 | stop_queue(q); | |
2142 | ||
2143 | __bind_mempools(md, t); | |
2144 | ||
d5b9dd04 MP |
2145 | merge_is_optional = dm_table_merge_is_optional(t); |
2146 | ||
523d9297 | 2147 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2148 | old_map = md->map; |
1da177e4 | 2149 | md->map = t; |
36a0456f AK |
2150 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2151 | ||
754c5fc7 | 2152 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2153 | if (merge_is_optional) |
2154 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2155 | else | |
2156 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2157 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2158 | |
042d2a9b | 2159 | return old_map; |
1da177e4 LT |
2160 | } |
2161 | ||
a7940155 AK |
2162 | /* |
2163 | * Returns unbound table for the caller to free. | |
2164 | */ | |
2165 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2166 | { |
2167 | struct dm_table *map = md->map; | |
523d9297 | 2168 | unsigned long flags; |
1da177e4 LT |
2169 | |
2170 | if (!map) | |
a7940155 | 2171 | return NULL; |
1da177e4 LT |
2172 | |
2173 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2174 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2175 | md->map = NULL; |
523d9297 | 2176 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2177 | |
2178 | return map; | |
1da177e4 LT |
2179 | } |
2180 | ||
2181 | /* | |
2182 | * Constructor for a new device. | |
2183 | */ | |
2b06cfff | 2184 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2185 | { |
2186 | struct mapped_device *md; | |
2187 | ||
2b06cfff | 2188 | md = alloc_dev(minor); |
1da177e4 LT |
2189 | if (!md) |
2190 | return -ENXIO; | |
2191 | ||
784aae73 MB |
2192 | dm_sysfs_init(md); |
2193 | ||
1da177e4 LT |
2194 | *result = md; |
2195 | return 0; | |
2196 | } | |
2197 | ||
a5664dad MS |
2198 | /* |
2199 | * Functions to manage md->type. | |
2200 | * All are required to hold md->type_lock. | |
2201 | */ | |
2202 | void dm_lock_md_type(struct mapped_device *md) | |
2203 | { | |
2204 | mutex_lock(&md->type_lock); | |
2205 | } | |
2206 | ||
2207 | void dm_unlock_md_type(struct mapped_device *md) | |
2208 | { | |
2209 | mutex_unlock(&md->type_lock); | |
2210 | } | |
2211 | ||
2212 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2213 | { | |
2214 | md->type = type; | |
2215 | } | |
2216 | ||
2217 | unsigned dm_get_md_type(struct mapped_device *md) | |
2218 | { | |
2219 | return md->type; | |
2220 | } | |
2221 | ||
36a0456f AK |
2222 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2223 | { | |
2224 | return md->immutable_target_type; | |
2225 | } | |
2226 | ||
4a0b4ddf MS |
2227 | /* |
2228 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2229 | */ | |
2230 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2231 | { | |
2232 | struct request_queue *q = NULL; | |
2233 | ||
2234 | if (md->queue->elevator) | |
2235 | return 1; | |
2236 | ||
2237 | /* Fully initialize the queue */ | |
2238 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2239 | if (!q) | |
2240 | return 0; | |
2241 | ||
2242 | md->queue = q; | |
4a0b4ddf MS |
2243 | dm_init_md_queue(md); |
2244 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2245 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2246 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2247 | |
2248 | elv_register_queue(md->queue); | |
2249 | ||
2250 | return 1; | |
2251 | } | |
2252 | ||
2253 | /* | |
2254 | * Setup the DM device's queue based on md's type | |
2255 | */ | |
2256 | int dm_setup_md_queue(struct mapped_device *md) | |
2257 | { | |
2258 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2259 | !dm_init_request_based_queue(md)) { | |
2260 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2261 | return -EINVAL; | |
2262 | } | |
2263 | ||
2264 | return 0; | |
2265 | } | |
2266 | ||
637842cf | 2267 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2268 | { |
2269 | struct mapped_device *md; | |
1da177e4 LT |
2270 | unsigned minor = MINOR(dev); |
2271 | ||
2272 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2273 | return NULL; | |
2274 | ||
f32c10b0 | 2275 | spin_lock(&_minor_lock); |
1da177e4 LT |
2276 | |
2277 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2278 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2279 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2280 | dm_deleting_md(md) || |
17b2f66f | 2281 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2282 | md = NULL; |
fba9f90e JM |
2283 | goto out; |
2284 | } | |
1da177e4 | 2285 | |
fba9f90e | 2286 | out: |
f32c10b0 | 2287 | spin_unlock(&_minor_lock); |
1da177e4 | 2288 | |
637842cf DT |
2289 | return md; |
2290 | } | |
2291 | ||
d229a958 DT |
2292 | struct mapped_device *dm_get_md(dev_t dev) |
2293 | { | |
2294 | struct mapped_device *md = dm_find_md(dev); | |
2295 | ||
2296 | if (md) | |
2297 | dm_get(md); | |
2298 | ||
2299 | return md; | |
2300 | } | |
3cf2e4ba | 2301 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2302 | |
9ade92a9 | 2303 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2304 | { |
9ade92a9 | 2305 | return md->interface_ptr; |
1da177e4 LT |
2306 | } |
2307 | ||
2308 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2309 | { | |
2310 | md->interface_ptr = ptr; | |
2311 | } | |
2312 | ||
2313 | void dm_get(struct mapped_device *md) | |
2314 | { | |
2315 | atomic_inc(&md->holders); | |
3f77316d | 2316 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2317 | } |
2318 | ||
72d94861 AK |
2319 | const char *dm_device_name(struct mapped_device *md) |
2320 | { | |
2321 | return md->name; | |
2322 | } | |
2323 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2324 | ||
3f77316d | 2325 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2326 | { |
1134e5ae | 2327 | struct dm_table *map; |
1da177e4 | 2328 | |
3f77316d | 2329 | might_sleep(); |
fba9f90e | 2330 | |
3f77316d KU |
2331 | spin_lock(&_minor_lock); |
2332 | map = dm_get_live_table(md); | |
2333 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2334 | set_bit(DMF_FREEING, &md->flags); | |
2335 | spin_unlock(&_minor_lock); | |
2336 | ||
2337 | if (!dm_suspended_md(md)) { | |
2338 | dm_table_presuspend_targets(map); | |
2339 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2340 | } |
3f77316d KU |
2341 | |
2342 | /* | |
2343 | * Rare, but there may be I/O requests still going to complete, | |
2344 | * for example. Wait for all references to disappear. | |
2345 | * No one should increment the reference count of the mapped_device, | |
2346 | * after the mapped_device state becomes DMF_FREEING. | |
2347 | */ | |
2348 | if (wait) | |
2349 | while (atomic_read(&md->holders)) | |
2350 | msleep(1); | |
2351 | else if (atomic_read(&md->holders)) | |
2352 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2353 | dm_device_name(md), atomic_read(&md->holders)); | |
2354 | ||
2355 | dm_sysfs_exit(md); | |
2356 | dm_table_put(map); | |
2357 | dm_table_destroy(__unbind(md)); | |
2358 | free_dev(md); | |
2359 | } | |
2360 | ||
2361 | void dm_destroy(struct mapped_device *md) | |
2362 | { | |
2363 | __dm_destroy(md, true); | |
2364 | } | |
2365 | ||
2366 | void dm_destroy_immediate(struct mapped_device *md) | |
2367 | { | |
2368 | __dm_destroy(md, false); | |
2369 | } | |
2370 | ||
2371 | void dm_put(struct mapped_device *md) | |
2372 | { | |
2373 | atomic_dec(&md->holders); | |
1da177e4 | 2374 | } |
79eb885c | 2375 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2376 | |
401600df | 2377 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2378 | { |
2379 | int r = 0; | |
b44ebeb0 MP |
2380 | DECLARE_WAITQUEUE(wait, current); |
2381 | ||
b44ebeb0 | 2382 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2383 | |
2384 | while (1) { | |
401600df | 2385 | set_current_state(interruptible); |
46125c1c | 2386 | |
b4324fee | 2387 | if (!md_in_flight(md)) |
46125c1c MB |
2388 | break; |
2389 | ||
401600df MP |
2390 | if (interruptible == TASK_INTERRUPTIBLE && |
2391 | signal_pending(current)) { | |
46125c1c MB |
2392 | r = -EINTR; |
2393 | break; | |
2394 | } | |
2395 | ||
2396 | io_schedule(); | |
2397 | } | |
2398 | set_current_state(TASK_RUNNING); | |
2399 | ||
b44ebeb0 MP |
2400 | remove_wait_queue(&md->wait, &wait); |
2401 | ||
46125c1c MB |
2402 | return r; |
2403 | } | |
2404 | ||
1da177e4 LT |
2405 | /* |
2406 | * Process the deferred bios | |
2407 | */ | |
ef208587 | 2408 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2409 | { |
ef208587 MP |
2410 | struct mapped_device *md = container_of(work, struct mapped_device, |
2411 | work); | |
6d6f10df | 2412 | struct bio *c; |
1da177e4 | 2413 | |
6a8736d1 | 2414 | down_read(&md->io_lock); |
ef208587 | 2415 | |
3b00b203 | 2416 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2417 | spin_lock_irq(&md->deferred_lock); |
2418 | c = bio_list_pop(&md->deferred); | |
2419 | spin_unlock_irq(&md->deferred_lock); | |
2420 | ||
6a8736d1 | 2421 | if (!c) |
df12ee99 | 2422 | break; |
022c2611 | 2423 | |
6a8736d1 | 2424 | up_read(&md->io_lock); |
3b00b203 | 2425 | |
e6ee8c0b KU |
2426 | if (dm_request_based(md)) |
2427 | generic_make_request(c); | |
6a8736d1 TH |
2428 | else |
2429 | __split_and_process_bio(md, c); | |
3b00b203 | 2430 | |
6a8736d1 | 2431 | down_read(&md->io_lock); |
022c2611 | 2432 | } |
73d410c0 | 2433 | |
6a8736d1 | 2434 | up_read(&md->io_lock); |
1da177e4 LT |
2435 | } |
2436 | ||
9a1fb464 | 2437 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2438 | { |
3b00b203 MP |
2439 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2440 | smp_mb__after_clear_bit(); | |
53d5914f | 2441 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2442 | } |
2443 | ||
1da177e4 | 2444 | /* |
042d2a9b | 2445 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2446 | */ |
042d2a9b | 2447 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2448 | { |
3ae70656 | 2449 | struct dm_table *live_map, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2450 | struct queue_limits limits; |
042d2a9b | 2451 | int r; |
1da177e4 | 2452 | |
e61290a4 | 2453 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2454 | |
2455 | /* device must be suspended */ | |
4f186f8b | 2456 | if (!dm_suspended_md(md)) |
93c534ae | 2457 | goto out; |
1da177e4 | 2458 | |
3ae70656 MS |
2459 | /* |
2460 | * If the new table has no data devices, retain the existing limits. | |
2461 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2462 | * then new I/O is queued based on these limits, and then some paths | |
2463 | * reappear. | |
2464 | */ | |
2465 | if (dm_table_has_no_data_devices(table)) { | |
2466 | live_map = dm_get_live_table(md); | |
2467 | if (live_map) | |
2468 | limits = md->queue->limits; | |
2469 | dm_table_put(live_map); | |
2470 | } | |
2471 | ||
754c5fc7 | 2472 | r = dm_calculate_queue_limits(table, &limits); |
042d2a9b AK |
2473 | if (r) { |
2474 | map = ERR_PTR(r); | |
754c5fc7 | 2475 | goto out; |
042d2a9b | 2476 | } |
754c5fc7 | 2477 | |
042d2a9b | 2478 | map = __bind(md, table, &limits); |
1da177e4 | 2479 | |
93c534ae | 2480 | out: |
e61290a4 | 2481 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2482 | return map; |
1da177e4 LT |
2483 | } |
2484 | ||
2485 | /* | |
2486 | * Functions to lock and unlock any filesystem running on the | |
2487 | * device. | |
2488 | */ | |
2ca3310e | 2489 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2490 | { |
e39e2e95 | 2491 | int r; |
1da177e4 LT |
2492 | |
2493 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2494 | |
db8fef4f | 2495 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2496 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2497 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2498 | md->frozen_sb = NULL; |
2499 | return r; | |
dfbe03f6 AK |
2500 | } |
2501 | ||
aa8d7c2f AK |
2502 | set_bit(DMF_FROZEN, &md->flags); |
2503 | ||
1da177e4 LT |
2504 | return 0; |
2505 | } | |
2506 | ||
2ca3310e | 2507 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2508 | { |
aa8d7c2f AK |
2509 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2510 | return; | |
2511 | ||
db8fef4f | 2512 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2513 | md->frozen_sb = NULL; |
aa8d7c2f | 2514 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2515 | } |
2516 | ||
2517 | /* | |
2518 | * We need to be able to change a mapping table under a mounted | |
2519 | * filesystem. For example we might want to move some data in | |
2520 | * the background. Before the table can be swapped with | |
2521 | * dm_bind_table, dm_suspend must be called to flush any in | |
2522 | * flight bios and ensure that any further io gets deferred. | |
2523 | */ | |
cec47e3d KU |
2524 | /* |
2525 | * Suspend mechanism in request-based dm. | |
2526 | * | |
9f518b27 KU |
2527 | * 1. Flush all I/Os by lock_fs() if needed. |
2528 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2529 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2530 | * |
9f518b27 | 2531 | * To abort suspend, start the request_queue. |
cec47e3d | 2532 | */ |
a3d77d35 | 2533 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2534 | { |
2ca3310e | 2535 | struct dm_table *map = NULL; |
46125c1c | 2536 | int r = 0; |
a3d77d35 | 2537 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2538 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2539 | |
e61290a4 | 2540 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2541 | |
4f186f8b | 2542 | if (dm_suspended_md(md)) { |
73d410c0 | 2543 | r = -EINVAL; |
d287483d | 2544 | goto out_unlock; |
73d410c0 | 2545 | } |
1da177e4 | 2546 | |
7c666411 | 2547 | map = dm_get_live_table(md); |
1da177e4 | 2548 | |
2e93ccc1 KU |
2549 | /* |
2550 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2551 | * This flag is cleared before dm_suspend returns. | |
2552 | */ | |
2553 | if (noflush) | |
2554 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2555 | ||
cf222b37 AK |
2556 | /* This does not get reverted if there's an error later. */ |
2557 | dm_table_presuspend_targets(map); | |
2558 | ||
32a926da | 2559 | /* |
9f518b27 KU |
2560 | * Flush I/O to the device. |
2561 | * Any I/O submitted after lock_fs() may not be flushed. | |
2562 | * noflush takes precedence over do_lockfs. | |
2563 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2564 | */ |
2565 | if (!noflush && do_lockfs) { | |
2566 | r = lock_fs(md); | |
2567 | if (r) | |
f431d966 | 2568 | goto out; |
aa8d7c2f | 2569 | } |
1da177e4 LT |
2570 | |
2571 | /* | |
3b00b203 MP |
2572 | * Here we must make sure that no processes are submitting requests |
2573 | * to target drivers i.e. no one may be executing | |
2574 | * __split_and_process_bio. This is called from dm_request and | |
2575 | * dm_wq_work. | |
2576 | * | |
2577 | * To get all processes out of __split_and_process_bio in dm_request, | |
2578 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2579 | * __split_and_process_bio from dm_request and quiesce the thread |
2580 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2581 | * flush_workqueue(md->wq). | |
1da177e4 | 2582 | */ |
2ca3310e | 2583 | down_write(&md->io_lock); |
1eb787ec | 2584 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2585 | up_write(&md->io_lock); |
1da177e4 | 2586 | |
d0bcb878 | 2587 | /* |
29e4013d TH |
2588 | * Stop md->queue before flushing md->wq in case request-based |
2589 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2590 | */ |
cec47e3d | 2591 | if (dm_request_based(md)) |
9f518b27 | 2592 | stop_queue(md->queue); |
cec47e3d | 2593 | |
d0bcb878 KU |
2594 | flush_workqueue(md->wq); |
2595 | ||
1da177e4 | 2596 | /* |
3b00b203 MP |
2597 | * At this point no more requests are entering target request routines. |
2598 | * We call dm_wait_for_completion to wait for all existing requests | |
2599 | * to finish. | |
1da177e4 | 2600 | */ |
401600df | 2601 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2602 | |
2ca3310e | 2603 | down_write(&md->io_lock); |
6d6f10df | 2604 | if (noflush) |
022c2611 | 2605 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2606 | up_write(&md->io_lock); |
2e93ccc1 | 2607 | |
1da177e4 | 2608 | /* were we interrupted ? */ |
46125c1c | 2609 | if (r < 0) { |
9a1fb464 | 2610 | dm_queue_flush(md); |
73d410c0 | 2611 | |
cec47e3d | 2612 | if (dm_request_based(md)) |
9f518b27 | 2613 | start_queue(md->queue); |
cec47e3d | 2614 | |
2ca3310e | 2615 | unlock_fs(md); |
2e93ccc1 | 2616 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2617 | } |
1da177e4 | 2618 | |
3b00b203 MP |
2619 | /* |
2620 | * If dm_wait_for_completion returned 0, the device is completely | |
2621 | * quiescent now. There is no request-processing activity. All new | |
2622 | * requests are being added to md->deferred list. | |
2623 | */ | |
2624 | ||
2ca3310e | 2625 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2626 | |
4d4471cb KU |
2627 | dm_table_postsuspend_targets(map); |
2628 | ||
2ca3310e AK |
2629 | out: |
2630 | dm_table_put(map); | |
d287483d AK |
2631 | |
2632 | out_unlock: | |
e61290a4 | 2633 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2634 | return r; |
1da177e4 LT |
2635 | } |
2636 | ||
2637 | int dm_resume(struct mapped_device *md) | |
2638 | { | |
cf222b37 | 2639 | int r = -EINVAL; |
cf222b37 | 2640 | struct dm_table *map = NULL; |
1da177e4 | 2641 | |
e61290a4 | 2642 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2643 | if (!dm_suspended_md(md)) |
cf222b37 | 2644 | goto out; |
cf222b37 | 2645 | |
7c666411 | 2646 | map = dm_get_live_table(md); |
2ca3310e | 2647 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2648 | goto out; |
1da177e4 | 2649 | |
8757b776 MB |
2650 | r = dm_table_resume_targets(map); |
2651 | if (r) | |
2652 | goto out; | |
2ca3310e | 2653 | |
9a1fb464 | 2654 | dm_queue_flush(md); |
2ca3310e | 2655 | |
cec47e3d KU |
2656 | /* |
2657 | * Flushing deferred I/Os must be done after targets are resumed | |
2658 | * so that mapping of targets can work correctly. | |
2659 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2660 | */ | |
2661 | if (dm_request_based(md)) | |
2662 | start_queue(md->queue); | |
2663 | ||
2ca3310e AK |
2664 | unlock_fs(md); |
2665 | ||
2666 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2667 | ||
cf222b37 AK |
2668 | r = 0; |
2669 | out: | |
2670 | dm_table_put(map); | |
e61290a4 | 2671 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2672 | |
cf222b37 | 2673 | return r; |
1da177e4 LT |
2674 | } |
2675 | ||
2676 | /*----------------------------------------------------------------- | |
2677 | * Event notification. | |
2678 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2679 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2680 | unsigned cookie) |
69267a30 | 2681 | { |
60935eb2 MB |
2682 | char udev_cookie[DM_COOKIE_LENGTH]; |
2683 | char *envp[] = { udev_cookie, NULL }; | |
2684 | ||
2685 | if (!cookie) | |
3abf85b5 | 2686 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2687 | else { |
2688 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2689 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2690 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2691 | action, envp); | |
60935eb2 | 2692 | } |
69267a30 AK |
2693 | } |
2694 | ||
7a8c3d3b MA |
2695 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2696 | { | |
2697 | return atomic_add_return(1, &md->uevent_seq); | |
2698 | } | |
2699 | ||
1da177e4 LT |
2700 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2701 | { | |
2702 | return atomic_read(&md->event_nr); | |
2703 | } | |
2704 | ||
2705 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2706 | { | |
2707 | return wait_event_interruptible(md->eventq, | |
2708 | (event_nr != atomic_read(&md->event_nr))); | |
2709 | } | |
2710 | ||
7a8c3d3b MA |
2711 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2712 | { | |
2713 | unsigned long flags; | |
2714 | ||
2715 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2716 | list_add(elist, &md->uevent_list); | |
2717 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2718 | } | |
2719 | ||
1da177e4 LT |
2720 | /* |
2721 | * The gendisk is only valid as long as you have a reference | |
2722 | * count on 'md'. | |
2723 | */ | |
2724 | struct gendisk *dm_disk(struct mapped_device *md) | |
2725 | { | |
2726 | return md->disk; | |
2727 | } | |
2728 | ||
784aae73 MB |
2729 | struct kobject *dm_kobject(struct mapped_device *md) |
2730 | { | |
2731 | return &md->kobj; | |
2732 | } | |
2733 | ||
2734 | /* | |
2735 | * struct mapped_device should not be exported outside of dm.c | |
2736 | * so use this check to verify that kobj is part of md structure | |
2737 | */ | |
2738 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2739 | { | |
2740 | struct mapped_device *md; | |
2741 | ||
2742 | md = container_of(kobj, struct mapped_device, kobj); | |
2743 | if (&md->kobj != kobj) | |
2744 | return NULL; | |
2745 | ||
4d89b7b4 | 2746 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2747 | dm_deleting_md(md)) |
4d89b7b4 MB |
2748 | return NULL; |
2749 | ||
784aae73 MB |
2750 | dm_get(md); |
2751 | return md; | |
2752 | } | |
2753 | ||
4f186f8b | 2754 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2755 | { |
2756 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2757 | } | |
2758 | ||
64dbce58 KU |
2759 | int dm_suspended(struct dm_target *ti) |
2760 | { | |
ecdb2e25 | 2761 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2762 | } |
2763 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2764 | ||
2e93ccc1 KU |
2765 | int dm_noflush_suspending(struct dm_target *ti) |
2766 | { | |
ecdb2e25 | 2767 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2768 | } |
2769 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2770 | ||
c0820cf5 | 2771 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b KU |
2772 | { |
2773 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
a91a2785 | 2774 | unsigned int pool_size = (type == DM_TYPE_BIO_BASED) ? 16 : MIN_IOS; |
e6ee8c0b KU |
2775 | |
2776 | if (!pools) | |
2777 | return NULL; | |
2778 | ||
c0820cf5 MP |
2779 | per_bio_data_size = roundup(per_bio_data_size, __alignof__(struct dm_target_io)); |
2780 | ||
e6ee8c0b KU |
2781 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? |
2782 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2783 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2784 | if (!pools->io_pool) | |
2785 | goto free_pools_and_out; | |
2786 | ||
dba14160 MP |
2787 | pools->tio_pool = NULL; |
2788 | if (type == DM_TYPE_REQUEST_BASED) { | |
2789 | pools->tio_pool = mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2790 | if (!pools->tio_pool) | |
2791 | goto free_io_pool_and_out; | |
2792 | } | |
e6ee8c0b | 2793 | |
94818742 | 2794 | pools->bs = (type == DM_TYPE_BIO_BASED) ? |
dba14160 | 2795 | bioset_create(pool_size, |
c0820cf5 | 2796 | per_bio_data_size + offsetof(struct dm_target_io, clone)) : |
94818742 KO |
2797 | bioset_create(pool_size, |
2798 | offsetof(struct dm_rq_clone_bio_info, clone)); | |
e6ee8c0b KU |
2799 | if (!pools->bs) |
2800 | goto free_tio_pool_and_out; | |
2801 | ||
a91a2785 MP |
2802 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
2803 | goto free_bioset_and_out; | |
2804 | ||
e6ee8c0b KU |
2805 | return pools; |
2806 | ||
a91a2785 MP |
2807 | free_bioset_and_out: |
2808 | bioset_free(pools->bs); | |
2809 | ||
e6ee8c0b | 2810 | free_tio_pool_and_out: |
dba14160 MP |
2811 | if (pools->tio_pool) |
2812 | mempool_destroy(pools->tio_pool); | |
e6ee8c0b KU |
2813 | |
2814 | free_io_pool_and_out: | |
2815 | mempool_destroy(pools->io_pool); | |
2816 | ||
2817 | free_pools_and_out: | |
2818 | kfree(pools); | |
2819 | ||
2820 | return NULL; | |
2821 | } | |
2822 | ||
2823 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2824 | { | |
2825 | if (!pools) | |
2826 | return; | |
2827 | ||
2828 | if (pools->io_pool) | |
2829 | mempool_destroy(pools->io_pool); | |
2830 | ||
2831 | if (pools->tio_pool) | |
2832 | mempool_destroy(pools->tio_pool); | |
2833 | ||
2834 | if (pools->bs) | |
2835 | bioset_free(pools->bs); | |
2836 | ||
2837 | kfree(pools); | |
2838 | } | |
2839 | ||
83d5cde4 | 2840 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2841 | .open = dm_blk_open, |
2842 | .release = dm_blk_close, | |
aa129a22 | 2843 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2844 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2845 | .owner = THIS_MODULE |
2846 | }; | |
2847 | ||
2848 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2849 | ||
2850 | /* | |
2851 | * module hooks | |
2852 | */ | |
2853 | module_init(dm_init); | |
2854 | module_exit(dm_exit); | |
2855 | ||
2856 | module_param(major, uint, 0); | |
2857 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2858 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2859 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2860 | MODULE_LICENSE("GPL"); |