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; | |
1da177e4 | 166 | |
9faf400f SB |
167 | struct bio_set *bs; |
168 | ||
1da177e4 LT |
169 | /* |
170 | * Event handling. | |
171 | */ | |
172 | atomic_t event_nr; | |
173 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
174 | atomic_t uevent_seq; |
175 | struct list_head uevent_list; | |
176 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
177 | |
178 | /* | |
179 | * freeze/thaw support require holding onto a super block | |
180 | */ | |
181 | struct super_block *frozen_sb; | |
db8fef4f | 182 | struct block_device *bdev; |
3ac51e74 DW |
183 | |
184 | /* forced geometry settings */ | |
185 | struct hd_geometry geometry; | |
784aae73 MB |
186 | |
187 | /* sysfs handle */ | |
188 | struct kobject kobj; | |
52b1fd5a | 189 | |
d87f4c14 TH |
190 | /* zero-length flush that will be cloned and submitted to targets */ |
191 | struct bio flush_bio; | |
1da177e4 LT |
192 | }; |
193 | ||
e6ee8c0b KU |
194 | /* |
195 | * For mempools pre-allocation at the table loading time. | |
196 | */ | |
197 | struct dm_md_mempools { | |
198 | mempool_t *io_pool; | |
e6ee8c0b KU |
199 | struct bio_set *bs; |
200 | }; | |
201 | ||
1da177e4 | 202 | #define MIN_IOS 256 |
e18b890b | 203 | static struct kmem_cache *_io_cache; |
8fbf26ad | 204 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 205 | |
1da177e4 LT |
206 | static int __init local_init(void) |
207 | { | |
51157b4a | 208 | int r = -ENOMEM; |
1da177e4 | 209 | |
1da177e4 | 210 | /* allocate a slab for the dm_ios */ |
028867ac | 211 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 212 | if (!_io_cache) |
51157b4a | 213 | return r; |
1da177e4 | 214 | |
8fbf26ad KU |
215 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
216 | if (!_rq_tio_cache) | |
dba14160 | 217 | goto out_free_io_cache; |
8fbf26ad | 218 | |
51e5b2bd | 219 | r = dm_uevent_init(); |
51157b4a | 220 | if (r) |
23e5083b | 221 | goto out_free_rq_tio_cache; |
51e5b2bd | 222 | |
1da177e4 LT |
223 | _major = major; |
224 | r = register_blkdev(_major, _name); | |
51157b4a KU |
225 | if (r < 0) |
226 | goto out_uevent_exit; | |
1da177e4 LT |
227 | |
228 | if (!_major) | |
229 | _major = r; | |
230 | ||
231 | return 0; | |
51157b4a KU |
232 | |
233 | out_uevent_exit: | |
234 | dm_uevent_exit(); | |
8fbf26ad KU |
235 | out_free_rq_tio_cache: |
236 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
237 | out_free_io_cache: |
238 | kmem_cache_destroy(_io_cache); | |
239 | ||
240 | return r; | |
1da177e4 LT |
241 | } |
242 | ||
243 | static void local_exit(void) | |
244 | { | |
8fbf26ad | 245 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 246 | kmem_cache_destroy(_io_cache); |
00d59405 | 247 | unregister_blkdev(_major, _name); |
51e5b2bd | 248 | dm_uevent_exit(); |
1da177e4 LT |
249 | |
250 | _major = 0; | |
251 | ||
252 | DMINFO("cleaned up"); | |
253 | } | |
254 | ||
b9249e55 | 255 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
256 | local_init, |
257 | dm_target_init, | |
258 | dm_linear_init, | |
259 | dm_stripe_init, | |
952b3557 | 260 | dm_io_init, |
945fa4d2 | 261 | dm_kcopyd_init, |
1da177e4 LT |
262 | dm_interface_init, |
263 | }; | |
264 | ||
b9249e55 | 265 | static void (*_exits[])(void) = { |
1da177e4 LT |
266 | local_exit, |
267 | dm_target_exit, | |
268 | dm_linear_exit, | |
269 | dm_stripe_exit, | |
952b3557 | 270 | dm_io_exit, |
945fa4d2 | 271 | dm_kcopyd_exit, |
1da177e4 LT |
272 | dm_interface_exit, |
273 | }; | |
274 | ||
275 | static int __init dm_init(void) | |
276 | { | |
277 | const int count = ARRAY_SIZE(_inits); | |
278 | ||
279 | int r, i; | |
280 | ||
281 | for (i = 0; i < count; i++) { | |
282 | r = _inits[i](); | |
283 | if (r) | |
284 | goto bad; | |
285 | } | |
286 | ||
287 | return 0; | |
288 | ||
289 | bad: | |
290 | while (i--) | |
291 | _exits[i](); | |
292 | ||
293 | return r; | |
294 | } | |
295 | ||
296 | static void __exit dm_exit(void) | |
297 | { | |
298 | int i = ARRAY_SIZE(_exits); | |
299 | ||
300 | while (i--) | |
301 | _exits[i](); | |
d15b774c AK |
302 | |
303 | /* | |
304 | * Should be empty by this point. | |
305 | */ | |
d15b774c | 306 | idr_destroy(&_minor_idr); |
1da177e4 LT |
307 | } |
308 | ||
309 | /* | |
310 | * Block device functions | |
311 | */ | |
432a212c MA |
312 | int dm_deleting_md(struct mapped_device *md) |
313 | { | |
314 | return test_bit(DMF_DELETING, &md->flags); | |
315 | } | |
316 | ||
fe5f9f2c | 317 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
318 | { |
319 | struct mapped_device *md; | |
320 | ||
fba9f90e JM |
321 | spin_lock(&_minor_lock); |
322 | ||
fe5f9f2c | 323 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
324 | if (!md) |
325 | goto out; | |
326 | ||
5c6bd75d | 327 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 328 | dm_deleting_md(md)) { |
fba9f90e JM |
329 | md = NULL; |
330 | goto out; | |
331 | } | |
332 | ||
1da177e4 | 333 | dm_get(md); |
5c6bd75d | 334 | atomic_inc(&md->open_count); |
fba9f90e JM |
335 | |
336 | out: | |
337 | spin_unlock(&_minor_lock); | |
338 | ||
339 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
340 | } |
341 | ||
db2a144b | 342 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 343 | { |
fe5f9f2c | 344 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 345 | |
4a1aeb98 MB |
346 | spin_lock(&_minor_lock); |
347 | ||
5c6bd75d | 348 | atomic_dec(&md->open_count); |
1da177e4 | 349 | dm_put(md); |
4a1aeb98 MB |
350 | |
351 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
352 | } |
353 | ||
5c6bd75d AK |
354 | int dm_open_count(struct mapped_device *md) |
355 | { | |
356 | return atomic_read(&md->open_count); | |
357 | } | |
358 | ||
359 | /* | |
360 | * Guarantees nothing is using the device before it's deleted. | |
361 | */ | |
362 | int dm_lock_for_deletion(struct mapped_device *md) | |
363 | { | |
364 | int r = 0; | |
365 | ||
366 | spin_lock(&_minor_lock); | |
367 | ||
368 | if (dm_open_count(md)) | |
369 | r = -EBUSY; | |
370 | else | |
371 | set_bit(DMF_DELETING, &md->flags); | |
372 | ||
373 | spin_unlock(&_minor_lock); | |
374 | ||
375 | return r; | |
376 | } | |
377 | ||
3ac51e74 DW |
378 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
379 | { | |
380 | struct mapped_device *md = bdev->bd_disk->private_data; | |
381 | ||
382 | return dm_get_geometry(md, geo); | |
383 | } | |
384 | ||
fe5f9f2c | 385 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
386 | unsigned int cmd, unsigned long arg) |
387 | { | |
fe5f9f2c | 388 | struct mapped_device *md = bdev->bd_disk->private_data; |
6c182cd8 | 389 | struct dm_table *map; |
aa129a22 MB |
390 | struct dm_target *tgt; |
391 | int r = -ENOTTY; | |
392 | ||
6c182cd8 HR |
393 | retry: |
394 | map = dm_get_live_table(md); | |
aa129a22 MB |
395 | if (!map || !dm_table_get_size(map)) |
396 | goto out; | |
397 | ||
398 | /* We only support devices that have a single target */ | |
399 | if (dm_table_get_num_targets(map) != 1) | |
400 | goto out; | |
401 | ||
402 | tgt = dm_table_get_target(map, 0); | |
403 | ||
4f186f8b | 404 | if (dm_suspended_md(md)) { |
aa129a22 MB |
405 | r = -EAGAIN; |
406 | goto out; | |
407 | } | |
408 | ||
409 | if (tgt->type->ioctl) | |
647b3d00 | 410 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
411 | |
412 | out: | |
413 | dm_table_put(map); | |
414 | ||
6c182cd8 HR |
415 | if (r == -ENOTCONN) { |
416 | msleep(10); | |
417 | goto retry; | |
418 | } | |
419 | ||
aa129a22 MB |
420 | return r; |
421 | } | |
422 | ||
028867ac | 423 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
424 | { |
425 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
426 | } | |
427 | ||
028867ac | 428 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
429 | { |
430 | mempool_free(io, md->io_pool); | |
431 | } | |
432 | ||
028867ac | 433 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 434 | { |
dba14160 | 435 | bio_put(&tio->clone); |
1da177e4 LT |
436 | } |
437 | ||
08885643 KU |
438 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
439 | gfp_t gfp_mask) | |
cec47e3d | 440 | { |
5f015204 | 441 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
442 | } |
443 | ||
444 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
445 | { | |
5f015204 | 446 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
447 | } |
448 | ||
90abb8c4 KU |
449 | static int md_in_flight(struct mapped_device *md) |
450 | { | |
451 | return atomic_read(&md->pending[READ]) + | |
452 | atomic_read(&md->pending[WRITE]); | |
453 | } | |
454 | ||
3eaf840e JNN |
455 | static void start_io_acct(struct dm_io *io) |
456 | { | |
457 | struct mapped_device *md = io->md; | |
c9959059 | 458 | int cpu; |
316d315b | 459 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
460 | |
461 | io->start_time = jiffies; | |
462 | ||
074a7aca TH |
463 | cpu = part_stat_lock(); |
464 | part_round_stats(cpu, &dm_disk(md)->part0); | |
465 | part_stat_unlock(); | |
1e9bb880 SL |
466 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
467 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
468 | } |
469 | ||
d221d2e7 | 470 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
471 | { |
472 | struct mapped_device *md = io->md; | |
473 | struct bio *bio = io->bio; | |
474 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 475 | int pending, cpu; |
3eaf840e JNN |
476 | int rw = bio_data_dir(bio); |
477 | ||
074a7aca TH |
478 | cpu = part_stat_lock(); |
479 | part_round_stats(cpu, &dm_disk(md)->part0); | |
480 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
481 | part_stat_unlock(); | |
3eaf840e | 482 | |
af7e466a MP |
483 | /* |
484 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 485 | * a flush. |
af7e466a | 486 | */ |
1e9bb880 SL |
487 | pending = atomic_dec_return(&md->pending[rw]); |
488 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 489 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 490 | |
d221d2e7 MP |
491 | /* nudge anyone waiting on suspend queue */ |
492 | if (!pending) | |
493 | wake_up(&md->wait); | |
3eaf840e JNN |
494 | } |
495 | ||
1da177e4 LT |
496 | /* |
497 | * Add the bio to the list of deferred io. | |
498 | */ | |
92c63902 | 499 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 500 | { |
05447420 | 501 | unsigned long flags; |
1da177e4 | 502 | |
05447420 | 503 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 504 | bio_list_add(&md->deferred, bio); |
05447420 | 505 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 506 | queue_work(md->wq, &md->work); |
1da177e4 LT |
507 | } |
508 | ||
509 | /* | |
510 | * Everyone (including functions in this file), should use this | |
511 | * function to access the md->map field, and make sure they call | |
512 | * dm_table_put() when finished. | |
513 | */ | |
7c666411 | 514 | struct dm_table *dm_get_live_table(struct mapped_device *md) |
1da177e4 LT |
515 | { |
516 | struct dm_table *t; | |
523d9297 | 517 | unsigned long flags; |
1da177e4 | 518 | |
523d9297 | 519 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
520 | t = md->map; |
521 | if (t) | |
522 | dm_table_get(t); | |
523d9297 | 523 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
524 | |
525 | return t; | |
526 | } | |
527 | ||
3ac51e74 DW |
528 | /* |
529 | * Get the geometry associated with a dm device | |
530 | */ | |
531 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
532 | { | |
533 | *geo = md->geometry; | |
534 | ||
535 | return 0; | |
536 | } | |
537 | ||
538 | /* | |
539 | * Set the geometry of a device. | |
540 | */ | |
541 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
542 | { | |
543 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
544 | ||
545 | if (geo->start > sz) { | |
546 | DMWARN("Start sector is beyond the geometry limits."); | |
547 | return -EINVAL; | |
548 | } | |
549 | ||
550 | md->geometry = *geo; | |
551 | ||
552 | return 0; | |
553 | } | |
554 | ||
1da177e4 LT |
555 | /*----------------------------------------------------------------- |
556 | * CRUD START: | |
557 | * A more elegant soln is in the works that uses the queue | |
558 | * merge fn, unfortunately there are a couple of changes to | |
559 | * the block layer that I want to make for this. So in the | |
560 | * interests of getting something for people to use I give | |
561 | * you this clearly demarcated crap. | |
562 | *---------------------------------------------------------------*/ | |
563 | ||
2e93ccc1 KU |
564 | static int __noflush_suspending(struct mapped_device *md) |
565 | { | |
566 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
567 | } | |
568 | ||
1da177e4 LT |
569 | /* |
570 | * Decrements the number of outstanding ios that a bio has been | |
571 | * cloned into, completing the original io if necc. | |
572 | */ | |
858119e1 | 573 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 574 | { |
2e93ccc1 | 575 | unsigned long flags; |
b35f8caa MB |
576 | int io_error; |
577 | struct bio *bio; | |
578 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
579 | |
580 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
581 | if (unlikely(error)) { |
582 | spin_lock_irqsave(&io->endio_lock, flags); | |
583 | if (!(io->error > 0 && __noflush_suspending(md))) | |
584 | io->error = error; | |
585 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
586 | } | |
1da177e4 LT |
587 | |
588 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
589 | if (io->error == DM_ENDIO_REQUEUE) { |
590 | /* | |
591 | * Target requested pushing back the I/O. | |
2e93ccc1 | 592 | */ |
022c2611 | 593 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
594 | if (__noflush_suspending(md)) |
595 | bio_list_add_head(&md->deferred, io->bio); | |
596 | else | |
2e93ccc1 KU |
597 | /* noflush suspend was interrupted. */ |
598 | io->error = -EIO; | |
022c2611 | 599 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
600 | } |
601 | ||
b35f8caa MB |
602 | io_error = io->error; |
603 | bio = io->bio; | |
6a8736d1 TH |
604 | end_io_acct(io); |
605 | free_io(md, io); | |
606 | ||
607 | if (io_error == DM_ENDIO_REQUEUE) | |
608 | return; | |
2e93ccc1 | 609 | |
b372d360 | 610 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 611 | /* |
6a8736d1 TH |
612 | * Preflush done for flush with data, reissue |
613 | * without REQ_FLUSH. | |
af7e466a | 614 | */ |
6a8736d1 TH |
615 | bio->bi_rw &= ~REQ_FLUSH; |
616 | queue_io(md, bio); | |
af7e466a | 617 | } else { |
b372d360 | 618 | /* done with normal IO or empty flush */ |
0a82a8d1 | 619 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 620 | bio_endio(bio, io_error); |
b35f8caa | 621 | } |
1da177e4 LT |
622 | } |
623 | } | |
624 | ||
6712ecf8 | 625 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
626 | { |
627 | int r = 0; | |
028867ac | 628 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 629 | struct dm_io *io = tio->io; |
9faf400f | 630 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
631 | dm_endio_fn endio = tio->ti->type->end_io; |
632 | ||
1da177e4 LT |
633 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
634 | error = -EIO; | |
635 | ||
636 | if (endio) { | |
7de3ee57 | 637 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
638 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
639 | /* | |
640 | * error and requeue request are handled | |
641 | * in dec_pending(). | |
642 | */ | |
1da177e4 | 643 | error = r; |
45cbcd79 KU |
644 | else if (r == DM_ENDIO_INCOMPLETE) |
645 | /* The target will handle the io */ | |
6712ecf8 | 646 | return; |
45cbcd79 KU |
647 | else if (r) { |
648 | DMWARN("unimplemented target endio return value: %d", r); | |
649 | BUG(); | |
650 | } | |
1da177e4 LT |
651 | } |
652 | ||
9faf400f | 653 | free_tio(md, tio); |
b35f8caa | 654 | dec_pending(io, error); |
1da177e4 LT |
655 | } |
656 | ||
cec47e3d KU |
657 | /* |
658 | * Partial completion handling for request-based dm | |
659 | */ | |
660 | static void end_clone_bio(struct bio *clone, int error) | |
661 | { | |
662 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
663 | struct dm_rq_target_io *tio = info->tio; | |
664 | struct bio *bio = info->orig; | |
665 | unsigned int nr_bytes = info->orig->bi_size; | |
666 | ||
667 | bio_put(clone); | |
668 | ||
669 | if (tio->error) | |
670 | /* | |
671 | * An error has already been detected on the request. | |
672 | * Once error occurred, just let clone->end_io() handle | |
673 | * the remainder. | |
674 | */ | |
675 | return; | |
676 | else if (error) { | |
677 | /* | |
678 | * Don't notice the error to the upper layer yet. | |
679 | * The error handling decision is made by the target driver, | |
680 | * when the request is completed. | |
681 | */ | |
682 | tio->error = error; | |
683 | return; | |
684 | } | |
685 | ||
686 | /* | |
687 | * I/O for the bio successfully completed. | |
688 | * Notice the data completion to the upper layer. | |
689 | */ | |
690 | ||
691 | /* | |
692 | * bios are processed from the head of the list. | |
693 | * So the completing bio should always be rq->bio. | |
694 | * If it's not, something wrong is happening. | |
695 | */ | |
696 | if (tio->orig->bio != bio) | |
697 | DMERR("bio completion is going in the middle of the request"); | |
698 | ||
699 | /* | |
700 | * Update the original request. | |
701 | * Do not use blk_end_request() here, because it may complete | |
702 | * the original request before the clone, and break the ordering. | |
703 | */ | |
704 | blk_update_request(tio->orig, 0, nr_bytes); | |
705 | } | |
706 | ||
707 | /* | |
708 | * Don't touch any member of the md after calling this function because | |
709 | * the md may be freed in dm_put() at the end of this function. | |
710 | * Or do dm_get() before calling this function and dm_put() later. | |
711 | */ | |
b4324fee | 712 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 713 | { |
b4324fee | 714 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
715 | |
716 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 717 | if (!md_in_flight(md)) |
cec47e3d KU |
718 | wake_up(&md->wait); |
719 | ||
a8c32a5c JA |
720 | /* |
721 | * Run this off this callpath, as drivers could invoke end_io while | |
722 | * inside their request_fn (and holding the queue lock). Calling | |
723 | * back into ->request_fn() could deadlock attempting to grab the | |
724 | * queue lock again. | |
725 | */ | |
cec47e3d | 726 | if (run_queue) |
a8c32a5c | 727 | blk_run_queue_async(md->queue); |
cec47e3d KU |
728 | |
729 | /* | |
730 | * dm_put() must be at the end of this function. See the comment above | |
731 | */ | |
732 | dm_put(md); | |
733 | } | |
734 | ||
a77e28c7 KU |
735 | static void free_rq_clone(struct request *clone) |
736 | { | |
737 | struct dm_rq_target_io *tio = clone->end_io_data; | |
738 | ||
739 | blk_rq_unprep_clone(clone); | |
740 | free_rq_tio(tio); | |
741 | } | |
742 | ||
980691e5 KU |
743 | /* |
744 | * Complete the clone and the original request. | |
745 | * Must be called without queue lock. | |
746 | */ | |
747 | static void dm_end_request(struct request *clone, int error) | |
748 | { | |
749 | int rw = rq_data_dir(clone); | |
750 | struct dm_rq_target_io *tio = clone->end_io_data; | |
751 | struct mapped_device *md = tio->md; | |
752 | struct request *rq = tio->orig; | |
753 | ||
29e4013d | 754 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
755 | rq->errors = clone->errors; |
756 | rq->resid_len = clone->resid_len; | |
757 | ||
758 | if (rq->sense) | |
759 | /* | |
760 | * We are using the sense buffer of the original | |
761 | * request. | |
762 | * So setting the length of the sense data is enough. | |
763 | */ | |
764 | rq->sense_len = clone->sense_len; | |
765 | } | |
766 | ||
767 | free_rq_clone(clone); | |
29e4013d TH |
768 | blk_end_request_all(rq, error); |
769 | rq_completed(md, rw, true); | |
980691e5 KU |
770 | } |
771 | ||
cec47e3d KU |
772 | static void dm_unprep_request(struct request *rq) |
773 | { | |
774 | struct request *clone = rq->special; | |
cec47e3d KU |
775 | |
776 | rq->special = NULL; | |
777 | rq->cmd_flags &= ~REQ_DONTPREP; | |
778 | ||
a77e28c7 | 779 | free_rq_clone(clone); |
cec47e3d KU |
780 | } |
781 | ||
782 | /* | |
783 | * Requeue the original request of a clone. | |
784 | */ | |
785 | void dm_requeue_unmapped_request(struct request *clone) | |
786 | { | |
b4324fee | 787 | int rw = rq_data_dir(clone); |
cec47e3d KU |
788 | struct dm_rq_target_io *tio = clone->end_io_data; |
789 | struct mapped_device *md = tio->md; | |
790 | struct request *rq = tio->orig; | |
791 | struct request_queue *q = rq->q; | |
792 | unsigned long flags; | |
793 | ||
794 | dm_unprep_request(rq); | |
795 | ||
796 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
797 | blk_requeue_request(q, rq); |
798 | spin_unlock_irqrestore(q->queue_lock, flags); | |
799 | ||
b4324fee | 800 | rq_completed(md, rw, 0); |
cec47e3d KU |
801 | } |
802 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
803 | ||
804 | static void __stop_queue(struct request_queue *q) | |
805 | { | |
806 | blk_stop_queue(q); | |
807 | } | |
808 | ||
809 | static void stop_queue(struct request_queue *q) | |
810 | { | |
811 | unsigned long flags; | |
812 | ||
813 | spin_lock_irqsave(q->queue_lock, flags); | |
814 | __stop_queue(q); | |
815 | spin_unlock_irqrestore(q->queue_lock, flags); | |
816 | } | |
817 | ||
818 | static void __start_queue(struct request_queue *q) | |
819 | { | |
820 | if (blk_queue_stopped(q)) | |
821 | blk_start_queue(q); | |
822 | } | |
823 | ||
824 | static void start_queue(struct request_queue *q) | |
825 | { | |
826 | unsigned long flags; | |
827 | ||
828 | spin_lock_irqsave(q->queue_lock, flags); | |
829 | __start_queue(q); | |
830 | spin_unlock_irqrestore(q->queue_lock, flags); | |
831 | } | |
832 | ||
11a68244 | 833 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 834 | { |
11a68244 | 835 | int r = error; |
cec47e3d | 836 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 837 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 838 | |
ba1cbad9 MS |
839 | if (tio->ti) { |
840 | rq_end_io = tio->ti->type->rq_end_io; | |
841 | ||
842 | if (mapped && rq_end_io) | |
843 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
844 | } | |
cec47e3d | 845 | |
11a68244 | 846 | if (r <= 0) |
cec47e3d | 847 | /* The target wants to complete the I/O */ |
11a68244 KU |
848 | dm_end_request(clone, r); |
849 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
850 | /* The target will handle the I/O */ |
851 | return; | |
11a68244 | 852 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
853 | /* The target wants to requeue the I/O */ |
854 | dm_requeue_unmapped_request(clone); | |
855 | else { | |
11a68244 | 856 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
857 | BUG(); |
858 | } | |
859 | } | |
860 | ||
11a68244 KU |
861 | /* |
862 | * Request completion handler for request-based dm | |
863 | */ | |
864 | static void dm_softirq_done(struct request *rq) | |
865 | { | |
866 | bool mapped = true; | |
867 | struct request *clone = rq->completion_data; | |
868 | struct dm_rq_target_io *tio = clone->end_io_data; | |
869 | ||
870 | if (rq->cmd_flags & REQ_FAILED) | |
871 | mapped = false; | |
872 | ||
873 | dm_done(clone, tio->error, mapped); | |
874 | } | |
875 | ||
cec47e3d KU |
876 | /* |
877 | * Complete the clone and the original request with the error status | |
878 | * through softirq context. | |
879 | */ | |
880 | static void dm_complete_request(struct request *clone, int error) | |
881 | { | |
882 | struct dm_rq_target_io *tio = clone->end_io_data; | |
883 | struct request *rq = tio->orig; | |
884 | ||
885 | tio->error = error; | |
886 | rq->completion_data = clone; | |
887 | blk_complete_request(rq); | |
888 | } | |
889 | ||
890 | /* | |
891 | * Complete the not-mapped clone and the original request with the error status | |
892 | * through softirq context. | |
893 | * Target's rq_end_io() function isn't called. | |
894 | * This may be used when the target's map_rq() function fails. | |
895 | */ | |
896 | void dm_kill_unmapped_request(struct request *clone, int error) | |
897 | { | |
898 | struct dm_rq_target_io *tio = clone->end_io_data; | |
899 | struct request *rq = tio->orig; | |
900 | ||
901 | rq->cmd_flags |= REQ_FAILED; | |
902 | dm_complete_request(clone, error); | |
903 | } | |
904 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
905 | ||
906 | /* | |
907 | * Called with the queue lock held | |
908 | */ | |
909 | static void end_clone_request(struct request *clone, int error) | |
910 | { | |
911 | /* | |
912 | * For just cleaning up the information of the queue in which | |
913 | * the clone was dispatched. | |
914 | * The clone is *NOT* freed actually here because it is alloced from | |
915 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
916 | */ | |
917 | __blk_put_request(clone->q, clone); | |
918 | ||
919 | /* | |
920 | * Actual request completion is done in a softirq context which doesn't | |
921 | * hold the queue lock. Otherwise, deadlock could occur because: | |
922 | * - another request may be submitted by the upper level driver | |
923 | * of the stacking during the completion | |
924 | * - the submission which requires queue lock may be done | |
925 | * against this queue | |
926 | */ | |
927 | dm_complete_request(clone, error); | |
928 | } | |
929 | ||
56a67df7 MS |
930 | /* |
931 | * Return maximum size of I/O possible at the supplied sector up to the current | |
932 | * target boundary. | |
933 | */ | |
934 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
935 | { | |
936 | sector_t target_offset = dm_target_offset(ti, sector); | |
937 | ||
938 | return ti->len - target_offset; | |
939 | } | |
940 | ||
941 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 942 | { |
56a67df7 | 943 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 944 | sector_t offset, max_len; |
1da177e4 LT |
945 | |
946 | /* | |
542f9038 | 947 | * Does the target need to split even further? |
1da177e4 | 948 | */ |
542f9038 MS |
949 | if (ti->max_io_len) { |
950 | offset = dm_target_offset(ti, sector); | |
951 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
952 | max_len = sector_div(offset, ti->max_io_len); | |
953 | else | |
954 | max_len = offset & (ti->max_io_len - 1); | |
955 | max_len = ti->max_io_len - max_len; | |
956 | ||
957 | if (len > max_len) | |
958 | len = max_len; | |
1da177e4 LT |
959 | } |
960 | ||
961 | return len; | |
962 | } | |
963 | ||
542f9038 MS |
964 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
965 | { | |
966 | if (len > UINT_MAX) { | |
967 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
968 | (unsigned long long)len, UINT_MAX); | |
969 | ti->error = "Maximum size of target IO is too large"; | |
970 | return -EINVAL; | |
971 | } | |
972 | ||
973 | ti->max_io_len = (uint32_t) len; | |
974 | ||
975 | return 0; | |
976 | } | |
977 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
978 | ||
bd2a49b8 | 979 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
980 | { |
981 | int r; | |
2056a782 | 982 | sector_t sector; |
9faf400f | 983 | struct mapped_device *md; |
dba14160 | 984 | struct bio *clone = &tio->clone; |
bd2a49b8 | 985 | struct dm_target *ti = tio->ti; |
1da177e4 | 986 | |
1da177e4 LT |
987 | clone->bi_end_io = clone_endio; |
988 | clone->bi_private = tio; | |
989 | ||
990 | /* | |
991 | * Map the clone. If r == 0 we don't need to do | |
992 | * anything, the target has assumed ownership of | |
993 | * this io. | |
994 | */ | |
995 | atomic_inc(&tio->io->io_count); | |
2056a782 | 996 | sector = clone->bi_sector; |
7de3ee57 | 997 | r = ti->type->map(ti, clone); |
45cbcd79 | 998 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 999 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1000 | |
d07335e5 MS |
1001 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1002 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1003 | |
1da177e4 | 1004 | generic_make_request(clone); |
2e93ccc1 KU |
1005 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1006 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1007 | md = tio->io->md; |
1008 | dec_pending(tio->io, r); | |
9faf400f | 1009 | free_tio(md, tio); |
45cbcd79 KU |
1010 | } else if (r) { |
1011 | DMWARN("unimplemented target map return value: %d", r); | |
1012 | BUG(); | |
1da177e4 LT |
1013 | } |
1014 | } | |
1015 | ||
1016 | struct clone_info { | |
1017 | struct mapped_device *md; | |
1018 | struct dm_table *map; | |
1019 | struct bio *bio; | |
1020 | struct dm_io *io; | |
1021 | sector_t sector; | |
1022 | sector_t sector_count; | |
1023 | unsigned short idx; | |
1024 | }; | |
1025 | ||
bd2a49b8 AK |
1026 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1027 | { | |
1028 | bio->bi_sector = sector; | |
1029 | bio->bi_size = to_bytes(len); | |
1030 | } | |
1031 | ||
1032 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1033 | { | |
1034 | bio->bi_idx = idx; | |
1035 | bio->bi_vcnt = idx + bv_count; | |
1036 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1037 | } | |
1038 | ||
1039 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1040 | unsigned short idx, unsigned len, unsigned offset, | |
1041 | unsigned trim) | |
1042 | { | |
1043 | if (!bio_integrity(bio)) | |
1044 | return; | |
1045 | ||
1046 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1047 | ||
1048 | if (trim) | |
1049 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1050 | } | |
1051 | ||
1da177e4 | 1052 | /* |
d87f4c14 | 1053 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1054 | */ |
14fe594d AK |
1055 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1056 | sector_t sector, unsigned short idx, | |
1057 | unsigned offset, unsigned len) | |
1da177e4 | 1058 | { |
dba14160 | 1059 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1060 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1061 | ||
1da177e4 LT |
1062 | *clone->bi_io_vec = *bv; |
1063 | ||
bd2a49b8 AK |
1064 | bio_setup_sector(clone, sector, len); |
1065 | ||
1da177e4 | 1066 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1067 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1068 | clone->bi_vcnt = 1; |
1da177e4 LT |
1069 | clone->bi_io_vec->bv_offset = offset; |
1070 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1071 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1072 | |
bd2a49b8 | 1073 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1074 | } |
1075 | ||
1076 | /* | |
1077 | * Creates a bio that consists of range of complete bvecs. | |
1078 | */ | |
dba14160 MP |
1079 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1080 | sector_t sector, unsigned short idx, | |
e4c93811 | 1081 | unsigned short bv_count, unsigned len) |
1da177e4 | 1082 | { |
dba14160 | 1083 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1084 | unsigned trim = 0; |
1da177e4 | 1085 | |
9faf400f | 1086 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1087 | bio_setup_sector(clone, sector, len); |
1088 | bio_setup_bv(clone, idx, bv_count); | |
1089 | ||
1090 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1091 | trim = 1; | |
1092 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1093 | } |
1094 | ||
9015df24 | 1095 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1096 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1097 | unsigned target_bio_nr) |
f9ab94ce | 1098 | { |
dba14160 MP |
1099 | struct dm_target_io *tio; |
1100 | struct bio *clone; | |
1101 | ||
1102 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1103 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1104 | |
1105 | tio->io = ci->io; | |
1106 | tio->ti = ti; | |
f9ab94ce | 1107 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1108 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1109 | |
1110 | return tio; | |
1111 | } | |
1112 | ||
14fe594d AK |
1113 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1114 | struct dm_target *ti, | |
1115 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1116 | { |
55a62eef | 1117 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1118 | struct bio *clone = &tio->clone; |
9015df24 | 1119 | |
06a426ce MS |
1120 | /* |
1121 | * Discard requests require the bio's inline iovecs be initialized. | |
1122 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1123 | * and discard, so no need for concern about wasted bvec allocations. | |
1124 | */ | |
dba14160 | 1125 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1126 | if (len) |
1127 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1128 | |
bd2a49b8 | 1129 | __map_bio(tio); |
f9ab94ce MP |
1130 | } |
1131 | ||
14fe594d AK |
1132 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1133 | unsigned num_bios, sector_t len) | |
06a426ce | 1134 | { |
55a62eef | 1135 | unsigned target_bio_nr; |
06a426ce | 1136 | |
55a62eef | 1137 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1138 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1139 | } |
1140 | ||
14fe594d | 1141 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1142 | { |
06a426ce | 1143 | unsigned target_nr = 0; |
f9ab94ce MP |
1144 | struct dm_target *ti; |
1145 | ||
b372d360 | 1146 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1147 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1148 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1149 | |
f9ab94ce MP |
1150 | return 0; |
1151 | } | |
1152 | ||
e4c93811 AK |
1153 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1154 | sector_t sector, int nr_iovecs, | |
1155 | unsigned short idx, unsigned short bv_count, | |
1156 | unsigned offset, unsigned len, | |
1157 | unsigned split_bvec) | |
5ae89a87 | 1158 | { |
dba14160 | 1159 | struct bio *bio = ci->bio; |
5ae89a87 | 1160 | struct dm_target_io *tio; |
b0d8ed4d AK |
1161 | unsigned target_bio_nr; |
1162 | unsigned num_target_bios = 1; | |
5ae89a87 | 1163 | |
b0d8ed4d AK |
1164 | /* |
1165 | * Does the target want to receive duplicate copies of the bio? | |
1166 | */ | |
1167 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1168 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1169 | |
b0d8ed4d AK |
1170 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
1171 | tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr); | |
1172 | if (split_bvec) | |
1173 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1174 | else | |
1175 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1176 | __map_bio(tio); | |
1177 | } | |
5ae89a87 MS |
1178 | } |
1179 | ||
55a62eef | 1180 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1181 | |
55a62eef | 1182 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1183 | { |
55a62eef | 1184 | return ti->num_discard_bios; |
23508a96 MS |
1185 | } |
1186 | ||
55a62eef | 1187 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1188 | { |
55a62eef | 1189 | return ti->num_write_same_bios; |
23508a96 MS |
1190 | } |
1191 | ||
1192 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1193 | ||
1194 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1195 | { | |
55a62eef | 1196 | return ti->split_discard_bios; |
23508a96 MS |
1197 | } |
1198 | ||
14fe594d AK |
1199 | static int __send_changing_extent_only(struct clone_info *ci, |
1200 | get_num_bios_fn get_num_bios, | |
1201 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1202 | { |
1203 | struct dm_target *ti; | |
a79245b3 | 1204 | sector_t len; |
55a62eef | 1205 | unsigned num_bios; |
5ae89a87 | 1206 | |
a79245b3 MS |
1207 | do { |
1208 | ti = dm_table_find_target(ci->map, ci->sector); | |
1209 | if (!dm_target_is_valid(ti)) | |
1210 | return -EIO; | |
5ae89a87 | 1211 | |
5ae89a87 | 1212 | /* |
23508a96 MS |
1213 | * Even though the device advertised support for this type of |
1214 | * request, that does not mean every target supports it, and | |
936688d7 | 1215 | * reconfiguration might also have changed that since the |
a79245b3 | 1216 | * check was performed. |
5ae89a87 | 1217 | */ |
55a62eef AK |
1218 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1219 | if (!num_bios) | |
a79245b3 | 1220 | return -EOPNOTSUPP; |
5ae89a87 | 1221 | |
23508a96 | 1222 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1223 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1224 | else | |
1225 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1226 | |
14fe594d | 1227 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1228 | |
1229 | ci->sector += len; | |
1230 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1231 | |
1232 | return 0; | |
1233 | } | |
1234 | ||
14fe594d | 1235 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1236 | { |
14fe594d AK |
1237 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1238 | is_split_required_for_discard); | |
23508a96 MS |
1239 | } |
1240 | ||
14fe594d | 1241 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1242 | { |
14fe594d | 1243 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1244 | } |
1245 | ||
e4c93811 AK |
1246 | /* |
1247 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1248 | */ | |
1249 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1250 | { | |
1251 | struct bio *bio = ci->bio; | |
1252 | sector_t bv_len, total_len = 0; | |
1253 | ||
1254 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1255 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1256 | ||
1257 | if (bv_len > max) | |
1258 | break; | |
1259 | ||
1260 | max -= bv_len; | |
1261 | total_len += bv_len; | |
1262 | } | |
1263 | ||
1264 | return total_len; | |
1265 | } | |
1266 | ||
1267 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1268 | struct dm_target *ti, sector_t max) | |
1269 | { | |
1270 | struct bio *bio = ci->bio; | |
1271 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1272 | sector_t remaining = to_sector(bv->bv_len); | |
1273 | unsigned offset = 0; | |
1274 | sector_t len; | |
1275 | ||
1276 | do { | |
1277 | if (offset) { | |
1278 | ti = dm_table_find_target(ci->map, ci->sector); | |
1279 | if (!dm_target_is_valid(ti)) | |
1280 | return -EIO; | |
1281 | ||
1282 | max = max_io_len(ci->sector, ti); | |
1283 | } | |
1284 | ||
1285 | len = min(remaining, max); | |
1286 | ||
1287 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1288 | bv->bv_offset + offset, len, 1); | |
1289 | ||
1290 | ci->sector += len; | |
1291 | ci->sector_count -= len; | |
1292 | offset += to_bytes(len); | |
1293 | } while (remaining -= len); | |
1294 | ||
1295 | ci->idx++; | |
1296 | ||
1297 | return 0; | |
1298 | } | |
1299 | ||
1300 | /* | |
1301 | * Select the correct strategy for processing a non-flush bio. | |
1302 | */ | |
14fe594d | 1303 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1304 | { |
dba14160 | 1305 | struct bio *bio = ci->bio; |
512875bd | 1306 | struct dm_target *ti; |
e4c93811 AK |
1307 | sector_t len, max; |
1308 | int idx; | |
1da177e4 | 1309 | |
5ae89a87 | 1310 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1311 | return __send_discard(ci); |
23508a96 | 1312 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1313 | return __send_write_same(ci); |
5ae89a87 | 1314 | |
512875bd JN |
1315 | ti = dm_table_find_target(ci->map, ci->sector); |
1316 | if (!dm_target_is_valid(ti)) | |
1317 | return -EIO; | |
1318 | ||
56a67df7 | 1319 | max = max_io_len(ci->sector, ti); |
512875bd | 1320 | |
e4c93811 AK |
1321 | /* |
1322 | * Optimise for the simple case where we can do all of | |
1323 | * the remaining io with a single clone. | |
1324 | */ | |
1da177e4 | 1325 | if (ci->sector_count <= max) { |
e4c93811 AK |
1326 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1327 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1328 | ci->sector_count, 0); | |
1329 | ci->sector_count = 0; | |
1330 | return 0; | |
1331 | } | |
1da177e4 | 1332 | |
e4c93811 AK |
1333 | /* |
1334 | * There are some bvecs that don't span targets. | |
1335 | * Do as many of these as possible. | |
1336 | */ | |
1337 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1338 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1339 | |
e4c93811 AK |
1340 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1341 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1342 | |
1343 | ci->sector += len; | |
1344 | ci->sector_count -= len; | |
e4c93811 | 1345 | ci->idx = idx; |
d2044a94 | 1346 | |
e4c93811 | 1347 | return 0; |
1da177e4 | 1348 | } |
512875bd | 1349 | |
e4c93811 AK |
1350 | /* |
1351 | * Handle a bvec that must be split between two or more targets. | |
1352 | */ | |
1353 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1354 | } |
1355 | ||
1356 | /* | |
14fe594d | 1357 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1358 | */ |
f0b9a450 | 1359 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1360 | { |
1361 | struct clone_info ci; | |
512875bd | 1362 | int error = 0; |
1da177e4 | 1363 | |
7c666411 | 1364 | ci.map = dm_get_live_table(md); |
f0b9a450 | 1365 | if (unlikely(!ci.map)) { |
6a8736d1 | 1366 | bio_io_error(bio); |
f0b9a450 MP |
1367 | return; |
1368 | } | |
692d0eb9 | 1369 | |
1da177e4 | 1370 | ci.md = md; |
1da177e4 LT |
1371 | ci.io = alloc_io(md); |
1372 | ci.io->error = 0; | |
1373 | atomic_set(&ci.io->io_count, 1); | |
1374 | ci.io->bio = bio; | |
1375 | ci.io->md = md; | |
f88fb981 | 1376 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1377 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1378 | ci.idx = bio->bi_idx; |
1379 | ||
3eaf840e | 1380 | start_io_acct(ci.io); |
bd2a49b8 | 1381 | |
b372d360 MS |
1382 | if (bio->bi_rw & REQ_FLUSH) { |
1383 | ci.bio = &ci.md->flush_bio; | |
1384 | ci.sector_count = 0; | |
14fe594d | 1385 | error = __send_empty_flush(&ci); |
b372d360 MS |
1386 | /* dec_pending submits any data associated with flush */ |
1387 | } else { | |
6a8736d1 | 1388 | ci.bio = bio; |
d87f4c14 | 1389 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1390 | while (ci.sector_count && !error) |
14fe594d | 1391 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1392 | } |
1da177e4 LT |
1393 | |
1394 | /* drop the extra reference count */ | |
512875bd | 1395 | dec_pending(ci.io, error); |
1da177e4 LT |
1396 | dm_table_put(ci.map); |
1397 | } | |
1398 | /*----------------------------------------------------------------- | |
1399 | * CRUD END | |
1400 | *---------------------------------------------------------------*/ | |
1401 | ||
f6fccb12 MB |
1402 | static int dm_merge_bvec(struct request_queue *q, |
1403 | struct bvec_merge_data *bvm, | |
1404 | struct bio_vec *biovec) | |
1405 | { | |
1406 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1407 | struct dm_table *map = dm_get_live_table(md); |
f6fccb12 MB |
1408 | struct dm_target *ti; |
1409 | sector_t max_sectors; | |
5037108a | 1410 | int max_size = 0; |
f6fccb12 MB |
1411 | |
1412 | if (unlikely(!map)) | |
5037108a | 1413 | goto out; |
f6fccb12 MB |
1414 | |
1415 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1416 | if (!dm_target_is_valid(ti)) |
1417 | goto out_table; | |
f6fccb12 MB |
1418 | |
1419 | /* | |
1420 | * Find maximum amount of I/O that won't need splitting | |
1421 | */ | |
56a67df7 | 1422 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1423 | (sector_t) BIO_MAX_SECTORS); |
1424 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1425 | if (max_size < 0) | |
1426 | max_size = 0; | |
1427 | ||
1428 | /* | |
1429 | * merge_bvec_fn() returns number of bytes | |
1430 | * it can accept at this offset | |
1431 | * max is precomputed maximal io size | |
1432 | */ | |
1433 | if (max_size && ti->type->merge) | |
1434 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1435 | /* |
1436 | * If the target doesn't support merge method and some of the devices | |
1437 | * provided their merge_bvec method (we know this by looking at | |
1438 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1439 | * entries. So always set max_size to 0, and the code below allows | |
1440 | * just one page. | |
1441 | */ | |
1442 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1443 | ||
1444 | max_size = 0; | |
f6fccb12 | 1445 | |
b01cd5ac | 1446 | out_table: |
5037108a MP |
1447 | dm_table_put(map); |
1448 | ||
1449 | out: | |
f6fccb12 MB |
1450 | /* |
1451 | * Always allow an entire first page | |
1452 | */ | |
1453 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1454 | max_size = biovec->bv_len; | |
1455 | ||
f6fccb12 MB |
1456 | return max_size; |
1457 | } | |
1458 | ||
1da177e4 LT |
1459 | /* |
1460 | * The request function that just remaps the bio built up by | |
1461 | * dm_merge_bvec. | |
1462 | */ | |
5a7bbad2 | 1463 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1464 | { |
12f03a49 | 1465 | int rw = bio_data_dir(bio); |
1da177e4 | 1466 | struct mapped_device *md = q->queuedata; |
c9959059 | 1467 | int cpu; |
1da177e4 | 1468 | |
2ca3310e | 1469 | down_read(&md->io_lock); |
1da177e4 | 1470 | |
074a7aca TH |
1471 | cpu = part_stat_lock(); |
1472 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1473 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1474 | part_stat_unlock(); | |
12f03a49 | 1475 | |
6a8736d1 TH |
1476 | /* if we're suspended, we have to queue this io for later */ |
1477 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
2ca3310e | 1478 | up_read(&md->io_lock); |
1da177e4 | 1479 | |
6a8736d1 TH |
1480 | if (bio_rw(bio) != READA) |
1481 | queue_io(md, bio); | |
1482 | else | |
54d9a1b4 | 1483 | bio_io_error(bio); |
5a7bbad2 | 1484 | return; |
1da177e4 LT |
1485 | } |
1486 | ||
f0b9a450 | 1487 | __split_and_process_bio(md, bio); |
2ca3310e | 1488 | up_read(&md->io_lock); |
5a7bbad2 | 1489 | return; |
cec47e3d KU |
1490 | } |
1491 | ||
1492 | static int dm_request_based(struct mapped_device *md) | |
1493 | { | |
1494 | return blk_queue_stackable(md->queue); | |
1495 | } | |
1496 | ||
5a7bbad2 | 1497 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1498 | { |
1499 | struct mapped_device *md = q->queuedata; | |
1500 | ||
1501 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1502 | blk_queue_bio(q, bio); |
1503 | else | |
1504 | _dm_request(q, bio); | |
cec47e3d KU |
1505 | } |
1506 | ||
1507 | void dm_dispatch_request(struct request *rq) | |
1508 | { | |
1509 | int r; | |
1510 | ||
1511 | if (blk_queue_io_stat(rq->q)) | |
1512 | rq->cmd_flags |= REQ_IO_STAT; | |
1513 | ||
1514 | rq->start_time = jiffies; | |
1515 | r = blk_insert_cloned_request(rq->q, rq); | |
1516 | if (r) | |
1517 | dm_complete_request(rq, r); | |
1518 | } | |
1519 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1520 | ||
cec47e3d KU |
1521 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1522 | void *data) | |
1523 | { | |
1524 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1525 | struct dm_rq_clone_bio_info *info = |
1526 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1527 | |
1528 | info->orig = bio_orig; | |
1529 | info->tio = tio; | |
1530 | bio->bi_end_io = end_clone_bio; | |
1531 | bio->bi_private = info; | |
cec47e3d KU |
1532 | |
1533 | return 0; | |
1534 | } | |
1535 | ||
1536 | static int setup_clone(struct request *clone, struct request *rq, | |
1537 | struct dm_rq_target_io *tio) | |
1538 | { | |
d0bcb878 | 1539 | int r; |
cec47e3d | 1540 | |
29e4013d TH |
1541 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1542 | dm_rq_bio_constructor, tio); | |
1543 | if (r) | |
1544 | return r; | |
cec47e3d | 1545 | |
29e4013d TH |
1546 | clone->cmd = rq->cmd; |
1547 | clone->cmd_len = rq->cmd_len; | |
1548 | clone->sense = rq->sense; | |
1549 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1550 | clone->end_io = end_clone_request; |
1551 | clone->end_io_data = tio; | |
1552 | ||
1553 | return 0; | |
1554 | } | |
1555 | ||
6facdaff KU |
1556 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1557 | gfp_t gfp_mask) | |
1558 | { | |
1559 | struct request *clone; | |
1560 | struct dm_rq_target_io *tio; | |
1561 | ||
1562 | tio = alloc_rq_tio(md, gfp_mask); | |
1563 | if (!tio) | |
1564 | return NULL; | |
1565 | ||
1566 | tio->md = md; | |
1567 | tio->ti = NULL; | |
1568 | tio->orig = rq; | |
1569 | tio->error = 0; | |
1570 | memset(&tio->info, 0, sizeof(tio->info)); | |
1571 | ||
1572 | clone = &tio->clone; | |
1573 | if (setup_clone(clone, rq, tio)) { | |
1574 | /* -ENOMEM */ | |
1575 | free_rq_tio(tio); | |
1576 | return NULL; | |
1577 | } | |
1578 | ||
1579 | return clone; | |
1580 | } | |
1581 | ||
cec47e3d KU |
1582 | /* |
1583 | * Called with the queue lock held. | |
1584 | */ | |
1585 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1586 | { | |
1587 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1588 | struct request *clone; |
1589 | ||
cec47e3d KU |
1590 | if (unlikely(rq->special)) { |
1591 | DMWARN("Already has something in rq->special."); | |
1592 | return BLKPREP_KILL; | |
1593 | } | |
1594 | ||
6facdaff KU |
1595 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1596 | if (!clone) | |
cec47e3d | 1597 | return BLKPREP_DEFER; |
cec47e3d KU |
1598 | |
1599 | rq->special = clone; | |
1600 | rq->cmd_flags |= REQ_DONTPREP; | |
1601 | ||
1602 | return BLKPREP_OK; | |
1603 | } | |
1604 | ||
9eef87da KU |
1605 | /* |
1606 | * Returns: | |
1607 | * 0 : the request has been processed (not requeued) | |
1608 | * !0 : the request has been requeued | |
1609 | */ | |
1610 | static int map_request(struct dm_target *ti, struct request *clone, | |
1611 | struct mapped_device *md) | |
cec47e3d | 1612 | { |
9eef87da | 1613 | int r, requeued = 0; |
cec47e3d KU |
1614 | struct dm_rq_target_io *tio = clone->end_io_data; |
1615 | ||
cec47e3d KU |
1616 | tio->ti = ti; |
1617 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1618 | switch (r) { | |
1619 | case DM_MAPIO_SUBMITTED: | |
1620 | /* The target has taken the I/O to submit by itself later */ | |
1621 | break; | |
1622 | case DM_MAPIO_REMAPPED: | |
1623 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1624 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1625 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1626 | dm_dispatch_request(clone); |
1627 | break; | |
1628 | case DM_MAPIO_REQUEUE: | |
1629 | /* The target wants to requeue the I/O */ | |
1630 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1631 | requeued = 1; |
cec47e3d KU |
1632 | break; |
1633 | default: | |
1634 | if (r > 0) { | |
1635 | DMWARN("unimplemented target map return value: %d", r); | |
1636 | BUG(); | |
1637 | } | |
1638 | ||
1639 | /* The target wants to complete the I/O */ | |
1640 | dm_kill_unmapped_request(clone, r); | |
1641 | break; | |
1642 | } | |
9eef87da KU |
1643 | |
1644 | return requeued; | |
cec47e3d KU |
1645 | } |
1646 | ||
ba1cbad9 MS |
1647 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1648 | { | |
1649 | struct request *clone; | |
1650 | ||
1651 | blk_start_request(orig); | |
1652 | clone = orig->special; | |
1653 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1654 | ||
1655 | /* | |
1656 | * Hold the md reference here for the in-flight I/O. | |
1657 | * We can't rely on the reference count by device opener, | |
1658 | * because the device may be closed during the request completion | |
1659 | * when all bios are completed. | |
1660 | * See the comment in rq_completed() too. | |
1661 | */ | |
1662 | dm_get(md); | |
1663 | ||
1664 | return clone; | |
1665 | } | |
1666 | ||
cec47e3d KU |
1667 | /* |
1668 | * q->request_fn for request-based dm. | |
1669 | * Called with the queue lock held. | |
1670 | */ | |
1671 | static void dm_request_fn(struct request_queue *q) | |
1672 | { | |
1673 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1674 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d | 1675 | struct dm_target *ti; |
b4324fee | 1676 | struct request *rq, *clone; |
29e4013d | 1677 | sector_t pos; |
cec47e3d KU |
1678 | |
1679 | /* | |
b4324fee KU |
1680 | * For suspend, check blk_queue_stopped() and increment |
1681 | * ->pending within a single queue_lock not to increment the | |
1682 | * number of in-flight I/Os after the queue is stopped in | |
1683 | * dm_suspend(). | |
cec47e3d | 1684 | */ |
7eaceacc | 1685 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1686 | rq = blk_peek_request(q); |
1687 | if (!rq) | |
7eaceacc | 1688 | goto delay_and_out; |
cec47e3d | 1689 | |
29e4013d TH |
1690 | /* always use block 0 to find the target for flushes for now */ |
1691 | pos = 0; | |
1692 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1693 | pos = blk_rq_pos(rq); | |
1694 | ||
1695 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1696 | if (!dm_target_is_valid(ti)) { |
1697 | /* | |
1698 | * Must perform setup, that dm_done() requires, | |
1699 | * before calling dm_kill_unmapped_request | |
1700 | */ | |
1701 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1702 | clone = dm_start_request(md, rq); | |
1703 | dm_kill_unmapped_request(clone, -EIO); | |
1704 | continue; | |
1705 | } | |
d0bcb878 | 1706 | |
cec47e3d | 1707 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1708 | goto delay_and_out; |
cec47e3d | 1709 | |
ba1cbad9 | 1710 | clone = dm_start_request(md, rq); |
b4324fee | 1711 | |
cec47e3d | 1712 | spin_unlock(q->queue_lock); |
9eef87da KU |
1713 | if (map_request(ti, clone, md)) |
1714 | goto requeued; | |
1715 | ||
052189a2 KU |
1716 | BUG_ON(!irqs_disabled()); |
1717 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1718 | } |
1719 | ||
1720 | goto out; | |
1721 | ||
9eef87da | 1722 | requeued: |
052189a2 KU |
1723 | BUG_ON(!irqs_disabled()); |
1724 | spin_lock(q->queue_lock); | |
9eef87da | 1725 | |
7eaceacc JA |
1726 | delay_and_out: |
1727 | blk_delay_queue(q, HZ / 10); | |
cec47e3d KU |
1728 | out: |
1729 | dm_table_put(map); | |
cec47e3d KU |
1730 | } |
1731 | ||
1732 | int dm_underlying_device_busy(struct request_queue *q) | |
1733 | { | |
1734 | return blk_lld_busy(q); | |
1735 | } | |
1736 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1737 | ||
1738 | static int dm_lld_busy(struct request_queue *q) | |
1739 | { | |
1740 | int r; | |
1741 | struct mapped_device *md = q->queuedata; | |
7c666411 | 1742 | struct dm_table *map = dm_get_live_table(md); |
cec47e3d KU |
1743 | |
1744 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1745 | r = 1; | |
1746 | else | |
1747 | r = dm_table_any_busy_target(map); | |
1748 | ||
1749 | dm_table_put(map); | |
1750 | ||
1751 | return r; | |
1752 | } | |
1753 | ||
1da177e4 LT |
1754 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1755 | { | |
8a57dfc6 CS |
1756 | int r = bdi_bits; |
1757 | struct mapped_device *md = congested_data; | |
1758 | struct dm_table *map; | |
1da177e4 | 1759 | |
1eb787ec | 1760 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
7c666411 | 1761 | map = dm_get_live_table(md); |
8a57dfc6 | 1762 | if (map) { |
cec47e3d KU |
1763 | /* |
1764 | * Request-based dm cares about only own queue for | |
1765 | * the query about congestion status of request_queue | |
1766 | */ | |
1767 | if (dm_request_based(md)) | |
1768 | r = md->queue->backing_dev_info.state & | |
1769 | bdi_bits; | |
1770 | else | |
1771 | r = dm_table_any_congested(map, bdi_bits); | |
1772 | ||
8a57dfc6 CS |
1773 | dm_table_put(map); |
1774 | } | |
1775 | } | |
1776 | ||
1da177e4 LT |
1777 | return r; |
1778 | } | |
1779 | ||
1780 | /*----------------------------------------------------------------- | |
1781 | * An IDR is used to keep track of allocated minor numbers. | |
1782 | *---------------------------------------------------------------*/ | |
2b06cfff | 1783 | static void free_minor(int minor) |
1da177e4 | 1784 | { |
f32c10b0 | 1785 | spin_lock(&_minor_lock); |
1da177e4 | 1786 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1787 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1788 | } |
1789 | ||
1790 | /* | |
1791 | * See if the device with a specific minor # is free. | |
1792 | */ | |
cf13ab8e | 1793 | static int specific_minor(int minor) |
1da177e4 | 1794 | { |
c9d76be6 | 1795 | int r; |
1da177e4 LT |
1796 | |
1797 | if (minor >= (1 << MINORBITS)) | |
1798 | return -EINVAL; | |
1799 | ||
c9d76be6 | 1800 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1801 | spin_lock(&_minor_lock); |
1da177e4 | 1802 | |
c9d76be6 | 1803 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1804 | |
f32c10b0 | 1805 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1806 | idr_preload_end(); |
1807 | if (r < 0) | |
1808 | return r == -ENOSPC ? -EBUSY : r; | |
1809 | return 0; | |
1da177e4 LT |
1810 | } |
1811 | ||
cf13ab8e | 1812 | static int next_free_minor(int *minor) |
1da177e4 | 1813 | { |
c9d76be6 | 1814 | int r; |
62f75c2f | 1815 | |
c9d76be6 | 1816 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1817 | spin_lock(&_minor_lock); |
1da177e4 | 1818 | |
c9d76be6 | 1819 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1820 | |
f32c10b0 | 1821 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1822 | idr_preload_end(); |
1823 | if (r < 0) | |
1824 | return r; | |
1825 | *minor = r; | |
1826 | return 0; | |
1da177e4 LT |
1827 | } |
1828 | ||
83d5cde4 | 1829 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1830 | |
53d5914f MP |
1831 | static void dm_wq_work(struct work_struct *work); |
1832 | ||
4a0b4ddf MS |
1833 | static void dm_init_md_queue(struct mapped_device *md) |
1834 | { | |
1835 | /* | |
1836 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1837 | * devices. The type of this dm device has not been decided yet. | |
1838 | * The type is decided at the first table loading time. | |
1839 | * To prevent problematic device stacking, clear the queue flag | |
1840 | * for request stacking support until then. | |
1841 | * | |
1842 | * This queue is new, so no concurrency on the queue_flags. | |
1843 | */ | |
1844 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1845 | ||
1846 | md->queue->queuedata = md; | |
1847 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1848 | md->queue->backing_dev_info.congested_data = md; | |
1849 | blk_queue_make_request(md->queue, dm_request); | |
1850 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1851 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1852 | } | |
1853 | ||
1da177e4 LT |
1854 | /* |
1855 | * Allocate and initialise a blank device with a given minor. | |
1856 | */ | |
2b06cfff | 1857 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1858 | { |
1859 | int r; | |
cf13ab8e | 1860 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1861 | void *old_md; |
1da177e4 LT |
1862 | |
1863 | if (!md) { | |
1864 | DMWARN("unable to allocate device, out of memory."); | |
1865 | return NULL; | |
1866 | } | |
1867 | ||
10da4f79 | 1868 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1869 | goto bad_module_get; |
10da4f79 | 1870 | |
1da177e4 | 1871 | /* get a minor number for the dev */ |
2b06cfff | 1872 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1873 | r = next_free_minor(&minor); |
2b06cfff | 1874 | else |
cf13ab8e | 1875 | r = specific_minor(minor); |
1da177e4 | 1876 | if (r < 0) |
6ed7ade8 | 1877 | goto bad_minor; |
1da177e4 | 1878 | |
a5664dad | 1879 | md->type = DM_TYPE_NONE; |
2ca3310e | 1880 | init_rwsem(&md->io_lock); |
e61290a4 | 1881 | mutex_init(&md->suspend_lock); |
a5664dad | 1882 | mutex_init(&md->type_lock); |
022c2611 | 1883 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1884 | rwlock_init(&md->map_lock); |
1885 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1886 | atomic_set(&md->open_count, 0); |
1da177e4 | 1887 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1888 | atomic_set(&md->uevent_seq, 0); |
1889 | INIT_LIST_HEAD(&md->uevent_list); | |
1890 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1891 | |
4a0b4ddf | 1892 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1893 | if (!md->queue) |
6ed7ade8 | 1894 | goto bad_queue; |
1da177e4 | 1895 | |
4a0b4ddf | 1896 | dm_init_md_queue(md); |
9faf400f | 1897 | |
1da177e4 LT |
1898 | md->disk = alloc_disk(1); |
1899 | if (!md->disk) | |
6ed7ade8 | 1900 | goto bad_disk; |
1da177e4 | 1901 | |
316d315b NK |
1902 | atomic_set(&md->pending[0], 0); |
1903 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1904 | init_waitqueue_head(&md->wait); |
53d5914f | 1905 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1906 | init_waitqueue_head(&md->eventq); |
1907 | ||
1da177e4 LT |
1908 | md->disk->major = _major; |
1909 | md->disk->first_minor = minor; | |
1910 | md->disk->fops = &dm_blk_dops; | |
1911 | md->disk->queue = md->queue; | |
1912 | md->disk->private_data = md; | |
1913 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1914 | add_disk(md->disk); | |
7e51f257 | 1915 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1916 | |
9c4376de TH |
1917 | md->wq = alloc_workqueue("kdmflush", |
1918 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1919 | if (!md->wq) |
1920 | goto bad_thread; | |
1921 | ||
32a926da MP |
1922 | md->bdev = bdget_disk(md->disk, 0); |
1923 | if (!md->bdev) | |
1924 | goto bad_bdev; | |
1925 | ||
6a8736d1 TH |
1926 | bio_init(&md->flush_bio); |
1927 | md->flush_bio.bi_bdev = md->bdev; | |
1928 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1929 | ||
ba61fdd1 | 1930 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1931 | spin_lock(&_minor_lock); |
ba61fdd1 | 1932 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1933 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1934 | |
1935 | BUG_ON(old_md != MINOR_ALLOCED); | |
1936 | ||
1da177e4 LT |
1937 | return md; |
1938 | ||
32a926da MP |
1939 | bad_bdev: |
1940 | destroy_workqueue(md->wq); | |
304f3f6a | 1941 | bad_thread: |
03022c54 | 1942 | del_gendisk(md->disk); |
304f3f6a | 1943 | put_disk(md->disk); |
6ed7ade8 | 1944 | bad_disk: |
1312f40e | 1945 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1946 | bad_queue: |
1da177e4 | 1947 | free_minor(minor); |
6ed7ade8 | 1948 | bad_minor: |
10da4f79 | 1949 | module_put(THIS_MODULE); |
6ed7ade8 | 1950 | bad_module_get: |
1da177e4 LT |
1951 | kfree(md); |
1952 | return NULL; | |
1953 | } | |
1954 | ||
ae9da83f JN |
1955 | static void unlock_fs(struct mapped_device *md); |
1956 | ||
1da177e4 LT |
1957 | static void free_dev(struct mapped_device *md) |
1958 | { | |
f331c029 | 1959 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1960 | |
32a926da MP |
1961 | unlock_fs(md); |
1962 | bdput(md->bdev); | |
304f3f6a | 1963 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1964 | if (md->io_pool) |
1965 | mempool_destroy(md->io_pool); | |
1966 | if (md->bs) | |
1967 | bioset_free(md->bs); | |
9c47008d | 1968 | blk_integrity_unregister(md->disk); |
1da177e4 | 1969 | del_gendisk(md->disk); |
63d94e48 | 1970 | free_minor(minor); |
fba9f90e JM |
1971 | |
1972 | spin_lock(&_minor_lock); | |
1973 | md->disk->private_data = NULL; | |
1974 | spin_unlock(&_minor_lock); | |
1975 | ||
1da177e4 | 1976 | put_disk(md->disk); |
1312f40e | 1977 | blk_cleanup_queue(md->queue); |
10da4f79 | 1978 | module_put(THIS_MODULE); |
1da177e4 LT |
1979 | kfree(md); |
1980 | } | |
1981 | ||
e6ee8c0b KU |
1982 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1983 | { | |
c0820cf5 | 1984 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 1985 | |
5f015204 | 1986 | if (md->io_pool && md->bs) { |
16245bdc JN |
1987 | /* The md already has necessary mempools. */ |
1988 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
1989 | /* | |
1990 | * Reload bioset because front_pad may have changed | |
1991 | * because a different table was loaded. | |
1992 | */ | |
1993 | bioset_free(md->bs); | |
1994 | md->bs = p->bs; | |
1995 | p->bs = NULL; | |
1996 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
1997 | /* |
1998 | * There's no need to reload with request-based dm | |
1999 | * because the size of front_pad doesn't change. | |
2000 | * Note for future: If you are to reload bioset, | |
2001 | * prep-ed requests in the queue may refer | |
2002 | * to bio from the old bioset, so you must walk | |
2003 | * through the queue to unprep. | |
2004 | */ | |
2005 | } | |
e6ee8c0b | 2006 | goto out; |
c0820cf5 | 2007 | } |
e6ee8c0b | 2008 | |
5f015204 | 2009 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2010 | |
2011 | md->io_pool = p->io_pool; | |
2012 | p->io_pool = NULL; | |
e6ee8c0b KU |
2013 | md->bs = p->bs; |
2014 | p->bs = NULL; | |
2015 | ||
2016 | out: | |
2017 | /* mempool bind completed, now no need any mempools in the table */ | |
2018 | dm_table_free_md_mempools(t); | |
2019 | } | |
2020 | ||
1da177e4 LT |
2021 | /* |
2022 | * Bind a table to the device. | |
2023 | */ | |
2024 | static void event_callback(void *context) | |
2025 | { | |
7a8c3d3b MA |
2026 | unsigned long flags; |
2027 | LIST_HEAD(uevents); | |
1da177e4 LT |
2028 | struct mapped_device *md = (struct mapped_device *) context; |
2029 | ||
7a8c3d3b MA |
2030 | spin_lock_irqsave(&md->uevent_lock, flags); |
2031 | list_splice_init(&md->uevent_list, &uevents); | |
2032 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2033 | ||
ed9e1982 | 2034 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2035 | |
1da177e4 LT |
2036 | atomic_inc(&md->event_nr); |
2037 | wake_up(&md->eventq); | |
2038 | } | |
2039 | ||
c217649b MS |
2040 | /* |
2041 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2042 | */ | |
4e90188b | 2043 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2044 | { |
4e90188b | 2045 | set_capacity(md->disk, size); |
1da177e4 | 2046 | |
db8fef4f | 2047 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2048 | } |
2049 | ||
d5b9dd04 MP |
2050 | /* |
2051 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2052 | * | |
2053 | * If this function returns 0, then the device is either a non-dm | |
2054 | * device without a merge_bvec_fn, or it is a dm device that is | |
2055 | * able to split any bios it receives that are too big. | |
2056 | */ | |
2057 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2058 | { | |
2059 | struct mapped_device *dev_md; | |
2060 | ||
2061 | if (!q->merge_bvec_fn) | |
2062 | return 0; | |
2063 | ||
2064 | if (q->make_request_fn == dm_request) { | |
2065 | dev_md = q->queuedata; | |
2066 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2067 | return 0; | |
2068 | } | |
2069 | ||
2070 | return 1; | |
2071 | } | |
2072 | ||
2073 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2074 | struct dm_dev *dev, sector_t start, | |
2075 | sector_t len, void *data) | |
2076 | { | |
2077 | struct block_device *bdev = dev->bdev; | |
2078 | struct request_queue *q = bdev_get_queue(bdev); | |
2079 | ||
2080 | return dm_queue_merge_is_compulsory(q); | |
2081 | } | |
2082 | ||
2083 | /* | |
2084 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2085 | * on the properties of the underlying devices. | |
2086 | */ | |
2087 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2088 | { | |
2089 | unsigned i = 0; | |
2090 | struct dm_target *ti; | |
2091 | ||
2092 | while (i < dm_table_get_num_targets(table)) { | |
2093 | ti = dm_table_get_target(table, i++); | |
2094 | ||
2095 | if (ti->type->iterate_devices && | |
2096 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2097 | return 0; | |
2098 | } | |
2099 | ||
2100 | return 1; | |
2101 | } | |
2102 | ||
042d2a9b AK |
2103 | /* |
2104 | * Returns old map, which caller must destroy. | |
2105 | */ | |
2106 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2107 | struct queue_limits *limits) | |
1da177e4 | 2108 | { |
042d2a9b | 2109 | struct dm_table *old_map; |
165125e1 | 2110 | struct request_queue *q = md->queue; |
1da177e4 | 2111 | sector_t size; |
523d9297 | 2112 | unsigned long flags; |
d5b9dd04 | 2113 | int merge_is_optional; |
1da177e4 LT |
2114 | |
2115 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2116 | |
2117 | /* | |
2118 | * Wipe any geometry if the size of the table changed. | |
2119 | */ | |
2120 | if (size != get_capacity(md->disk)) | |
2121 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2122 | ||
32a926da | 2123 | __set_size(md, size); |
d5816876 | 2124 | |
2ca3310e AK |
2125 | dm_table_event_callback(t, event_callback, md); |
2126 | ||
e6ee8c0b KU |
2127 | /* |
2128 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2129 | * for request-based during suspension. So stop it to prevent | |
2130 | * I/O mapping before resume. | |
2131 | * This must be done before setting the queue restrictions, | |
2132 | * because request-based dm may be run just after the setting. | |
2133 | */ | |
2134 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2135 | stop_queue(q); | |
2136 | ||
2137 | __bind_mempools(md, t); | |
2138 | ||
d5b9dd04 MP |
2139 | merge_is_optional = dm_table_merge_is_optional(t); |
2140 | ||
523d9297 | 2141 | write_lock_irqsave(&md->map_lock, flags); |
042d2a9b | 2142 | old_map = md->map; |
1da177e4 | 2143 | md->map = t; |
36a0456f AK |
2144 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2145 | ||
754c5fc7 | 2146 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2147 | if (merge_is_optional) |
2148 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2149 | else | |
2150 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
523d9297 | 2151 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 2152 | |
042d2a9b | 2153 | return old_map; |
1da177e4 LT |
2154 | } |
2155 | ||
a7940155 AK |
2156 | /* |
2157 | * Returns unbound table for the caller to free. | |
2158 | */ | |
2159 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2160 | { |
2161 | struct dm_table *map = md->map; | |
523d9297 | 2162 | unsigned long flags; |
1da177e4 LT |
2163 | |
2164 | if (!map) | |
a7940155 | 2165 | return NULL; |
1da177e4 LT |
2166 | |
2167 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 2168 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 2169 | md->map = NULL; |
523d9297 | 2170 | write_unlock_irqrestore(&md->map_lock, flags); |
a7940155 AK |
2171 | |
2172 | return map; | |
1da177e4 LT |
2173 | } |
2174 | ||
2175 | /* | |
2176 | * Constructor for a new device. | |
2177 | */ | |
2b06cfff | 2178 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2179 | { |
2180 | struct mapped_device *md; | |
2181 | ||
2b06cfff | 2182 | md = alloc_dev(minor); |
1da177e4 LT |
2183 | if (!md) |
2184 | return -ENXIO; | |
2185 | ||
784aae73 MB |
2186 | dm_sysfs_init(md); |
2187 | ||
1da177e4 LT |
2188 | *result = md; |
2189 | return 0; | |
2190 | } | |
2191 | ||
a5664dad MS |
2192 | /* |
2193 | * Functions to manage md->type. | |
2194 | * All are required to hold md->type_lock. | |
2195 | */ | |
2196 | void dm_lock_md_type(struct mapped_device *md) | |
2197 | { | |
2198 | mutex_lock(&md->type_lock); | |
2199 | } | |
2200 | ||
2201 | void dm_unlock_md_type(struct mapped_device *md) | |
2202 | { | |
2203 | mutex_unlock(&md->type_lock); | |
2204 | } | |
2205 | ||
2206 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2207 | { | |
2208 | md->type = type; | |
2209 | } | |
2210 | ||
2211 | unsigned dm_get_md_type(struct mapped_device *md) | |
2212 | { | |
2213 | return md->type; | |
2214 | } | |
2215 | ||
36a0456f AK |
2216 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2217 | { | |
2218 | return md->immutable_target_type; | |
2219 | } | |
2220 | ||
4a0b4ddf MS |
2221 | /* |
2222 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2223 | */ | |
2224 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2225 | { | |
2226 | struct request_queue *q = NULL; | |
2227 | ||
2228 | if (md->queue->elevator) | |
2229 | return 1; | |
2230 | ||
2231 | /* Fully initialize the queue */ | |
2232 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2233 | if (!q) | |
2234 | return 0; | |
2235 | ||
2236 | md->queue = q; | |
4a0b4ddf MS |
2237 | dm_init_md_queue(md); |
2238 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2239 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2240 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2241 | |
2242 | elv_register_queue(md->queue); | |
2243 | ||
2244 | return 1; | |
2245 | } | |
2246 | ||
2247 | /* | |
2248 | * Setup the DM device's queue based on md's type | |
2249 | */ | |
2250 | int dm_setup_md_queue(struct mapped_device *md) | |
2251 | { | |
2252 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2253 | !dm_init_request_based_queue(md)) { | |
2254 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2255 | return -EINVAL; | |
2256 | } | |
2257 | ||
2258 | return 0; | |
2259 | } | |
2260 | ||
637842cf | 2261 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2262 | { |
2263 | struct mapped_device *md; | |
1da177e4 LT |
2264 | unsigned minor = MINOR(dev); |
2265 | ||
2266 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2267 | return NULL; | |
2268 | ||
f32c10b0 | 2269 | spin_lock(&_minor_lock); |
1da177e4 LT |
2270 | |
2271 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2272 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2273 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2274 | dm_deleting_md(md) || |
17b2f66f | 2275 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2276 | md = NULL; |
fba9f90e JM |
2277 | goto out; |
2278 | } | |
1da177e4 | 2279 | |
fba9f90e | 2280 | out: |
f32c10b0 | 2281 | spin_unlock(&_minor_lock); |
1da177e4 | 2282 | |
637842cf DT |
2283 | return md; |
2284 | } | |
2285 | ||
d229a958 DT |
2286 | struct mapped_device *dm_get_md(dev_t dev) |
2287 | { | |
2288 | struct mapped_device *md = dm_find_md(dev); | |
2289 | ||
2290 | if (md) | |
2291 | dm_get(md); | |
2292 | ||
2293 | return md; | |
2294 | } | |
3cf2e4ba | 2295 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2296 | |
9ade92a9 | 2297 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2298 | { |
9ade92a9 | 2299 | return md->interface_ptr; |
1da177e4 LT |
2300 | } |
2301 | ||
2302 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2303 | { | |
2304 | md->interface_ptr = ptr; | |
2305 | } | |
2306 | ||
2307 | void dm_get(struct mapped_device *md) | |
2308 | { | |
2309 | atomic_inc(&md->holders); | |
3f77316d | 2310 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2311 | } |
2312 | ||
72d94861 AK |
2313 | const char *dm_device_name(struct mapped_device *md) |
2314 | { | |
2315 | return md->name; | |
2316 | } | |
2317 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2318 | ||
3f77316d | 2319 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2320 | { |
1134e5ae | 2321 | struct dm_table *map; |
1da177e4 | 2322 | |
3f77316d | 2323 | might_sleep(); |
fba9f90e | 2324 | |
3f77316d KU |
2325 | spin_lock(&_minor_lock); |
2326 | map = dm_get_live_table(md); | |
2327 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); | |
2328 | set_bit(DMF_FREEING, &md->flags); | |
2329 | spin_unlock(&_minor_lock); | |
2330 | ||
2331 | if (!dm_suspended_md(md)) { | |
2332 | dm_table_presuspend_targets(map); | |
2333 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2334 | } |
3f77316d KU |
2335 | |
2336 | /* | |
2337 | * Rare, but there may be I/O requests still going to complete, | |
2338 | * for example. Wait for all references to disappear. | |
2339 | * No one should increment the reference count of the mapped_device, | |
2340 | * after the mapped_device state becomes DMF_FREEING. | |
2341 | */ | |
2342 | if (wait) | |
2343 | while (atomic_read(&md->holders)) | |
2344 | msleep(1); | |
2345 | else if (atomic_read(&md->holders)) | |
2346 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2347 | dm_device_name(md), atomic_read(&md->holders)); | |
2348 | ||
2349 | dm_sysfs_exit(md); | |
2350 | dm_table_put(map); | |
2351 | dm_table_destroy(__unbind(md)); | |
2352 | free_dev(md); | |
2353 | } | |
2354 | ||
2355 | void dm_destroy(struct mapped_device *md) | |
2356 | { | |
2357 | __dm_destroy(md, true); | |
2358 | } | |
2359 | ||
2360 | void dm_destroy_immediate(struct mapped_device *md) | |
2361 | { | |
2362 | __dm_destroy(md, false); | |
2363 | } | |
2364 | ||
2365 | void dm_put(struct mapped_device *md) | |
2366 | { | |
2367 | atomic_dec(&md->holders); | |
1da177e4 | 2368 | } |
79eb885c | 2369 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2370 | |
401600df | 2371 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2372 | { |
2373 | int r = 0; | |
b44ebeb0 MP |
2374 | DECLARE_WAITQUEUE(wait, current); |
2375 | ||
b44ebeb0 | 2376 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2377 | |
2378 | while (1) { | |
401600df | 2379 | set_current_state(interruptible); |
46125c1c | 2380 | |
b4324fee | 2381 | if (!md_in_flight(md)) |
46125c1c MB |
2382 | break; |
2383 | ||
401600df MP |
2384 | if (interruptible == TASK_INTERRUPTIBLE && |
2385 | signal_pending(current)) { | |
46125c1c MB |
2386 | r = -EINTR; |
2387 | break; | |
2388 | } | |
2389 | ||
2390 | io_schedule(); | |
2391 | } | |
2392 | set_current_state(TASK_RUNNING); | |
2393 | ||
b44ebeb0 MP |
2394 | remove_wait_queue(&md->wait, &wait); |
2395 | ||
46125c1c MB |
2396 | return r; |
2397 | } | |
2398 | ||
1da177e4 LT |
2399 | /* |
2400 | * Process the deferred bios | |
2401 | */ | |
ef208587 | 2402 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2403 | { |
ef208587 MP |
2404 | struct mapped_device *md = container_of(work, struct mapped_device, |
2405 | work); | |
6d6f10df | 2406 | struct bio *c; |
1da177e4 | 2407 | |
6a8736d1 | 2408 | down_read(&md->io_lock); |
ef208587 | 2409 | |
3b00b203 | 2410 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2411 | spin_lock_irq(&md->deferred_lock); |
2412 | c = bio_list_pop(&md->deferred); | |
2413 | spin_unlock_irq(&md->deferred_lock); | |
2414 | ||
6a8736d1 | 2415 | if (!c) |
df12ee99 | 2416 | break; |
022c2611 | 2417 | |
6a8736d1 | 2418 | up_read(&md->io_lock); |
3b00b203 | 2419 | |
e6ee8c0b KU |
2420 | if (dm_request_based(md)) |
2421 | generic_make_request(c); | |
6a8736d1 TH |
2422 | else |
2423 | __split_and_process_bio(md, c); | |
3b00b203 | 2424 | |
6a8736d1 | 2425 | down_read(&md->io_lock); |
022c2611 | 2426 | } |
73d410c0 | 2427 | |
6a8736d1 | 2428 | up_read(&md->io_lock); |
1da177e4 LT |
2429 | } |
2430 | ||
9a1fb464 | 2431 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2432 | { |
3b00b203 MP |
2433 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2434 | smp_mb__after_clear_bit(); | |
53d5914f | 2435 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2436 | } |
2437 | ||
1da177e4 | 2438 | /* |
042d2a9b | 2439 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2440 | */ |
042d2a9b | 2441 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2442 | { |
87eb5b21 | 2443 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2444 | struct queue_limits limits; |
042d2a9b | 2445 | int r; |
1da177e4 | 2446 | |
e61290a4 | 2447 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2448 | |
2449 | /* device must be suspended */ | |
4f186f8b | 2450 | if (!dm_suspended_md(md)) |
93c534ae | 2451 | goto out; |
1da177e4 | 2452 | |
3ae70656 MS |
2453 | /* |
2454 | * If the new table has no data devices, retain the existing limits. | |
2455 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2456 | * then new I/O is queued based on these limits, and then some paths | |
2457 | * reappear. | |
2458 | */ | |
2459 | if (dm_table_has_no_data_devices(table)) { | |
2460 | live_map = dm_get_live_table(md); | |
2461 | if (live_map) | |
2462 | limits = md->queue->limits; | |
2463 | dm_table_put(live_map); | |
2464 | } | |
2465 | ||
87eb5b21 MC |
2466 | if (!live_map) { |
2467 | r = dm_calculate_queue_limits(table, &limits); | |
2468 | if (r) { | |
2469 | map = ERR_PTR(r); | |
2470 | goto out; | |
2471 | } | |
042d2a9b | 2472 | } |
754c5fc7 | 2473 | |
042d2a9b | 2474 | map = __bind(md, table, &limits); |
1da177e4 | 2475 | |
93c534ae | 2476 | out: |
e61290a4 | 2477 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2478 | return map; |
1da177e4 LT |
2479 | } |
2480 | ||
2481 | /* | |
2482 | * Functions to lock and unlock any filesystem running on the | |
2483 | * device. | |
2484 | */ | |
2ca3310e | 2485 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2486 | { |
e39e2e95 | 2487 | int r; |
1da177e4 LT |
2488 | |
2489 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2490 | |
db8fef4f | 2491 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2492 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2493 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2494 | md->frozen_sb = NULL; |
2495 | return r; | |
dfbe03f6 AK |
2496 | } |
2497 | ||
aa8d7c2f AK |
2498 | set_bit(DMF_FROZEN, &md->flags); |
2499 | ||
1da177e4 LT |
2500 | return 0; |
2501 | } | |
2502 | ||
2ca3310e | 2503 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2504 | { |
aa8d7c2f AK |
2505 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2506 | return; | |
2507 | ||
db8fef4f | 2508 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2509 | md->frozen_sb = NULL; |
aa8d7c2f | 2510 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2511 | } |
2512 | ||
2513 | /* | |
2514 | * We need to be able to change a mapping table under a mounted | |
2515 | * filesystem. For example we might want to move some data in | |
2516 | * the background. Before the table can be swapped with | |
2517 | * dm_bind_table, dm_suspend must be called to flush any in | |
2518 | * flight bios and ensure that any further io gets deferred. | |
2519 | */ | |
cec47e3d KU |
2520 | /* |
2521 | * Suspend mechanism in request-based dm. | |
2522 | * | |
9f518b27 KU |
2523 | * 1. Flush all I/Os by lock_fs() if needed. |
2524 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2525 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2526 | * |
9f518b27 | 2527 | * To abort suspend, start the request_queue. |
cec47e3d | 2528 | */ |
a3d77d35 | 2529 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2530 | { |
2ca3310e | 2531 | struct dm_table *map = NULL; |
46125c1c | 2532 | int r = 0; |
a3d77d35 | 2533 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2534 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2535 | |
e61290a4 | 2536 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2537 | |
4f186f8b | 2538 | if (dm_suspended_md(md)) { |
73d410c0 | 2539 | r = -EINVAL; |
d287483d | 2540 | goto out_unlock; |
73d410c0 | 2541 | } |
1da177e4 | 2542 | |
7c666411 | 2543 | map = dm_get_live_table(md); |
1da177e4 | 2544 | |
2e93ccc1 KU |
2545 | /* |
2546 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2547 | * This flag is cleared before dm_suspend returns. | |
2548 | */ | |
2549 | if (noflush) | |
2550 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2551 | ||
cf222b37 AK |
2552 | /* This does not get reverted if there's an error later. */ |
2553 | dm_table_presuspend_targets(map); | |
2554 | ||
32a926da | 2555 | /* |
9f518b27 KU |
2556 | * Flush I/O to the device. |
2557 | * Any I/O submitted after lock_fs() may not be flushed. | |
2558 | * noflush takes precedence over do_lockfs. | |
2559 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2560 | */ |
2561 | if (!noflush && do_lockfs) { | |
2562 | r = lock_fs(md); | |
2563 | if (r) | |
f431d966 | 2564 | goto out; |
aa8d7c2f | 2565 | } |
1da177e4 LT |
2566 | |
2567 | /* | |
3b00b203 MP |
2568 | * Here we must make sure that no processes are submitting requests |
2569 | * to target drivers i.e. no one may be executing | |
2570 | * __split_and_process_bio. This is called from dm_request and | |
2571 | * dm_wq_work. | |
2572 | * | |
2573 | * To get all processes out of __split_and_process_bio in dm_request, | |
2574 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2575 | * __split_and_process_bio from dm_request and quiesce the thread |
2576 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2577 | * flush_workqueue(md->wq). | |
1da177e4 | 2578 | */ |
2ca3310e | 2579 | down_write(&md->io_lock); |
1eb787ec | 2580 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2ca3310e | 2581 | up_write(&md->io_lock); |
1da177e4 | 2582 | |
d0bcb878 | 2583 | /* |
29e4013d TH |
2584 | * Stop md->queue before flushing md->wq in case request-based |
2585 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2586 | */ |
cec47e3d | 2587 | if (dm_request_based(md)) |
9f518b27 | 2588 | stop_queue(md->queue); |
cec47e3d | 2589 | |
d0bcb878 KU |
2590 | flush_workqueue(md->wq); |
2591 | ||
1da177e4 | 2592 | /* |
3b00b203 MP |
2593 | * At this point no more requests are entering target request routines. |
2594 | * We call dm_wait_for_completion to wait for all existing requests | |
2595 | * to finish. | |
1da177e4 | 2596 | */ |
401600df | 2597 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2598 | |
2ca3310e | 2599 | down_write(&md->io_lock); |
6d6f10df | 2600 | if (noflush) |
022c2611 | 2601 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2602 | up_write(&md->io_lock); |
2e93ccc1 | 2603 | |
1da177e4 | 2604 | /* were we interrupted ? */ |
46125c1c | 2605 | if (r < 0) { |
9a1fb464 | 2606 | dm_queue_flush(md); |
73d410c0 | 2607 | |
cec47e3d | 2608 | if (dm_request_based(md)) |
9f518b27 | 2609 | start_queue(md->queue); |
cec47e3d | 2610 | |
2ca3310e | 2611 | unlock_fs(md); |
2e93ccc1 | 2612 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2613 | } |
1da177e4 | 2614 | |
3b00b203 MP |
2615 | /* |
2616 | * If dm_wait_for_completion returned 0, the device is completely | |
2617 | * quiescent now. There is no request-processing activity. All new | |
2618 | * requests are being added to md->deferred list. | |
2619 | */ | |
2620 | ||
2ca3310e | 2621 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2622 | |
4d4471cb KU |
2623 | dm_table_postsuspend_targets(map); |
2624 | ||
2ca3310e AK |
2625 | out: |
2626 | dm_table_put(map); | |
d287483d AK |
2627 | |
2628 | out_unlock: | |
e61290a4 | 2629 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2630 | return r; |
1da177e4 LT |
2631 | } |
2632 | ||
2633 | int dm_resume(struct mapped_device *md) | |
2634 | { | |
cf222b37 | 2635 | int r = -EINVAL; |
cf222b37 | 2636 | struct dm_table *map = NULL; |
1da177e4 | 2637 | |
e61290a4 | 2638 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2639 | if (!dm_suspended_md(md)) |
cf222b37 | 2640 | goto out; |
cf222b37 | 2641 | |
7c666411 | 2642 | map = dm_get_live_table(md); |
2ca3310e | 2643 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2644 | goto out; |
1da177e4 | 2645 | |
8757b776 MB |
2646 | r = dm_table_resume_targets(map); |
2647 | if (r) | |
2648 | goto out; | |
2ca3310e | 2649 | |
9a1fb464 | 2650 | dm_queue_flush(md); |
2ca3310e | 2651 | |
cec47e3d KU |
2652 | /* |
2653 | * Flushing deferred I/Os must be done after targets are resumed | |
2654 | * so that mapping of targets can work correctly. | |
2655 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2656 | */ | |
2657 | if (dm_request_based(md)) | |
2658 | start_queue(md->queue); | |
2659 | ||
2ca3310e AK |
2660 | unlock_fs(md); |
2661 | ||
2662 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2663 | ||
cf222b37 AK |
2664 | r = 0; |
2665 | out: | |
2666 | dm_table_put(map); | |
e61290a4 | 2667 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2668 | |
cf222b37 | 2669 | return r; |
1da177e4 LT |
2670 | } |
2671 | ||
2672 | /*----------------------------------------------------------------- | |
2673 | * Event notification. | |
2674 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2675 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2676 | unsigned cookie) |
69267a30 | 2677 | { |
60935eb2 MB |
2678 | char udev_cookie[DM_COOKIE_LENGTH]; |
2679 | char *envp[] = { udev_cookie, NULL }; | |
2680 | ||
2681 | if (!cookie) | |
3abf85b5 | 2682 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2683 | else { |
2684 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2685 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2686 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2687 | action, envp); | |
60935eb2 | 2688 | } |
69267a30 AK |
2689 | } |
2690 | ||
7a8c3d3b MA |
2691 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2692 | { | |
2693 | return atomic_add_return(1, &md->uevent_seq); | |
2694 | } | |
2695 | ||
1da177e4 LT |
2696 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2697 | { | |
2698 | return atomic_read(&md->event_nr); | |
2699 | } | |
2700 | ||
2701 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2702 | { | |
2703 | return wait_event_interruptible(md->eventq, | |
2704 | (event_nr != atomic_read(&md->event_nr))); | |
2705 | } | |
2706 | ||
7a8c3d3b MA |
2707 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2708 | { | |
2709 | unsigned long flags; | |
2710 | ||
2711 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2712 | list_add(elist, &md->uevent_list); | |
2713 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2714 | } | |
2715 | ||
1da177e4 LT |
2716 | /* |
2717 | * The gendisk is only valid as long as you have a reference | |
2718 | * count on 'md'. | |
2719 | */ | |
2720 | struct gendisk *dm_disk(struct mapped_device *md) | |
2721 | { | |
2722 | return md->disk; | |
2723 | } | |
2724 | ||
784aae73 MB |
2725 | struct kobject *dm_kobject(struct mapped_device *md) |
2726 | { | |
2727 | return &md->kobj; | |
2728 | } | |
2729 | ||
2730 | /* | |
2731 | * struct mapped_device should not be exported outside of dm.c | |
2732 | * so use this check to verify that kobj is part of md structure | |
2733 | */ | |
2734 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2735 | { | |
2736 | struct mapped_device *md; | |
2737 | ||
2738 | md = container_of(kobj, struct mapped_device, kobj); | |
2739 | if (&md->kobj != kobj) | |
2740 | return NULL; | |
2741 | ||
4d89b7b4 | 2742 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2743 | dm_deleting_md(md)) |
4d89b7b4 MB |
2744 | return NULL; |
2745 | ||
784aae73 MB |
2746 | dm_get(md); |
2747 | return md; | |
2748 | } | |
2749 | ||
4f186f8b | 2750 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2751 | { |
2752 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2753 | } | |
2754 | ||
64dbce58 KU |
2755 | int dm_suspended(struct dm_target *ti) |
2756 | { | |
ecdb2e25 | 2757 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2758 | } |
2759 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2760 | ||
2e93ccc1 KU |
2761 | int dm_noflush_suspending(struct dm_target *ti) |
2762 | { | |
ecdb2e25 | 2763 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2764 | } |
2765 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2766 | ||
c0820cf5 | 2767 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 2768 | { |
5f015204 JN |
2769 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
2770 | struct kmem_cache *cachep; | |
2771 | unsigned int pool_size; | |
2772 | unsigned int front_pad; | |
e6ee8c0b KU |
2773 | |
2774 | if (!pools) | |
2775 | return NULL; | |
2776 | ||
23e5083b | 2777 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 JN |
2778 | cachep = _io_cache; |
2779 | pool_size = 16; | |
2780 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
2781 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
2782 | cachep = _rq_tio_cache; | |
2783 | pool_size = MIN_IOS; | |
2784 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
2785 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
2786 | WARN_ON(per_bio_data_size != 0); | |
2787 | } else | |
2788 | goto out; | |
e6ee8c0b | 2789 | |
5f015204 JN |
2790 | pools->io_pool = mempool_create_slab_pool(MIN_IOS, cachep); |
2791 | if (!pools->io_pool) | |
2792 | goto out; | |
e6ee8c0b | 2793 | |
5f015204 | 2794 | pools->bs = bioset_create(pool_size, front_pad); |
e6ee8c0b | 2795 | if (!pools->bs) |
5f015204 | 2796 | goto out; |
e6ee8c0b | 2797 | |
a91a2785 | 2798 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2799 | goto out; |
a91a2785 | 2800 | |
e6ee8c0b KU |
2801 | return pools; |
2802 | ||
5f015204 JN |
2803 | out: |
2804 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
2805 | |
2806 | return NULL; | |
2807 | } | |
2808 | ||
2809 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2810 | { | |
2811 | if (!pools) | |
2812 | return; | |
2813 | ||
2814 | if (pools->io_pool) | |
2815 | mempool_destroy(pools->io_pool); | |
2816 | ||
e6ee8c0b KU |
2817 | if (pools->bs) |
2818 | bioset_free(pools->bs); | |
2819 | ||
2820 | kfree(pools); | |
2821 | } | |
2822 | ||
83d5cde4 | 2823 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2824 | .open = dm_blk_open, |
2825 | .release = dm_blk_close, | |
aa129a22 | 2826 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2827 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2828 | .owner = THIS_MODULE |
2829 | }; | |
2830 | ||
2831 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2832 | ||
2833 | /* | |
2834 | * module hooks | |
2835 | */ | |
2836 | module_init(dm_init); | |
2837 | module_exit(dm_exit); | |
2838 | ||
2839 | module_param(major, uint, 0); | |
2840 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2841 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2842 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2843 | MODULE_LICENSE("GPL"); |