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