]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
2c140a24 MP |
52 | |
53 | static void do_deferred_remove(struct work_struct *w); | |
54 | ||
55 | static DECLARE_WORK(deferred_remove_work, do_deferred_remove); | |
56 | ||
acfe0ad7 MP |
57 | static struct workqueue_struct *deferred_remove_workqueue; |
58 | ||
1da177e4 | 59 | /* |
8fbf26ad | 60 | * For bio-based dm. |
1da177e4 LT |
61 | * One of these is allocated per bio. |
62 | */ | |
63 | struct dm_io { | |
64 | struct mapped_device *md; | |
65 | int error; | |
1da177e4 | 66 | atomic_t io_count; |
6ae2fa67 | 67 | struct bio *bio; |
3eaf840e | 68 | unsigned long start_time; |
f88fb981 | 69 | spinlock_t endio_lock; |
fd2ed4d2 | 70 | struct dm_stats_aux stats_aux; |
1da177e4 LT |
71 | }; |
72 | ||
8fbf26ad KU |
73 | /* |
74 | * For request-based dm. | |
75 | * One of these is allocated per request. | |
76 | */ | |
77 | struct dm_rq_target_io { | |
78 | struct mapped_device *md; | |
79 | struct dm_target *ti; | |
80 | struct request *orig, clone; | |
81 | int error; | |
82 | union map_info info; | |
83 | }; | |
84 | ||
85 | /* | |
94818742 KO |
86 | * For request-based dm - the bio clones we allocate are embedded in these |
87 | * structs. | |
88 | * | |
89 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
90 | * the bioset is created - this means the bio has to come at the end of the | |
91 | * struct. | |
8fbf26ad KU |
92 | */ |
93 | struct dm_rq_clone_bio_info { | |
94 | struct bio *orig; | |
cec47e3d | 95 | struct dm_rq_target_io *tio; |
94818742 | 96 | struct bio clone; |
8fbf26ad KU |
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 |
2c140a24 | 119 | #define DMF_DEFERRED_REMOVE 7 |
1da177e4 | 120 | |
83d5e5b0 MP |
121 | /* |
122 | * A dummy definition to make RCU happy. | |
123 | * struct dm_table should never be dereferenced in this file. | |
124 | */ | |
125 | struct dm_table { | |
126 | int undefined__; | |
127 | }; | |
128 | ||
304f3f6a MB |
129 | /* |
130 | * Work processed by per-device workqueue. | |
131 | */ | |
1da177e4 | 132 | struct mapped_device { |
83d5e5b0 | 133 | struct srcu_struct io_barrier; |
e61290a4 | 134 | struct mutex suspend_lock; |
1da177e4 | 135 | atomic_t holders; |
5c6bd75d | 136 | atomic_t open_count; |
1da177e4 | 137 | |
2a7faeb1 MP |
138 | /* |
139 | * The current mapping. | |
140 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
141 | * dereference. | |
142 | */ | |
143 | struct dm_table *map; | |
144 | ||
1da177e4 LT |
145 | unsigned long flags; |
146 | ||
165125e1 | 147 | struct request_queue *queue; |
a5664dad | 148 | unsigned type; |
4a0b4ddf | 149 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
150 | struct mutex type_lock; |
151 | ||
36a0456f AK |
152 | struct target_type *immutable_target_type; |
153 | ||
1da177e4 | 154 | struct gendisk *disk; |
7e51f257 | 155 | char name[16]; |
1da177e4 LT |
156 | |
157 | void *interface_ptr; | |
158 | ||
159 | /* | |
160 | * A list of ios that arrived while we were suspended. | |
161 | */ | |
316d315b | 162 | atomic_t pending[2]; |
1da177e4 | 163 | wait_queue_head_t wait; |
53d5914f | 164 | struct work_struct work; |
74859364 | 165 | struct bio_list deferred; |
022c2611 | 166 | spinlock_t deferred_lock; |
1da177e4 | 167 | |
af7e466a | 168 | /* |
29e4013d | 169 | * Processing queue (flush) |
304f3f6a MB |
170 | */ |
171 | struct workqueue_struct *wq; | |
172 | ||
1da177e4 LT |
173 | /* |
174 | * io objects are allocated from here. | |
175 | */ | |
176 | mempool_t *io_pool; | |
1da177e4 | 177 | |
9faf400f SB |
178 | struct bio_set *bs; |
179 | ||
1da177e4 LT |
180 | /* |
181 | * Event handling. | |
182 | */ | |
183 | atomic_t event_nr; | |
184 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
185 | atomic_t uevent_seq; |
186 | struct list_head uevent_list; | |
187 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
188 | |
189 | /* | |
190 | * freeze/thaw support require holding onto a super block | |
191 | */ | |
192 | struct super_block *frozen_sb; | |
db8fef4f | 193 | struct block_device *bdev; |
3ac51e74 DW |
194 | |
195 | /* forced geometry settings */ | |
196 | struct hd_geometry geometry; | |
784aae73 | 197 | |
2995fa78 MP |
198 | /* kobject and completion */ |
199 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 200 | |
d87f4c14 TH |
201 | /* zero-length flush that will be cloned and submitted to targets */ |
202 | struct bio flush_bio; | |
fd2ed4d2 MP |
203 | |
204 | struct dm_stats stats; | |
1da177e4 LT |
205 | }; |
206 | ||
e6ee8c0b KU |
207 | /* |
208 | * For mempools pre-allocation at the table loading time. | |
209 | */ | |
210 | struct dm_md_mempools { | |
211 | mempool_t *io_pool; | |
e6ee8c0b KU |
212 | struct bio_set *bs; |
213 | }; | |
214 | ||
6cfa5857 MS |
215 | #define RESERVED_BIO_BASED_IOS 16 |
216 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 217 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 218 | static struct kmem_cache *_io_cache; |
8fbf26ad | 219 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 220 | |
e8603136 MS |
221 | /* |
222 | * Bio-based DM's mempools' reserved IOs set by the user. | |
223 | */ | |
224 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
225 | ||
f4790826 MS |
226 | /* |
227 | * Request-based DM's mempools' reserved IOs set by the user. | |
228 | */ | |
229 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
230 | ||
231 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
232 | unsigned def, unsigned max) | |
233 | { | |
234 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
235 | unsigned modified_ios = 0; | |
236 | ||
237 | if (!ios) | |
238 | modified_ios = def; | |
239 | else if (ios > max) | |
240 | modified_ios = max; | |
241 | ||
242 | if (modified_ios) { | |
243 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
244 | ios = modified_ios; | |
245 | } | |
246 | ||
247 | return ios; | |
248 | } | |
249 | ||
e8603136 MS |
250 | unsigned dm_get_reserved_bio_based_ios(void) |
251 | { | |
252 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
253 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
254 | } | |
255 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
256 | ||
f4790826 MS |
257 | unsigned dm_get_reserved_rq_based_ios(void) |
258 | { | |
259 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
260 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
261 | } | |
262 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
263 | ||
1da177e4 LT |
264 | static int __init local_init(void) |
265 | { | |
51157b4a | 266 | int r = -ENOMEM; |
1da177e4 | 267 | |
1da177e4 | 268 | /* allocate a slab for the dm_ios */ |
028867ac | 269 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 270 | if (!_io_cache) |
51157b4a | 271 | return r; |
1da177e4 | 272 | |
8fbf26ad KU |
273 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
274 | if (!_rq_tio_cache) | |
dba14160 | 275 | goto out_free_io_cache; |
8fbf26ad | 276 | |
51e5b2bd | 277 | r = dm_uevent_init(); |
51157b4a | 278 | if (r) |
23e5083b | 279 | goto out_free_rq_tio_cache; |
51e5b2bd | 280 | |
acfe0ad7 MP |
281 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
282 | if (!deferred_remove_workqueue) { | |
283 | r = -ENOMEM; | |
284 | goto out_uevent_exit; | |
285 | } | |
286 | ||
1da177e4 LT |
287 | _major = major; |
288 | r = register_blkdev(_major, _name); | |
51157b4a | 289 | if (r < 0) |
acfe0ad7 | 290 | goto out_free_workqueue; |
1da177e4 LT |
291 | |
292 | if (!_major) | |
293 | _major = r; | |
294 | ||
295 | return 0; | |
51157b4a | 296 | |
acfe0ad7 MP |
297 | out_free_workqueue: |
298 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
299 | out_uevent_exit: |
300 | dm_uevent_exit(); | |
8fbf26ad KU |
301 | out_free_rq_tio_cache: |
302 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
303 | out_free_io_cache: |
304 | kmem_cache_destroy(_io_cache); | |
305 | ||
306 | return r; | |
1da177e4 LT |
307 | } |
308 | ||
309 | static void local_exit(void) | |
310 | { | |
2c140a24 | 311 | flush_scheduled_work(); |
acfe0ad7 | 312 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 313 | |
8fbf26ad | 314 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 315 | kmem_cache_destroy(_io_cache); |
00d59405 | 316 | unregister_blkdev(_major, _name); |
51e5b2bd | 317 | dm_uevent_exit(); |
1da177e4 LT |
318 | |
319 | _major = 0; | |
320 | ||
321 | DMINFO("cleaned up"); | |
322 | } | |
323 | ||
b9249e55 | 324 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
325 | local_init, |
326 | dm_target_init, | |
327 | dm_linear_init, | |
328 | dm_stripe_init, | |
952b3557 | 329 | dm_io_init, |
945fa4d2 | 330 | dm_kcopyd_init, |
1da177e4 | 331 | dm_interface_init, |
fd2ed4d2 | 332 | dm_statistics_init, |
1da177e4 LT |
333 | }; |
334 | ||
b9249e55 | 335 | static void (*_exits[])(void) = { |
1da177e4 LT |
336 | local_exit, |
337 | dm_target_exit, | |
338 | dm_linear_exit, | |
339 | dm_stripe_exit, | |
952b3557 | 340 | dm_io_exit, |
945fa4d2 | 341 | dm_kcopyd_exit, |
1da177e4 | 342 | dm_interface_exit, |
fd2ed4d2 | 343 | dm_statistics_exit, |
1da177e4 LT |
344 | }; |
345 | ||
346 | static int __init dm_init(void) | |
347 | { | |
348 | const int count = ARRAY_SIZE(_inits); | |
349 | ||
350 | int r, i; | |
351 | ||
352 | for (i = 0; i < count; i++) { | |
353 | r = _inits[i](); | |
354 | if (r) | |
355 | goto bad; | |
356 | } | |
357 | ||
358 | return 0; | |
359 | ||
360 | bad: | |
361 | while (i--) | |
362 | _exits[i](); | |
363 | ||
364 | return r; | |
365 | } | |
366 | ||
367 | static void __exit dm_exit(void) | |
368 | { | |
369 | int i = ARRAY_SIZE(_exits); | |
370 | ||
371 | while (i--) | |
372 | _exits[i](); | |
d15b774c AK |
373 | |
374 | /* | |
375 | * Should be empty by this point. | |
376 | */ | |
d15b774c | 377 | idr_destroy(&_minor_idr); |
1da177e4 LT |
378 | } |
379 | ||
380 | /* | |
381 | * Block device functions | |
382 | */ | |
432a212c MA |
383 | int dm_deleting_md(struct mapped_device *md) |
384 | { | |
385 | return test_bit(DMF_DELETING, &md->flags); | |
386 | } | |
387 | ||
fe5f9f2c | 388 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
389 | { |
390 | struct mapped_device *md; | |
391 | ||
fba9f90e JM |
392 | spin_lock(&_minor_lock); |
393 | ||
fe5f9f2c | 394 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
395 | if (!md) |
396 | goto out; | |
397 | ||
5c6bd75d | 398 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 399 | dm_deleting_md(md)) { |
fba9f90e JM |
400 | md = NULL; |
401 | goto out; | |
402 | } | |
403 | ||
1da177e4 | 404 | dm_get(md); |
5c6bd75d | 405 | atomic_inc(&md->open_count); |
fba9f90e JM |
406 | |
407 | out: | |
408 | spin_unlock(&_minor_lock); | |
409 | ||
410 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
411 | } |
412 | ||
db2a144b | 413 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 414 | { |
fe5f9f2c | 415 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 416 | |
4a1aeb98 MB |
417 | spin_lock(&_minor_lock); |
418 | ||
2c140a24 MP |
419 | if (atomic_dec_and_test(&md->open_count) && |
420 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 421 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 422 | |
1da177e4 | 423 | dm_put(md); |
4a1aeb98 MB |
424 | |
425 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
426 | } |
427 | ||
5c6bd75d AK |
428 | int dm_open_count(struct mapped_device *md) |
429 | { | |
430 | return atomic_read(&md->open_count); | |
431 | } | |
432 | ||
433 | /* | |
434 | * Guarantees nothing is using the device before it's deleted. | |
435 | */ | |
2c140a24 | 436 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
437 | { |
438 | int r = 0; | |
439 | ||
440 | spin_lock(&_minor_lock); | |
441 | ||
2c140a24 | 442 | if (dm_open_count(md)) { |
5c6bd75d | 443 | r = -EBUSY; |
2c140a24 MP |
444 | if (mark_deferred) |
445 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
446 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
447 | r = -EEXIST; | |
5c6bd75d AK |
448 | else |
449 | set_bit(DMF_DELETING, &md->flags); | |
450 | ||
451 | spin_unlock(&_minor_lock); | |
452 | ||
453 | return r; | |
454 | } | |
455 | ||
2c140a24 MP |
456 | int dm_cancel_deferred_remove(struct mapped_device *md) |
457 | { | |
458 | int r = 0; | |
459 | ||
460 | spin_lock(&_minor_lock); | |
461 | ||
462 | if (test_bit(DMF_DELETING, &md->flags)) | |
463 | r = -EBUSY; | |
464 | else | |
465 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
466 | ||
467 | spin_unlock(&_minor_lock); | |
468 | ||
469 | return r; | |
470 | } | |
471 | ||
472 | static void do_deferred_remove(struct work_struct *w) | |
473 | { | |
474 | dm_deferred_remove(); | |
475 | } | |
476 | ||
fd2ed4d2 MP |
477 | sector_t dm_get_size(struct mapped_device *md) |
478 | { | |
479 | return get_capacity(md->disk); | |
480 | } | |
481 | ||
9974fa2c MS |
482 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
483 | { | |
484 | return md->queue; | |
485 | } | |
486 | ||
fd2ed4d2 MP |
487 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
488 | { | |
489 | return &md->stats; | |
490 | } | |
491 | ||
3ac51e74 DW |
492 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
493 | { | |
494 | struct mapped_device *md = bdev->bd_disk->private_data; | |
495 | ||
496 | return dm_get_geometry(md, geo); | |
497 | } | |
498 | ||
fe5f9f2c | 499 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
500 | unsigned int cmd, unsigned long arg) |
501 | { | |
fe5f9f2c | 502 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 503 | int srcu_idx; |
6c182cd8 | 504 | struct dm_table *map; |
aa129a22 MB |
505 | struct dm_target *tgt; |
506 | int r = -ENOTTY; | |
507 | ||
6c182cd8 | 508 | retry: |
83d5e5b0 MP |
509 | map = dm_get_live_table(md, &srcu_idx); |
510 | ||
aa129a22 MB |
511 | if (!map || !dm_table_get_size(map)) |
512 | goto out; | |
513 | ||
514 | /* We only support devices that have a single target */ | |
515 | if (dm_table_get_num_targets(map) != 1) | |
516 | goto out; | |
517 | ||
518 | tgt = dm_table_get_target(map, 0); | |
519 | ||
4f186f8b | 520 | if (dm_suspended_md(md)) { |
aa129a22 MB |
521 | r = -EAGAIN; |
522 | goto out; | |
523 | } | |
524 | ||
525 | if (tgt->type->ioctl) | |
647b3d00 | 526 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
527 | |
528 | out: | |
83d5e5b0 | 529 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 530 | |
6c182cd8 HR |
531 | if (r == -ENOTCONN) { |
532 | msleep(10); | |
533 | goto retry; | |
534 | } | |
535 | ||
aa129a22 MB |
536 | return r; |
537 | } | |
538 | ||
028867ac | 539 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
540 | { |
541 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
542 | } | |
543 | ||
028867ac | 544 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
545 | { |
546 | mempool_free(io, md->io_pool); | |
547 | } | |
548 | ||
028867ac | 549 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 550 | { |
dba14160 | 551 | bio_put(&tio->clone); |
1da177e4 LT |
552 | } |
553 | ||
08885643 KU |
554 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
555 | gfp_t gfp_mask) | |
cec47e3d | 556 | { |
5f015204 | 557 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
558 | } |
559 | ||
560 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
561 | { | |
5f015204 | 562 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
563 | } |
564 | ||
90abb8c4 KU |
565 | static int md_in_flight(struct mapped_device *md) |
566 | { | |
567 | return atomic_read(&md->pending[READ]) + | |
568 | atomic_read(&md->pending[WRITE]); | |
569 | } | |
570 | ||
3eaf840e JNN |
571 | static void start_io_acct(struct dm_io *io) |
572 | { | |
573 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 574 | struct bio *bio = io->bio; |
c9959059 | 575 | int cpu; |
fd2ed4d2 | 576 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
577 | |
578 | io->start_time = jiffies; | |
579 | ||
074a7aca TH |
580 | cpu = part_stat_lock(); |
581 | part_round_stats(cpu, &dm_disk(md)->part0); | |
582 | part_stat_unlock(); | |
1e9bb880 SL |
583 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
584 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
585 | |
586 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 587 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 588 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
589 | } |
590 | ||
d221d2e7 | 591 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
592 | { |
593 | struct mapped_device *md = io->md; | |
594 | struct bio *bio = io->bio; | |
595 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 596 | int pending, cpu; |
3eaf840e JNN |
597 | int rw = bio_data_dir(bio); |
598 | ||
074a7aca TH |
599 | cpu = part_stat_lock(); |
600 | part_round_stats(cpu, &dm_disk(md)->part0); | |
601 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
602 | part_stat_unlock(); | |
3eaf840e | 603 | |
fd2ed4d2 | 604 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 605 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
606 | bio_sectors(bio), true, duration, &io->stats_aux); |
607 | ||
af7e466a MP |
608 | /* |
609 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 610 | * a flush. |
af7e466a | 611 | */ |
1e9bb880 SL |
612 | pending = atomic_dec_return(&md->pending[rw]); |
613 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 614 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 615 | |
d221d2e7 MP |
616 | /* nudge anyone waiting on suspend queue */ |
617 | if (!pending) | |
618 | wake_up(&md->wait); | |
3eaf840e JNN |
619 | } |
620 | ||
1da177e4 LT |
621 | /* |
622 | * Add the bio to the list of deferred io. | |
623 | */ | |
92c63902 | 624 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 625 | { |
05447420 | 626 | unsigned long flags; |
1da177e4 | 627 | |
05447420 | 628 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 629 | bio_list_add(&md->deferred, bio); |
05447420 | 630 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 631 | queue_work(md->wq, &md->work); |
1da177e4 LT |
632 | } |
633 | ||
634 | /* | |
635 | * Everyone (including functions in this file), should use this | |
636 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 637 | * dm_put_live_table() when finished. |
1da177e4 | 638 | */ |
83d5e5b0 | 639 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 640 | { |
83d5e5b0 MP |
641 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
642 | ||
643 | return srcu_dereference(md->map, &md->io_barrier); | |
644 | } | |
1da177e4 | 645 | |
83d5e5b0 MP |
646 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
647 | { | |
648 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
649 | } | |
650 | ||
651 | void dm_sync_table(struct mapped_device *md) | |
652 | { | |
653 | synchronize_srcu(&md->io_barrier); | |
654 | synchronize_rcu_expedited(); | |
655 | } | |
656 | ||
657 | /* | |
658 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
659 | * The caller must not block between these two functions. | |
660 | */ | |
661 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
662 | { | |
663 | rcu_read_lock(); | |
664 | return rcu_dereference(md->map); | |
665 | } | |
1da177e4 | 666 | |
83d5e5b0 MP |
667 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
668 | { | |
669 | rcu_read_unlock(); | |
1da177e4 LT |
670 | } |
671 | ||
3ac51e74 DW |
672 | /* |
673 | * Get the geometry associated with a dm device | |
674 | */ | |
675 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
676 | { | |
677 | *geo = md->geometry; | |
678 | ||
679 | return 0; | |
680 | } | |
681 | ||
682 | /* | |
683 | * Set the geometry of a device. | |
684 | */ | |
685 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
686 | { | |
687 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
688 | ||
689 | if (geo->start > sz) { | |
690 | DMWARN("Start sector is beyond the geometry limits."); | |
691 | return -EINVAL; | |
692 | } | |
693 | ||
694 | md->geometry = *geo; | |
695 | ||
696 | return 0; | |
697 | } | |
698 | ||
1da177e4 LT |
699 | /*----------------------------------------------------------------- |
700 | * CRUD START: | |
701 | * A more elegant soln is in the works that uses the queue | |
702 | * merge fn, unfortunately there are a couple of changes to | |
703 | * the block layer that I want to make for this. So in the | |
704 | * interests of getting something for people to use I give | |
705 | * you this clearly demarcated crap. | |
706 | *---------------------------------------------------------------*/ | |
707 | ||
2e93ccc1 KU |
708 | static int __noflush_suspending(struct mapped_device *md) |
709 | { | |
710 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
711 | } | |
712 | ||
1da177e4 LT |
713 | /* |
714 | * Decrements the number of outstanding ios that a bio has been | |
715 | * cloned into, completing the original io if necc. | |
716 | */ | |
858119e1 | 717 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 718 | { |
2e93ccc1 | 719 | unsigned long flags; |
b35f8caa MB |
720 | int io_error; |
721 | struct bio *bio; | |
722 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
723 | |
724 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
725 | if (unlikely(error)) { |
726 | spin_lock_irqsave(&io->endio_lock, flags); | |
727 | if (!(io->error > 0 && __noflush_suspending(md))) | |
728 | io->error = error; | |
729 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
730 | } | |
1da177e4 LT |
731 | |
732 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
733 | if (io->error == DM_ENDIO_REQUEUE) { |
734 | /* | |
735 | * Target requested pushing back the I/O. | |
2e93ccc1 | 736 | */ |
022c2611 | 737 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
738 | if (__noflush_suspending(md)) |
739 | bio_list_add_head(&md->deferred, io->bio); | |
740 | else | |
2e93ccc1 KU |
741 | /* noflush suspend was interrupted. */ |
742 | io->error = -EIO; | |
022c2611 | 743 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
744 | } |
745 | ||
b35f8caa MB |
746 | io_error = io->error; |
747 | bio = io->bio; | |
6a8736d1 TH |
748 | end_io_acct(io); |
749 | free_io(md, io); | |
750 | ||
751 | if (io_error == DM_ENDIO_REQUEUE) | |
752 | return; | |
2e93ccc1 | 753 | |
4f024f37 | 754 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 755 | /* |
6a8736d1 TH |
756 | * Preflush done for flush with data, reissue |
757 | * without REQ_FLUSH. | |
af7e466a | 758 | */ |
6a8736d1 TH |
759 | bio->bi_rw &= ~REQ_FLUSH; |
760 | queue_io(md, bio); | |
af7e466a | 761 | } else { |
b372d360 | 762 | /* done with normal IO or empty flush */ |
0a82a8d1 | 763 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 764 | bio_endio(bio, io_error); |
b35f8caa | 765 | } |
1da177e4 LT |
766 | } |
767 | } | |
768 | ||
7eee4ae2 MS |
769 | static void disable_write_same(struct mapped_device *md) |
770 | { | |
771 | struct queue_limits *limits = dm_get_queue_limits(md); | |
772 | ||
773 | /* device doesn't really support WRITE SAME, disable it */ | |
774 | limits->max_write_same_sectors = 0; | |
775 | } | |
776 | ||
6712ecf8 | 777 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
778 | { |
779 | int r = 0; | |
bfc6d41c | 780 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 781 | struct dm_io *io = tio->io; |
9faf400f | 782 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
783 | dm_endio_fn endio = tio->ti->type->end_io; |
784 | ||
1da177e4 LT |
785 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
786 | error = -EIO; | |
787 | ||
788 | if (endio) { | |
7de3ee57 | 789 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
790 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
791 | /* | |
792 | * error and requeue request are handled | |
793 | * in dec_pending(). | |
794 | */ | |
1da177e4 | 795 | error = r; |
45cbcd79 KU |
796 | else if (r == DM_ENDIO_INCOMPLETE) |
797 | /* The target will handle the io */ | |
6712ecf8 | 798 | return; |
45cbcd79 KU |
799 | else if (r) { |
800 | DMWARN("unimplemented target endio return value: %d", r); | |
801 | BUG(); | |
802 | } | |
1da177e4 LT |
803 | } |
804 | ||
7eee4ae2 MS |
805 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
806 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
807 | disable_write_same(md); | |
808 | ||
9faf400f | 809 | free_tio(md, tio); |
b35f8caa | 810 | dec_pending(io, error); |
1da177e4 LT |
811 | } |
812 | ||
cec47e3d KU |
813 | /* |
814 | * Partial completion handling for request-based dm | |
815 | */ | |
816 | static void end_clone_bio(struct bio *clone, int error) | |
817 | { | |
bfc6d41c MP |
818 | struct dm_rq_clone_bio_info *info = |
819 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
820 | struct dm_rq_target_io *tio = info->tio; |
821 | struct bio *bio = info->orig; | |
4f024f37 | 822 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
823 | |
824 | bio_put(clone); | |
825 | ||
826 | if (tio->error) | |
827 | /* | |
828 | * An error has already been detected on the request. | |
829 | * Once error occurred, just let clone->end_io() handle | |
830 | * the remainder. | |
831 | */ | |
832 | return; | |
833 | else if (error) { | |
834 | /* | |
835 | * Don't notice the error to the upper layer yet. | |
836 | * The error handling decision is made by the target driver, | |
837 | * when the request is completed. | |
838 | */ | |
839 | tio->error = error; | |
840 | return; | |
841 | } | |
842 | ||
843 | /* | |
844 | * I/O for the bio successfully completed. | |
845 | * Notice the data completion to the upper layer. | |
846 | */ | |
847 | ||
848 | /* | |
849 | * bios are processed from the head of the list. | |
850 | * So the completing bio should always be rq->bio. | |
851 | * If it's not, something wrong is happening. | |
852 | */ | |
853 | if (tio->orig->bio != bio) | |
854 | DMERR("bio completion is going in the middle of the request"); | |
855 | ||
856 | /* | |
857 | * Update the original request. | |
858 | * Do not use blk_end_request() here, because it may complete | |
859 | * the original request before the clone, and break the ordering. | |
860 | */ | |
861 | blk_update_request(tio->orig, 0, nr_bytes); | |
862 | } | |
863 | ||
864 | /* | |
865 | * Don't touch any member of the md after calling this function because | |
866 | * the md may be freed in dm_put() at the end of this function. | |
867 | * Or do dm_get() before calling this function and dm_put() later. | |
868 | */ | |
b4324fee | 869 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 870 | { |
b4324fee | 871 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
872 | |
873 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 874 | if (!md_in_flight(md)) |
cec47e3d KU |
875 | wake_up(&md->wait); |
876 | ||
a8c32a5c JA |
877 | /* |
878 | * Run this off this callpath, as drivers could invoke end_io while | |
879 | * inside their request_fn (and holding the queue lock). Calling | |
880 | * back into ->request_fn() could deadlock attempting to grab the | |
881 | * queue lock again. | |
882 | */ | |
cec47e3d | 883 | if (run_queue) |
a8c32a5c | 884 | blk_run_queue_async(md->queue); |
cec47e3d KU |
885 | |
886 | /* | |
887 | * dm_put() must be at the end of this function. See the comment above | |
888 | */ | |
889 | dm_put(md); | |
890 | } | |
891 | ||
a77e28c7 KU |
892 | static void free_rq_clone(struct request *clone) |
893 | { | |
894 | struct dm_rq_target_io *tio = clone->end_io_data; | |
895 | ||
896 | blk_rq_unprep_clone(clone); | |
897 | free_rq_tio(tio); | |
898 | } | |
899 | ||
980691e5 KU |
900 | /* |
901 | * Complete the clone and the original request. | |
902 | * Must be called without queue lock. | |
903 | */ | |
904 | static void dm_end_request(struct request *clone, int error) | |
905 | { | |
906 | int rw = rq_data_dir(clone); | |
907 | struct dm_rq_target_io *tio = clone->end_io_data; | |
908 | struct mapped_device *md = tio->md; | |
909 | struct request *rq = tio->orig; | |
910 | ||
29e4013d | 911 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
912 | rq->errors = clone->errors; |
913 | rq->resid_len = clone->resid_len; | |
914 | ||
915 | if (rq->sense) | |
916 | /* | |
917 | * We are using the sense buffer of the original | |
918 | * request. | |
919 | * So setting the length of the sense data is enough. | |
920 | */ | |
921 | rq->sense_len = clone->sense_len; | |
922 | } | |
923 | ||
924 | free_rq_clone(clone); | |
29e4013d TH |
925 | blk_end_request_all(rq, error); |
926 | rq_completed(md, rw, true); | |
980691e5 KU |
927 | } |
928 | ||
cec47e3d KU |
929 | static void dm_unprep_request(struct request *rq) |
930 | { | |
931 | struct request *clone = rq->special; | |
cec47e3d KU |
932 | |
933 | rq->special = NULL; | |
934 | rq->cmd_flags &= ~REQ_DONTPREP; | |
935 | ||
a77e28c7 | 936 | free_rq_clone(clone); |
cec47e3d KU |
937 | } |
938 | ||
939 | /* | |
940 | * Requeue the original request of a clone. | |
941 | */ | |
942 | void dm_requeue_unmapped_request(struct request *clone) | |
943 | { | |
b4324fee | 944 | int rw = rq_data_dir(clone); |
cec47e3d KU |
945 | struct dm_rq_target_io *tio = clone->end_io_data; |
946 | struct mapped_device *md = tio->md; | |
947 | struct request *rq = tio->orig; | |
948 | struct request_queue *q = rq->q; | |
949 | unsigned long flags; | |
950 | ||
951 | dm_unprep_request(rq); | |
952 | ||
953 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
954 | blk_requeue_request(q, rq); |
955 | spin_unlock_irqrestore(q->queue_lock, flags); | |
956 | ||
b4324fee | 957 | rq_completed(md, rw, 0); |
cec47e3d KU |
958 | } |
959 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
960 | ||
961 | static void __stop_queue(struct request_queue *q) | |
962 | { | |
963 | blk_stop_queue(q); | |
964 | } | |
965 | ||
966 | static void stop_queue(struct request_queue *q) | |
967 | { | |
968 | unsigned long flags; | |
969 | ||
970 | spin_lock_irqsave(q->queue_lock, flags); | |
971 | __stop_queue(q); | |
972 | spin_unlock_irqrestore(q->queue_lock, flags); | |
973 | } | |
974 | ||
975 | static void __start_queue(struct request_queue *q) | |
976 | { | |
977 | if (blk_queue_stopped(q)) | |
978 | blk_start_queue(q); | |
979 | } | |
980 | ||
981 | static void start_queue(struct request_queue *q) | |
982 | { | |
983 | unsigned long flags; | |
984 | ||
985 | spin_lock_irqsave(q->queue_lock, flags); | |
986 | __start_queue(q); | |
987 | spin_unlock_irqrestore(q->queue_lock, flags); | |
988 | } | |
989 | ||
11a68244 | 990 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 991 | { |
11a68244 | 992 | int r = error; |
cec47e3d | 993 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 994 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 995 | |
ba1cbad9 MS |
996 | if (tio->ti) { |
997 | rq_end_io = tio->ti->type->rq_end_io; | |
998 | ||
999 | if (mapped && rq_end_io) | |
1000 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1001 | } | |
cec47e3d | 1002 | |
7eee4ae2 MS |
1003 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1004 | !clone->q->limits.max_write_same_sectors)) | |
1005 | disable_write_same(tio->md); | |
1006 | ||
11a68244 | 1007 | if (r <= 0) |
cec47e3d | 1008 | /* The target wants to complete the I/O */ |
11a68244 KU |
1009 | dm_end_request(clone, r); |
1010 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1011 | /* The target will handle the I/O */ |
1012 | return; | |
11a68244 | 1013 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1014 | /* The target wants to requeue the I/O */ |
1015 | dm_requeue_unmapped_request(clone); | |
1016 | else { | |
11a68244 | 1017 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1018 | BUG(); |
1019 | } | |
1020 | } | |
1021 | ||
11a68244 KU |
1022 | /* |
1023 | * Request completion handler for request-based dm | |
1024 | */ | |
1025 | static void dm_softirq_done(struct request *rq) | |
1026 | { | |
1027 | bool mapped = true; | |
1028 | struct request *clone = rq->completion_data; | |
1029 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1030 | ||
1031 | if (rq->cmd_flags & REQ_FAILED) | |
1032 | mapped = false; | |
1033 | ||
1034 | dm_done(clone, tio->error, mapped); | |
1035 | } | |
1036 | ||
cec47e3d KU |
1037 | /* |
1038 | * Complete the clone and the original request with the error status | |
1039 | * through softirq context. | |
1040 | */ | |
1041 | static void dm_complete_request(struct request *clone, int error) | |
1042 | { | |
1043 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1044 | struct request *rq = tio->orig; | |
1045 | ||
1046 | tio->error = error; | |
1047 | rq->completion_data = clone; | |
1048 | blk_complete_request(rq); | |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * Complete the not-mapped clone and the original request with the error status | |
1053 | * through softirq context. | |
1054 | * Target's rq_end_io() function isn't called. | |
1055 | * This may be used when the target's map_rq() function fails. | |
1056 | */ | |
1057 | void dm_kill_unmapped_request(struct request *clone, int error) | |
1058 | { | |
1059 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1060 | struct request *rq = tio->orig; | |
1061 | ||
1062 | rq->cmd_flags |= REQ_FAILED; | |
1063 | dm_complete_request(clone, error); | |
1064 | } | |
1065 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1066 | ||
1067 | /* | |
1068 | * Called with the queue lock held | |
1069 | */ | |
1070 | static void end_clone_request(struct request *clone, int error) | |
1071 | { | |
1072 | /* | |
1073 | * For just cleaning up the information of the queue in which | |
1074 | * the clone was dispatched. | |
1075 | * The clone is *NOT* freed actually here because it is alloced from | |
1076 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1077 | */ | |
1078 | __blk_put_request(clone->q, clone); | |
1079 | ||
1080 | /* | |
1081 | * Actual request completion is done in a softirq context which doesn't | |
1082 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1083 | * - another request may be submitted by the upper level driver | |
1084 | * of the stacking during the completion | |
1085 | * - the submission which requires queue lock may be done | |
1086 | * against this queue | |
1087 | */ | |
1088 | dm_complete_request(clone, error); | |
1089 | } | |
1090 | ||
56a67df7 MS |
1091 | /* |
1092 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1093 | * target boundary. | |
1094 | */ | |
1095 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1096 | { | |
1097 | sector_t target_offset = dm_target_offset(ti, sector); | |
1098 | ||
1099 | return ti->len - target_offset; | |
1100 | } | |
1101 | ||
1102 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1103 | { |
56a67df7 | 1104 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1105 | sector_t offset, max_len; |
1da177e4 LT |
1106 | |
1107 | /* | |
542f9038 | 1108 | * Does the target need to split even further? |
1da177e4 | 1109 | */ |
542f9038 MS |
1110 | if (ti->max_io_len) { |
1111 | offset = dm_target_offset(ti, sector); | |
1112 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1113 | max_len = sector_div(offset, ti->max_io_len); | |
1114 | else | |
1115 | max_len = offset & (ti->max_io_len - 1); | |
1116 | max_len = ti->max_io_len - max_len; | |
1117 | ||
1118 | if (len > max_len) | |
1119 | len = max_len; | |
1da177e4 LT |
1120 | } |
1121 | ||
1122 | return len; | |
1123 | } | |
1124 | ||
542f9038 MS |
1125 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1126 | { | |
1127 | if (len > UINT_MAX) { | |
1128 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1129 | (unsigned long long)len, UINT_MAX); | |
1130 | ti->error = "Maximum size of target IO is too large"; | |
1131 | return -EINVAL; | |
1132 | } | |
1133 | ||
1134 | ti->max_io_len = (uint32_t) len; | |
1135 | ||
1136 | return 0; | |
1137 | } | |
1138 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1139 | ||
1dd40c3e MP |
1140 | /* |
1141 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1142 | * allowed for all bio types except REQ_FLUSH. | |
1143 | * | |
1144 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1145 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1146 | * sent in a next bio. | |
1147 | * | |
1148 | * A diagram that explains the arithmetics: | |
1149 | * +--------------------+---------------+-------+ | |
1150 | * | 1 | 2 | 3 | | |
1151 | * +--------------------+---------------+-------+ | |
1152 | * | |
1153 | * <-------------- *tio->len_ptr ---------------> | |
1154 | * <------- bi_size -------> | |
1155 | * <-- n_sectors --> | |
1156 | * | |
1157 | * Region 1 was already iterated over with bio_advance or similar function. | |
1158 | * (it may be empty if the target doesn't use bio_advance) | |
1159 | * Region 2 is the remaining bio size that the target wants to process. | |
1160 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1161 | * to make it empty) | |
1162 | * The target requires that region 3 is to be sent in the next bio. | |
1163 | * | |
1164 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1165 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1166 | * copies of the bio. | |
1167 | */ | |
1168 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1169 | { | |
1170 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1171 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1172 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1173 | BUG_ON(bi_size > *tio->len_ptr); | |
1174 | BUG_ON(n_sectors > bi_size); | |
1175 | *tio->len_ptr -= bi_size - n_sectors; | |
1176 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1177 | } | |
1178 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1179 | ||
bd2a49b8 | 1180 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1181 | { |
1182 | int r; | |
2056a782 | 1183 | sector_t sector; |
9faf400f | 1184 | struct mapped_device *md; |
dba14160 | 1185 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1186 | struct dm_target *ti = tio->ti; |
1da177e4 | 1187 | |
1da177e4 | 1188 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1189 | |
1190 | /* | |
1191 | * Map the clone. If r == 0 we don't need to do | |
1192 | * anything, the target has assumed ownership of | |
1193 | * this io. | |
1194 | */ | |
1195 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1196 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1197 | r = ti->type->map(ti, clone); |
45cbcd79 | 1198 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1199 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1200 | |
d07335e5 MS |
1201 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1202 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1203 | |
1da177e4 | 1204 | generic_make_request(clone); |
2e93ccc1 KU |
1205 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1206 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1207 | md = tio->io->md; |
1208 | dec_pending(tio->io, r); | |
9faf400f | 1209 | free_tio(md, tio); |
45cbcd79 KU |
1210 | } else if (r) { |
1211 | DMWARN("unimplemented target map return value: %d", r); | |
1212 | BUG(); | |
1da177e4 LT |
1213 | } |
1214 | } | |
1215 | ||
1216 | struct clone_info { | |
1217 | struct mapped_device *md; | |
1218 | struct dm_table *map; | |
1219 | struct bio *bio; | |
1220 | struct dm_io *io; | |
1221 | sector_t sector; | |
e0d6609a | 1222 | unsigned sector_count; |
1da177e4 LT |
1223 | }; |
1224 | ||
e0d6609a | 1225 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1226 | { |
4f024f37 KO |
1227 | bio->bi_iter.bi_sector = sector; |
1228 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1229 | } |
1230 | ||
1231 | /* | |
1232 | * Creates a bio that consists of range of complete bvecs. | |
1233 | */ | |
dba14160 | 1234 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1235 | sector_t sector, unsigned len) |
1da177e4 | 1236 | { |
dba14160 | 1237 | struct bio *clone = &tio->clone; |
1da177e4 | 1238 | |
1c3b13e6 KO |
1239 | __bio_clone_fast(clone, bio); |
1240 | ||
1241 | if (bio_integrity(bio)) | |
1242 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1243 | |
1c3b13e6 KO |
1244 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1245 | clone->bi_iter.bi_size = to_bytes(len); | |
1246 | ||
1247 | if (bio_integrity(bio)) | |
1248 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1249 | } |
1250 | ||
9015df24 | 1251 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1252 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1253 | unsigned target_bio_nr) |
f9ab94ce | 1254 | { |
dba14160 MP |
1255 | struct dm_target_io *tio; |
1256 | struct bio *clone; | |
1257 | ||
1258 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1259 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1260 | |
1261 | tio->io = ci->io; | |
1262 | tio->ti = ti; | |
55a62eef | 1263 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1264 | |
1265 | return tio; | |
1266 | } | |
1267 | ||
14fe594d AK |
1268 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1269 | struct dm_target *ti, | |
1dd40c3e | 1270 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1271 | { |
55a62eef | 1272 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1273 | struct bio *clone = &tio->clone; |
9015df24 | 1274 | |
1dd40c3e MP |
1275 | tio->len_ptr = len; |
1276 | ||
06a426ce MS |
1277 | /* |
1278 | * Discard requests require the bio's inline iovecs be initialized. | |
1279 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1280 | * and discard, so no need for concern about wasted bvec allocations. | |
1281 | */ | |
1c3b13e6 | 1282 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1283 | if (len) |
1dd40c3e | 1284 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1285 | |
bd2a49b8 | 1286 | __map_bio(tio); |
f9ab94ce MP |
1287 | } |
1288 | ||
14fe594d | 1289 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1290 | unsigned num_bios, unsigned *len) |
06a426ce | 1291 | { |
55a62eef | 1292 | unsigned target_bio_nr; |
06a426ce | 1293 | |
55a62eef | 1294 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1295 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1296 | } |
1297 | ||
14fe594d | 1298 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1299 | { |
06a426ce | 1300 | unsigned target_nr = 0; |
f9ab94ce MP |
1301 | struct dm_target *ti; |
1302 | ||
b372d360 | 1303 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1304 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1305 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1306 | |
f9ab94ce MP |
1307 | return 0; |
1308 | } | |
1309 | ||
e4c93811 | 1310 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1311 | sector_t sector, unsigned *len) |
5ae89a87 | 1312 | { |
dba14160 | 1313 | struct bio *bio = ci->bio; |
5ae89a87 | 1314 | struct dm_target_io *tio; |
b0d8ed4d AK |
1315 | unsigned target_bio_nr; |
1316 | unsigned num_target_bios = 1; | |
5ae89a87 | 1317 | |
b0d8ed4d AK |
1318 | /* |
1319 | * Does the target want to receive duplicate copies of the bio? | |
1320 | */ | |
1321 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1322 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1323 | |
b0d8ed4d | 1324 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
1c3b13e6 | 1325 | tio = alloc_tio(ci, ti, 0, target_bio_nr); |
1dd40c3e MP |
1326 | tio->len_ptr = len; |
1327 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1328 | __map_bio(tio); |
1329 | } | |
5ae89a87 MS |
1330 | } |
1331 | ||
55a62eef | 1332 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1333 | |
55a62eef | 1334 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1335 | { |
55a62eef | 1336 | return ti->num_discard_bios; |
23508a96 MS |
1337 | } |
1338 | ||
55a62eef | 1339 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1340 | { |
55a62eef | 1341 | return ti->num_write_same_bios; |
23508a96 MS |
1342 | } |
1343 | ||
1344 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1345 | ||
1346 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1347 | { | |
55a62eef | 1348 | return ti->split_discard_bios; |
23508a96 MS |
1349 | } |
1350 | ||
14fe594d AK |
1351 | static int __send_changing_extent_only(struct clone_info *ci, |
1352 | get_num_bios_fn get_num_bios, | |
1353 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1354 | { |
1355 | struct dm_target *ti; | |
e0d6609a | 1356 | unsigned len; |
55a62eef | 1357 | unsigned num_bios; |
5ae89a87 | 1358 | |
a79245b3 MS |
1359 | do { |
1360 | ti = dm_table_find_target(ci->map, ci->sector); | |
1361 | if (!dm_target_is_valid(ti)) | |
1362 | return -EIO; | |
5ae89a87 | 1363 | |
5ae89a87 | 1364 | /* |
23508a96 MS |
1365 | * Even though the device advertised support for this type of |
1366 | * request, that does not mean every target supports it, and | |
936688d7 | 1367 | * reconfiguration might also have changed that since the |
a79245b3 | 1368 | * check was performed. |
5ae89a87 | 1369 | */ |
55a62eef AK |
1370 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1371 | if (!num_bios) | |
a79245b3 | 1372 | return -EOPNOTSUPP; |
5ae89a87 | 1373 | |
23508a96 | 1374 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1375 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1376 | else |
e0d6609a | 1377 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1378 | |
1dd40c3e | 1379 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1380 | |
1381 | ci->sector += len; | |
1382 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1383 | |
1384 | return 0; | |
1385 | } | |
1386 | ||
14fe594d | 1387 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1388 | { |
14fe594d AK |
1389 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1390 | is_split_required_for_discard); | |
23508a96 MS |
1391 | } |
1392 | ||
14fe594d | 1393 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1394 | { |
14fe594d | 1395 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1396 | } |
1397 | ||
e4c93811 AK |
1398 | /* |
1399 | * Select the correct strategy for processing a non-flush bio. | |
1400 | */ | |
14fe594d | 1401 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1402 | { |
dba14160 | 1403 | struct bio *bio = ci->bio; |
512875bd | 1404 | struct dm_target *ti; |
1c3b13e6 | 1405 | unsigned len; |
1da177e4 | 1406 | |
5ae89a87 | 1407 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1408 | return __send_discard(ci); |
23508a96 | 1409 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1410 | return __send_write_same(ci); |
5ae89a87 | 1411 | |
512875bd JN |
1412 | ti = dm_table_find_target(ci->map, ci->sector); |
1413 | if (!dm_target_is_valid(ti)) | |
1414 | return -EIO; | |
1415 | ||
1c3b13e6 | 1416 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1417 | |
1dd40c3e | 1418 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1419 | |
1c3b13e6 KO |
1420 | ci->sector += len; |
1421 | ci->sector_count -= len; | |
1da177e4 | 1422 | |
1c3b13e6 | 1423 | return 0; |
1da177e4 LT |
1424 | } |
1425 | ||
1426 | /* | |
14fe594d | 1427 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1428 | */ |
83d5e5b0 MP |
1429 | static void __split_and_process_bio(struct mapped_device *md, |
1430 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1431 | { |
1432 | struct clone_info ci; | |
512875bd | 1433 | int error = 0; |
1da177e4 | 1434 | |
83d5e5b0 | 1435 | if (unlikely(!map)) { |
6a8736d1 | 1436 | bio_io_error(bio); |
f0b9a450 MP |
1437 | return; |
1438 | } | |
692d0eb9 | 1439 | |
83d5e5b0 | 1440 | ci.map = map; |
1da177e4 | 1441 | ci.md = md; |
1da177e4 LT |
1442 | ci.io = alloc_io(md); |
1443 | ci.io->error = 0; | |
1444 | atomic_set(&ci.io->io_count, 1); | |
1445 | ci.io->bio = bio; | |
1446 | ci.io->md = md; | |
f88fb981 | 1447 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1448 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1449 | |
3eaf840e | 1450 | start_io_acct(ci.io); |
bd2a49b8 | 1451 | |
b372d360 MS |
1452 | if (bio->bi_rw & REQ_FLUSH) { |
1453 | ci.bio = &ci.md->flush_bio; | |
1454 | ci.sector_count = 0; | |
14fe594d | 1455 | error = __send_empty_flush(&ci); |
b372d360 MS |
1456 | /* dec_pending submits any data associated with flush */ |
1457 | } else { | |
6a8736d1 | 1458 | ci.bio = bio; |
d87f4c14 | 1459 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1460 | while (ci.sector_count && !error) |
14fe594d | 1461 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1462 | } |
1da177e4 LT |
1463 | |
1464 | /* drop the extra reference count */ | |
512875bd | 1465 | dec_pending(ci.io, error); |
1da177e4 LT |
1466 | } |
1467 | /*----------------------------------------------------------------- | |
1468 | * CRUD END | |
1469 | *---------------------------------------------------------------*/ | |
1470 | ||
f6fccb12 MB |
1471 | static int dm_merge_bvec(struct request_queue *q, |
1472 | struct bvec_merge_data *bvm, | |
1473 | struct bio_vec *biovec) | |
1474 | { | |
1475 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1476 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1477 | struct dm_target *ti; |
1478 | sector_t max_sectors; | |
5037108a | 1479 | int max_size = 0; |
f6fccb12 MB |
1480 | |
1481 | if (unlikely(!map)) | |
5037108a | 1482 | goto out; |
f6fccb12 MB |
1483 | |
1484 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1485 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1486 | goto out; |
f6fccb12 MB |
1487 | |
1488 | /* | |
1489 | * Find maximum amount of I/O that won't need splitting | |
1490 | */ | |
56a67df7 | 1491 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1492 | (sector_t) BIO_MAX_SECTORS); |
1493 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1494 | if (max_size < 0) | |
1495 | max_size = 0; | |
1496 | ||
1497 | /* | |
1498 | * merge_bvec_fn() returns number of bytes | |
1499 | * it can accept at this offset | |
1500 | * max is precomputed maximal io size | |
1501 | */ | |
1502 | if (max_size && ti->type->merge) | |
1503 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1504 | /* |
1505 | * If the target doesn't support merge method and some of the devices | |
1506 | * provided their merge_bvec method (we know this by looking at | |
1507 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1508 | * entries. So always set max_size to 0, and the code below allows | |
1509 | * just one page. | |
1510 | */ | |
1511 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1512 | max_size = 0; |
f6fccb12 | 1513 | |
5037108a | 1514 | out: |
83d5e5b0 | 1515 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1516 | /* |
1517 | * Always allow an entire first page | |
1518 | */ | |
1519 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1520 | max_size = biovec->bv_len; | |
1521 | ||
f6fccb12 MB |
1522 | return max_size; |
1523 | } | |
1524 | ||
1da177e4 LT |
1525 | /* |
1526 | * The request function that just remaps the bio built up by | |
1527 | * dm_merge_bvec. | |
1528 | */ | |
5a7bbad2 | 1529 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1530 | { |
12f03a49 | 1531 | int rw = bio_data_dir(bio); |
1da177e4 | 1532 | struct mapped_device *md = q->queuedata; |
c9959059 | 1533 | int cpu; |
83d5e5b0 MP |
1534 | int srcu_idx; |
1535 | struct dm_table *map; | |
1da177e4 | 1536 | |
83d5e5b0 | 1537 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1538 | |
074a7aca TH |
1539 | cpu = part_stat_lock(); |
1540 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1541 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1542 | part_stat_unlock(); | |
12f03a49 | 1543 | |
6a8736d1 TH |
1544 | /* if we're suspended, we have to queue this io for later */ |
1545 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1546 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1547 | |
6a8736d1 TH |
1548 | if (bio_rw(bio) != READA) |
1549 | queue_io(md, bio); | |
1550 | else | |
54d9a1b4 | 1551 | bio_io_error(bio); |
5a7bbad2 | 1552 | return; |
1da177e4 LT |
1553 | } |
1554 | ||
83d5e5b0 MP |
1555 | __split_and_process_bio(md, map, bio); |
1556 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1557 | return; |
cec47e3d KU |
1558 | } |
1559 | ||
fd2ed4d2 | 1560 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1561 | { |
1562 | return blk_queue_stackable(md->queue); | |
1563 | } | |
1564 | ||
5a7bbad2 | 1565 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1566 | { |
1567 | struct mapped_device *md = q->queuedata; | |
1568 | ||
1569 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1570 | blk_queue_bio(q, bio); |
1571 | else | |
1572 | _dm_request(q, bio); | |
cec47e3d KU |
1573 | } |
1574 | ||
1575 | void dm_dispatch_request(struct request *rq) | |
1576 | { | |
1577 | int r; | |
1578 | ||
1579 | if (blk_queue_io_stat(rq->q)) | |
1580 | rq->cmd_flags |= REQ_IO_STAT; | |
1581 | ||
1582 | rq->start_time = jiffies; | |
1583 | r = blk_insert_cloned_request(rq->q, rq); | |
1584 | if (r) | |
1585 | dm_complete_request(rq, r); | |
1586 | } | |
1587 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1588 | ||
cec47e3d KU |
1589 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1590 | void *data) | |
1591 | { | |
1592 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1593 | struct dm_rq_clone_bio_info *info = |
1594 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1595 | |
1596 | info->orig = bio_orig; | |
1597 | info->tio = tio; | |
1598 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1599 | |
1600 | return 0; | |
1601 | } | |
1602 | ||
1603 | static int setup_clone(struct request *clone, struct request *rq, | |
1604 | struct dm_rq_target_io *tio) | |
1605 | { | |
d0bcb878 | 1606 | int r; |
cec47e3d | 1607 | |
29e4013d TH |
1608 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1609 | dm_rq_bio_constructor, tio); | |
1610 | if (r) | |
1611 | return r; | |
cec47e3d | 1612 | |
29e4013d TH |
1613 | clone->cmd = rq->cmd; |
1614 | clone->cmd_len = rq->cmd_len; | |
1615 | clone->sense = rq->sense; | |
cec47e3d KU |
1616 | clone->end_io = end_clone_request; |
1617 | clone->end_io_data = tio; | |
1618 | ||
1619 | return 0; | |
1620 | } | |
1621 | ||
6facdaff KU |
1622 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1623 | gfp_t gfp_mask) | |
1624 | { | |
1625 | struct request *clone; | |
1626 | struct dm_rq_target_io *tio; | |
1627 | ||
1628 | tio = alloc_rq_tio(md, gfp_mask); | |
1629 | if (!tio) | |
1630 | return NULL; | |
1631 | ||
1632 | tio->md = md; | |
1633 | tio->ti = NULL; | |
1634 | tio->orig = rq; | |
1635 | tio->error = 0; | |
1636 | memset(&tio->info, 0, sizeof(tio->info)); | |
1637 | ||
1638 | clone = &tio->clone; | |
1639 | if (setup_clone(clone, rq, tio)) { | |
1640 | /* -ENOMEM */ | |
1641 | free_rq_tio(tio); | |
1642 | return NULL; | |
1643 | } | |
1644 | ||
1645 | return clone; | |
1646 | } | |
1647 | ||
cec47e3d KU |
1648 | /* |
1649 | * Called with the queue lock held. | |
1650 | */ | |
1651 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1652 | { | |
1653 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1654 | struct request *clone; |
1655 | ||
cec47e3d KU |
1656 | if (unlikely(rq->special)) { |
1657 | DMWARN("Already has something in rq->special."); | |
1658 | return BLKPREP_KILL; | |
1659 | } | |
1660 | ||
6facdaff KU |
1661 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1662 | if (!clone) | |
cec47e3d | 1663 | return BLKPREP_DEFER; |
cec47e3d KU |
1664 | |
1665 | rq->special = clone; | |
1666 | rq->cmd_flags |= REQ_DONTPREP; | |
1667 | ||
1668 | return BLKPREP_OK; | |
1669 | } | |
1670 | ||
9eef87da KU |
1671 | /* |
1672 | * Returns: | |
1673 | * 0 : the request has been processed (not requeued) | |
1674 | * !0 : the request has been requeued | |
1675 | */ | |
1676 | static int map_request(struct dm_target *ti, struct request *clone, | |
1677 | struct mapped_device *md) | |
cec47e3d | 1678 | { |
9eef87da | 1679 | int r, requeued = 0; |
cec47e3d KU |
1680 | struct dm_rq_target_io *tio = clone->end_io_data; |
1681 | ||
cec47e3d KU |
1682 | tio->ti = ti; |
1683 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1684 | switch (r) { | |
1685 | case DM_MAPIO_SUBMITTED: | |
1686 | /* The target has taken the I/O to submit by itself later */ | |
1687 | break; | |
1688 | case DM_MAPIO_REMAPPED: | |
1689 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1690 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1691 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1692 | dm_dispatch_request(clone); |
1693 | break; | |
1694 | case DM_MAPIO_REQUEUE: | |
1695 | /* The target wants to requeue the I/O */ | |
1696 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1697 | requeued = 1; |
cec47e3d KU |
1698 | break; |
1699 | default: | |
1700 | if (r > 0) { | |
1701 | DMWARN("unimplemented target map return value: %d", r); | |
1702 | BUG(); | |
1703 | } | |
1704 | ||
1705 | /* The target wants to complete the I/O */ | |
1706 | dm_kill_unmapped_request(clone, r); | |
1707 | break; | |
1708 | } | |
9eef87da KU |
1709 | |
1710 | return requeued; | |
cec47e3d KU |
1711 | } |
1712 | ||
ba1cbad9 MS |
1713 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1714 | { | |
1715 | struct request *clone; | |
1716 | ||
1717 | blk_start_request(orig); | |
1718 | clone = orig->special; | |
1719 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1720 | ||
1721 | /* | |
1722 | * Hold the md reference here for the in-flight I/O. | |
1723 | * We can't rely on the reference count by device opener, | |
1724 | * because the device may be closed during the request completion | |
1725 | * when all bios are completed. | |
1726 | * See the comment in rq_completed() too. | |
1727 | */ | |
1728 | dm_get(md); | |
1729 | ||
1730 | return clone; | |
1731 | } | |
1732 | ||
cec47e3d KU |
1733 | /* |
1734 | * q->request_fn for request-based dm. | |
1735 | * Called with the queue lock held. | |
1736 | */ | |
1737 | static void dm_request_fn(struct request_queue *q) | |
1738 | { | |
1739 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1740 | int srcu_idx; |
1741 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1742 | struct dm_target *ti; |
b4324fee | 1743 | struct request *rq, *clone; |
29e4013d | 1744 | sector_t pos; |
cec47e3d KU |
1745 | |
1746 | /* | |
b4324fee KU |
1747 | * For suspend, check blk_queue_stopped() and increment |
1748 | * ->pending within a single queue_lock not to increment the | |
1749 | * number of in-flight I/Os after the queue is stopped in | |
1750 | * dm_suspend(). | |
cec47e3d | 1751 | */ |
7eaceacc | 1752 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1753 | rq = blk_peek_request(q); |
1754 | if (!rq) | |
7eaceacc | 1755 | goto delay_and_out; |
cec47e3d | 1756 | |
29e4013d TH |
1757 | /* always use block 0 to find the target for flushes for now */ |
1758 | pos = 0; | |
1759 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1760 | pos = blk_rq_pos(rq); | |
1761 | ||
1762 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1763 | if (!dm_target_is_valid(ti)) { |
1764 | /* | |
1765 | * Must perform setup, that dm_done() requires, | |
1766 | * before calling dm_kill_unmapped_request | |
1767 | */ | |
1768 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1769 | clone = dm_start_request(md, rq); | |
1770 | dm_kill_unmapped_request(clone, -EIO); | |
1771 | continue; | |
1772 | } | |
d0bcb878 | 1773 | |
cec47e3d | 1774 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1775 | goto delay_and_out; |
cec47e3d | 1776 | |
ba1cbad9 | 1777 | clone = dm_start_request(md, rq); |
b4324fee | 1778 | |
cec47e3d | 1779 | spin_unlock(q->queue_lock); |
9eef87da KU |
1780 | if (map_request(ti, clone, md)) |
1781 | goto requeued; | |
1782 | ||
052189a2 KU |
1783 | BUG_ON(!irqs_disabled()); |
1784 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1785 | } |
1786 | ||
1787 | goto out; | |
1788 | ||
9eef87da | 1789 | requeued: |
052189a2 KU |
1790 | BUG_ON(!irqs_disabled()); |
1791 | spin_lock(q->queue_lock); | |
9eef87da | 1792 | |
7eaceacc JA |
1793 | delay_and_out: |
1794 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1795 | out: |
83d5e5b0 | 1796 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1797 | } |
1798 | ||
1799 | int dm_underlying_device_busy(struct request_queue *q) | |
1800 | { | |
1801 | return blk_lld_busy(q); | |
1802 | } | |
1803 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1804 | ||
1805 | static int dm_lld_busy(struct request_queue *q) | |
1806 | { | |
1807 | int r; | |
1808 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1809 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1810 | |
1811 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1812 | r = 1; | |
1813 | else | |
1814 | r = dm_table_any_busy_target(map); | |
1815 | ||
83d5e5b0 | 1816 | dm_put_live_table_fast(md); |
cec47e3d KU |
1817 | |
1818 | return r; | |
1819 | } | |
1820 | ||
1da177e4 LT |
1821 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1822 | { | |
8a57dfc6 CS |
1823 | int r = bdi_bits; |
1824 | struct mapped_device *md = congested_data; | |
1825 | struct dm_table *map; | |
1da177e4 | 1826 | |
1eb787ec | 1827 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1828 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1829 | if (map) { |
cec47e3d KU |
1830 | /* |
1831 | * Request-based dm cares about only own queue for | |
1832 | * the query about congestion status of request_queue | |
1833 | */ | |
1834 | if (dm_request_based(md)) | |
1835 | r = md->queue->backing_dev_info.state & | |
1836 | bdi_bits; | |
1837 | else | |
1838 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1839 | } |
83d5e5b0 | 1840 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1841 | } |
1842 | ||
1da177e4 LT |
1843 | return r; |
1844 | } | |
1845 | ||
1846 | /*----------------------------------------------------------------- | |
1847 | * An IDR is used to keep track of allocated minor numbers. | |
1848 | *---------------------------------------------------------------*/ | |
2b06cfff | 1849 | static void free_minor(int minor) |
1da177e4 | 1850 | { |
f32c10b0 | 1851 | spin_lock(&_minor_lock); |
1da177e4 | 1852 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1853 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1854 | } |
1855 | ||
1856 | /* | |
1857 | * See if the device with a specific minor # is free. | |
1858 | */ | |
cf13ab8e | 1859 | static int specific_minor(int minor) |
1da177e4 | 1860 | { |
c9d76be6 | 1861 | int r; |
1da177e4 LT |
1862 | |
1863 | if (minor >= (1 << MINORBITS)) | |
1864 | return -EINVAL; | |
1865 | ||
c9d76be6 | 1866 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1867 | spin_lock(&_minor_lock); |
1da177e4 | 1868 | |
c9d76be6 | 1869 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1870 | |
f32c10b0 | 1871 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1872 | idr_preload_end(); |
1873 | if (r < 0) | |
1874 | return r == -ENOSPC ? -EBUSY : r; | |
1875 | return 0; | |
1da177e4 LT |
1876 | } |
1877 | ||
cf13ab8e | 1878 | static int next_free_minor(int *minor) |
1da177e4 | 1879 | { |
c9d76be6 | 1880 | int r; |
62f75c2f | 1881 | |
c9d76be6 | 1882 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1883 | spin_lock(&_minor_lock); |
1da177e4 | 1884 | |
c9d76be6 | 1885 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1886 | |
f32c10b0 | 1887 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1888 | idr_preload_end(); |
1889 | if (r < 0) | |
1890 | return r; | |
1891 | *minor = r; | |
1892 | return 0; | |
1da177e4 LT |
1893 | } |
1894 | ||
83d5cde4 | 1895 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1896 | |
53d5914f MP |
1897 | static void dm_wq_work(struct work_struct *work); |
1898 | ||
4a0b4ddf MS |
1899 | static void dm_init_md_queue(struct mapped_device *md) |
1900 | { | |
1901 | /* | |
1902 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1903 | * devices. The type of this dm device has not been decided yet. | |
1904 | * The type is decided at the first table loading time. | |
1905 | * To prevent problematic device stacking, clear the queue flag | |
1906 | * for request stacking support until then. | |
1907 | * | |
1908 | * This queue is new, so no concurrency on the queue_flags. | |
1909 | */ | |
1910 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1911 | ||
1912 | md->queue->queuedata = md; | |
1913 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1914 | md->queue->backing_dev_info.congested_data = md; | |
1915 | blk_queue_make_request(md->queue, dm_request); | |
1916 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1917 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1918 | } | |
1919 | ||
1da177e4 LT |
1920 | /* |
1921 | * Allocate and initialise a blank device with a given minor. | |
1922 | */ | |
2b06cfff | 1923 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1924 | { |
1925 | int r; | |
cf13ab8e | 1926 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1927 | void *old_md; |
1da177e4 LT |
1928 | |
1929 | if (!md) { | |
1930 | DMWARN("unable to allocate device, out of memory."); | |
1931 | return NULL; | |
1932 | } | |
1933 | ||
10da4f79 | 1934 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1935 | goto bad_module_get; |
10da4f79 | 1936 | |
1da177e4 | 1937 | /* get a minor number for the dev */ |
2b06cfff | 1938 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1939 | r = next_free_minor(&minor); |
2b06cfff | 1940 | else |
cf13ab8e | 1941 | r = specific_minor(minor); |
1da177e4 | 1942 | if (r < 0) |
6ed7ade8 | 1943 | goto bad_minor; |
1da177e4 | 1944 | |
83d5e5b0 MP |
1945 | r = init_srcu_struct(&md->io_barrier); |
1946 | if (r < 0) | |
1947 | goto bad_io_barrier; | |
1948 | ||
a5664dad | 1949 | md->type = DM_TYPE_NONE; |
e61290a4 | 1950 | mutex_init(&md->suspend_lock); |
a5664dad | 1951 | mutex_init(&md->type_lock); |
022c2611 | 1952 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1953 | atomic_set(&md->holders, 1); |
5c6bd75d | 1954 | atomic_set(&md->open_count, 0); |
1da177e4 | 1955 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1956 | atomic_set(&md->uevent_seq, 0); |
1957 | INIT_LIST_HEAD(&md->uevent_list); | |
1958 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1959 | |
4a0b4ddf | 1960 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1961 | if (!md->queue) |
6ed7ade8 | 1962 | goto bad_queue; |
1da177e4 | 1963 | |
4a0b4ddf | 1964 | dm_init_md_queue(md); |
9faf400f | 1965 | |
1da177e4 LT |
1966 | md->disk = alloc_disk(1); |
1967 | if (!md->disk) | |
6ed7ade8 | 1968 | goto bad_disk; |
1da177e4 | 1969 | |
316d315b NK |
1970 | atomic_set(&md->pending[0], 0); |
1971 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1972 | init_waitqueue_head(&md->wait); |
53d5914f | 1973 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 1974 | init_waitqueue_head(&md->eventq); |
2995fa78 | 1975 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 1976 | |
1da177e4 LT |
1977 | md->disk->major = _major; |
1978 | md->disk->first_minor = minor; | |
1979 | md->disk->fops = &dm_blk_dops; | |
1980 | md->disk->queue = md->queue; | |
1981 | md->disk->private_data = md; | |
1982 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1983 | add_disk(md->disk); | |
7e51f257 | 1984 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1985 | |
670368a8 | 1986 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
1987 | if (!md->wq) |
1988 | goto bad_thread; | |
1989 | ||
32a926da MP |
1990 | md->bdev = bdget_disk(md->disk, 0); |
1991 | if (!md->bdev) | |
1992 | goto bad_bdev; | |
1993 | ||
6a8736d1 TH |
1994 | bio_init(&md->flush_bio); |
1995 | md->flush_bio.bi_bdev = md->bdev; | |
1996 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1997 | ||
fd2ed4d2 MP |
1998 | dm_stats_init(&md->stats); |
1999 | ||
ba61fdd1 | 2000 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2001 | spin_lock(&_minor_lock); |
ba61fdd1 | 2002 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2003 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2004 | |
2005 | BUG_ON(old_md != MINOR_ALLOCED); | |
2006 | ||
1da177e4 LT |
2007 | return md; |
2008 | ||
32a926da MP |
2009 | bad_bdev: |
2010 | destroy_workqueue(md->wq); | |
304f3f6a | 2011 | bad_thread: |
03022c54 | 2012 | del_gendisk(md->disk); |
304f3f6a | 2013 | put_disk(md->disk); |
6ed7ade8 | 2014 | bad_disk: |
1312f40e | 2015 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2016 | bad_queue: |
83d5e5b0 MP |
2017 | cleanup_srcu_struct(&md->io_barrier); |
2018 | bad_io_barrier: | |
1da177e4 | 2019 | free_minor(minor); |
6ed7ade8 | 2020 | bad_minor: |
10da4f79 | 2021 | module_put(THIS_MODULE); |
6ed7ade8 | 2022 | bad_module_get: |
1da177e4 LT |
2023 | kfree(md); |
2024 | return NULL; | |
2025 | } | |
2026 | ||
ae9da83f JN |
2027 | static void unlock_fs(struct mapped_device *md); |
2028 | ||
1da177e4 LT |
2029 | static void free_dev(struct mapped_device *md) |
2030 | { | |
f331c029 | 2031 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2032 | |
32a926da MP |
2033 | unlock_fs(md); |
2034 | bdput(md->bdev); | |
304f3f6a | 2035 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
2036 | if (md->io_pool) |
2037 | mempool_destroy(md->io_pool); | |
2038 | if (md->bs) | |
2039 | bioset_free(md->bs); | |
9c47008d | 2040 | blk_integrity_unregister(md->disk); |
1da177e4 | 2041 | del_gendisk(md->disk); |
83d5e5b0 | 2042 | cleanup_srcu_struct(&md->io_barrier); |
63d94e48 | 2043 | free_minor(minor); |
fba9f90e JM |
2044 | |
2045 | spin_lock(&_minor_lock); | |
2046 | md->disk->private_data = NULL; | |
2047 | spin_unlock(&_minor_lock); | |
2048 | ||
1da177e4 | 2049 | put_disk(md->disk); |
1312f40e | 2050 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2051 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2052 | module_put(THIS_MODULE); |
1da177e4 LT |
2053 | kfree(md); |
2054 | } | |
2055 | ||
e6ee8c0b KU |
2056 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2057 | { | |
c0820cf5 | 2058 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2059 | |
5f015204 | 2060 | if (md->io_pool && md->bs) { |
16245bdc JN |
2061 | /* The md already has necessary mempools. */ |
2062 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2063 | /* | |
2064 | * Reload bioset because front_pad may have changed | |
2065 | * because a different table was loaded. | |
2066 | */ | |
2067 | bioset_free(md->bs); | |
2068 | md->bs = p->bs; | |
2069 | p->bs = NULL; | |
2070 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2071 | /* |
2072 | * There's no need to reload with request-based dm | |
2073 | * because the size of front_pad doesn't change. | |
2074 | * Note for future: If you are to reload bioset, | |
2075 | * prep-ed requests in the queue may refer | |
2076 | * to bio from the old bioset, so you must walk | |
2077 | * through the queue to unprep. | |
2078 | */ | |
2079 | } | |
e6ee8c0b | 2080 | goto out; |
c0820cf5 | 2081 | } |
e6ee8c0b | 2082 | |
5f015204 | 2083 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2084 | |
2085 | md->io_pool = p->io_pool; | |
2086 | p->io_pool = NULL; | |
e6ee8c0b KU |
2087 | md->bs = p->bs; |
2088 | p->bs = NULL; | |
2089 | ||
2090 | out: | |
2091 | /* mempool bind completed, now no need any mempools in the table */ | |
2092 | dm_table_free_md_mempools(t); | |
2093 | } | |
2094 | ||
1da177e4 LT |
2095 | /* |
2096 | * Bind a table to the device. | |
2097 | */ | |
2098 | static void event_callback(void *context) | |
2099 | { | |
7a8c3d3b MA |
2100 | unsigned long flags; |
2101 | LIST_HEAD(uevents); | |
1da177e4 LT |
2102 | struct mapped_device *md = (struct mapped_device *) context; |
2103 | ||
7a8c3d3b MA |
2104 | spin_lock_irqsave(&md->uevent_lock, flags); |
2105 | list_splice_init(&md->uevent_list, &uevents); | |
2106 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2107 | ||
ed9e1982 | 2108 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2109 | |
1da177e4 LT |
2110 | atomic_inc(&md->event_nr); |
2111 | wake_up(&md->eventq); | |
2112 | } | |
2113 | ||
c217649b MS |
2114 | /* |
2115 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2116 | */ | |
4e90188b | 2117 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2118 | { |
4e90188b | 2119 | set_capacity(md->disk, size); |
1da177e4 | 2120 | |
db8fef4f | 2121 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2122 | } |
2123 | ||
d5b9dd04 MP |
2124 | /* |
2125 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2126 | * | |
2127 | * If this function returns 0, then the device is either a non-dm | |
2128 | * device without a merge_bvec_fn, or it is a dm device that is | |
2129 | * able to split any bios it receives that are too big. | |
2130 | */ | |
2131 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2132 | { | |
2133 | struct mapped_device *dev_md; | |
2134 | ||
2135 | if (!q->merge_bvec_fn) | |
2136 | return 0; | |
2137 | ||
2138 | if (q->make_request_fn == dm_request) { | |
2139 | dev_md = q->queuedata; | |
2140 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2141 | return 0; | |
2142 | } | |
2143 | ||
2144 | return 1; | |
2145 | } | |
2146 | ||
2147 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2148 | struct dm_dev *dev, sector_t start, | |
2149 | sector_t len, void *data) | |
2150 | { | |
2151 | struct block_device *bdev = dev->bdev; | |
2152 | struct request_queue *q = bdev_get_queue(bdev); | |
2153 | ||
2154 | return dm_queue_merge_is_compulsory(q); | |
2155 | } | |
2156 | ||
2157 | /* | |
2158 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2159 | * on the properties of the underlying devices. | |
2160 | */ | |
2161 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2162 | { | |
2163 | unsigned i = 0; | |
2164 | struct dm_target *ti; | |
2165 | ||
2166 | while (i < dm_table_get_num_targets(table)) { | |
2167 | ti = dm_table_get_target(table, i++); | |
2168 | ||
2169 | if (ti->type->iterate_devices && | |
2170 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2171 | return 0; | |
2172 | } | |
2173 | ||
2174 | return 1; | |
2175 | } | |
2176 | ||
042d2a9b AK |
2177 | /* |
2178 | * Returns old map, which caller must destroy. | |
2179 | */ | |
2180 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2181 | struct queue_limits *limits) | |
1da177e4 | 2182 | { |
042d2a9b | 2183 | struct dm_table *old_map; |
165125e1 | 2184 | struct request_queue *q = md->queue; |
1da177e4 | 2185 | sector_t size; |
d5b9dd04 | 2186 | int merge_is_optional; |
1da177e4 LT |
2187 | |
2188 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2189 | |
2190 | /* | |
2191 | * Wipe any geometry if the size of the table changed. | |
2192 | */ | |
fd2ed4d2 | 2193 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2194 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2195 | ||
32a926da | 2196 | __set_size(md, size); |
d5816876 | 2197 | |
2ca3310e AK |
2198 | dm_table_event_callback(t, event_callback, md); |
2199 | ||
e6ee8c0b KU |
2200 | /* |
2201 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2202 | * for request-based during suspension. So stop it to prevent | |
2203 | * I/O mapping before resume. | |
2204 | * This must be done before setting the queue restrictions, | |
2205 | * because request-based dm may be run just after the setting. | |
2206 | */ | |
2207 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2208 | stop_queue(q); | |
2209 | ||
2210 | __bind_mempools(md, t); | |
2211 | ||
d5b9dd04 MP |
2212 | merge_is_optional = dm_table_merge_is_optional(t); |
2213 | ||
042d2a9b | 2214 | old_map = md->map; |
83d5e5b0 | 2215 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2216 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2217 | ||
754c5fc7 | 2218 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2219 | if (merge_is_optional) |
2220 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2221 | else | |
2222 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
83d5e5b0 | 2223 | dm_sync_table(md); |
1da177e4 | 2224 | |
042d2a9b | 2225 | return old_map; |
1da177e4 LT |
2226 | } |
2227 | ||
a7940155 AK |
2228 | /* |
2229 | * Returns unbound table for the caller to free. | |
2230 | */ | |
2231 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2232 | { |
2233 | struct dm_table *map = md->map; | |
2234 | ||
2235 | if (!map) | |
a7940155 | 2236 | return NULL; |
1da177e4 LT |
2237 | |
2238 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2239 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2240 | dm_sync_table(md); |
a7940155 AK |
2241 | |
2242 | return map; | |
1da177e4 LT |
2243 | } |
2244 | ||
2245 | /* | |
2246 | * Constructor for a new device. | |
2247 | */ | |
2b06cfff | 2248 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2249 | { |
2250 | struct mapped_device *md; | |
2251 | ||
2b06cfff | 2252 | md = alloc_dev(minor); |
1da177e4 LT |
2253 | if (!md) |
2254 | return -ENXIO; | |
2255 | ||
784aae73 MB |
2256 | dm_sysfs_init(md); |
2257 | ||
1da177e4 LT |
2258 | *result = md; |
2259 | return 0; | |
2260 | } | |
2261 | ||
a5664dad MS |
2262 | /* |
2263 | * Functions to manage md->type. | |
2264 | * All are required to hold md->type_lock. | |
2265 | */ | |
2266 | void dm_lock_md_type(struct mapped_device *md) | |
2267 | { | |
2268 | mutex_lock(&md->type_lock); | |
2269 | } | |
2270 | ||
2271 | void dm_unlock_md_type(struct mapped_device *md) | |
2272 | { | |
2273 | mutex_unlock(&md->type_lock); | |
2274 | } | |
2275 | ||
2276 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2277 | { | |
00c4fc3b | 2278 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2279 | md->type = type; |
2280 | } | |
2281 | ||
2282 | unsigned dm_get_md_type(struct mapped_device *md) | |
2283 | { | |
00c4fc3b | 2284 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2285 | return md->type; |
2286 | } | |
2287 | ||
36a0456f AK |
2288 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2289 | { | |
2290 | return md->immutable_target_type; | |
2291 | } | |
2292 | ||
f84cb8a4 MS |
2293 | /* |
2294 | * The queue_limits are only valid as long as you have a reference | |
2295 | * count on 'md'. | |
2296 | */ | |
2297 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2298 | { | |
2299 | BUG_ON(!atomic_read(&md->holders)); | |
2300 | return &md->queue->limits; | |
2301 | } | |
2302 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2303 | ||
4a0b4ddf MS |
2304 | /* |
2305 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2306 | */ | |
2307 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2308 | { | |
2309 | struct request_queue *q = NULL; | |
2310 | ||
2311 | if (md->queue->elevator) | |
2312 | return 1; | |
2313 | ||
2314 | /* Fully initialize the queue */ | |
2315 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2316 | if (!q) | |
2317 | return 0; | |
2318 | ||
2319 | md->queue = q; | |
4a0b4ddf MS |
2320 | dm_init_md_queue(md); |
2321 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2322 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2323 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2324 | |
2325 | elv_register_queue(md->queue); | |
2326 | ||
2327 | return 1; | |
2328 | } | |
2329 | ||
2330 | /* | |
2331 | * Setup the DM device's queue based on md's type | |
2332 | */ | |
2333 | int dm_setup_md_queue(struct mapped_device *md) | |
2334 | { | |
2335 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2336 | !dm_init_request_based_queue(md)) { | |
2337 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2338 | return -EINVAL; | |
2339 | } | |
2340 | ||
2341 | return 0; | |
2342 | } | |
2343 | ||
637842cf | 2344 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2345 | { |
2346 | struct mapped_device *md; | |
1da177e4 LT |
2347 | unsigned minor = MINOR(dev); |
2348 | ||
2349 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2350 | return NULL; | |
2351 | ||
f32c10b0 | 2352 | spin_lock(&_minor_lock); |
1da177e4 LT |
2353 | |
2354 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2355 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2356 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2357 | dm_deleting_md(md) || |
17b2f66f | 2358 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2359 | md = NULL; |
fba9f90e JM |
2360 | goto out; |
2361 | } | |
1da177e4 | 2362 | |
fba9f90e | 2363 | out: |
f32c10b0 | 2364 | spin_unlock(&_minor_lock); |
1da177e4 | 2365 | |
637842cf DT |
2366 | return md; |
2367 | } | |
2368 | ||
d229a958 DT |
2369 | struct mapped_device *dm_get_md(dev_t dev) |
2370 | { | |
2371 | struct mapped_device *md = dm_find_md(dev); | |
2372 | ||
2373 | if (md) | |
2374 | dm_get(md); | |
2375 | ||
2376 | return md; | |
2377 | } | |
3cf2e4ba | 2378 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2379 | |
9ade92a9 | 2380 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2381 | { |
9ade92a9 | 2382 | return md->interface_ptr; |
1da177e4 LT |
2383 | } |
2384 | ||
2385 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2386 | { | |
2387 | md->interface_ptr = ptr; | |
2388 | } | |
2389 | ||
2390 | void dm_get(struct mapped_device *md) | |
2391 | { | |
2392 | atomic_inc(&md->holders); | |
3f77316d | 2393 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2394 | } |
2395 | ||
72d94861 AK |
2396 | const char *dm_device_name(struct mapped_device *md) |
2397 | { | |
2398 | return md->name; | |
2399 | } | |
2400 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2401 | ||
3f77316d | 2402 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2403 | { |
1134e5ae | 2404 | struct dm_table *map; |
83d5e5b0 | 2405 | int srcu_idx; |
1da177e4 | 2406 | |
3f77316d | 2407 | might_sleep(); |
fba9f90e | 2408 | |
3f77316d | 2409 | spin_lock(&_minor_lock); |
83d5e5b0 | 2410 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2411 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2412 | set_bit(DMF_FREEING, &md->flags); | |
2413 | spin_unlock(&_minor_lock); | |
2414 | ||
2415 | if (!dm_suspended_md(md)) { | |
2416 | dm_table_presuspend_targets(map); | |
2417 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2418 | } |
3f77316d | 2419 | |
83d5e5b0 MP |
2420 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2421 | dm_put_live_table(md, srcu_idx); | |
2422 | ||
3f77316d KU |
2423 | /* |
2424 | * Rare, but there may be I/O requests still going to complete, | |
2425 | * for example. Wait for all references to disappear. | |
2426 | * No one should increment the reference count of the mapped_device, | |
2427 | * after the mapped_device state becomes DMF_FREEING. | |
2428 | */ | |
2429 | if (wait) | |
2430 | while (atomic_read(&md->holders)) | |
2431 | msleep(1); | |
2432 | else if (atomic_read(&md->holders)) | |
2433 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2434 | dm_device_name(md), atomic_read(&md->holders)); | |
2435 | ||
2436 | dm_sysfs_exit(md); | |
3f77316d KU |
2437 | dm_table_destroy(__unbind(md)); |
2438 | free_dev(md); | |
2439 | } | |
2440 | ||
2441 | void dm_destroy(struct mapped_device *md) | |
2442 | { | |
2443 | __dm_destroy(md, true); | |
2444 | } | |
2445 | ||
2446 | void dm_destroy_immediate(struct mapped_device *md) | |
2447 | { | |
2448 | __dm_destroy(md, false); | |
2449 | } | |
2450 | ||
2451 | void dm_put(struct mapped_device *md) | |
2452 | { | |
2453 | atomic_dec(&md->holders); | |
1da177e4 | 2454 | } |
79eb885c | 2455 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2456 | |
401600df | 2457 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2458 | { |
2459 | int r = 0; | |
b44ebeb0 MP |
2460 | DECLARE_WAITQUEUE(wait, current); |
2461 | ||
b44ebeb0 | 2462 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2463 | |
2464 | while (1) { | |
401600df | 2465 | set_current_state(interruptible); |
46125c1c | 2466 | |
b4324fee | 2467 | if (!md_in_flight(md)) |
46125c1c MB |
2468 | break; |
2469 | ||
401600df MP |
2470 | if (interruptible == TASK_INTERRUPTIBLE && |
2471 | signal_pending(current)) { | |
46125c1c MB |
2472 | r = -EINTR; |
2473 | break; | |
2474 | } | |
2475 | ||
2476 | io_schedule(); | |
2477 | } | |
2478 | set_current_state(TASK_RUNNING); | |
2479 | ||
b44ebeb0 MP |
2480 | remove_wait_queue(&md->wait, &wait); |
2481 | ||
46125c1c MB |
2482 | return r; |
2483 | } | |
2484 | ||
1da177e4 LT |
2485 | /* |
2486 | * Process the deferred bios | |
2487 | */ | |
ef208587 | 2488 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2489 | { |
ef208587 MP |
2490 | struct mapped_device *md = container_of(work, struct mapped_device, |
2491 | work); | |
6d6f10df | 2492 | struct bio *c; |
83d5e5b0 MP |
2493 | int srcu_idx; |
2494 | struct dm_table *map; | |
1da177e4 | 2495 | |
83d5e5b0 | 2496 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2497 | |
3b00b203 | 2498 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2499 | spin_lock_irq(&md->deferred_lock); |
2500 | c = bio_list_pop(&md->deferred); | |
2501 | spin_unlock_irq(&md->deferred_lock); | |
2502 | ||
6a8736d1 | 2503 | if (!c) |
df12ee99 | 2504 | break; |
022c2611 | 2505 | |
e6ee8c0b KU |
2506 | if (dm_request_based(md)) |
2507 | generic_make_request(c); | |
6a8736d1 | 2508 | else |
83d5e5b0 | 2509 | __split_and_process_bio(md, map, c); |
022c2611 | 2510 | } |
73d410c0 | 2511 | |
83d5e5b0 | 2512 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2513 | } |
2514 | ||
9a1fb464 | 2515 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2516 | { |
3b00b203 | 2517 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2518 | smp_mb__after_atomic(); |
53d5914f | 2519 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2520 | } |
2521 | ||
1da177e4 | 2522 | /* |
042d2a9b | 2523 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2524 | */ |
042d2a9b | 2525 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2526 | { |
87eb5b21 | 2527 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2528 | struct queue_limits limits; |
042d2a9b | 2529 | int r; |
1da177e4 | 2530 | |
e61290a4 | 2531 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2532 | |
2533 | /* device must be suspended */ | |
4f186f8b | 2534 | if (!dm_suspended_md(md)) |
93c534ae | 2535 | goto out; |
1da177e4 | 2536 | |
3ae70656 MS |
2537 | /* |
2538 | * If the new table has no data devices, retain the existing limits. | |
2539 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2540 | * then new I/O is queued based on these limits, and then some paths | |
2541 | * reappear. | |
2542 | */ | |
2543 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2544 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2545 | if (live_map) |
2546 | limits = md->queue->limits; | |
83d5e5b0 | 2547 | dm_put_live_table_fast(md); |
3ae70656 MS |
2548 | } |
2549 | ||
87eb5b21 MC |
2550 | if (!live_map) { |
2551 | r = dm_calculate_queue_limits(table, &limits); | |
2552 | if (r) { | |
2553 | map = ERR_PTR(r); | |
2554 | goto out; | |
2555 | } | |
042d2a9b | 2556 | } |
754c5fc7 | 2557 | |
042d2a9b | 2558 | map = __bind(md, table, &limits); |
1da177e4 | 2559 | |
93c534ae | 2560 | out: |
e61290a4 | 2561 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2562 | return map; |
1da177e4 LT |
2563 | } |
2564 | ||
2565 | /* | |
2566 | * Functions to lock and unlock any filesystem running on the | |
2567 | * device. | |
2568 | */ | |
2ca3310e | 2569 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2570 | { |
e39e2e95 | 2571 | int r; |
1da177e4 LT |
2572 | |
2573 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2574 | |
db8fef4f | 2575 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2576 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2577 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2578 | md->frozen_sb = NULL; |
2579 | return r; | |
dfbe03f6 AK |
2580 | } |
2581 | ||
aa8d7c2f AK |
2582 | set_bit(DMF_FROZEN, &md->flags); |
2583 | ||
1da177e4 LT |
2584 | return 0; |
2585 | } | |
2586 | ||
2ca3310e | 2587 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2588 | { |
aa8d7c2f AK |
2589 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2590 | return; | |
2591 | ||
db8fef4f | 2592 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2593 | md->frozen_sb = NULL; |
aa8d7c2f | 2594 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2595 | } |
2596 | ||
2597 | /* | |
2598 | * We need to be able to change a mapping table under a mounted | |
2599 | * filesystem. For example we might want to move some data in | |
2600 | * the background. Before the table can be swapped with | |
2601 | * dm_bind_table, dm_suspend must be called to flush any in | |
2602 | * flight bios and ensure that any further io gets deferred. | |
2603 | */ | |
cec47e3d KU |
2604 | /* |
2605 | * Suspend mechanism in request-based dm. | |
2606 | * | |
9f518b27 KU |
2607 | * 1. Flush all I/Os by lock_fs() if needed. |
2608 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2609 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2610 | * |
9f518b27 | 2611 | * To abort suspend, start the request_queue. |
cec47e3d | 2612 | */ |
a3d77d35 | 2613 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2614 | { |
2ca3310e | 2615 | struct dm_table *map = NULL; |
46125c1c | 2616 | int r = 0; |
a3d77d35 | 2617 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2618 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2619 | |
e61290a4 | 2620 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2621 | |
4f186f8b | 2622 | if (dm_suspended_md(md)) { |
73d410c0 | 2623 | r = -EINVAL; |
d287483d | 2624 | goto out_unlock; |
73d410c0 | 2625 | } |
1da177e4 | 2626 | |
83d5e5b0 | 2627 | map = md->map; |
1da177e4 | 2628 | |
2e93ccc1 KU |
2629 | /* |
2630 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2631 | * This flag is cleared before dm_suspend returns. | |
2632 | */ | |
2633 | if (noflush) | |
2634 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2635 | ||
cf222b37 AK |
2636 | /* This does not get reverted if there's an error later. */ |
2637 | dm_table_presuspend_targets(map); | |
2638 | ||
32a926da | 2639 | /* |
9f518b27 KU |
2640 | * Flush I/O to the device. |
2641 | * Any I/O submitted after lock_fs() may not be flushed. | |
2642 | * noflush takes precedence over do_lockfs. | |
2643 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2644 | */ |
2645 | if (!noflush && do_lockfs) { | |
2646 | r = lock_fs(md); | |
2647 | if (r) | |
83d5e5b0 | 2648 | goto out_unlock; |
aa8d7c2f | 2649 | } |
1da177e4 LT |
2650 | |
2651 | /* | |
3b00b203 MP |
2652 | * Here we must make sure that no processes are submitting requests |
2653 | * to target drivers i.e. no one may be executing | |
2654 | * __split_and_process_bio. This is called from dm_request and | |
2655 | * dm_wq_work. | |
2656 | * | |
2657 | * To get all processes out of __split_and_process_bio in dm_request, | |
2658 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2659 | * __split_and_process_bio from dm_request and quiesce the thread |
2660 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2661 | * flush_workqueue(md->wq). | |
1da177e4 | 2662 | */ |
1eb787ec | 2663 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
83d5e5b0 | 2664 | synchronize_srcu(&md->io_barrier); |
1da177e4 | 2665 | |
d0bcb878 | 2666 | /* |
29e4013d TH |
2667 | * Stop md->queue before flushing md->wq in case request-based |
2668 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2669 | */ |
cec47e3d | 2670 | if (dm_request_based(md)) |
9f518b27 | 2671 | stop_queue(md->queue); |
cec47e3d | 2672 | |
d0bcb878 KU |
2673 | flush_workqueue(md->wq); |
2674 | ||
1da177e4 | 2675 | /* |
3b00b203 MP |
2676 | * At this point no more requests are entering target request routines. |
2677 | * We call dm_wait_for_completion to wait for all existing requests | |
2678 | * to finish. | |
1da177e4 | 2679 | */ |
401600df | 2680 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2681 | |
6d6f10df | 2682 | if (noflush) |
022c2611 | 2683 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
83d5e5b0 | 2684 | synchronize_srcu(&md->io_barrier); |
2e93ccc1 | 2685 | |
1da177e4 | 2686 | /* were we interrupted ? */ |
46125c1c | 2687 | if (r < 0) { |
9a1fb464 | 2688 | dm_queue_flush(md); |
73d410c0 | 2689 | |
cec47e3d | 2690 | if (dm_request_based(md)) |
9f518b27 | 2691 | start_queue(md->queue); |
cec47e3d | 2692 | |
2ca3310e | 2693 | unlock_fs(md); |
83d5e5b0 | 2694 | goto out_unlock; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2695 | } |
1da177e4 | 2696 | |
3b00b203 MP |
2697 | /* |
2698 | * If dm_wait_for_completion returned 0, the device is completely | |
2699 | * quiescent now. There is no request-processing activity. All new | |
2700 | * requests are being added to md->deferred list. | |
2701 | */ | |
2702 | ||
2ca3310e | 2703 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2704 | |
4d4471cb KU |
2705 | dm_table_postsuspend_targets(map); |
2706 | ||
d287483d | 2707 | out_unlock: |
e61290a4 | 2708 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2709 | return r; |
1da177e4 LT |
2710 | } |
2711 | ||
2712 | int dm_resume(struct mapped_device *md) | |
2713 | { | |
cf222b37 | 2714 | int r = -EINVAL; |
cf222b37 | 2715 | struct dm_table *map = NULL; |
1da177e4 | 2716 | |
e61290a4 | 2717 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2718 | if (!dm_suspended_md(md)) |
cf222b37 | 2719 | goto out; |
cf222b37 | 2720 | |
83d5e5b0 | 2721 | map = md->map; |
2ca3310e | 2722 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2723 | goto out; |
1da177e4 | 2724 | |
8757b776 MB |
2725 | r = dm_table_resume_targets(map); |
2726 | if (r) | |
2727 | goto out; | |
2ca3310e | 2728 | |
9a1fb464 | 2729 | dm_queue_flush(md); |
2ca3310e | 2730 | |
cec47e3d KU |
2731 | /* |
2732 | * Flushing deferred I/Os must be done after targets are resumed | |
2733 | * so that mapping of targets can work correctly. | |
2734 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2735 | */ | |
2736 | if (dm_request_based(md)) | |
2737 | start_queue(md->queue); | |
2738 | ||
2ca3310e AK |
2739 | unlock_fs(md); |
2740 | ||
2741 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2742 | ||
cf222b37 AK |
2743 | r = 0; |
2744 | out: | |
e61290a4 | 2745 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2746 | |
cf222b37 | 2747 | return r; |
1da177e4 LT |
2748 | } |
2749 | ||
fd2ed4d2 MP |
2750 | /* |
2751 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2752 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2753 | * It may be used only from the kernel. | |
2754 | * | |
2755 | * Internal suspend holds md->suspend_lock, which prevents interaction with | |
2756 | * userspace-driven suspend. | |
2757 | */ | |
2758 | ||
2759 | void dm_internal_suspend(struct mapped_device *md) | |
2760 | { | |
2761 | mutex_lock(&md->suspend_lock); | |
2762 | if (dm_suspended_md(md)) | |
2763 | return; | |
2764 | ||
2765 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2766 | synchronize_srcu(&md->io_barrier); | |
2767 | flush_workqueue(md->wq); | |
2768 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2769 | } | |
2770 | ||
2771 | void dm_internal_resume(struct mapped_device *md) | |
2772 | { | |
2773 | if (dm_suspended_md(md)) | |
2774 | goto done; | |
2775 | ||
2776 | dm_queue_flush(md); | |
2777 | ||
2778 | done: | |
2779 | mutex_unlock(&md->suspend_lock); | |
2780 | } | |
2781 | ||
1da177e4 LT |
2782 | /*----------------------------------------------------------------- |
2783 | * Event notification. | |
2784 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2785 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2786 | unsigned cookie) |
69267a30 | 2787 | { |
60935eb2 MB |
2788 | char udev_cookie[DM_COOKIE_LENGTH]; |
2789 | char *envp[] = { udev_cookie, NULL }; | |
2790 | ||
2791 | if (!cookie) | |
3abf85b5 | 2792 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2793 | else { |
2794 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2795 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2796 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2797 | action, envp); | |
60935eb2 | 2798 | } |
69267a30 AK |
2799 | } |
2800 | ||
7a8c3d3b MA |
2801 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2802 | { | |
2803 | return atomic_add_return(1, &md->uevent_seq); | |
2804 | } | |
2805 | ||
1da177e4 LT |
2806 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2807 | { | |
2808 | return atomic_read(&md->event_nr); | |
2809 | } | |
2810 | ||
2811 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2812 | { | |
2813 | return wait_event_interruptible(md->eventq, | |
2814 | (event_nr != atomic_read(&md->event_nr))); | |
2815 | } | |
2816 | ||
7a8c3d3b MA |
2817 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2818 | { | |
2819 | unsigned long flags; | |
2820 | ||
2821 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2822 | list_add(elist, &md->uevent_list); | |
2823 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2824 | } | |
2825 | ||
1da177e4 LT |
2826 | /* |
2827 | * The gendisk is only valid as long as you have a reference | |
2828 | * count on 'md'. | |
2829 | */ | |
2830 | struct gendisk *dm_disk(struct mapped_device *md) | |
2831 | { | |
2832 | return md->disk; | |
2833 | } | |
2834 | ||
784aae73 MB |
2835 | struct kobject *dm_kobject(struct mapped_device *md) |
2836 | { | |
2995fa78 | 2837 | return &md->kobj_holder.kobj; |
784aae73 MB |
2838 | } |
2839 | ||
784aae73 MB |
2840 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2841 | { | |
2842 | struct mapped_device *md; | |
2843 | ||
2995fa78 | 2844 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2845 | |
4d89b7b4 | 2846 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2847 | dm_deleting_md(md)) |
4d89b7b4 MB |
2848 | return NULL; |
2849 | ||
784aae73 MB |
2850 | dm_get(md); |
2851 | return md; | |
2852 | } | |
2853 | ||
4f186f8b | 2854 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2855 | { |
2856 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2857 | } | |
2858 | ||
2c140a24 MP |
2859 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2860 | { | |
2861 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2862 | } | |
2863 | ||
64dbce58 KU |
2864 | int dm_suspended(struct dm_target *ti) |
2865 | { | |
ecdb2e25 | 2866 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2867 | } |
2868 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2869 | ||
2e93ccc1 KU |
2870 | int dm_noflush_suspending(struct dm_target *ti) |
2871 | { | |
ecdb2e25 | 2872 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2873 | } |
2874 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2875 | ||
c0820cf5 | 2876 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 2877 | { |
5f015204 JN |
2878 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
2879 | struct kmem_cache *cachep; | |
2880 | unsigned int pool_size; | |
2881 | unsigned int front_pad; | |
e6ee8c0b KU |
2882 | |
2883 | if (!pools) | |
2884 | return NULL; | |
2885 | ||
23e5083b | 2886 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 | 2887 | cachep = _io_cache; |
e8603136 | 2888 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 JN |
2889 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
2890 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
2891 | cachep = _rq_tio_cache; | |
f4790826 | 2892 | pool_size = dm_get_reserved_rq_based_ios(); |
5f015204 JN |
2893 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
2894 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
2895 | WARN_ON(per_bio_data_size != 0); | |
2896 | } else | |
2897 | goto out; | |
e6ee8c0b | 2898 | |
6cfa5857 | 2899 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
2900 | if (!pools->io_pool) |
2901 | goto out; | |
e6ee8c0b | 2902 | |
5f015204 | 2903 | pools->bs = bioset_create(pool_size, front_pad); |
e6ee8c0b | 2904 | if (!pools->bs) |
5f015204 | 2905 | goto out; |
e6ee8c0b | 2906 | |
a91a2785 | 2907 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2908 | goto out; |
a91a2785 | 2909 | |
e6ee8c0b KU |
2910 | return pools; |
2911 | ||
5f015204 JN |
2912 | out: |
2913 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
2914 | |
2915 | return NULL; | |
2916 | } | |
2917 | ||
2918 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2919 | { | |
2920 | if (!pools) | |
2921 | return; | |
2922 | ||
2923 | if (pools->io_pool) | |
2924 | mempool_destroy(pools->io_pool); | |
2925 | ||
e6ee8c0b KU |
2926 | if (pools->bs) |
2927 | bioset_free(pools->bs); | |
2928 | ||
2929 | kfree(pools); | |
2930 | } | |
2931 | ||
83d5cde4 | 2932 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2933 | .open = dm_blk_open, |
2934 | .release = dm_blk_close, | |
aa129a22 | 2935 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2936 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2937 | .owner = THIS_MODULE |
2938 | }; | |
2939 | ||
1da177e4 LT |
2940 | /* |
2941 | * module hooks | |
2942 | */ | |
2943 | module_init(dm_init); | |
2944 | module_exit(dm_exit); | |
2945 | ||
2946 | module_param(major, uint, 0); | |
2947 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 2948 | |
e8603136 MS |
2949 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
2950 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
2951 | ||
f4790826 MS |
2952 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
2953 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
2954 | ||
1da177e4 LT |
2955 | MODULE_DESCRIPTION(DM_NAME " driver"); |
2956 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2957 | MODULE_LICENSE("GPL"); |