]>
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 | */ | |
6fa99520 | 143 | struct dm_table __rcu *map; |
2a7faeb1 | 144 | |
86f1152b BM |
145 | struct list_head table_devices; |
146 | struct mutex table_devices_lock; | |
147 | ||
1da177e4 LT |
148 | unsigned long flags; |
149 | ||
165125e1 | 150 | struct request_queue *queue; |
a5664dad | 151 | unsigned type; |
4a0b4ddf | 152 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
153 | struct mutex type_lock; |
154 | ||
36a0456f AK |
155 | struct target_type *immutable_target_type; |
156 | ||
1da177e4 | 157 | struct gendisk *disk; |
7e51f257 | 158 | char name[16]; |
1da177e4 LT |
159 | |
160 | void *interface_ptr; | |
161 | ||
162 | /* | |
163 | * A list of ios that arrived while we were suspended. | |
164 | */ | |
316d315b | 165 | atomic_t pending[2]; |
1da177e4 | 166 | wait_queue_head_t wait; |
53d5914f | 167 | struct work_struct work; |
74859364 | 168 | struct bio_list deferred; |
022c2611 | 169 | spinlock_t deferred_lock; |
1da177e4 | 170 | |
af7e466a | 171 | /* |
29e4013d | 172 | * Processing queue (flush) |
304f3f6a MB |
173 | */ |
174 | struct workqueue_struct *wq; | |
175 | ||
1da177e4 LT |
176 | /* |
177 | * io objects are allocated from here. | |
178 | */ | |
179 | mempool_t *io_pool; | |
1da177e4 | 180 | |
9faf400f SB |
181 | struct bio_set *bs; |
182 | ||
1da177e4 LT |
183 | /* |
184 | * Event handling. | |
185 | */ | |
186 | atomic_t event_nr; | |
187 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
188 | atomic_t uevent_seq; |
189 | struct list_head uevent_list; | |
190 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
191 | |
192 | /* | |
193 | * freeze/thaw support require holding onto a super block | |
194 | */ | |
195 | struct super_block *frozen_sb; | |
db8fef4f | 196 | struct block_device *bdev; |
3ac51e74 DW |
197 | |
198 | /* forced geometry settings */ | |
199 | struct hd_geometry geometry; | |
784aae73 | 200 | |
2995fa78 MP |
201 | /* kobject and completion */ |
202 | struct dm_kobject_holder kobj_holder; | |
be35f486 | 203 | |
d87f4c14 TH |
204 | /* zero-length flush that will be cloned and submitted to targets */ |
205 | struct bio flush_bio; | |
fd2ed4d2 MP |
206 | |
207 | struct dm_stats stats; | |
1da177e4 LT |
208 | }; |
209 | ||
e6ee8c0b KU |
210 | /* |
211 | * For mempools pre-allocation at the table loading time. | |
212 | */ | |
213 | struct dm_md_mempools { | |
214 | mempool_t *io_pool; | |
e6ee8c0b KU |
215 | struct bio_set *bs; |
216 | }; | |
217 | ||
86f1152b BM |
218 | struct table_device { |
219 | struct list_head list; | |
220 | atomic_t count; | |
221 | struct dm_dev dm_dev; | |
222 | }; | |
223 | ||
6cfa5857 MS |
224 | #define RESERVED_BIO_BASED_IOS 16 |
225 | #define RESERVED_REQUEST_BASED_IOS 256 | |
f4790826 | 226 | #define RESERVED_MAX_IOS 1024 |
e18b890b | 227 | static struct kmem_cache *_io_cache; |
8fbf26ad | 228 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 229 | |
e8603136 MS |
230 | /* |
231 | * Bio-based DM's mempools' reserved IOs set by the user. | |
232 | */ | |
233 | static unsigned reserved_bio_based_ios = RESERVED_BIO_BASED_IOS; | |
234 | ||
f4790826 MS |
235 | /* |
236 | * Request-based DM's mempools' reserved IOs set by the user. | |
237 | */ | |
238 | static unsigned reserved_rq_based_ios = RESERVED_REQUEST_BASED_IOS; | |
239 | ||
240 | static unsigned __dm_get_reserved_ios(unsigned *reserved_ios, | |
241 | unsigned def, unsigned max) | |
242 | { | |
243 | unsigned ios = ACCESS_ONCE(*reserved_ios); | |
244 | unsigned modified_ios = 0; | |
245 | ||
246 | if (!ios) | |
247 | modified_ios = def; | |
248 | else if (ios > max) | |
249 | modified_ios = max; | |
250 | ||
251 | if (modified_ios) { | |
252 | (void)cmpxchg(reserved_ios, ios, modified_ios); | |
253 | ios = modified_ios; | |
254 | } | |
255 | ||
256 | return ios; | |
257 | } | |
258 | ||
e8603136 MS |
259 | unsigned dm_get_reserved_bio_based_ios(void) |
260 | { | |
261 | return __dm_get_reserved_ios(&reserved_bio_based_ios, | |
262 | RESERVED_BIO_BASED_IOS, RESERVED_MAX_IOS); | |
263 | } | |
264 | EXPORT_SYMBOL_GPL(dm_get_reserved_bio_based_ios); | |
265 | ||
f4790826 MS |
266 | unsigned dm_get_reserved_rq_based_ios(void) |
267 | { | |
268 | return __dm_get_reserved_ios(&reserved_rq_based_ios, | |
269 | RESERVED_REQUEST_BASED_IOS, RESERVED_MAX_IOS); | |
270 | } | |
271 | EXPORT_SYMBOL_GPL(dm_get_reserved_rq_based_ios); | |
272 | ||
1da177e4 LT |
273 | static int __init local_init(void) |
274 | { | |
51157b4a | 275 | int r = -ENOMEM; |
1da177e4 | 276 | |
1da177e4 | 277 | /* allocate a slab for the dm_ios */ |
028867ac | 278 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 279 | if (!_io_cache) |
51157b4a | 280 | return r; |
1da177e4 | 281 | |
8fbf26ad KU |
282 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
283 | if (!_rq_tio_cache) | |
dba14160 | 284 | goto out_free_io_cache; |
8fbf26ad | 285 | |
51e5b2bd | 286 | r = dm_uevent_init(); |
51157b4a | 287 | if (r) |
23e5083b | 288 | goto out_free_rq_tio_cache; |
51e5b2bd | 289 | |
acfe0ad7 MP |
290 | deferred_remove_workqueue = alloc_workqueue("kdmremove", WQ_UNBOUND, 1); |
291 | if (!deferred_remove_workqueue) { | |
292 | r = -ENOMEM; | |
293 | goto out_uevent_exit; | |
294 | } | |
295 | ||
1da177e4 LT |
296 | _major = major; |
297 | r = register_blkdev(_major, _name); | |
51157b4a | 298 | if (r < 0) |
acfe0ad7 | 299 | goto out_free_workqueue; |
1da177e4 LT |
300 | |
301 | if (!_major) | |
302 | _major = r; | |
303 | ||
304 | return 0; | |
51157b4a | 305 | |
acfe0ad7 MP |
306 | out_free_workqueue: |
307 | destroy_workqueue(deferred_remove_workqueue); | |
51157b4a KU |
308 | out_uevent_exit: |
309 | dm_uevent_exit(); | |
8fbf26ad KU |
310 | out_free_rq_tio_cache: |
311 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
312 | out_free_io_cache: |
313 | kmem_cache_destroy(_io_cache); | |
314 | ||
315 | return r; | |
1da177e4 LT |
316 | } |
317 | ||
318 | static void local_exit(void) | |
319 | { | |
2c140a24 | 320 | flush_scheduled_work(); |
acfe0ad7 | 321 | destroy_workqueue(deferred_remove_workqueue); |
2c140a24 | 322 | |
8fbf26ad | 323 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 324 | kmem_cache_destroy(_io_cache); |
00d59405 | 325 | unregister_blkdev(_major, _name); |
51e5b2bd | 326 | dm_uevent_exit(); |
1da177e4 LT |
327 | |
328 | _major = 0; | |
329 | ||
330 | DMINFO("cleaned up"); | |
331 | } | |
332 | ||
b9249e55 | 333 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
334 | local_init, |
335 | dm_target_init, | |
336 | dm_linear_init, | |
337 | dm_stripe_init, | |
952b3557 | 338 | dm_io_init, |
945fa4d2 | 339 | dm_kcopyd_init, |
1da177e4 | 340 | dm_interface_init, |
fd2ed4d2 | 341 | dm_statistics_init, |
1da177e4 LT |
342 | }; |
343 | ||
b9249e55 | 344 | static void (*_exits[])(void) = { |
1da177e4 LT |
345 | local_exit, |
346 | dm_target_exit, | |
347 | dm_linear_exit, | |
348 | dm_stripe_exit, | |
952b3557 | 349 | dm_io_exit, |
945fa4d2 | 350 | dm_kcopyd_exit, |
1da177e4 | 351 | dm_interface_exit, |
fd2ed4d2 | 352 | dm_statistics_exit, |
1da177e4 LT |
353 | }; |
354 | ||
355 | static int __init dm_init(void) | |
356 | { | |
357 | const int count = ARRAY_SIZE(_inits); | |
358 | ||
359 | int r, i; | |
360 | ||
361 | for (i = 0; i < count; i++) { | |
362 | r = _inits[i](); | |
363 | if (r) | |
364 | goto bad; | |
365 | } | |
366 | ||
367 | return 0; | |
368 | ||
369 | bad: | |
370 | while (i--) | |
371 | _exits[i](); | |
372 | ||
373 | return r; | |
374 | } | |
375 | ||
376 | static void __exit dm_exit(void) | |
377 | { | |
378 | int i = ARRAY_SIZE(_exits); | |
379 | ||
380 | while (i--) | |
381 | _exits[i](); | |
d15b774c AK |
382 | |
383 | /* | |
384 | * Should be empty by this point. | |
385 | */ | |
d15b774c | 386 | idr_destroy(&_minor_idr); |
1da177e4 LT |
387 | } |
388 | ||
389 | /* | |
390 | * Block device functions | |
391 | */ | |
432a212c MA |
392 | int dm_deleting_md(struct mapped_device *md) |
393 | { | |
394 | return test_bit(DMF_DELETING, &md->flags); | |
395 | } | |
396 | ||
fe5f9f2c | 397 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
398 | { |
399 | struct mapped_device *md; | |
400 | ||
fba9f90e JM |
401 | spin_lock(&_minor_lock); |
402 | ||
fe5f9f2c | 403 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
404 | if (!md) |
405 | goto out; | |
406 | ||
5c6bd75d | 407 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 408 | dm_deleting_md(md)) { |
fba9f90e JM |
409 | md = NULL; |
410 | goto out; | |
411 | } | |
412 | ||
1da177e4 | 413 | dm_get(md); |
5c6bd75d | 414 | atomic_inc(&md->open_count); |
fba9f90e JM |
415 | |
416 | out: | |
417 | spin_unlock(&_minor_lock); | |
418 | ||
419 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
420 | } |
421 | ||
db2a144b | 422 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 423 | { |
fe5f9f2c | 424 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 425 | |
4a1aeb98 MB |
426 | spin_lock(&_minor_lock); |
427 | ||
2c140a24 MP |
428 | if (atomic_dec_and_test(&md->open_count) && |
429 | (test_bit(DMF_DEFERRED_REMOVE, &md->flags))) | |
acfe0ad7 | 430 | queue_work(deferred_remove_workqueue, &deferred_remove_work); |
2c140a24 | 431 | |
1da177e4 | 432 | dm_put(md); |
4a1aeb98 MB |
433 | |
434 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
435 | } |
436 | ||
5c6bd75d AK |
437 | int dm_open_count(struct mapped_device *md) |
438 | { | |
439 | return atomic_read(&md->open_count); | |
440 | } | |
441 | ||
442 | /* | |
443 | * Guarantees nothing is using the device before it's deleted. | |
444 | */ | |
2c140a24 | 445 | int dm_lock_for_deletion(struct mapped_device *md, bool mark_deferred, bool only_deferred) |
5c6bd75d AK |
446 | { |
447 | int r = 0; | |
448 | ||
449 | spin_lock(&_minor_lock); | |
450 | ||
2c140a24 | 451 | if (dm_open_count(md)) { |
5c6bd75d | 452 | r = -EBUSY; |
2c140a24 MP |
453 | if (mark_deferred) |
454 | set_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
455 | } else if (only_deferred && !test_bit(DMF_DEFERRED_REMOVE, &md->flags)) | |
456 | r = -EEXIST; | |
5c6bd75d AK |
457 | else |
458 | set_bit(DMF_DELETING, &md->flags); | |
459 | ||
460 | spin_unlock(&_minor_lock); | |
461 | ||
462 | return r; | |
463 | } | |
464 | ||
2c140a24 MP |
465 | int dm_cancel_deferred_remove(struct mapped_device *md) |
466 | { | |
467 | int r = 0; | |
468 | ||
469 | spin_lock(&_minor_lock); | |
470 | ||
471 | if (test_bit(DMF_DELETING, &md->flags)) | |
472 | r = -EBUSY; | |
473 | else | |
474 | clear_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
475 | ||
476 | spin_unlock(&_minor_lock); | |
477 | ||
478 | return r; | |
479 | } | |
480 | ||
481 | static void do_deferred_remove(struct work_struct *w) | |
482 | { | |
483 | dm_deferred_remove(); | |
484 | } | |
485 | ||
fd2ed4d2 MP |
486 | sector_t dm_get_size(struct mapped_device *md) |
487 | { | |
488 | return get_capacity(md->disk); | |
489 | } | |
490 | ||
9974fa2c MS |
491 | struct request_queue *dm_get_md_queue(struct mapped_device *md) |
492 | { | |
493 | return md->queue; | |
494 | } | |
495 | ||
fd2ed4d2 MP |
496 | struct dm_stats *dm_get_stats(struct mapped_device *md) |
497 | { | |
498 | return &md->stats; | |
499 | } | |
500 | ||
3ac51e74 DW |
501 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
502 | { | |
503 | struct mapped_device *md = bdev->bd_disk->private_data; | |
504 | ||
505 | return dm_get_geometry(md, geo); | |
506 | } | |
507 | ||
fe5f9f2c | 508 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
509 | unsigned int cmd, unsigned long arg) |
510 | { | |
fe5f9f2c | 511 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 512 | int srcu_idx; |
6c182cd8 | 513 | struct dm_table *map; |
aa129a22 MB |
514 | struct dm_target *tgt; |
515 | int r = -ENOTTY; | |
516 | ||
6c182cd8 | 517 | retry: |
83d5e5b0 MP |
518 | map = dm_get_live_table(md, &srcu_idx); |
519 | ||
aa129a22 MB |
520 | if (!map || !dm_table_get_size(map)) |
521 | goto out; | |
522 | ||
523 | /* We only support devices that have a single target */ | |
524 | if (dm_table_get_num_targets(map) != 1) | |
525 | goto out; | |
526 | ||
527 | tgt = dm_table_get_target(map, 0); | |
4d341d82 MS |
528 | if (!tgt->type->ioctl) |
529 | goto out; | |
aa129a22 | 530 | |
4f186f8b | 531 | if (dm_suspended_md(md)) { |
aa129a22 MB |
532 | r = -EAGAIN; |
533 | goto out; | |
534 | } | |
535 | ||
4d341d82 | 536 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
537 | |
538 | out: | |
83d5e5b0 | 539 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 540 | |
6c182cd8 HR |
541 | if (r == -ENOTCONN) { |
542 | msleep(10); | |
543 | goto retry; | |
544 | } | |
545 | ||
aa129a22 MB |
546 | return r; |
547 | } | |
548 | ||
028867ac | 549 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
550 | { |
551 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
552 | } | |
553 | ||
028867ac | 554 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
555 | { |
556 | mempool_free(io, md->io_pool); | |
557 | } | |
558 | ||
028867ac | 559 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 560 | { |
dba14160 | 561 | bio_put(&tio->clone); |
1da177e4 LT |
562 | } |
563 | ||
08885643 KU |
564 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
565 | gfp_t gfp_mask) | |
cec47e3d | 566 | { |
5f015204 | 567 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
568 | } |
569 | ||
570 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
571 | { | |
5f015204 | 572 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
573 | } |
574 | ||
90abb8c4 KU |
575 | static int md_in_flight(struct mapped_device *md) |
576 | { | |
577 | return atomic_read(&md->pending[READ]) + | |
578 | atomic_read(&md->pending[WRITE]); | |
579 | } | |
580 | ||
3eaf840e JNN |
581 | static void start_io_acct(struct dm_io *io) |
582 | { | |
583 | struct mapped_device *md = io->md; | |
fd2ed4d2 | 584 | struct bio *bio = io->bio; |
c9959059 | 585 | int cpu; |
fd2ed4d2 | 586 | int rw = bio_data_dir(bio); |
3eaf840e JNN |
587 | |
588 | io->start_time = jiffies; | |
589 | ||
074a7aca TH |
590 | cpu = part_stat_lock(); |
591 | part_round_stats(cpu, &dm_disk(md)->part0); | |
592 | part_stat_unlock(); | |
1e9bb880 SL |
593 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
594 | atomic_inc_return(&md->pending[rw])); | |
fd2ed4d2 MP |
595 | |
596 | if (unlikely(dm_stats_used(&md->stats))) | |
4f024f37 | 597 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 | 598 | bio_sectors(bio), false, 0, &io->stats_aux); |
3eaf840e JNN |
599 | } |
600 | ||
d221d2e7 | 601 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
602 | { |
603 | struct mapped_device *md = io->md; | |
604 | struct bio *bio = io->bio; | |
605 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 606 | int pending, cpu; |
3eaf840e JNN |
607 | int rw = bio_data_dir(bio); |
608 | ||
074a7aca TH |
609 | cpu = part_stat_lock(); |
610 | part_round_stats(cpu, &dm_disk(md)->part0); | |
611 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
612 | part_stat_unlock(); | |
3eaf840e | 613 | |
fd2ed4d2 | 614 | if (unlikely(dm_stats_used(&md->stats))) |
4f024f37 | 615 | dm_stats_account_io(&md->stats, bio->bi_rw, bio->bi_iter.bi_sector, |
fd2ed4d2 MP |
616 | bio_sectors(bio), true, duration, &io->stats_aux); |
617 | ||
af7e466a MP |
618 | /* |
619 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 620 | * a flush. |
af7e466a | 621 | */ |
1e9bb880 SL |
622 | pending = atomic_dec_return(&md->pending[rw]); |
623 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 624 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 625 | |
d221d2e7 MP |
626 | /* nudge anyone waiting on suspend queue */ |
627 | if (!pending) | |
628 | wake_up(&md->wait); | |
3eaf840e JNN |
629 | } |
630 | ||
1da177e4 LT |
631 | /* |
632 | * Add the bio to the list of deferred io. | |
633 | */ | |
92c63902 | 634 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 635 | { |
05447420 | 636 | unsigned long flags; |
1da177e4 | 637 | |
05447420 | 638 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 639 | bio_list_add(&md->deferred, bio); |
05447420 | 640 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 641 | queue_work(md->wq, &md->work); |
1da177e4 LT |
642 | } |
643 | ||
644 | /* | |
645 | * Everyone (including functions in this file), should use this | |
646 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 647 | * dm_put_live_table() when finished. |
1da177e4 | 648 | */ |
83d5e5b0 | 649 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 650 | { |
83d5e5b0 MP |
651 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
652 | ||
653 | return srcu_dereference(md->map, &md->io_barrier); | |
654 | } | |
1da177e4 | 655 | |
83d5e5b0 MP |
656 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
657 | { | |
658 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
659 | } | |
660 | ||
661 | void dm_sync_table(struct mapped_device *md) | |
662 | { | |
663 | synchronize_srcu(&md->io_barrier); | |
664 | synchronize_rcu_expedited(); | |
665 | } | |
666 | ||
667 | /* | |
668 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
669 | * The caller must not block between these two functions. | |
670 | */ | |
671 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
672 | { | |
673 | rcu_read_lock(); | |
674 | return rcu_dereference(md->map); | |
675 | } | |
1da177e4 | 676 | |
83d5e5b0 MP |
677 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
678 | { | |
679 | rcu_read_unlock(); | |
1da177e4 LT |
680 | } |
681 | ||
86f1152b BM |
682 | /* |
683 | * Open a table device so we can use it as a map destination. | |
684 | */ | |
685 | static int open_table_device(struct table_device *td, dev_t dev, | |
686 | struct mapped_device *md) | |
687 | { | |
688 | static char *_claim_ptr = "I belong to device-mapper"; | |
689 | struct block_device *bdev; | |
690 | ||
691 | int r; | |
692 | ||
693 | BUG_ON(td->dm_dev.bdev); | |
694 | ||
695 | bdev = blkdev_get_by_dev(dev, td->dm_dev.mode | FMODE_EXCL, _claim_ptr); | |
696 | if (IS_ERR(bdev)) | |
697 | return PTR_ERR(bdev); | |
698 | ||
699 | r = bd_link_disk_holder(bdev, dm_disk(md)); | |
700 | if (r) { | |
701 | blkdev_put(bdev, td->dm_dev.mode | FMODE_EXCL); | |
702 | return r; | |
703 | } | |
704 | ||
705 | td->dm_dev.bdev = bdev; | |
706 | return 0; | |
707 | } | |
708 | ||
709 | /* | |
710 | * Close a table device that we've been using. | |
711 | */ | |
712 | static void close_table_device(struct table_device *td, struct mapped_device *md) | |
713 | { | |
714 | if (!td->dm_dev.bdev) | |
715 | return; | |
716 | ||
717 | bd_unlink_disk_holder(td->dm_dev.bdev, dm_disk(md)); | |
718 | blkdev_put(td->dm_dev.bdev, td->dm_dev.mode | FMODE_EXCL); | |
719 | td->dm_dev.bdev = NULL; | |
720 | } | |
721 | ||
722 | static struct table_device *find_table_device(struct list_head *l, dev_t dev, | |
723 | fmode_t mode) { | |
724 | struct table_device *td; | |
725 | ||
726 | list_for_each_entry(td, l, list) | |
727 | if (td->dm_dev.bdev->bd_dev == dev && td->dm_dev.mode == mode) | |
728 | return td; | |
729 | ||
730 | return NULL; | |
731 | } | |
732 | ||
733 | int dm_get_table_device(struct mapped_device *md, dev_t dev, fmode_t mode, | |
734 | struct dm_dev **result) { | |
735 | int r; | |
736 | struct table_device *td; | |
737 | ||
738 | mutex_lock(&md->table_devices_lock); | |
739 | td = find_table_device(&md->table_devices, dev, mode); | |
740 | if (!td) { | |
741 | td = kmalloc(sizeof(*td), GFP_KERNEL); | |
742 | if (!td) { | |
743 | mutex_unlock(&md->table_devices_lock); | |
744 | return -ENOMEM; | |
745 | } | |
746 | ||
747 | td->dm_dev.mode = mode; | |
748 | td->dm_dev.bdev = NULL; | |
749 | ||
750 | if ((r = open_table_device(td, dev, md))) { | |
751 | mutex_unlock(&md->table_devices_lock); | |
752 | kfree(td); | |
753 | return r; | |
754 | } | |
755 | ||
756 | format_dev_t(td->dm_dev.name, dev); | |
757 | ||
758 | atomic_set(&td->count, 0); | |
759 | list_add(&td->list, &md->table_devices); | |
760 | } | |
761 | atomic_inc(&td->count); | |
762 | mutex_unlock(&md->table_devices_lock); | |
763 | ||
764 | *result = &td->dm_dev; | |
765 | return 0; | |
766 | } | |
767 | EXPORT_SYMBOL_GPL(dm_get_table_device); | |
768 | ||
769 | void dm_put_table_device(struct mapped_device *md, struct dm_dev *d) | |
770 | { | |
771 | struct table_device *td = container_of(d, struct table_device, dm_dev); | |
772 | ||
773 | mutex_lock(&md->table_devices_lock); | |
774 | if (atomic_dec_and_test(&td->count)) { | |
775 | close_table_device(td, md); | |
776 | list_del(&td->list); | |
777 | kfree(td); | |
778 | } | |
779 | mutex_unlock(&md->table_devices_lock); | |
780 | } | |
781 | EXPORT_SYMBOL(dm_put_table_device); | |
782 | ||
783 | static void free_table_devices(struct list_head *devices) | |
784 | { | |
785 | struct list_head *tmp, *next; | |
786 | ||
787 | list_for_each_safe(tmp, next, devices) { | |
788 | struct table_device *td = list_entry(tmp, struct table_device, list); | |
789 | ||
790 | DMWARN("dm_destroy: %s still exists with %d references", | |
791 | td->dm_dev.name, atomic_read(&td->count)); | |
792 | kfree(td); | |
793 | } | |
794 | } | |
795 | ||
3ac51e74 DW |
796 | /* |
797 | * Get the geometry associated with a dm device | |
798 | */ | |
799 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
800 | { | |
801 | *geo = md->geometry; | |
802 | ||
803 | return 0; | |
804 | } | |
805 | ||
806 | /* | |
807 | * Set the geometry of a device. | |
808 | */ | |
809 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
810 | { | |
811 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
812 | ||
813 | if (geo->start > sz) { | |
814 | DMWARN("Start sector is beyond the geometry limits."); | |
815 | return -EINVAL; | |
816 | } | |
817 | ||
818 | md->geometry = *geo; | |
819 | ||
820 | return 0; | |
821 | } | |
822 | ||
1da177e4 LT |
823 | /*----------------------------------------------------------------- |
824 | * CRUD START: | |
825 | * A more elegant soln is in the works that uses the queue | |
826 | * merge fn, unfortunately there are a couple of changes to | |
827 | * the block layer that I want to make for this. So in the | |
828 | * interests of getting something for people to use I give | |
829 | * you this clearly demarcated crap. | |
830 | *---------------------------------------------------------------*/ | |
831 | ||
2e93ccc1 KU |
832 | static int __noflush_suspending(struct mapped_device *md) |
833 | { | |
834 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
835 | } | |
836 | ||
1da177e4 LT |
837 | /* |
838 | * Decrements the number of outstanding ios that a bio has been | |
839 | * cloned into, completing the original io if necc. | |
840 | */ | |
858119e1 | 841 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 842 | { |
2e93ccc1 | 843 | unsigned long flags; |
b35f8caa MB |
844 | int io_error; |
845 | struct bio *bio; | |
846 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
847 | |
848 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
849 | if (unlikely(error)) { |
850 | spin_lock_irqsave(&io->endio_lock, flags); | |
851 | if (!(io->error > 0 && __noflush_suspending(md))) | |
852 | io->error = error; | |
853 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
854 | } | |
1da177e4 LT |
855 | |
856 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
857 | if (io->error == DM_ENDIO_REQUEUE) { |
858 | /* | |
859 | * Target requested pushing back the I/O. | |
2e93ccc1 | 860 | */ |
022c2611 | 861 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
862 | if (__noflush_suspending(md)) |
863 | bio_list_add_head(&md->deferred, io->bio); | |
864 | else | |
2e93ccc1 KU |
865 | /* noflush suspend was interrupted. */ |
866 | io->error = -EIO; | |
022c2611 | 867 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
868 | } |
869 | ||
b35f8caa MB |
870 | io_error = io->error; |
871 | bio = io->bio; | |
6a8736d1 TH |
872 | end_io_acct(io); |
873 | free_io(md, io); | |
874 | ||
875 | if (io_error == DM_ENDIO_REQUEUE) | |
876 | return; | |
2e93ccc1 | 877 | |
4f024f37 | 878 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_iter.bi_size) { |
af7e466a | 879 | /* |
6a8736d1 TH |
880 | * Preflush done for flush with data, reissue |
881 | * without REQ_FLUSH. | |
af7e466a | 882 | */ |
6a8736d1 TH |
883 | bio->bi_rw &= ~REQ_FLUSH; |
884 | queue_io(md, bio); | |
af7e466a | 885 | } else { |
b372d360 | 886 | /* done with normal IO or empty flush */ |
0a82a8d1 | 887 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 888 | bio_endio(bio, io_error); |
b35f8caa | 889 | } |
1da177e4 LT |
890 | } |
891 | } | |
892 | ||
7eee4ae2 MS |
893 | static void disable_write_same(struct mapped_device *md) |
894 | { | |
895 | struct queue_limits *limits = dm_get_queue_limits(md); | |
896 | ||
897 | /* device doesn't really support WRITE SAME, disable it */ | |
898 | limits->max_write_same_sectors = 0; | |
899 | } | |
900 | ||
6712ecf8 | 901 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
902 | { |
903 | int r = 0; | |
bfc6d41c | 904 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); |
b35f8caa | 905 | struct dm_io *io = tio->io; |
9faf400f | 906 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
907 | dm_endio_fn endio = tio->ti->type->end_io; |
908 | ||
1da177e4 LT |
909 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
910 | error = -EIO; | |
911 | ||
912 | if (endio) { | |
7de3ee57 | 913 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
914 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
915 | /* | |
916 | * error and requeue request are handled | |
917 | * in dec_pending(). | |
918 | */ | |
1da177e4 | 919 | error = r; |
45cbcd79 KU |
920 | else if (r == DM_ENDIO_INCOMPLETE) |
921 | /* The target will handle the io */ | |
6712ecf8 | 922 | return; |
45cbcd79 KU |
923 | else if (r) { |
924 | DMWARN("unimplemented target endio return value: %d", r); | |
925 | BUG(); | |
926 | } | |
1da177e4 LT |
927 | } |
928 | ||
7eee4ae2 MS |
929 | if (unlikely(r == -EREMOTEIO && (bio->bi_rw & REQ_WRITE_SAME) && |
930 | !bdev_get_queue(bio->bi_bdev)->limits.max_write_same_sectors)) | |
931 | disable_write_same(md); | |
932 | ||
9faf400f | 933 | free_tio(md, tio); |
b35f8caa | 934 | dec_pending(io, error); |
1da177e4 LT |
935 | } |
936 | ||
cec47e3d KU |
937 | /* |
938 | * Partial completion handling for request-based dm | |
939 | */ | |
940 | static void end_clone_bio(struct bio *clone, int error) | |
941 | { | |
bfc6d41c MP |
942 | struct dm_rq_clone_bio_info *info = |
943 | container_of(clone, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
944 | struct dm_rq_target_io *tio = info->tio; |
945 | struct bio *bio = info->orig; | |
4f024f37 | 946 | unsigned int nr_bytes = info->orig->bi_iter.bi_size; |
cec47e3d KU |
947 | |
948 | bio_put(clone); | |
949 | ||
950 | if (tio->error) | |
951 | /* | |
952 | * An error has already been detected on the request. | |
953 | * Once error occurred, just let clone->end_io() handle | |
954 | * the remainder. | |
955 | */ | |
956 | return; | |
957 | else if (error) { | |
958 | /* | |
959 | * Don't notice the error to the upper layer yet. | |
960 | * The error handling decision is made by the target driver, | |
961 | * when the request is completed. | |
962 | */ | |
963 | tio->error = error; | |
964 | return; | |
965 | } | |
966 | ||
967 | /* | |
968 | * I/O for the bio successfully completed. | |
969 | * Notice the data completion to the upper layer. | |
970 | */ | |
971 | ||
972 | /* | |
973 | * bios are processed from the head of the list. | |
974 | * So the completing bio should always be rq->bio. | |
975 | * If it's not, something wrong is happening. | |
976 | */ | |
977 | if (tio->orig->bio != bio) | |
978 | DMERR("bio completion is going in the middle of the request"); | |
979 | ||
980 | /* | |
981 | * Update the original request. | |
982 | * Do not use blk_end_request() here, because it may complete | |
983 | * the original request before the clone, and break the ordering. | |
984 | */ | |
985 | blk_update_request(tio->orig, 0, nr_bytes); | |
986 | } | |
987 | ||
988 | /* | |
989 | * Don't touch any member of the md after calling this function because | |
990 | * the md may be freed in dm_put() at the end of this function. | |
991 | * Or do dm_get() before calling this function and dm_put() later. | |
992 | */ | |
b4324fee | 993 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 994 | { |
b4324fee | 995 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
996 | |
997 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 998 | if (!md_in_flight(md)) |
cec47e3d KU |
999 | wake_up(&md->wait); |
1000 | ||
a8c32a5c JA |
1001 | /* |
1002 | * Run this off this callpath, as drivers could invoke end_io while | |
1003 | * inside their request_fn (and holding the queue lock). Calling | |
1004 | * back into ->request_fn() could deadlock attempting to grab the | |
1005 | * queue lock again. | |
1006 | */ | |
cec47e3d | 1007 | if (run_queue) |
a8c32a5c | 1008 | blk_run_queue_async(md->queue); |
cec47e3d KU |
1009 | |
1010 | /* | |
1011 | * dm_put() must be at the end of this function. See the comment above | |
1012 | */ | |
1013 | dm_put(md); | |
1014 | } | |
1015 | ||
a77e28c7 KU |
1016 | static void free_rq_clone(struct request *clone) |
1017 | { | |
1018 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1019 | ||
1020 | blk_rq_unprep_clone(clone); | |
1021 | free_rq_tio(tio); | |
1022 | } | |
1023 | ||
980691e5 KU |
1024 | /* |
1025 | * Complete the clone and the original request. | |
1026 | * Must be called without queue lock. | |
1027 | */ | |
1028 | static void dm_end_request(struct request *clone, int error) | |
1029 | { | |
1030 | int rw = rq_data_dir(clone); | |
1031 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1032 | struct mapped_device *md = tio->md; | |
1033 | struct request *rq = tio->orig; | |
1034 | ||
29e4013d | 1035 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
1036 | rq->errors = clone->errors; |
1037 | rq->resid_len = clone->resid_len; | |
1038 | ||
1039 | if (rq->sense) | |
1040 | /* | |
1041 | * We are using the sense buffer of the original | |
1042 | * request. | |
1043 | * So setting the length of the sense data is enough. | |
1044 | */ | |
1045 | rq->sense_len = clone->sense_len; | |
1046 | } | |
1047 | ||
1048 | free_rq_clone(clone); | |
29e4013d TH |
1049 | blk_end_request_all(rq, error); |
1050 | rq_completed(md, rw, true); | |
980691e5 KU |
1051 | } |
1052 | ||
cec47e3d KU |
1053 | static void dm_unprep_request(struct request *rq) |
1054 | { | |
1055 | struct request *clone = rq->special; | |
cec47e3d KU |
1056 | |
1057 | rq->special = NULL; | |
1058 | rq->cmd_flags &= ~REQ_DONTPREP; | |
1059 | ||
a77e28c7 | 1060 | free_rq_clone(clone); |
cec47e3d KU |
1061 | } |
1062 | ||
1063 | /* | |
1064 | * Requeue the original request of a clone. | |
1065 | */ | |
1066 | void dm_requeue_unmapped_request(struct request *clone) | |
1067 | { | |
b4324fee | 1068 | int rw = rq_data_dir(clone); |
cec47e3d KU |
1069 | struct dm_rq_target_io *tio = clone->end_io_data; |
1070 | struct mapped_device *md = tio->md; | |
1071 | struct request *rq = tio->orig; | |
1072 | struct request_queue *q = rq->q; | |
1073 | unsigned long flags; | |
1074 | ||
1075 | dm_unprep_request(rq); | |
1076 | ||
1077 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
1078 | blk_requeue_request(q, rq); |
1079 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1080 | ||
b4324fee | 1081 | rq_completed(md, rw, 0); |
cec47e3d KU |
1082 | } |
1083 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
1084 | ||
1085 | static void __stop_queue(struct request_queue *q) | |
1086 | { | |
1087 | blk_stop_queue(q); | |
1088 | } | |
1089 | ||
1090 | static void stop_queue(struct request_queue *q) | |
1091 | { | |
1092 | unsigned long flags; | |
1093 | ||
1094 | spin_lock_irqsave(q->queue_lock, flags); | |
1095 | __stop_queue(q); | |
1096 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1097 | } | |
1098 | ||
1099 | static void __start_queue(struct request_queue *q) | |
1100 | { | |
1101 | if (blk_queue_stopped(q)) | |
1102 | blk_start_queue(q); | |
1103 | } | |
1104 | ||
1105 | static void start_queue(struct request_queue *q) | |
1106 | { | |
1107 | unsigned long flags; | |
1108 | ||
1109 | spin_lock_irqsave(q->queue_lock, flags); | |
1110 | __start_queue(q); | |
1111 | spin_unlock_irqrestore(q->queue_lock, flags); | |
1112 | } | |
1113 | ||
11a68244 | 1114 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 1115 | { |
11a68244 | 1116 | int r = error; |
cec47e3d | 1117 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 1118 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 1119 | |
ba1cbad9 MS |
1120 | if (tio->ti) { |
1121 | rq_end_io = tio->ti->type->rq_end_io; | |
1122 | ||
1123 | if (mapped && rq_end_io) | |
1124 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
1125 | } | |
cec47e3d | 1126 | |
7eee4ae2 MS |
1127 | if (unlikely(r == -EREMOTEIO && (clone->cmd_flags & REQ_WRITE_SAME) && |
1128 | !clone->q->limits.max_write_same_sectors)) | |
1129 | disable_write_same(tio->md); | |
1130 | ||
11a68244 | 1131 | if (r <= 0) |
cec47e3d | 1132 | /* The target wants to complete the I/O */ |
11a68244 KU |
1133 | dm_end_request(clone, r); |
1134 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
1135 | /* The target will handle the I/O */ |
1136 | return; | |
11a68244 | 1137 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
1138 | /* The target wants to requeue the I/O */ |
1139 | dm_requeue_unmapped_request(clone); | |
1140 | else { | |
11a68244 | 1141 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
1142 | BUG(); |
1143 | } | |
1144 | } | |
1145 | ||
11a68244 KU |
1146 | /* |
1147 | * Request completion handler for request-based dm | |
1148 | */ | |
1149 | static void dm_softirq_done(struct request *rq) | |
1150 | { | |
1151 | bool mapped = true; | |
1152 | struct request *clone = rq->completion_data; | |
1153 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1154 | ||
1155 | if (rq->cmd_flags & REQ_FAILED) | |
1156 | mapped = false; | |
1157 | ||
1158 | dm_done(clone, tio->error, mapped); | |
1159 | } | |
1160 | ||
cec47e3d KU |
1161 | /* |
1162 | * Complete the clone and the original request with the error status | |
1163 | * through softirq context. | |
1164 | */ | |
1165 | static void dm_complete_request(struct request *clone, int error) | |
1166 | { | |
1167 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1168 | struct request *rq = tio->orig; | |
1169 | ||
1170 | tio->error = error; | |
1171 | rq->completion_data = clone; | |
1172 | blk_complete_request(rq); | |
1173 | } | |
1174 | ||
1175 | /* | |
1176 | * Complete the not-mapped clone and the original request with the error status | |
1177 | * through softirq context. | |
1178 | * Target's rq_end_io() function isn't called. | |
1179 | * This may be used when the target's map_rq() function fails. | |
1180 | */ | |
1181 | void dm_kill_unmapped_request(struct request *clone, int error) | |
1182 | { | |
1183 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1184 | struct request *rq = tio->orig; | |
1185 | ||
1186 | rq->cmd_flags |= REQ_FAILED; | |
1187 | dm_complete_request(clone, error); | |
1188 | } | |
1189 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
1190 | ||
1191 | /* | |
1192 | * Called with the queue lock held | |
1193 | */ | |
1194 | static void end_clone_request(struct request *clone, int error) | |
1195 | { | |
1196 | /* | |
1197 | * For just cleaning up the information of the queue in which | |
1198 | * the clone was dispatched. | |
1199 | * The clone is *NOT* freed actually here because it is alloced from | |
1200 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
1201 | */ | |
1202 | __blk_put_request(clone->q, clone); | |
1203 | ||
1204 | /* | |
1205 | * Actual request completion is done in a softirq context which doesn't | |
1206 | * hold the queue lock. Otherwise, deadlock could occur because: | |
1207 | * - another request may be submitted by the upper level driver | |
1208 | * of the stacking during the completion | |
1209 | * - the submission which requires queue lock may be done | |
1210 | * against this queue | |
1211 | */ | |
1212 | dm_complete_request(clone, error); | |
1213 | } | |
1214 | ||
56a67df7 MS |
1215 | /* |
1216 | * Return maximum size of I/O possible at the supplied sector up to the current | |
1217 | * target boundary. | |
1218 | */ | |
1219 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
1220 | { | |
1221 | sector_t target_offset = dm_target_offset(ti, sector); | |
1222 | ||
1223 | return ti->len - target_offset; | |
1224 | } | |
1225 | ||
1226 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 1227 | { |
56a67df7 | 1228 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 1229 | sector_t offset, max_len; |
1da177e4 LT |
1230 | |
1231 | /* | |
542f9038 | 1232 | * Does the target need to split even further? |
1da177e4 | 1233 | */ |
542f9038 MS |
1234 | if (ti->max_io_len) { |
1235 | offset = dm_target_offset(ti, sector); | |
1236 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
1237 | max_len = sector_div(offset, ti->max_io_len); | |
1238 | else | |
1239 | max_len = offset & (ti->max_io_len - 1); | |
1240 | max_len = ti->max_io_len - max_len; | |
1241 | ||
1242 | if (len > max_len) | |
1243 | len = max_len; | |
1da177e4 LT |
1244 | } |
1245 | ||
1246 | return len; | |
1247 | } | |
1248 | ||
542f9038 MS |
1249 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
1250 | { | |
1251 | if (len > UINT_MAX) { | |
1252 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
1253 | (unsigned long long)len, UINT_MAX); | |
1254 | ti->error = "Maximum size of target IO is too large"; | |
1255 | return -EINVAL; | |
1256 | } | |
1257 | ||
1258 | ti->max_io_len = (uint32_t) len; | |
1259 | ||
1260 | return 0; | |
1261 | } | |
1262 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1263 | ||
1dd40c3e MP |
1264 | /* |
1265 | * A target may call dm_accept_partial_bio only from the map routine. It is | |
1266 | * allowed for all bio types except REQ_FLUSH. | |
1267 | * | |
1268 | * dm_accept_partial_bio informs the dm that the target only wants to process | |
1269 | * additional n_sectors sectors of the bio and the rest of the data should be | |
1270 | * sent in a next bio. | |
1271 | * | |
1272 | * A diagram that explains the arithmetics: | |
1273 | * +--------------------+---------------+-------+ | |
1274 | * | 1 | 2 | 3 | | |
1275 | * +--------------------+---------------+-------+ | |
1276 | * | |
1277 | * <-------------- *tio->len_ptr ---------------> | |
1278 | * <------- bi_size -------> | |
1279 | * <-- n_sectors --> | |
1280 | * | |
1281 | * Region 1 was already iterated over with bio_advance or similar function. | |
1282 | * (it may be empty if the target doesn't use bio_advance) | |
1283 | * Region 2 is the remaining bio size that the target wants to process. | |
1284 | * (it may be empty if region 1 is non-empty, although there is no reason | |
1285 | * to make it empty) | |
1286 | * The target requires that region 3 is to be sent in the next bio. | |
1287 | * | |
1288 | * If the target wants to receive multiple copies of the bio (via num_*bios, etc), | |
1289 | * the partially processed part (the sum of regions 1+2) must be the same for all | |
1290 | * copies of the bio. | |
1291 | */ | |
1292 | void dm_accept_partial_bio(struct bio *bio, unsigned n_sectors) | |
1293 | { | |
1294 | struct dm_target_io *tio = container_of(bio, struct dm_target_io, clone); | |
1295 | unsigned bi_size = bio->bi_iter.bi_size >> SECTOR_SHIFT; | |
1296 | BUG_ON(bio->bi_rw & REQ_FLUSH); | |
1297 | BUG_ON(bi_size > *tio->len_ptr); | |
1298 | BUG_ON(n_sectors > bi_size); | |
1299 | *tio->len_ptr -= bi_size - n_sectors; | |
1300 | bio->bi_iter.bi_size = n_sectors << SECTOR_SHIFT; | |
1301 | } | |
1302 | EXPORT_SYMBOL_GPL(dm_accept_partial_bio); | |
1303 | ||
bd2a49b8 | 1304 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1305 | { |
1306 | int r; | |
2056a782 | 1307 | sector_t sector; |
9faf400f | 1308 | struct mapped_device *md; |
dba14160 | 1309 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1310 | struct dm_target *ti = tio->ti; |
1da177e4 | 1311 | |
1da177e4 | 1312 | clone->bi_end_io = clone_endio; |
1da177e4 LT |
1313 | |
1314 | /* | |
1315 | * Map the clone. If r == 0 we don't need to do | |
1316 | * anything, the target has assumed ownership of | |
1317 | * this io. | |
1318 | */ | |
1319 | atomic_inc(&tio->io->io_count); | |
4f024f37 | 1320 | sector = clone->bi_iter.bi_sector; |
7de3ee57 | 1321 | r = ti->type->map(ti, clone); |
45cbcd79 | 1322 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1323 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1324 | |
d07335e5 MS |
1325 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1326 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1327 | |
1da177e4 | 1328 | generic_make_request(clone); |
2e93ccc1 KU |
1329 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1330 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1331 | md = tio->io->md; |
1332 | dec_pending(tio->io, r); | |
9faf400f | 1333 | free_tio(md, tio); |
45cbcd79 KU |
1334 | } else if (r) { |
1335 | DMWARN("unimplemented target map return value: %d", r); | |
1336 | BUG(); | |
1da177e4 LT |
1337 | } |
1338 | } | |
1339 | ||
1340 | struct clone_info { | |
1341 | struct mapped_device *md; | |
1342 | struct dm_table *map; | |
1343 | struct bio *bio; | |
1344 | struct dm_io *io; | |
1345 | sector_t sector; | |
e0d6609a | 1346 | unsigned sector_count; |
1da177e4 LT |
1347 | }; |
1348 | ||
e0d6609a | 1349 | static void bio_setup_sector(struct bio *bio, sector_t sector, unsigned len) |
bd2a49b8 | 1350 | { |
4f024f37 KO |
1351 | bio->bi_iter.bi_sector = sector; |
1352 | bio->bi_iter.bi_size = to_bytes(len); | |
1da177e4 LT |
1353 | } |
1354 | ||
1355 | /* | |
1356 | * Creates a bio that consists of range of complete bvecs. | |
1357 | */ | |
dba14160 | 1358 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1c3b13e6 | 1359 | sector_t sector, unsigned len) |
1da177e4 | 1360 | { |
dba14160 | 1361 | struct bio *clone = &tio->clone; |
1da177e4 | 1362 | |
1c3b13e6 KO |
1363 | __bio_clone_fast(clone, bio); |
1364 | ||
1365 | if (bio_integrity(bio)) | |
1366 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
bd2a49b8 | 1367 | |
1c3b13e6 KO |
1368 | bio_advance(clone, to_bytes(sector - clone->bi_iter.bi_sector)); |
1369 | clone->bi_iter.bi_size = to_bytes(len); | |
1370 | ||
1371 | if (bio_integrity(bio)) | |
1372 | bio_integrity_trim(clone, 0, len); | |
1da177e4 LT |
1373 | } |
1374 | ||
9015df24 | 1375 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
99778273 | 1376 | struct dm_target *ti, |
55a62eef | 1377 | unsigned target_bio_nr) |
f9ab94ce | 1378 | { |
dba14160 MP |
1379 | struct dm_target_io *tio; |
1380 | struct bio *clone; | |
1381 | ||
99778273 | 1382 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); |
dba14160 | 1383 | tio = container_of(clone, struct dm_target_io, clone); |
f9ab94ce MP |
1384 | |
1385 | tio->io = ci->io; | |
1386 | tio->ti = ti; | |
55a62eef | 1387 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1388 | |
1389 | return tio; | |
1390 | } | |
1391 | ||
14fe594d AK |
1392 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1393 | struct dm_target *ti, | |
1dd40c3e | 1394 | unsigned target_bio_nr, unsigned *len) |
9015df24 | 1395 | { |
99778273 | 1396 | struct dm_target_io *tio = alloc_tio(ci, ti, target_bio_nr); |
dba14160 | 1397 | struct bio *clone = &tio->clone; |
9015df24 | 1398 | |
1dd40c3e MP |
1399 | tio->len_ptr = len; |
1400 | ||
99778273 | 1401 | __bio_clone_fast(clone, ci->bio); |
bd2a49b8 | 1402 | if (len) |
1dd40c3e | 1403 | bio_setup_sector(clone, ci->sector, *len); |
f9ab94ce | 1404 | |
bd2a49b8 | 1405 | __map_bio(tio); |
f9ab94ce MP |
1406 | } |
1407 | ||
14fe594d | 1408 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1409 | unsigned num_bios, unsigned *len) |
06a426ce | 1410 | { |
55a62eef | 1411 | unsigned target_bio_nr; |
06a426ce | 1412 | |
55a62eef | 1413 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1414 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1415 | } |
1416 | ||
14fe594d | 1417 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1418 | { |
06a426ce | 1419 | unsigned target_nr = 0; |
f9ab94ce MP |
1420 | struct dm_target *ti; |
1421 | ||
b372d360 | 1422 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1423 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
1dd40c3e | 1424 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, NULL); |
f9ab94ce | 1425 | |
f9ab94ce MP |
1426 | return 0; |
1427 | } | |
1428 | ||
e4c93811 | 1429 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1dd40c3e | 1430 | sector_t sector, unsigned *len) |
5ae89a87 | 1431 | { |
dba14160 | 1432 | struct bio *bio = ci->bio; |
5ae89a87 | 1433 | struct dm_target_io *tio; |
b0d8ed4d AK |
1434 | unsigned target_bio_nr; |
1435 | unsigned num_target_bios = 1; | |
5ae89a87 | 1436 | |
b0d8ed4d AK |
1437 | /* |
1438 | * Does the target want to receive duplicate copies of the bio? | |
1439 | */ | |
1440 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1441 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1442 | |
b0d8ed4d | 1443 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
99778273 | 1444 | tio = alloc_tio(ci, ti, target_bio_nr); |
1dd40c3e MP |
1445 | tio->len_ptr = len; |
1446 | clone_bio(tio, bio, sector, *len); | |
b0d8ed4d AK |
1447 | __map_bio(tio); |
1448 | } | |
5ae89a87 MS |
1449 | } |
1450 | ||
55a62eef | 1451 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1452 | |
55a62eef | 1453 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1454 | { |
55a62eef | 1455 | return ti->num_discard_bios; |
23508a96 MS |
1456 | } |
1457 | ||
55a62eef | 1458 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1459 | { |
55a62eef | 1460 | return ti->num_write_same_bios; |
23508a96 MS |
1461 | } |
1462 | ||
1463 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1464 | ||
1465 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1466 | { | |
55a62eef | 1467 | return ti->split_discard_bios; |
23508a96 MS |
1468 | } |
1469 | ||
14fe594d AK |
1470 | static int __send_changing_extent_only(struct clone_info *ci, |
1471 | get_num_bios_fn get_num_bios, | |
1472 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1473 | { |
1474 | struct dm_target *ti; | |
e0d6609a | 1475 | unsigned len; |
55a62eef | 1476 | unsigned num_bios; |
5ae89a87 | 1477 | |
a79245b3 MS |
1478 | do { |
1479 | ti = dm_table_find_target(ci->map, ci->sector); | |
1480 | if (!dm_target_is_valid(ti)) | |
1481 | return -EIO; | |
5ae89a87 | 1482 | |
5ae89a87 | 1483 | /* |
23508a96 MS |
1484 | * Even though the device advertised support for this type of |
1485 | * request, that does not mean every target supports it, and | |
936688d7 | 1486 | * reconfiguration might also have changed that since the |
a79245b3 | 1487 | * check was performed. |
5ae89a87 | 1488 | */ |
55a62eef AK |
1489 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1490 | if (!num_bios) | |
a79245b3 | 1491 | return -EOPNOTSUPP; |
5ae89a87 | 1492 | |
23508a96 | 1493 | if (is_split_required && !is_split_required(ti)) |
e0d6609a | 1494 | len = min((sector_t)ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
7acf0277 | 1495 | else |
e0d6609a | 1496 | len = min((sector_t)ci->sector_count, max_io_len(ci->sector, ti)); |
06a426ce | 1497 | |
1dd40c3e | 1498 | __send_duplicate_bios(ci, ti, num_bios, &len); |
a79245b3 MS |
1499 | |
1500 | ci->sector += len; | |
1501 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1502 | |
1503 | return 0; | |
1504 | } | |
1505 | ||
14fe594d | 1506 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1507 | { |
14fe594d AK |
1508 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1509 | is_split_required_for_discard); | |
23508a96 MS |
1510 | } |
1511 | ||
14fe594d | 1512 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1513 | { |
14fe594d | 1514 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1515 | } |
1516 | ||
e4c93811 AK |
1517 | /* |
1518 | * Select the correct strategy for processing a non-flush bio. | |
1519 | */ | |
14fe594d | 1520 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1521 | { |
dba14160 | 1522 | struct bio *bio = ci->bio; |
512875bd | 1523 | struct dm_target *ti; |
1c3b13e6 | 1524 | unsigned len; |
1da177e4 | 1525 | |
5ae89a87 | 1526 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1527 | return __send_discard(ci); |
23508a96 | 1528 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1529 | return __send_write_same(ci); |
5ae89a87 | 1530 | |
512875bd JN |
1531 | ti = dm_table_find_target(ci->map, ci->sector); |
1532 | if (!dm_target_is_valid(ti)) | |
1533 | return -EIO; | |
1534 | ||
1c3b13e6 | 1535 | len = min_t(sector_t, max_io_len(ci->sector, ti), ci->sector_count); |
1da177e4 | 1536 | |
1dd40c3e | 1537 | __clone_and_map_data_bio(ci, ti, ci->sector, &len); |
1da177e4 | 1538 | |
1c3b13e6 KO |
1539 | ci->sector += len; |
1540 | ci->sector_count -= len; | |
1da177e4 | 1541 | |
1c3b13e6 | 1542 | return 0; |
1da177e4 LT |
1543 | } |
1544 | ||
1545 | /* | |
14fe594d | 1546 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1547 | */ |
83d5e5b0 MP |
1548 | static void __split_and_process_bio(struct mapped_device *md, |
1549 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1550 | { |
1551 | struct clone_info ci; | |
512875bd | 1552 | int error = 0; |
1da177e4 | 1553 | |
83d5e5b0 | 1554 | if (unlikely(!map)) { |
6a8736d1 | 1555 | bio_io_error(bio); |
f0b9a450 MP |
1556 | return; |
1557 | } | |
692d0eb9 | 1558 | |
83d5e5b0 | 1559 | ci.map = map; |
1da177e4 | 1560 | ci.md = md; |
1da177e4 LT |
1561 | ci.io = alloc_io(md); |
1562 | ci.io->error = 0; | |
1563 | atomic_set(&ci.io->io_count, 1); | |
1564 | ci.io->bio = bio; | |
1565 | ci.io->md = md; | |
f88fb981 | 1566 | spin_lock_init(&ci.io->endio_lock); |
4f024f37 | 1567 | ci.sector = bio->bi_iter.bi_sector; |
1da177e4 | 1568 | |
3eaf840e | 1569 | start_io_acct(ci.io); |
bd2a49b8 | 1570 | |
b372d360 MS |
1571 | if (bio->bi_rw & REQ_FLUSH) { |
1572 | ci.bio = &ci.md->flush_bio; | |
1573 | ci.sector_count = 0; | |
14fe594d | 1574 | error = __send_empty_flush(&ci); |
b372d360 MS |
1575 | /* dec_pending submits any data associated with flush */ |
1576 | } else { | |
6a8736d1 | 1577 | ci.bio = bio; |
d87f4c14 | 1578 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1579 | while (ci.sector_count && !error) |
14fe594d | 1580 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1581 | } |
1da177e4 LT |
1582 | |
1583 | /* drop the extra reference count */ | |
512875bd | 1584 | dec_pending(ci.io, error); |
1da177e4 LT |
1585 | } |
1586 | /*----------------------------------------------------------------- | |
1587 | * CRUD END | |
1588 | *---------------------------------------------------------------*/ | |
1589 | ||
f6fccb12 MB |
1590 | static int dm_merge_bvec(struct request_queue *q, |
1591 | struct bvec_merge_data *bvm, | |
1592 | struct bio_vec *biovec) | |
1593 | { | |
1594 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1595 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1596 | struct dm_target *ti; |
1597 | sector_t max_sectors; | |
5037108a | 1598 | int max_size = 0; |
f6fccb12 MB |
1599 | |
1600 | if (unlikely(!map)) | |
5037108a | 1601 | goto out; |
f6fccb12 MB |
1602 | |
1603 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1604 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1605 | goto out; |
f6fccb12 MB |
1606 | |
1607 | /* | |
1608 | * Find maximum amount of I/O that won't need splitting | |
1609 | */ | |
56a67df7 | 1610 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
148e51ba | 1611 | (sector_t) queue_max_sectors(q)); |
f6fccb12 | 1612 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; |
148e51ba | 1613 | if (unlikely(max_size < 0)) /* this shouldn't _ever_ happen */ |
f6fccb12 MB |
1614 | max_size = 0; |
1615 | ||
1616 | /* | |
1617 | * merge_bvec_fn() returns number of bytes | |
1618 | * it can accept at this offset | |
1619 | * max is precomputed maximal io size | |
1620 | */ | |
1621 | if (max_size && ti->type->merge) | |
1622 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1623 | /* |
1624 | * If the target doesn't support merge method and some of the devices | |
148e51ba MS |
1625 | * provided their merge_bvec method (we know this by looking for the |
1626 | * max_hw_sectors that dm_set_device_limits may set), then we can't | |
1627 | * allow bios with multiple vector entries. So always set max_size | |
1628 | * to 0, and the code below allows just one page. | |
8cbeb67a MP |
1629 | */ |
1630 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
8cbeb67a | 1631 | max_size = 0; |
f6fccb12 | 1632 | |
5037108a | 1633 | out: |
83d5e5b0 | 1634 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1635 | /* |
1636 | * Always allow an entire first page | |
1637 | */ | |
1638 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1639 | max_size = biovec->bv_len; | |
1640 | ||
f6fccb12 MB |
1641 | return max_size; |
1642 | } | |
1643 | ||
1da177e4 LT |
1644 | /* |
1645 | * The request function that just remaps the bio built up by | |
1646 | * dm_merge_bvec. | |
1647 | */ | |
5a7bbad2 | 1648 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1649 | { |
12f03a49 | 1650 | int rw = bio_data_dir(bio); |
1da177e4 | 1651 | struct mapped_device *md = q->queuedata; |
c9959059 | 1652 | int cpu; |
83d5e5b0 MP |
1653 | int srcu_idx; |
1654 | struct dm_table *map; | |
1da177e4 | 1655 | |
83d5e5b0 | 1656 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1657 | |
074a7aca TH |
1658 | cpu = part_stat_lock(); |
1659 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1660 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1661 | part_stat_unlock(); | |
12f03a49 | 1662 | |
6a8736d1 TH |
1663 | /* if we're suspended, we have to queue this io for later */ |
1664 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1665 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1666 | |
6a8736d1 TH |
1667 | if (bio_rw(bio) != READA) |
1668 | queue_io(md, bio); | |
1669 | else | |
54d9a1b4 | 1670 | bio_io_error(bio); |
5a7bbad2 | 1671 | return; |
1da177e4 LT |
1672 | } |
1673 | ||
83d5e5b0 MP |
1674 | __split_and_process_bio(md, map, bio); |
1675 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1676 | return; |
cec47e3d KU |
1677 | } |
1678 | ||
fd2ed4d2 | 1679 | int dm_request_based(struct mapped_device *md) |
cec47e3d KU |
1680 | { |
1681 | return blk_queue_stackable(md->queue); | |
1682 | } | |
1683 | ||
5a7bbad2 | 1684 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1685 | { |
1686 | struct mapped_device *md = q->queuedata; | |
1687 | ||
1688 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1689 | blk_queue_bio(q, bio); |
1690 | else | |
1691 | _dm_request(q, bio); | |
cec47e3d KU |
1692 | } |
1693 | ||
1694 | void dm_dispatch_request(struct request *rq) | |
1695 | { | |
1696 | int r; | |
1697 | ||
1698 | if (blk_queue_io_stat(rq->q)) | |
1699 | rq->cmd_flags |= REQ_IO_STAT; | |
1700 | ||
1701 | rq->start_time = jiffies; | |
1702 | r = blk_insert_cloned_request(rq->q, rq); | |
1703 | if (r) | |
1704 | dm_complete_request(rq, r); | |
1705 | } | |
1706 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1707 | ||
cec47e3d KU |
1708 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1709 | void *data) | |
1710 | { | |
1711 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1712 | struct dm_rq_clone_bio_info *info = |
1713 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1714 | |
1715 | info->orig = bio_orig; | |
1716 | info->tio = tio; | |
1717 | bio->bi_end_io = end_clone_bio; | |
cec47e3d KU |
1718 | |
1719 | return 0; | |
1720 | } | |
1721 | ||
1722 | static int setup_clone(struct request *clone, struct request *rq, | |
1723 | struct dm_rq_target_io *tio) | |
1724 | { | |
d0bcb878 | 1725 | int r; |
cec47e3d | 1726 | |
29e4013d TH |
1727 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1728 | dm_rq_bio_constructor, tio); | |
1729 | if (r) | |
1730 | return r; | |
cec47e3d | 1731 | |
29e4013d TH |
1732 | clone->cmd = rq->cmd; |
1733 | clone->cmd_len = rq->cmd_len; | |
1734 | clone->sense = rq->sense; | |
cec47e3d KU |
1735 | clone->end_io = end_clone_request; |
1736 | clone->end_io_data = tio; | |
1737 | ||
1738 | return 0; | |
1739 | } | |
1740 | ||
6facdaff KU |
1741 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1742 | gfp_t gfp_mask) | |
1743 | { | |
1744 | struct request *clone; | |
1745 | struct dm_rq_target_io *tio; | |
1746 | ||
1747 | tio = alloc_rq_tio(md, gfp_mask); | |
1748 | if (!tio) | |
1749 | return NULL; | |
1750 | ||
1751 | tio->md = md; | |
1752 | tio->ti = NULL; | |
1753 | tio->orig = rq; | |
1754 | tio->error = 0; | |
1755 | memset(&tio->info, 0, sizeof(tio->info)); | |
1756 | ||
1757 | clone = &tio->clone; | |
1758 | if (setup_clone(clone, rq, tio)) { | |
1759 | /* -ENOMEM */ | |
1760 | free_rq_tio(tio); | |
1761 | return NULL; | |
1762 | } | |
1763 | ||
1764 | return clone; | |
1765 | } | |
1766 | ||
cec47e3d KU |
1767 | /* |
1768 | * Called with the queue lock held. | |
1769 | */ | |
1770 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1771 | { | |
1772 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1773 | struct request *clone; |
1774 | ||
cec47e3d KU |
1775 | if (unlikely(rq->special)) { |
1776 | DMWARN("Already has something in rq->special."); | |
1777 | return BLKPREP_KILL; | |
1778 | } | |
1779 | ||
6facdaff KU |
1780 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1781 | if (!clone) | |
cec47e3d | 1782 | return BLKPREP_DEFER; |
cec47e3d KU |
1783 | |
1784 | rq->special = clone; | |
1785 | rq->cmd_flags |= REQ_DONTPREP; | |
1786 | ||
1787 | return BLKPREP_OK; | |
1788 | } | |
1789 | ||
9eef87da KU |
1790 | /* |
1791 | * Returns: | |
1792 | * 0 : the request has been processed (not requeued) | |
1793 | * !0 : the request has been requeued | |
1794 | */ | |
1795 | static int map_request(struct dm_target *ti, struct request *clone, | |
1796 | struct mapped_device *md) | |
cec47e3d | 1797 | { |
9eef87da | 1798 | int r, requeued = 0; |
cec47e3d KU |
1799 | struct dm_rq_target_io *tio = clone->end_io_data; |
1800 | ||
cec47e3d KU |
1801 | tio->ti = ti; |
1802 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1803 | switch (r) { | |
1804 | case DM_MAPIO_SUBMITTED: | |
1805 | /* The target has taken the I/O to submit by itself later */ | |
1806 | break; | |
1807 | case DM_MAPIO_REMAPPED: | |
1808 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1809 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1810 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1811 | dm_dispatch_request(clone); |
1812 | break; | |
1813 | case DM_MAPIO_REQUEUE: | |
1814 | /* The target wants to requeue the I/O */ | |
1815 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1816 | requeued = 1; |
cec47e3d KU |
1817 | break; |
1818 | default: | |
1819 | if (r > 0) { | |
1820 | DMWARN("unimplemented target map return value: %d", r); | |
1821 | BUG(); | |
1822 | } | |
1823 | ||
1824 | /* The target wants to complete the I/O */ | |
1825 | dm_kill_unmapped_request(clone, r); | |
1826 | break; | |
1827 | } | |
9eef87da KU |
1828 | |
1829 | return requeued; | |
cec47e3d KU |
1830 | } |
1831 | ||
ba1cbad9 MS |
1832 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1833 | { | |
1834 | struct request *clone; | |
1835 | ||
1836 | blk_start_request(orig); | |
1837 | clone = orig->special; | |
1838 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1839 | ||
1840 | /* | |
1841 | * Hold the md reference here for the in-flight I/O. | |
1842 | * We can't rely on the reference count by device opener, | |
1843 | * because the device may be closed during the request completion | |
1844 | * when all bios are completed. | |
1845 | * See the comment in rq_completed() too. | |
1846 | */ | |
1847 | dm_get(md); | |
1848 | ||
1849 | return clone; | |
1850 | } | |
1851 | ||
cec47e3d KU |
1852 | /* |
1853 | * q->request_fn for request-based dm. | |
1854 | * Called with the queue lock held. | |
1855 | */ | |
1856 | static void dm_request_fn(struct request_queue *q) | |
1857 | { | |
1858 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1859 | int srcu_idx; |
1860 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1861 | struct dm_target *ti; |
b4324fee | 1862 | struct request *rq, *clone; |
29e4013d | 1863 | sector_t pos; |
cec47e3d KU |
1864 | |
1865 | /* | |
b4324fee KU |
1866 | * For suspend, check blk_queue_stopped() and increment |
1867 | * ->pending within a single queue_lock not to increment the | |
1868 | * number of in-flight I/Os after the queue is stopped in | |
1869 | * dm_suspend(). | |
cec47e3d | 1870 | */ |
7eaceacc | 1871 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1872 | rq = blk_peek_request(q); |
1873 | if (!rq) | |
7eaceacc | 1874 | goto delay_and_out; |
cec47e3d | 1875 | |
29e4013d TH |
1876 | /* always use block 0 to find the target for flushes for now */ |
1877 | pos = 0; | |
1878 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1879 | pos = blk_rq_pos(rq); | |
1880 | ||
1881 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1882 | if (!dm_target_is_valid(ti)) { |
1883 | /* | |
1884 | * Must perform setup, that dm_done() requires, | |
1885 | * before calling dm_kill_unmapped_request | |
1886 | */ | |
1887 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1888 | clone = dm_start_request(md, rq); | |
1889 | dm_kill_unmapped_request(clone, -EIO); | |
1890 | continue; | |
1891 | } | |
d0bcb878 | 1892 | |
cec47e3d | 1893 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1894 | goto delay_and_out; |
cec47e3d | 1895 | |
ba1cbad9 | 1896 | clone = dm_start_request(md, rq); |
b4324fee | 1897 | |
cec47e3d | 1898 | spin_unlock(q->queue_lock); |
9eef87da KU |
1899 | if (map_request(ti, clone, md)) |
1900 | goto requeued; | |
1901 | ||
052189a2 KU |
1902 | BUG_ON(!irqs_disabled()); |
1903 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1904 | } |
1905 | ||
1906 | goto out; | |
1907 | ||
9eef87da | 1908 | requeued: |
052189a2 KU |
1909 | BUG_ON(!irqs_disabled()); |
1910 | spin_lock(q->queue_lock); | |
9eef87da | 1911 | |
7eaceacc JA |
1912 | delay_and_out: |
1913 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1914 | out: |
83d5e5b0 | 1915 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1916 | } |
1917 | ||
1918 | int dm_underlying_device_busy(struct request_queue *q) | |
1919 | { | |
1920 | return blk_lld_busy(q); | |
1921 | } | |
1922 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1923 | ||
1924 | static int dm_lld_busy(struct request_queue *q) | |
1925 | { | |
1926 | int r; | |
1927 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1928 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1929 | |
1930 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1931 | r = 1; | |
1932 | else | |
1933 | r = dm_table_any_busy_target(map); | |
1934 | ||
83d5e5b0 | 1935 | dm_put_live_table_fast(md); |
cec47e3d KU |
1936 | |
1937 | return r; | |
1938 | } | |
1939 | ||
1da177e4 LT |
1940 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1941 | { | |
8a57dfc6 CS |
1942 | int r = bdi_bits; |
1943 | struct mapped_device *md = congested_data; | |
1944 | struct dm_table *map; | |
1da177e4 | 1945 | |
1eb787ec | 1946 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1947 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1948 | if (map) { |
cec47e3d KU |
1949 | /* |
1950 | * Request-based dm cares about only own queue for | |
1951 | * the query about congestion status of request_queue | |
1952 | */ | |
1953 | if (dm_request_based(md)) | |
1954 | r = md->queue->backing_dev_info.state & | |
1955 | bdi_bits; | |
1956 | else | |
1957 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1958 | } |
83d5e5b0 | 1959 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1960 | } |
1961 | ||
1da177e4 LT |
1962 | return r; |
1963 | } | |
1964 | ||
1965 | /*----------------------------------------------------------------- | |
1966 | * An IDR is used to keep track of allocated minor numbers. | |
1967 | *---------------------------------------------------------------*/ | |
2b06cfff | 1968 | static void free_minor(int minor) |
1da177e4 | 1969 | { |
f32c10b0 | 1970 | spin_lock(&_minor_lock); |
1da177e4 | 1971 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1972 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1973 | } |
1974 | ||
1975 | /* | |
1976 | * See if the device with a specific minor # is free. | |
1977 | */ | |
cf13ab8e | 1978 | static int specific_minor(int minor) |
1da177e4 | 1979 | { |
c9d76be6 | 1980 | int r; |
1da177e4 LT |
1981 | |
1982 | if (minor >= (1 << MINORBITS)) | |
1983 | return -EINVAL; | |
1984 | ||
c9d76be6 | 1985 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1986 | spin_lock(&_minor_lock); |
1da177e4 | 1987 | |
c9d76be6 | 1988 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1989 | |
f32c10b0 | 1990 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1991 | idr_preload_end(); |
1992 | if (r < 0) | |
1993 | return r == -ENOSPC ? -EBUSY : r; | |
1994 | return 0; | |
1da177e4 LT |
1995 | } |
1996 | ||
cf13ab8e | 1997 | static int next_free_minor(int *minor) |
1da177e4 | 1998 | { |
c9d76be6 | 1999 | int r; |
62f75c2f | 2000 | |
c9d76be6 | 2001 | idr_preload(GFP_KERNEL); |
f32c10b0 | 2002 | spin_lock(&_minor_lock); |
1da177e4 | 2003 | |
c9d76be6 | 2004 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 2005 | |
f32c10b0 | 2006 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
2007 | idr_preload_end(); |
2008 | if (r < 0) | |
2009 | return r; | |
2010 | *minor = r; | |
2011 | return 0; | |
1da177e4 LT |
2012 | } |
2013 | ||
83d5cde4 | 2014 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 2015 | |
53d5914f MP |
2016 | static void dm_wq_work(struct work_struct *work); |
2017 | ||
4a0b4ddf MS |
2018 | static void dm_init_md_queue(struct mapped_device *md) |
2019 | { | |
2020 | /* | |
2021 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
2022 | * devices. The type of this dm device has not been decided yet. | |
2023 | * The type is decided at the first table loading time. | |
2024 | * To prevent problematic device stacking, clear the queue flag | |
2025 | * for request stacking support until then. | |
2026 | * | |
2027 | * This queue is new, so no concurrency on the queue_flags. | |
2028 | */ | |
2029 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
2030 | ||
2031 | md->queue->queuedata = md; | |
2032 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
2033 | md->queue->backing_dev_info.congested_data = md; | |
2034 | blk_queue_make_request(md->queue, dm_request); | |
2035 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
2036 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
2037 | } | |
2038 | ||
1da177e4 LT |
2039 | /* |
2040 | * Allocate and initialise a blank device with a given minor. | |
2041 | */ | |
2b06cfff | 2042 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
2043 | { |
2044 | int r; | |
cf13ab8e | 2045 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 2046 | void *old_md; |
1da177e4 LT |
2047 | |
2048 | if (!md) { | |
2049 | DMWARN("unable to allocate device, out of memory."); | |
2050 | return NULL; | |
2051 | } | |
2052 | ||
10da4f79 | 2053 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 2054 | goto bad_module_get; |
10da4f79 | 2055 | |
1da177e4 | 2056 | /* get a minor number for the dev */ |
2b06cfff | 2057 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 2058 | r = next_free_minor(&minor); |
2b06cfff | 2059 | else |
cf13ab8e | 2060 | r = specific_minor(minor); |
1da177e4 | 2061 | if (r < 0) |
6ed7ade8 | 2062 | goto bad_minor; |
1da177e4 | 2063 | |
83d5e5b0 MP |
2064 | r = init_srcu_struct(&md->io_barrier); |
2065 | if (r < 0) | |
2066 | goto bad_io_barrier; | |
2067 | ||
a5664dad | 2068 | md->type = DM_TYPE_NONE; |
e61290a4 | 2069 | mutex_init(&md->suspend_lock); |
a5664dad | 2070 | mutex_init(&md->type_lock); |
86f1152b | 2071 | mutex_init(&md->table_devices_lock); |
022c2611 | 2072 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 2073 | atomic_set(&md->holders, 1); |
5c6bd75d | 2074 | atomic_set(&md->open_count, 0); |
1da177e4 | 2075 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
2076 | atomic_set(&md->uevent_seq, 0); |
2077 | INIT_LIST_HEAD(&md->uevent_list); | |
86f1152b | 2078 | INIT_LIST_HEAD(&md->table_devices); |
7a8c3d3b | 2079 | spin_lock_init(&md->uevent_lock); |
1da177e4 | 2080 | |
4a0b4ddf | 2081 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 2082 | if (!md->queue) |
6ed7ade8 | 2083 | goto bad_queue; |
1da177e4 | 2084 | |
4a0b4ddf | 2085 | dm_init_md_queue(md); |
9faf400f | 2086 | |
1da177e4 LT |
2087 | md->disk = alloc_disk(1); |
2088 | if (!md->disk) | |
6ed7ade8 | 2089 | goto bad_disk; |
1da177e4 | 2090 | |
316d315b NK |
2091 | atomic_set(&md->pending[0], 0); |
2092 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 2093 | init_waitqueue_head(&md->wait); |
53d5914f | 2094 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 | 2095 | init_waitqueue_head(&md->eventq); |
2995fa78 | 2096 | init_completion(&md->kobj_holder.completion); |
f0b04115 | 2097 | |
1da177e4 LT |
2098 | md->disk->major = _major; |
2099 | md->disk->first_minor = minor; | |
2100 | md->disk->fops = &dm_blk_dops; | |
2101 | md->disk->queue = md->queue; | |
2102 | md->disk->private_data = md; | |
2103 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
2104 | add_disk(md->disk); | |
7e51f257 | 2105 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 2106 | |
670368a8 | 2107 | md->wq = alloc_workqueue("kdmflush", WQ_MEM_RECLAIM, 0); |
304f3f6a MB |
2108 | if (!md->wq) |
2109 | goto bad_thread; | |
2110 | ||
32a926da MP |
2111 | md->bdev = bdget_disk(md->disk, 0); |
2112 | if (!md->bdev) | |
2113 | goto bad_bdev; | |
2114 | ||
6a8736d1 TH |
2115 | bio_init(&md->flush_bio); |
2116 | md->flush_bio.bi_bdev = md->bdev; | |
2117 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
2118 | ||
fd2ed4d2 MP |
2119 | dm_stats_init(&md->stats); |
2120 | ||
ba61fdd1 | 2121 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 2122 | spin_lock(&_minor_lock); |
ba61fdd1 | 2123 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 2124 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
2125 | |
2126 | BUG_ON(old_md != MINOR_ALLOCED); | |
2127 | ||
1da177e4 LT |
2128 | return md; |
2129 | ||
32a926da MP |
2130 | bad_bdev: |
2131 | destroy_workqueue(md->wq); | |
304f3f6a | 2132 | bad_thread: |
03022c54 | 2133 | del_gendisk(md->disk); |
304f3f6a | 2134 | put_disk(md->disk); |
6ed7ade8 | 2135 | bad_disk: |
1312f40e | 2136 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 2137 | bad_queue: |
83d5e5b0 MP |
2138 | cleanup_srcu_struct(&md->io_barrier); |
2139 | bad_io_barrier: | |
1da177e4 | 2140 | free_minor(minor); |
6ed7ade8 | 2141 | bad_minor: |
10da4f79 | 2142 | module_put(THIS_MODULE); |
6ed7ade8 | 2143 | bad_module_get: |
1da177e4 LT |
2144 | kfree(md); |
2145 | return NULL; | |
2146 | } | |
2147 | ||
ae9da83f JN |
2148 | static void unlock_fs(struct mapped_device *md); |
2149 | ||
1da177e4 LT |
2150 | static void free_dev(struct mapped_device *md) |
2151 | { | |
f331c029 | 2152 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 2153 | |
32a926da MP |
2154 | unlock_fs(md); |
2155 | bdput(md->bdev); | |
304f3f6a | 2156 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
2157 | if (md->io_pool) |
2158 | mempool_destroy(md->io_pool); | |
2159 | if (md->bs) | |
2160 | bioset_free(md->bs); | |
9c47008d | 2161 | blk_integrity_unregister(md->disk); |
1da177e4 | 2162 | del_gendisk(md->disk); |
83d5e5b0 | 2163 | cleanup_srcu_struct(&md->io_barrier); |
86f1152b | 2164 | free_table_devices(&md->table_devices); |
63d94e48 | 2165 | free_minor(minor); |
fba9f90e JM |
2166 | |
2167 | spin_lock(&_minor_lock); | |
2168 | md->disk->private_data = NULL; | |
2169 | spin_unlock(&_minor_lock); | |
2170 | ||
1da177e4 | 2171 | put_disk(md->disk); |
1312f40e | 2172 | blk_cleanup_queue(md->queue); |
fd2ed4d2 | 2173 | dm_stats_cleanup(&md->stats); |
10da4f79 | 2174 | module_put(THIS_MODULE); |
1da177e4 LT |
2175 | kfree(md); |
2176 | } | |
2177 | ||
e6ee8c0b KU |
2178 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2179 | { | |
c0820cf5 | 2180 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2181 | |
5f015204 | 2182 | if (md->io_pool && md->bs) { |
16245bdc JN |
2183 | /* The md already has necessary mempools. */ |
2184 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2185 | /* | |
2186 | * Reload bioset because front_pad may have changed | |
2187 | * because a different table was loaded. | |
2188 | */ | |
2189 | bioset_free(md->bs); | |
2190 | md->bs = p->bs; | |
2191 | p->bs = NULL; | |
2192 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2193 | /* |
2194 | * There's no need to reload with request-based dm | |
2195 | * because the size of front_pad doesn't change. | |
2196 | * Note for future: If you are to reload bioset, | |
2197 | * prep-ed requests in the queue may refer | |
2198 | * to bio from the old bioset, so you must walk | |
2199 | * through the queue to unprep. | |
2200 | */ | |
2201 | } | |
e6ee8c0b | 2202 | goto out; |
c0820cf5 | 2203 | } |
e6ee8c0b | 2204 | |
5f015204 | 2205 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2206 | |
2207 | md->io_pool = p->io_pool; | |
2208 | p->io_pool = NULL; | |
e6ee8c0b KU |
2209 | md->bs = p->bs; |
2210 | p->bs = NULL; | |
2211 | ||
2212 | out: | |
2213 | /* mempool bind completed, now no need any mempools in the table */ | |
2214 | dm_table_free_md_mempools(t); | |
2215 | } | |
2216 | ||
1da177e4 LT |
2217 | /* |
2218 | * Bind a table to the device. | |
2219 | */ | |
2220 | static void event_callback(void *context) | |
2221 | { | |
7a8c3d3b MA |
2222 | unsigned long flags; |
2223 | LIST_HEAD(uevents); | |
1da177e4 LT |
2224 | struct mapped_device *md = (struct mapped_device *) context; |
2225 | ||
7a8c3d3b MA |
2226 | spin_lock_irqsave(&md->uevent_lock, flags); |
2227 | list_splice_init(&md->uevent_list, &uevents); | |
2228 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2229 | ||
ed9e1982 | 2230 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2231 | |
1da177e4 LT |
2232 | atomic_inc(&md->event_nr); |
2233 | wake_up(&md->eventq); | |
2234 | } | |
2235 | ||
c217649b MS |
2236 | /* |
2237 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2238 | */ | |
4e90188b | 2239 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2240 | { |
4e90188b | 2241 | set_capacity(md->disk, size); |
1da177e4 | 2242 | |
db8fef4f | 2243 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2244 | } |
2245 | ||
d5b9dd04 MP |
2246 | /* |
2247 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2248 | * | |
2249 | * If this function returns 0, then the device is either a non-dm | |
2250 | * device without a merge_bvec_fn, or it is a dm device that is | |
2251 | * able to split any bios it receives that are too big. | |
2252 | */ | |
2253 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2254 | { | |
2255 | struct mapped_device *dev_md; | |
2256 | ||
2257 | if (!q->merge_bvec_fn) | |
2258 | return 0; | |
2259 | ||
2260 | if (q->make_request_fn == dm_request) { | |
2261 | dev_md = q->queuedata; | |
2262 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2263 | return 0; | |
2264 | } | |
2265 | ||
2266 | return 1; | |
2267 | } | |
2268 | ||
2269 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2270 | struct dm_dev *dev, sector_t start, | |
2271 | sector_t len, void *data) | |
2272 | { | |
2273 | struct block_device *bdev = dev->bdev; | |
2274 | struct request_queue *q = bdev_get_queue(bdev); | |
2275 | ||
2276 | return dm_queue_merge_is_compulsory(q); | |
2277 | } | |
2278 | ||
2279 | /* | |
2280 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2281 | * on the properties of the underlying devices. | |
2282 | */ | |
2283 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2284 | { | |
2285 | unsigned i = 0; | |
2286 | struct dm_target *ti; | |
2287 | ||
2288 | while (i < dm_table_get_num_targets(table)) { | |
2289 | ti = dm_table_get_target(table, i++); | |
2290 | ||
2291 | if (ti->type->iterate_devices && | |
2292 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2293 | return 0; | |
2294 | } | |
2295 | ||
2296 | return 1; | |
2297 | } | |
2298 | ||
042d2a9b AK |
2299 | /* |
2300 | * Returns old map, which caller must destroy. | |
2301 | */ | |
2302 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2303 | struct queue_limits *limits) | |
1da177e4 | 2304 | { |
042d2a9b | 2305 | struct dm_table *old_map; |
165125e1 | 2306 | struct request_queue *q = md->queue; |
1da177e4 | 2307 | sector_t size; |
d5b9dd04 | 2308 | int merge_is_optional; |
1da177e4 LT |
2309 | |
2310 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2311 | |
2312 | /* | |
2313 | * Wipe any geometry if the size of the table changed. | |
2314 | */ | |
fd2ed4d2 | 2315 | if (size != dm_get_size(md)) |
3ac51e74 DW |
2316 | memset(&md->geometry, 0, sizeof(md->geometry)); |
2317 | ||
32a926da | 2318 | __set_size(md, size); |
d5816876 | 2319 | |
2ca3310e AK |
2320 | dm_table_event_callback(t, event_callback, md); |
2321 | ||
e6ee8c0b KU |
2322 | /* |
2323 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2324 | * for request-based during suspension. So stop it to prevent | |
2325 | * I/O mapping before resume. | |
2326 | * This must be done before setting the queue restrictions, | |
2327 | * because request-based dm may be run just after the setting. | |
2328 | */ | |
2329 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2330 | stop_queue(q); | |
2331 | ||
2332 | __bind_mempools(md, t); | |
2333 | ||
d5b9dd04 MP |
2334 | merge_is_optional = dm_table_merge_is_optional(t); |
2335 | ||
33423974 | 2336 | old_map = rcu_dereference(md->map); |
83d5e5b0 | 2337 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2338 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2339 | ||
754c5fc7 | 2340 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2341 | if (merge_is_optional) |
2342 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2343 | else | |
2344 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
41abc4e1 HR |
2345 | if (old_map) |
2346 | dm_sync_table(md); | |
1da177e4 | 2347 | |
042d2a9b | 2348 | return old_map; |
1da177e4 LT |
2349 | } |
2350 | ||
a7940155 AK |
2351 | /* |
2352 | * Returns unbound table for the caller to free. | |
2353 | */ | |
2354 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 | 2355 | { |
33423974 | 2356 | struct dm_table *map = rcu_dereference(md->map); |
1da177e4 LT |
2357 | |
2358 | if (!map) | |
a7940155 | 2359 | return NULL; |
1da177e4 LT |
2360 | |
2361 | dm_table_event_callback(map, NULL, NULL); | |
9cdb8520 | 2362 | RCU_INIT_POINTER(md->map, NULL); |
83d5e5b0 | 2363 | dm_sync_table(md); |
a7940155 AK |
2364 | |
2365 | return map; | |
1da177e4 LT |
2366 | } |
2367 | ||
2368 | /* | |
2369 | * Constructor for a new device. | |
2370 | */ | |
2b06cfff | 2371 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2372 | { |
2373 | struct mapped_device *md; | |
2374 | ||
2b06cfff | 2375 | md = alloc_dev(minor); |
1da177e4 LT |
2376 | if (!md) |
2377 | return -ENXIO; | |
2378 | ||
784aae73 MB |
2379 | dm_sysfs_init(md); |
2380 | ||
1da177e4 LT |
2381 | *result = md; |
2382 | return 0; | |
2383 | } | |
2384 | ||
a5664dad MS |
2385 | /* |
2386 | * Functions to manage md->type. | |
2387 | * All are required to hold md->type_lock. | |
2388 | */ | |
2389 | void dm_lock_md_type(struct mapped_device *md) | |
2390 | { | |
2391 | mutex_lock(&md->type_lock); | |
2392 | } | |
2393 | ||
2394 | void dm_unlock_md_type(struct mapped_device *md) | |
2395 | { | |
2396 | mutex_unlock(&md->type_lock); | |
2397 | } | |
2398 | ||
2399 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2400 | { | |
00c4fc3b | 2401 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2402 | md->type = type; |
2403 | } | |
2404 | ||
2405 | unsigned dm_get_md_type(struct mapped_device *md) | |
2406 | { | |
00c4fc3b | 2407 | BUG_ON(!mutex_is_locked(&md->type_lock)); |
a5664dad MS |
2408 | return md->type; |
2409 | } | |
2410 | ||
36a0456f AK |
2411 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2412 | { | |
2413 | return md->immutable_target_type; | |
2414 | } | |
2415 | ||
f84cb8a4 MS |
2416 | /* |
2417 | * The queue_limits are only valid as long as you have a reference | |
2418 | * count on 'md'. | |
2419 | */ | |
2420 | struct queue_limits *dm_get_queue_limits(struct mapped_device *md) | |
2421 | { | |
2422 | BUG_ON(!atomic_read(&md->holders)); | |
2423 | return &md->queue->limits; | |
2424 | } | |
2425 | EXPORT_SYMBOL_GPL(dm_get_queue_limits); | |
2426 | ||
4a0b4ddf MS |
2427 | /* |
2428 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2429 | */ | |
2430 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2431 | { | |
2432 | struct request_queue *q = NULL; | |
2433 | ||
2434 | if (md->queue->elevator) | |
2435 | return 1; | |
2436 | ||
2437 | /* Fully initialize the queue */ | |
2438 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2439 | if (!q) | |
2440 | return 0; | |
2441 | ||
2442 | md->queue = q; | |
4a0b4ddf MS |
2443 | dm_init_md_queue(md); |
2444 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2445 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2446 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2447 | |
2448 | elv_register_queue(md->queue); | |
2449 | ||
2450 | return 1; | |
2451 | } | |
2452 | ||
2453 | /* | |
2454 | * Setup the DM device's queue based on md's type | |
2455 | */ | |
2456 | int dm_setup_md_queue(struct mapped_device *md) | |
2457 | { | |
2458 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2459 | !dm_init_request_based_queue(md)) { | |
2460 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2461 | return -EINVAL; | |
2462 | } | |
2463 | ||
2464 | return 0; | |
2465 | } | |
2466 | ||
637842cf | 2467 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2468 | { |
2469 | struct mapped_device *md; | |
1da177e4 LT |
2470 | unsigned minor = MINOR(dev); |
2471 | ||
2472 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2473 | return NULL; | |
2474 | ||
f32c10b0 | 2475 | spin_lock(&_minor_lock); |
1da177e4 LT |
2476 | |
2477 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2478 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2479 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2480 | dm_deleting_md(md) || |
17b2f66f | 2481 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2482 | md = NULL; |
fba9f90e JM |
2483 | goto out; |
2484 | } | |
1da177e4 | 2485 | |
fba9f90e | 2486 | out: |
f32c10b0 | 2487 | spin_unlock(&_minor_lock); |
1da177e4 | 2488 | |
637842cf DT |
2489 | return md; |
2490 | } | |
2491 | ||
d229a958 DT |
2492 | struct mapped_device *dm_get_md(dev_t dev) |
2493 | { | |
2494 | struct mapped_device *md = dm_find_md(dev); | |
2495 | ||
2496 | if (md) | |
2497 | dm_get(md); | |
2498 | ||
2499 | return md; | |
2500 | } | |
3cf2e4ba | 2501 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2502 | |
9ade92a9 | 2503 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2504 | { |
9ade92a9 | 2505 | return md->interface_ptr; |
1da177e4 LT |
2506 | } |
2507 | ||
2508 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2509 | { | |
2510 | md->interface_ptr = ptr; | |
2511 | } | |
2512 | ||
2513 | void dm_get(struct mapped_device *md) | |
2514 | { | |
2515 | atomic_inc(&md->holders); | |
3f77316d | 2516 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2517 | } |
2518 | ||
72d94861 AK |
2519 | const char *dm_device_name(struct mapped_device *md) |
2520 | { | |
2521 | return md->name; | |
2522 | } | |
2523 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2524 | ||
3f77316d | 2525 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2526 | { |
1134e5ae | 2527 | struct dm_table *map; |
83d5e5b0 | 2528 | int srcu_idx; |
1da177e4 | 2529 | |
3f77316d | 2530 | might_sleep(); |
fba9f90e | 2531 | |
3f77316d | 2532 | spin_lock(&_minor_lock); |
83d5e5b0 | 2533 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2534 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2535 | set_bit(DMF_FREEING, &md->flags); | |
2536 | spin_unlock(&_minor_lock); | |
2537 | ||
2538 | if (!dm_suspended_md(md)) { | |
2539 | dm_table_presuspend_targets(map); | |
2540 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2541 | } |
3f77316d | 2542 | |
83d5e5b0 MP |
2543 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2544 | dm_put_live_table(md, srcu_idx); | |
2545 | ||
3f77316d KU |
2546 | /* |
2547 | * Rare, but there may be I/O requests still going to complete, | |
2548 | * for example. Wait for all references to disappear. | |
2549 | * No one should increment the reference count of the mapped_device, | |
2550 | * after the mapped_device state becomes DMF_FREEING. | |
2551 | */ | |
2552 | if (wait) | |
2553 | while (atomic_read(&md->holders)) | |
2554 | msleep(1); | |
2555 | else if (atomic_read(&md->holders)) | |
2556 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2557 | dm_device_name(md), atomic_read(&md->holders)); | |
2558 | ||
2559 | dm_sysfs_exit(md); | |
3f77316d KU |
2560 | dm_table_destroy(__unbind(md)); |
2561 | free_dev(md); | |
2562 | } | |
2563 | ||
2564 | void dm_destroy(struct mapped_device *md) | |
2565 | { | |
2566 | __dm_destroy(md, true); | |
2567 | } | |
2568 | ||
2569 | void dm_destroy_immediate(struct mapped_device *md) | |
2570 | { | |
2571 | __dm_destroy(md, false); | |
2572 | } | |
2573 | ||
2574 | void dm_put(struct mapped_device *md) | |
2575 | { | |
2576 | atomic_dec(&md->holders); | |
1da177e4 | 2577 | } |
79eb885c | 2578 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2579 | |
401600df | 2580 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2581 | { |
2582 | int r = 0; | |
b44ebeb0 MP |
2583 | DECLARE_WAITQUEUE(wait, current); |
2584 | ||
b44ebeb0 | 2585 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2586 | |
2587 | while (1) { | |
401600df | 2588 | set_current_state(interruptible); |
46125c1c | 2589 | |
b4324fee | 2590 | if (!md_in_flight(md)) |
46125c1c MB |
2591 | break; |
2592 | ||
401600df MP |
2593 | if (interruptible == TASK_INTERRUPTIBLE && |
2594 | signal_pending(current)) { | |
46125c1c MB |
2595 | r = -EINTR; |
2596 | break; | |
2597 | } | |
2598 | ||
2599 | io_schedule(); | |
2600 | } | |
2601 | set_current_state(TASK_RUNNING); | |
2602 | ||
b44ebeb0 MP |
2603 | remove_wait_queue(&md->wait, &wait); |
2604 | ||
46125c1c MB |
2605 | return r; |
2606 | } | |
2607 | ||
1da177e4 LT |
2608 | /* |
2609 | * Process the deferred bios | |
2610 | */ | |
ef208587 | 2611 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2612 | { |
ef208587 MP |
2613 | struct mapped_device *md = container_of(work, struct mapped_device, |
2614 | work); | |
6d6f10df | 2615 | struct bio *c; |
83d5e5b0 MP |
2616 | int srcu_idx; |
2617 | struct dm_table *map; | |
1da177e4 | 2618 | |
83d5e5b0 | 2619 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2620 | |
3b00b203 | 2621 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2622 | spin_lock_irq(&md->deferred_lock); |
2623 | c = bio_list_pop(&md->deferred); | |
2624 | spin_unlock_irq(&md->deferred_lock); | |
2625 | ||
6a8736d1 | 2626 | if (!c) |
df12ee99 | 2627 | break; |
022c2611 | 2628 | |
e6ee8c0b KU |
2629 | if (dm_request_based(md)) |
2630 | generic_make_request(c); | |
6a8736d1 | 2631 | else |
83d5e5b0 | 2632 | __split_and_process_bio(md, map, c); |
022c2611 | 2633 | } |
73d410c0 | 2634 | |
83d5e5b0 | 2635 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2636 | } |
2637 | ||
9a1fb464 | 2638 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2639 | { |
3b00b203 | 2640 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
4e857c58 | 2641 | smp_mb__after_atomic(); |
53d5914f | 2642 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2643 | } |
2644 | ||
1da177e4 | 2645 | /* |
042d2a9b | 2646 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2647 | */ |
042d2a9b | 2648 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2649 | { |
87eb5b21 | 2650 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2651 | struct queue_limits limits; |
042d2a9b | 2652 | int r; |
1da177e4 | 2653 | |
e61290a4 | 2654 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2655 | |
2656 | /* device must be suspended */ | |
4f186f8b | 2657 | if (!dm_suspended_md(md)) |
93c534ae | 2658 | goto out; |
1da177e4 | 2659 | |
3ae70656 MS |
2660 | /* |
2661 | * If the new table has no data devices, retain the existing limits. | |
2662 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2663 | * then new I/O is queued based on these limits, and then some paths | |
2664 | * reappear. | |
2665 | */ | |
2666 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2667 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2668 | if (live_map) |
2669 | limits = md->queue->limits; | |
83d5e5b0 | 2670 | dm_put_live_table_fast(md); |
3ae70656 MS |
2671 | } |
2672 | ||
87eb5b21 MC |
2673 | if (!live_map) { |
2674 | r = dm_calculate_queue_limits(table, &limits); | |
2675 | if (r) { | |
2676 | map = ERR_PTR(r); | |
2677 | goto out; | |
2678 | } | |
042d2a9b | 2679 | } |
754c5fc7 | 2680 | |
042d2a9b | 2681 | map = __bind(md, table, &limits); |
1da177e4 | 2682 | |
93c534ae | 2683 | out: |
e61290a4 | 2684 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2685 | return map; |
1da177e4 LT |
2686 | } |
2687 | ||
2688 | /* | |
2689 | * Functions to lock and unlock any filesystem running on the | |
2690 | * device. | |
2691 | */ | |
2ca3310e | 2692 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2693 | { |
e39e2e95 | 2694 | int r; |
1da177e4 LT |
2695 | |
2696 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2697 | |
db8fef4f | 2698 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2699 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2700 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2701 | md->frozen_sb = NULL; |
2702 | return r; | |
dfbe03f6 AK |
2703 | } |
2704 | ||
aa8d7c2f AK |
2705 | set_bit(DMF_FROZEN, &md->flags); |
2706 | ||
1da177e4 LT |
2707 | return 0; |
2708 | } | |
2709 | ||
2ca3310e | 2710 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2711 | { |
aa8d7c2f AK |
2712 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2713 | return; | |
2714 | ||
db8fef4f | 2715 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2716 | md->frozen_sb = NULL; |
aa8d7c2f | 2717 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2718 | } |
2719 | ||
2720 | /* | |
2721 | * We need to be able to change a mapping table under a mounted | |
2722 | * filesystem. For example we might want to move some data in | |
2723 | * the background. Before the table can be swapped with | |
2724 | * dm_bind_table, dm_suspend must be called to flush any in | |
2725 | * flight bios and ensure that any further io gets deferred. | |
2726 | */ | |
cec47e3d KU |
2727 | /* |
2728 | * Suspend mechanism in request-based dm. | |
2729 | * | |
9f518b27 KU |
2730 | * 1. Flush all I/Os by lock_fs() if needed. |
2731 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2732 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2733 | * |
9f518b27 | 2734 | * To abort suspend, start the request_queue. |
cec47e3d | 2735 | */ |
a3d77d35 | 2736 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2737 | { |
2ca3310e | 2738 | struct dm_table *map = NULL; |
46125c1c | 2739 | int r = 0; |
a3d77d35 | 2740 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2741 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2742 | |
e61290a4 | 2743 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2744 | |
4f186f8b | 2745 | if (dm_suspended_md(md)) { |
73d410c0 | 2746 | r = -EINVAL; |
d287483d | 2747 | goto out_unlock; |
73d410c0 | 2748 | } |
1da177e4 | 2749 | |
33423974 | 2750 | map = rcu_dereference(md->map); |
1da177e4 | 2751 | |
2e93ccc1 KU |
2752 | /* |
2753 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2754 | * This flag is cleared before dm_suspend returns. | |
2755 | */ | |
2756 | if (noflush) | |
2757 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2758 | ||
d67ee213 MS |
2759 | /* |
2760 | * This gets reverted if there's an error later and the targets | |
2761 | * provide the .presuspend_undo hook. | |
2762 | */ | |
cf222b37 AK |
2763 | dm_table_presuspend_targets(map); |
2764 | ||
32a926da | 2765 | /* |
9f518b27 KU |
2766 | * Flush I/O to the device. |
2767 | * Any I/O submitted after lock_fs() may not be flushed. | |
2768 | * noflush takes precedence over do_lockfs. | |
2769 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2770 | */ |
2771 | if (!noflush && do_lockfs) { | |
2772 | r = lock_fs(md); | |
d67ee213 MS |
2773 | if (r) { |
2774 | dm_table_presuspend_undo_targets(map); | |
83d5e5b0 | 2775 | goto out_unlock; |
d67ee213 | 2776 | } |
aa8d7c2f | 2777 | } |
1da177e4 LT |
2778 | |
2779 | /* | |
3b00b203 MP |
2780 | * Here we must make sure that no processes are submitting requests |
2781 | * to target drivers i.e. no one may be executing | |
2782 | * __split_and_process_bio. This is called from dm_request and | |
2783 | * dm_wq_work. | |
2784 | * | |
2785 | * To get all processes out of __split_and_process_bio in dm_request, | |
2786 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2787 | * __split_and_process_bio from dm_request and quiesce the thread |
2788 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2789 | * flush_workqueue(md->wq). | |
1da177e4 | 2790 | */ |
1eb787ec | 2791 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
41abc4e1 HR |
2792 | if (map) |
2793 | synchronize_srcu(&md->io_barrier); | |
1da177e4 | 2794 | |
d0bcb878 | 2795 | /* |
29e4013d TH |
2796 | * Stop md->queue before flushing md->wq in case request-based |
2797 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2798 | */ |
cec47e3d | 2799 | if (dm_request_based(md)) |
9f518b27 | 2800 | stop_queue(md->queue); |
cec47e3d | 2801 | |
d0bcb878 KU |
2802 | flush_workqueue(md->wq); |
2803 | ||
1da177e4 | 2804 | /* |
3b00b203 MP |
2805 | * At this point no more requests are entering target request routines. |
2806 | * We call dm_wait_for_completion to wait for all existing requests | |
2807 | * to finish. | |
1da177e4 | 2808 | */ |
401600df | 2809 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2810 | |
6d6f10df | 2811 | if (noflush) |
022c2611 | 2812 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
41abc4e1 HR |
2813 | if (map) |
2814 | synchronize_srcu(&md->io_barrier); | |
2e93ccc1 | 2815 | |
1da177e4 | 2816 | /* were we interrupted ? */ |
46125c1c | 2817 | if (r < 0) { |
9a1fb464 | 2818 | dm_queue_flush(md); |
73d410c0 | 2819 | |
cec47e3d | 2820 | if (dm_request_based(md)) |
9f518b27 | 2821 | start_queue(md->queue); |
cec47e3d | 2822 | |
2ca3310e | 2823 | unlock_fs(md); |
d67ee213 | 2824 | dm_table_presuspend_undo_targets(map); |
83d5e5b0 | 2825 | goto out_unlock; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2826 | } |
1da177e4 | 2827 | |
3b00b203 MP |
2828 | /* |
2829 | * If dm_wait_for_completion returned 0, the device is completely | |
2830 | * quiescent now. There is no request-processing activity. All new | |
2831 | * requests are being added to md->deferred list. | |
2832 | */ | |
2833 | ||
2ca3310e | 2834 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2835 | |
4d4471cb KU |
2836 | dm_table_postsuspend_targets(map); |
2837 | ||
d287483d | 2838 | out_unlock: |
e61290a4 | 2839 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2840 | return r; |
1da177e4 LT |
2841 | } |
2842 | ||
2843 | int dm_resume(struct mapped_device *md) | |
2844 | { | |
cf222b37 | 2845 | int r = -EINVAL; |
cf222b37 | 2846 | struct dm_table *map = NULL; |
1da177e4 | 2847 | |
e61290a4 | 2848 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2849 | if (!dm_suspended_md(md)) |
cf222b37 | 2850 | goto out; |
cf222b37 | 2851 | |
33423974 | 2852 | map = rcu_dereference(md->map); |
2ca3310e | 2853 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2854 | goto out; |
1da177e4 | 2855 | |
8757b776 MB |
2856 | r = dm_table_resume_targets(map); |
2857 | if (r) | |
2858 | goto out; | |
2ca3310e | 2859 | |
9a1fb464 | 2860 | dm_queue_flush(md); |
2ca3310e | 2861 | |
cec47e3d KU |
2862 | /* |
2863 | * Flushing deferred I/Os must be done after targets are resumed | |
2864 | * so that mapping of targets can work correctly. | |
2865 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2866 | */ | |
2867 | if (dm_request_based(md)) | |
2868 | start_queue(md->queue); | |
2869 | ||
2ca3310e AK |
2870 | unlock_fs(md); |
2871 | ||
2872 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2873 | ||
cf222b37 AK |
2874 | r = 0; |
2875 | out: | |
e61290a4 | 2876 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2877 | |
cf222b37 | 2878 | return r; |
1da177e4 LT |
2879 | } |
2880 | ||
fd2ed4d2 MP |
2881 | /* |
2882 | * Internal suspend/resume works like userspace-driven suspend. It waits | |
2883 | * until all bios finish and prevents issuing new bios to the target drivers. | |
2884 | * It may be used only from the kernel. | |
2885 | * | |
2886 | * Internal suspend holds md->suspend_lock, which prevents interaction with | |
2887 | * userspace-driven suspend. | |
2888 | */ | |
2889 | ||
2890 | void dm_internal_suspend(struct mapped_device *md) | |
2891 | { | |
2892 | mutex_lock(&md->suspend_lock); | |
2893 | if (dm_suspended_md(md)) | |
2894 | return; | |
2895 | ||
2896 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); | |
2897 | synchronize_srcu(&md->io_barrier); | |
2898 | flush_workqueue(md->wq); | |
2899 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
2900 | } | |
2901 | ||
2902 | void dm_internal_resume(struct mapped_device *md) | |
2903 | { | |
2904 | if (dm_suspended_md(md)) | |
2905 | goto done; | |
2906 | ||
2907 | dm_queue_flush(md); | |
2908 | ||
2909 | done: | |
2910 | mutex_unlock(&md->suspend_lock); | |
2911 | } | |
2912 | ||
1da177e4 LT |
2913 | /*----------------------------------------------------------------- |
2914 | * Event notification. | |
2915 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2916 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2917 | unsigned cookie) |
69267a30 | 2918 | { |
60935eb2 MB |
2919 | char udev_cookie[DM_COOKIE_LENGTH]; |
2920 | char *envp[] = { udev_cookie, NULL }; | |
2921 | ||
2922 | if (!cookie) | |
3abf85b5 | 2923 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2924 | else { |
2925 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2926 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2927 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2928 | action, envp); | |
60935eb2 | 2929 | } |
69267a30 AK |
2930 | } |
2931 | ||
7a8c3d3b MA |
2932 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2933 | { | |
2934 | return atomic_add_return(1, &md->uevent_seq); | |
2935 | } | |
2936 | ||
1da177e4 LT |
2937 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2938 | { | |
2939 | return atomic_read(&md->event_nr); | |
2940 | } | |
2941 | ||
2942 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2943 | { | |
2944 | return wait_event_interruptible(md->eventq, | |
2945 | (event_nr != atomic_read(&md->event_nr))); | |
2946 | } | |
2947 | ||
7a8c3d3b MA |
2948 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2949 | { | |
2950 | unsigned long flags; | |
2951 | ||
2952 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2953 | list_add(elist, &md->uevent_list); | |
2954 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2955 | } | |
2956 | ||
1da177e4 LT |
2957 | /* |
2958 | * The gendisk is only valid as long as you have a reference | |
2959 | * count on 'md'. | |
2960 | */ | |
2961 | struct gendisk *dm_disk(struct mapped_device *md) | |
2962 | { | |
2963 | return md->disk; | |
2964 | } | |
2965 | ||
784aae73 MB |
2966 | struct kobject *dm_kobject(struct mapped_device *md) |
2967 | { | |
2995fa78 | 2968 | return &md->kobj_holder.kobj; |
784aae73 MB |
2969 | } |
2970 | ||
784aae73 MB |
2971 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) |
2972 | { | |
2973 | struct mapped_device *md; | |
2974 | ||
2995fa78 | 2975 | md = container_of(kobj, struct mapped_device, kobj_holder.kobj); |
784aae73 | 2976 | |
4d89b7b4 | 2977 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2978 | dm_deleting_md(md)) |
4d89b7b4 MB |
2979 | return NULL; |
2980 | ||
784aae73 MB |
2981 | dm_get(md); |
2982 | return md; | |
2983 | } | |
2984 | ||
4f186f8b | 2985 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2986 | { |
2987 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2988 | } | |
2989 | ||
2c140a24 MP |
2990 | int dm_test_deferred_remove_flag(struct mapped_device *md) |
2991 | { | |
2992 | return test_bit(DMF_DEFERRED_REMOVE, &md->flags); | |
2993 | } | |
2994 | ||
64dbce58 KU |
2995 | int dm_suspended(struct dm_target *ti) |
2996 | { | |
ecdb2e25 | 2997 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2998 | } |
2999 | EXPORT_SYMBOL_GPL(dm_suspended); | |
3000 | ||
2e93ccc1 KU |
3001 | int dm_noflush_suspending(struct dm_target *ti) |
3002 | { | |
ecdb2e25 | 3003 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
3004 | } |
3005 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
3006 | ||
c0820cf5 | 3007 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 3008 | { |
5f015204 JN |
3009 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
3010 | struct kmem_cache *cachep; | |
3011 | unsigned int pool_size; | |
3012 | unsigned int front_pad; | |
e6ee8c0b KU |
3013 | |
3014 | if (!pools) | |
3015 | return NULL; | |
3016 | ||
23e5083b | 3017 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 | 3018 | cachep = _io_cache; |
e8603136 | 3019 | pool_size = dm_get_reserved_bio_based_ios(); |
5f015204 JN |
3020 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); |
3021 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
3022 | cachep = _rq_tio_cache; | |
f4790826 | 3023 | pool_size = dm_get_reserved_rq_based_ios(); |
5f015204 JN |
3024 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); |
3025 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
3026 | WARN_ON(per_bio_data_size != 0); | |
3027 | } else | |
3028 | goto out; | |
e6ee8c0b | 3029 | |
6cfa5857 | 3030 | pools->io_pool = mempool_create_slab_pool(pool_size, cachep); |
5f015204 JN |
3031 | if (!pools->io_pool) |
3032 | goto out; | |
e6ee8c0b | 3033 | |
3d8aab2d | 3034 | pools->bs = bioset_create_nobvec(pool_size, front_pad); |
e6ee8c0b | 3035 | if (!pools->bs) |
5f015204 | 3036 | goto out; |
e6ee8c0b | 3037 | |
a91a2785 | 3038 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 3039 | goto out; |
a91a2785 | 3040 | |
e6ee8c0b KU |
3041 | return pools; |
3042 | ||
5f015204 JN |
3043 | out: |
3044 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
3045 | |
3046 | return NULL; | |
3047 | } | |
3048 | ||
3049 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
3050 | { | |
3051 | if (!pools) | |
3052 | return; | |
3053 | ||
3054 | if (pools->io_pool) | |
3055 | mempool_destroy(pools->io_pool); | |
3056 | ||
e6ee8c0b KU |
3057 | if (pools->bs) |
3058 | bioset_free(pools->bs); | |
3059 | ||
3060 | kfree(pools); | |
3061 | } | |
3062 | ||
83d5cde4 | 3063 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
3064 | .open = dm_blk_open, |
3065 | .release = dm_blk_close, | |
aa129a22 | 3066 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 3067 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
3068 | .owner = THIS_MODULE |
3069 | }; | |
3070 | ||
1da177e4 LT |
3071 | /* |
3072 | * module hooks | |
3073 | */ | |
3074 | module_init(dm_init); | |
3075 | module_exit(dm_exit); | |
3076 | ||
3077 | module_param(major, uint, 0); | |
3078 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
f4790826 | 3079 | |
e8603136 MS |
3080 | module_param(reserved_bio_based_ios, uint, S_IRUGO | S_IWUSR); |
3081 | MODULE_PARM_DESC(reserved_bio_based_ios, "Reserved IOs in bio-based mempools"); | |
3082 | ||
f4790826 MS |
3083 | module_param(reserved_rq_based_ios, uint, S_IRUGO | S_IWUSR); |
3084 | MODULE_PARM_DESC(reserved_rq_based_ios, "Reserved IOs in request-based mempools"); | |
3085 | ||
1da177e4 LT |
3086 | MODULE_DESCRIPTION(DM_NAME " driver"); |
3087 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
3088 | MODULE_LICENSE("GPL"); |