]>
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> | |
17 | #include <linux/buffer_head.h> | |
18 | #include <linux/mempool.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/idr.h> | |
3ac51e74 | 21 | #include <linux/hdreg.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
60935eb2 MB |
27 | /* |
28 | * Cookies are numeric values sent with CHANGE and REMOVE | |
29 | * uevents while resuming, removing or renaming the device. | |
30 | */ | |
31 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
32 | #define DM_COOKIE_LENGTH 24 | |
33 | ||
1da177e4 LT |
34 | static const char *_name = DM_NAME; |
35 | ||
36 | static unsigned int major = 0; | |
37 | static unsigned int _major = 0; | |
38 | ||
f32c10b0 | 39 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 40 | /* |
8fbf26ad | 41 | * For bio-based dm. |
1da177e4 LT |
42 | * One of these is allocated per bio. |
43 | */ | |
44 | struct dm_io { | |
45 | struct mapped_device *md; | |
46 | int error; | |
1da177e4 | 47 | atomic_t io_count; |
6ae2fa67 | 48 | struct bio *bio; |
3eaf840e | 49 | unsigned long start_time; |
1da177e4 LT |
50 | }; |
51 | ||
52 | /* | |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per target within a bio. Hopefully |
55 | * this will be simplified out one day. | |
56 | */ | |
028867ac | 57 | struct dm_target_io { |
1da177e4 LT |
58 | struct dm_io *io; |
59 | struct dm_target *ti; | |
60 | union map_info info; | |
61 | }; | |
62 | ||
8fbf26ad KU |
63 | /* |
64 | * For request-based dm. | |
65 | * One of these is allocated per request. | |
66 | */ | |
67 | struct dm_rq_target_io { | |
68 | struct mapped_device *md; | |
69 | struct dm_target *ti; | |
70 | struct request *orig, clone; | |
71 | int error; | |
72 | union map_info info; | |
73 | }; | |
74 | ||
75 | /* | |
76 | * For request-based dm. | |
77 | * One of these is allocated per bio. | |
78 | */ | |
79 | struct dm_rq_clone_bio_info { | |
80 | struct bio *orig; | |
cec47e3d | 81 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
82 | }; |
83 | ||
1da177e4 LT |
84 | union map_info *dm_get_mapinfo(struct bio *bio) |
85 | { | |
17b2f66f | 86 | if (bio && bio->bi_private) |
028867ac | 87 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 88 | return NULL; |
1da177e4 LT |
89 | } |
90 | ||
cec47e3d KU |
91 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
92 | { | |
93 | if (rq && rq->end_io_data) | |
94 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
95 | return NULL; | |
96 | } | |
97 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
98 | ||
ba61fdd1 JM |
99 | #define MINOR_ALLOCED ((void *)-1) |
100 | ||
1da177e4 LT |
101 | /* |
102 | * Bits for the md->flags field. | |
103 | */ | |
1eb787ec | 104 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 105 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 106 | #define DMF_FROZEN 2 |
fba9f90e | 107 | #define DMF_FREEING 3 |
5c6bd75d | 108 | #define DMF_DELETING 4 |
2e93ccc1 | 109 | #define DMF_NOFLUSH_SUSPENDING 5 |
1eb787ec | 110 | #define DMF_QUEUE_IO_TO_THREAD 6 |
1da177e4 | 111 | |
304f3f6a MB |
112 | /* |
113 | * Work processed by per-device workqueue. | |
114 | */ | |
1da177e4 | 115 | struct mapped_device { |
2ca3310e | 116 | struct rw_semaphore io_lock; |
e61290a4 | 117 | struct mutex suspend_lock; |
1da177e4 LT |
118 | rwlock_t map_lock; |
119 | atomic_t holders; | |
5c6bd75d | 120 | atomic_t open_count; |
1da177e4 LT |
121 | |
122 | unsigned long flags; | |
123 | ||
165125e1 | 124 | struct request_queue *queue; |
1da177e4 | 125 | struct gendisk *disk; |
7e51f257 | 126 | char name[16]; |
1da177e4 LT |
127 | |
128 | void *interface_ptr; | |
129 | ||
130 | /* | |
131 | * A list of ios that arrived while we were suspended. | |
132 | */ | |
133 | atomic_t pending; | |
134 | wait_queue_head_t wait; | |
53d5914f | 135 | struct work_struct work; |
74859364 | 136 | struct bio_list deferred; |
022c2611 | 137 | spinlock_t deferred_lock; |
1da177e4 | 138 | |
af7e466a MP |
139 | /* |
140 | * An error from the barrier request currently being processed. | |
141 | */ | |
142 | int barrier_error; | |
143 | ||
304f3f6a MB |
144 | /* |
145 | * Processing queue (flush/barriers) | |
146 | */ | |
147 | struct workqueue_struct *wq; | |
148 | ||
1da177e4 LT |
149 | /* |
150 | * The current mapping. | |
151 | */ | |
152 | struct dm_table *map; | |
153 | ||
154 | /* | |
155 | * io objects are allocated from here. | |
156 | */ | |
157 | mempool_t *io_pool; | |
158 | mempool_t *tio_pool; | |
159 | ||
9faf400f SB |
160 | struct bio_set *bs; |
161 | ||
1da177e4 LT |
162 | /* |
163 | * Event handling. | |
164 | */ | |
165 | atomic_t event_nr; | |
166 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
167 | atomic_t uevent_seq; |
168 | struct list_head uevent_list; | |
169 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
170 | |
171 | /* | |
172 | * freeze/thaw support require holding onto a super block | |
173 | */ | |
174 | struct super_block *frozen_sb; | |
db8fef4f | 175 | struct block_device *bdev; |
3ac51e74 DW |
176 | |
177 | /* forced geometry settings */ | |
178 | struct hd_geometry geometry; | |
784aae73 | 179 | |
cec47e3d KU |
180 | /* marker of flush suspend for request-based dm */ |
181 | struct request suspend_rq; | |
182 | ||
183 | /* For saving the address of __make_request for request based dm */ | |
184 | make_request_fn *saved_make_request_fn; | |
185 | ||
784aae73 MB |
186 | /* sysfs handle */ |
187 | struct kobject kobj; | |
52b1fd5a MP |
188 | |
189 | /* zero-length barrier that will be cloned and submitted to targets */ | |
190 | struct bio barrier_bio; | |
1da177e4 LT |
191 | }; |
192 | ||
e6ee8c0b KU |
193 | /* |
194 | * For mempools pre-allocation at the table loading time. | |
195 | */ | |
196 | struct dm_md_mempools { | |
197 | mempool_t *io_pool; | |
198 | mempool_t *tio_pool; | |
199 | struct bio_set *bs; | |
200 | }; | |
201 | ||
1da177e4 | 202 | #define MIN_IOS 256 |
e18b890b CL |
203 | static struct kmem_cache *_io_cache; |
204 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
205 | static struct kmem_cache *_rq_tio_cache; |
206 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 207 | |
1da177e4 LT |
208 | static int __init local_init(void) |
209 | { | |
51157b4a | 210 | int r = -ENOMEM; |
1da177e4 | 211 | |
1da177e4 | 212 | /* allocate a slab for the dm_ios */ |
028867ac | 213 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 214 | if (!_io_cache) |
51157b4a | 215 | return r; |
1da177e4 LT |
216 | |
217 | /* allocate a slab for the target ios */ | |
028867ac | 218 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
219 | if (!_tio_cache) |
220 | goto out_free_io_cache; | |
1da177e4 | 221 | |
8fbf26ad KU |
222 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
223 | if (!_rq_tio_cache) | |
224 | goto out_free_tio_cache; | |
225 | ||
226 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
227 | if (!_rq_bio_info_cache) | |
228 | goto out_free_rq_tio_cache; | |
229 | ||
51e5b2bd | 230 | r = dm_uevent_init(); |
51157b4a | 231 | if (r) |
8fbf26ad | 232 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 233 | |
1da177e4 LT |
234 | _major = major; |
235 | r = register_blkdev(_major, _name); | |
51157b4a KU |
236 | if (r < 0) |
237 | goto out_uevent_exit; | |
1da177e4 LT |
238 | |
239 | if (!_major) | |
240 | _major = r; | |
241 | ||
242 | return 0; | |
51157b4a KU |
243 | |
244 | out_uevent_exit: | |
245 | dm_uevent_exit(); | |
8fbf26ad KU |
246 | out_free_rq_bio_info_cache: |
247 | kmem_cache_destroy(_rq_bio_info_cache); | |
248 | out_free_rq_tio_cache: | |
249 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
250 | out_free_tio_cache: |
251 | kmem_cache_destroy(_tio_cache); | |
252 | out_free_io_cache: | |
253 | kmem_cache_destroy(_io_cache); | |
254 | ||
255 | return r; | |
1da177e4 LT |
256 | } |
257 | ||
258 | static void local_exit(void) | |
259 | { | |
8fbf26ad KU |
260 | kmem_cache_destroy(_rq_bio_info_cache); |
261 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
262 | kmem_cache_destroy(_tio_cache); |
263 | kmem_cache_destroy(_io_cache); | |
00d59405 | 264 | unregister_blkdev(_major, _name); |
51e5b2bd | 265 | dm_uevent_exit(); |
1da177e4 LT |
266 | |
267 | _major = 0; | |
268 | ||
269 | DMINFO("cleaned up"); | |
270 | } | |
271 | ||
b9249e55 | 272 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
273 | local_init, |
274 | dm_target_init, | |
275 | dm_linear_init, | |
276 | dm_stripe_init, | |
945fa4d2 | 277 | dm_kcopyd_init, |
1da177e4 LT |
278 | dm_interface_init, |
279 | }; | |
280 | ||
b9249e55 | 281 | static void (*_exits[])(void) = { |
1da177e4 LT |
282 | local_exit, |
283 | dm_target_exit, | |
284 | dm_linear_exit, | |
285 | dm_stripe_exit, | |
945fa4d2 | 286 | dm_kcopyd_exit, |
1da177e4 LT |
287 | dm_interface_exit, |
288 | }; | |
289 | ||
290 | static int __init dm_init(void) | |
291 | { | |
292 | const int count = ARRAY_SIZE(_inits); | |
293 | ||
294 | int r, i; | |
295 | ||
296 | for (i = 0; i < count; i++) { | |
297 | r = _inits[i](); | |
298 | if (r) | |
299 | goto bad; | |
300 | } | |
301 | ||
302 | return 0; | |
303 | ||
304 | bad: | |
305 | while (i--) | |
306 | _exits[i](); | |
307 | ||
308 | return r; | |
309 | } | |
310 | ||
311 | static void __exit dm_exit(void) | |
312 | { | |
313 | int i = ARRAY_SIZE(_exits); | |
314 | ||
315 | while (i--) | |
316 | _exits[i](); | |
317 | } | |
318 | ||
319 | /* | |
320 | * Block device functions | |
321 | */ | |
fe5f9f2c | 322 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
323 | { |
324 | struct mapped_device *md; | |
325 | ||
fba9f90e JM |
326 | spin_lock(&_minor_lock); |
327 | ||
fe5f9f2c | 328 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
329 | if (!md) |
330 | goto out; | |
331 | ||
5c6bd75d AK |
332 | if (test_bit(DMF_FREEING, &md->flags) || |
333 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
334 | md = NULL; |
335 | goto out; | |
336 | } | |
337 | ||
1da177e4 | 338 | dm_get(md); |
5c6bd75d | 339 | atomic_inc(&md->open_count); |
fba9f90e JM |
340 | |
341 | out: | |
342 | spin_unlock(&_minor_lock); | |
343 | ||
344 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
345 | } |
346 | ||
fe5f9f2c | 347 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 348 | { |
fe5f9f2c | 349 | struct mapped_device *md = disk->private_data; |
5c6bd75d | 350 | atomic_dec(&md->open_count); |
1da177e4 LT |
351 | dm_put(md); |
352 | return 0; | |
353 | } | |
354 | ||
5c6bd75d AK |
355 | int dm_open_count(struct mapped_device *md) |
356 | { | |
357 | return atomic_read(&md->open_count); | |
358 | } | |
359 | ||
360 | /* | |
361 | * Guarantees nothing is using the device before it's deleted. | |
362 | */ | |
363 | int dm_lock_for_deletion(struct mapped_device *md) | |
364 | { | |
365 | int r = 0; | |
366 | ||
367 | spin_lock(&_minor_lock); | |
368 | ||
369 | if (dm_open_count(md)) | |
370 | r = -EBUSY; | |
371 | else | |
372 | set_bit(DMF_DELETING, &md->flags); | |
373 | ||
374 | spin_unlock(&_minor_lock); | |
375 | ||
376 | return r; | |
377 | } | |
378 | ||
3ac51e74 DW |
379 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
380 | { | |
381 | struct mapped_device *md = bdev->bd_disk->private_data; | |
382 | ||
383 | return dm_get_geometry(md, geo); | |
384 | } | |
385 | ||
fe5f9f2c | 386 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
387 | unsigned int cmd, unsigned long arg) |
388 | { | |
fe5f9f2c AV |
389 | struct mapped_device *md = bdev->bd_disk->private_data; |
390 | struct dm_table *map = dm_get_table(md); | |
aa129a22 MB |
391 | struct dm_target *tgt; |
392 | int r = -ENOTTY; | |
393 | ||
aa129a22 MB |
394 | if (!map || !dm_table_get_size(map)) |
395 | goto out; | |
396 | ||
397 | /* We only support devices that have a single target */ | |
398 | if (dm_table_get_num_targets(map) != 1) | |
399 | goto out; | |
400 | ||
401 | tgt = dm_table_get_target(map, 0); | |
402 | ||
403 | if (dm_suspended(md)) { | |
404 | r = -EAGAIN; | |
405 | goto out; | |
406 | } | |
407 | ||
408 | if (tgt->type->ioctl) | |
647b3d00 | 409 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
410 | |
411 | out: | |
412 | dm_table_put(map); | |
413 | ||
aa129a22 MB |
414 | return r; |
415 | } | |
416 | ||
028867ac | 417 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
418 | { |
419 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
420 | } | |
421 | ||
028867ac | 422 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
423 | { |
424 | mempool_free(io, md->io_pool); | |
425 | } | |
426 | ||
028867ac | 427 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
428 | { |
429 | mempool_free(tio, md->tio_pool); | |
430 | } | |
431 | ||
cec47e3d KU |
432 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md) |
433 | { | |
434 | return mempool_alloc(md->tio_pool, GFP_ATOMIC); | |
435 | } | |
436 | ||
437 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
438 | { | |
439 | mempool_free(tio, tio->md->tio_pool); | |
440 | } | |
441 | ||
442 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
443 | { | |
444 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
445 | } | |
446 | ||
447 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
448 | { | |
449 | mempool_free(info, info->tio->md->io_pool); | |
450 | } | |
451 | ||
3eaf840e JNN |
452 | static void start_io_acct(struct dm_io *io) |
453 | { | |
454 | struct mapped_device *md = io->md; | |
c9959059 | 455 | int cpu; |
3eaf840e JNN |
456 | |
457 | io->start_time = jiffies; | |
458 | ||
074a7aca TH |
459 | cpu = part_stat_lock(); |
460 | part_round_stats(cpu, &dm_disk(md)->part0); | |
461 | part_stat_unlock(); | |
462 | dm_disk(md)->part0.in_flight = atomic_inc_return(&md->pending); | |
3eaf840e JNN |
463 | } |
464 | ||
d221d2e7 | 465 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
466 | { |
467 | struct mapped_device *md = io->md; | |
468 | struct bio *bio = io->bio; | |
469 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 470 | int pending, cpu; |
3eaf840e JNN |
471 | int rw = bio_data_dir(bio); |
472 | ||
074a7aca TH |
473 | cpu = part_stat_lock(); |
474 | part_round_stats(cpu, &dm_disk(md)->part0); | |
475 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
476 | part_stat_unlock(); | |
3eaf840e | 477 | |
af7e466a MP |
478 | /* |
479 | * After this is decremented the bio must not be touched if it is | |
480 | * a barrier. | |
481 | */ | |
074a7aca TH |
482 | dm_disk(md)->part0.in_flight = pending = |
483 | atomic_dec_return(&md->pending); | |
3eaf840e | 484 | |
d221d2e7 MP |
485 | /* nudge anyone waiting on suspend queue */ |
486 | if (!pending) | |
487 | wake_up(&md->wait); | |
3eaf840e JNN |
488 | } |
489 | ||
1da177e4 LT |
490 | /* |
491 | * Add the bio to the list of deferred io. | |
492 | */ | |
92c63902 | 493 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 494 | { |
2ca3310e | 495 | down_write(&md->io_lock); |
1da177e4 | 496 | |
022c2611 | 497 | spin_lock_irq(&md->deferred_lock); |
1da177e4 | 498 | bio_list_add(&md->deferred, bio); |
022c2611 | 499 | spin_unlock_irq(&md->deferred_lock); |
1da177e4 | 500 | |
92c63902 MP |
501 | if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) |
502 | queue_work(md->wq, &md->work); | |
503 | ||
2ca3310e | 504 | up_write(&md->io_lock); |
1da177e4 LT |
505 | } |
506 | ||
507 | /* | |
508 | * Everyone (including functions in this file), should use this | |
509 | * function to access the md->map field, and make sure they call | |
510 | * dm_table_put() when finished. | |
511 | */ | |
512 | struct dm_table *dm_get_table(struct mapped_device *md) | |
513 | { | |
514 | struct dm_table *t; | |
523d9297 | 515 | unsigned long flags; |
1da177e4 | 516 | |
523d9297 | 517 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
518 | t = md->map; |
519 | if (t) | |
520 | dm_table_get(t); | |
523d9297 | 521 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
522 | |
523 | return t; | |
524 | } | |
525 | ||
3ac51e74 DW |
526 | /* |
527 | * Get the geometry associated with a dm device | |
528 | */ | |
529 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
530 | { | |
531 | *geo = md->geometry; | |
532 | ||
533 | return 0; | |
534 | } | |
535 | ||
536 | /* | |
537 | * Set the geometry of a device. | |
538 | */ | |
539 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
540 | { | |
541 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
542 | ||
543 | if (geo->start > sz) { | |
544 | DMWARN("Start sector is beyond the geometry limits."); | |
545 | return -EINVAL; | |
546 | } | |
547 | ||
548 | md->geometry = *geo; | |
549 | ||
550 | return 0; | |
551 | } | |
552 | ||
1da177e4 LT |
553 | /*----------------------------------------------------------------- |
554 | * CRUD START: | |
555 | * A more elegant soln is in the works that uses the queue | |
556 | * merge fn, unfortunately there are a couple of changes to | |
557 | * the block layer that I want to make for this. So in the | |
558 | * interests of getting something for people to use I give | |
559 | * you this clearly demarcated crap. | |
560 | *---------------------------------------------------------------*/ | |
561 | ||
2e93ccc1 KU |
562 | static int __noflush_suspending(struct mapped_device *md) |
563 | { | |
564 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
565 | } | |
566 | ||
1da177e4 LT |
567 | /* |
568 | * Decrements the number of outstanding ios that a bio has been | |
569 | * cloned into, completing the original io if necc. | |
570 | */ | |
858119e1 | 571 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 572 | { |
2e93ccc1 | 573 | unsigned long flags; |
b35f8caa MB |
574 | int io_error; |
575 | struct bio *bio; | |
576 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
577 | |
578 | /* Push-back supersedes any I/O errors */ | |
b35f8caa | 579 | if (error && !(io->error > 0 && __noflush_suspending(md))) |
1da177e4 LT |
580 | io->error = error; |
581 | ||
582 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
583 | if (io->error == DM_ENDIO_REQUEUE) { |
584 | /* | |
585 | * Target requested pushing back the I/O. | |
2e93ccc1 | 586 | */ |
022c2611 | 587 | spin_lock_irqsave(&md->deferred_lock, flags); |
2761e95f MP |
588 | if (__noflush_suspending(md)) { |
589 | if (!bio_barrier(io->bio)) | |
590 | bio_list_add_head(&md->deferred, | |
591 | io->bio); | |
592 | } else | |
2e93ccc1 KU |
593 | /* noflush suspend was interrupted. */ |
594 | io->error = -EIO; | |
022c2611 | 595 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
596 | } |
597 | ||
b35f8caa MB |
598 | io_error = io->error; |
599 | bio = io->bio; | |
2e93ccc1 | 600 | |
af7e466a MP |
601 | if (bio_barrier(bio)) { |
602 | /* | |
603 | * There can be just one barrier request so we use | |
604 | * a per-device variable for error reporting. | |
605 | * Note that you can't touch the bio after end_io_acct | |
606 | */ | |
fdb9572b | 607 | if (!md->barrier_error && io_error != -EOPNOTSUPP) |
5aa2781d | 608 | md->barrier_error = io_error; |
af7e466a MP |
609 | end_io_acct(io); |
610 | } else { | |
611 | end_io_acct(io); | |
b35f8caa | 612 | |
af7e466a MP |
613 | if (io_error != DM_ENDIO_REQUEUE) { |
614 | trace_block_bio_complete(md->queue, bio); | |
2056a782 | 615 | |
af7e466a MP |
616 | bio_endio(bio, io_error); |
617 | } | |
b35f8caa | 618 | } |
af7e466a MP |
619 | |
620 | free_io(md, io); | |
1da177e4 LT |
621 | } |
622 | } | |
623 | ||
6712ecf8 | 624 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
625 | { |
626 | int r = 0; | |
028867ac | 627 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 628 | struct dm_io *io = tio->io; |
9faf400f | 629 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
630 | dm_endio_fn endio = tio->ti->type->end_io; |
631 | ||
1da177e4 LT |
632 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
633 | error = -EIO; | |
634 | ||
635 | if (endio) { | |
636 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
637 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
638 | /* | |
639 | * error and requeue request are handled | |
640 | * in dec_pending(). | |
641 | */ | |
1da177e4 | 642 | error = r; |
45cbcd79 KU |
643 | else if (r == DM_ENDIO_INCOMPLETE) |
644 | /* The target will handle the io */ | |
6712ecf8 | 645 | return; |
45cbcd79 KU |
646 | else if (r) { |
647 | DMWARN("unimplemented target endio return value: %d", r); | |
648 | BUG(); | |
649 | } | |
1da177e4 LT |
650 | } |
651 | ||
9faf400f SB |
652 | /* |
653 | * Store md for cleanup instead of tio which is about to get freed. | |
654 | */ | |
655 | bio->bi_private = md->bs; | |
656 | ||
9faf400f | 657 | free_tio(md, tio); |
b35f8caa MB |
658 | bio_put(bio); |
659 | dec_pending(io, error); | |
1da177e4 LT |
660 | } |
661 | ||
cec47e3d KU |
662 | /* |
663 | * Partial completion handling for request-based dm | |
664 | */ | |
665 | static void end_clone_bio(struct bio *clone, int error) | |
666 | { | |
667 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
668 | struct dm_rq_target_io *tio = info->tio; | |
669 | struct bio *bio = info->orig; | |
670 | unsigned int nr_bytes = info->orig->bi_size; | |
671 | ||
672 | bio_put(clone); | |
673 | ||
674 | if (tio->error) | |
675 | /* | |
676 | * An error has already been detected on the request. | |
677 | * Once error occurred, just let clone->end_io() handle | |
678 | * the remainder. | |
679 | */ | |
680 | return; | |
681 | else if (error) { | |
682 | /* | |
683 | * Don't notice the error to the upper layer yet. | |
684 | * The error handling decision is made by the target driver, | |
685 | * when the request is completed. | |
686 | */ | |
687 | tio->error = error; | |
688 | return; | |
689 | } | |
690 | ||
691 | /* | |
692 | * I/O for the bio successfully completed. | |
693 | * Notice the data completion to the upper layer. | |
694 | */ | |
695 | ||
696 | /* | |
697 | * bios are processed from the head of the list. | |
698 | * So the completing bio should always be rq->bio. | |
699 | * If it's not, something wrong is happening. | |
700 | */ | |
701 | if (tio->orig->bio != bio) | |
702 | DMERR("bio completion is going in the middle of the request"); | |
703 | ||
704 | /* | |
705 | * Update the original request. | |
706 | * Do not use blk_end_request() here, because it may complete | |
707 | * the original request before the clone, and break the ordering. | |
708 | */ | |
709 | blk_update_request(tio->orig, 0, nr_bytes); | |
710 | } | |
711 | ||
712 | /* | |
713 | * Don't touch any member of the md after calling this function because | |
714 | * the md may be freed in dm_put() at the end of this function. | |
715 | * Or do dm_get() before calling this function and dm_put() later. | |
716 | */ | |
717 | static void rq_completed(struct mapped_device *md, int run_queue) | |
718 | { | |
719 | int wakeup_waiters = 0; | |
720 | struct request_queue *q = md->queue; | |
721 | unsigned long flags; | |
722 | ||
723 | spin_lock_irqsave(q->queue_lock, flags); | |
724 | if (!queue_in_flight(q)) | |
725 | wakeup_waiters = 1; | |
726 | spin_unlock_irqrestore(q->queue_lock, flags); | |
727 | ||
728 | /* nudge anyone waiting on suspend queue */ | |
729 | if (wakeup_waiters) | |
730 | wake_up(&md->wait); | |
731 | ||
732 | if (run_queue) | |
733 | blk_run_queue(q); | |
734 | ||
735 | /* | |
736 | * dm_put() must be at the end of this function. See the comment above | |
737 | */ | |
738 | dm_put(md); | |
739 | } | |
740 | ||
a77e28c7 KU |
741 | static void free_rq_clone(struct request *clone) |
742 | { | |
743 | struct dm_rq_target_io *tio = clone->end_io_data; | |
744 | ||
745 | blk_rq_unprep_clone(clone); | |
746 | free_rq_tio(tio); | |
747 | } | |
748 | ||
cec47e3d KU |
749 | static void dm_unprep_request(struct request *rq) |
750 | { | |
751 | struct request *clone = rq->special; | |
cec47e3d KU |
752 | |
753 | rq->special = NULL; | |
754 | rq->cmd_flags &= ~REQ_DONTPREP; | |
755 | ||
a77e28c7 | 756 | free_rq_clone(clone); |
cec47e3d KU |
757 | } |
758 | ||
759 | /* | |
760 | * Requeue the original request of a clone. | |
761 | */ | |
762 | void dm_requeue_unmapped_request(struct request *clone) | |
763 | { | |
764 | struct dm_rq_target_io *tio = clone->end_io_data; | |
765 | struct mapped_device *md = tio->md; | |
766 | struct request *rq = tio->orig; | |
767 | struct request_queue *q = rq->q; | |
768 | unsigned long flags; | |
769 | ||
770 | dm_unprep_request(rq); | |
771 | ||
772 | spin_lock_irqsave(q->queue_lock, flags); | |
773 | if (elv_queue_empty(q)) | |
774 | blk_plug_device(q); | |
775 | blk_requeue_request(q, rq); | |
776 | spin_unlock_irqrestore(q->queue_lock, flags); | |
777 | ||
778 | rq_completed(md, 0); | |
779 | } | |
780 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
781 | ||
782 | static void __stop_queue(struct request_queue *q) | |
783 | { | |
784 | blk_stop_queue(q); | |
785 | } | |
786 | ||
787 | static void stop_queue(struct request_queue *q) | |
788 | { | |
789 | unsigned long flags; | |
790 | ||
791 | spin_lock_irqsave(q->queue_lock, flags); | |
792 | __stop_queue(q); | |
793 | spin_unlock_irqrestore(q->queue_lock, flags); | |
794 | } | |
795 | ||
796 | static void __start_queue(struct request_queue *q) | |
797 | { | |
798 | if (blk_queue_stopped(q)) | |
799 | blk_start_queue(q); | |
800 | } | |
801 | ||
802 | static void start_queue(struct request_queue *q) | |
803 | { | |
804 | unsigned long flags; | |
805 | ||
806 | spin_lock_irqsave(q->queue_lock, flags); | |
807 | __start_queue(q); | |
808 | spin_unlock_irqrestore(q->queue_lock, flags); | |
809 | } | |
810 | ||
811 | /* | |
812 | * Complete the clone and the original request. | |
813 | * Must be called without queue lock. | |
814 | */ | |
815 | static void dm_end_request(struct request *clone, int error) | |
816 | { | |
817 | struct dm_rq_target_io *tio = clone->end_io_data; | |
818 | struct mapped_device *md = tio->md; | |
819 | struct request *rq = tio->orig; | |
820 | ||
821 | if (blk_pc_request(rq)) { | |
822 | rq->errors = clone->errors; | |
823 | rq->resid_len = clone->resid_len; | |
824 | ||
825 | if (rq->sense) | |
826 | /* | |
827 | * We are using the sense buffer of the original | |
828 | * request. | |
829 | * So setting the length of the sense data is enough. | |
830 | */ | |
831 | rq->sense_len = clone->sense_len; | |
832 | } | |
833 | ||
a77e28c7 | 834 | free_rq_clone(clone); |
cec47e3d KU |
835 | |
836 | blk_end_request_all(rq, error); | |
837 | ||
838 | rq_completed(md, 1); | |
839 | } | |
840 | ||
841 | /* | |
842 | * Request completion handler for request-based dm | |
843 | */ | |
844 | static void dm_softirq_done(struct request *rq) | |
845 | { | |
846 | struct request *clone = rq->completion_data; | |
847 | struct dm_rq_target_io *tio = clone->end_io_data; | |
848 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
849 | int error = tio->error; | |
850 | ||
851 | if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io) | |
852 | error = rq_end_io(tio->ti, clone, error, &tio->info); | |
853 | ||
854 | if (error <= 0) | |
855 | /* The target wants to complete the I/O */ | |
856 | dm_end_request(clone, error); | |
857 | else if (error == DM_ENDIO_INCOMPLETE) | |
858 | /* The target will handle the I/O */ | |
859 | return; | |
860 | else if (error == DM_ENDIO_REQUEUE) | |
861 | /* The target wants to requeue the I/O */ | |
862 | dm_requeue_unmapped_request(clone); | |
863 | else { | |
864 | DMWARN("unimplemented target endio return value: %d", error); | |
865 | BUG(); | |
866 | } | |
867 | } | |
868 | ||
869 | /* | |
870 | * Complete the clone and the original request with the error status | |
871 | * through softirq context. | |
872 | */ | |
873 | static void dm_complete_request(struct request *clone, int error) | |
874 | { | |
875 | struct dm_rq_target_io *tio = clone->end_io_data; | |
876 | struct request *rq = tio->orig; | |
877 | ||
878 | tio->error = error; | |
879 | rq->completion_data = clone; | |
880 | blk_complete_request(rq); | |
881 | } | |
882 | ||
883 | /* | |
884 | * Complete the not-mapped clone and the original request with the error status | |
885 | * through softirq context. | |
886 | * Target's rq_end_io() function isn't called. | |
887 | * This may be used when the target's map_rq() function fails. | |
888 | */ | |
889 | void dm_kill_unmapped_request(struct request *clone, int error) | |
890 | { | |
891 | struct dm_rq_target_io *tio = clone->end_io_data; | |
892 | struct request *rq = tio->orig; | |
893 | ||
894 | rq->cmd_flags |= REQ_FAILED; | |
895 | dm_complete_request(clone, error); | |
896 | } | |
897 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
898 | ||
899 | /* | |
900 | * Called with the queue lock held | |
901 | */ | |
902 | static void end_clone_request(struct request *clone, int error) | |
903 | { | |
904 | /* | |
905 | * For just cleaning up the information of the queue in which | |
906 | * the clone was dispatched. | |
907 | * The clone is *NOT* freed actually here because it is alloced from | |
908 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
909 | */ | |
910 | __blk_put_request(clone->q, clone); | |
911 | ||
912 | /* | |
913 | * Actual request completion is done in a softirq context which doesn't | |
914 | * hold the queue lock. Otherwise, deadlock could occur because: | |
915 | * - another request may be submitted by the upper level driver | |
916 | * of the stacking during the completion | |
917 | * - the submission which requires queue lock may be done | |
918 | * against this queue | |
919 | */ | |
920 | dm_complete_request(clone, error); | |
921 | } | |
922 | ||
1da177e4 LT |
923 | static sector_t max_io_len(struct mapped_device *md, |
924 | sector_t sector, struct dm_target *ti) | |
925 | { | |
926 | sector_t offset = sector - ti->begin; | |
927 | sector_t len = ti->len - offset; | |
928 | ||
929 | /* | |
930 | * Does the target need to split even further ? | |
931 | */ | |
932 | if (ti->split_io) { | |
933 | sector_t boundary; | |
934 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
935 | - offset; | |
936 | if (len > boundary) | |
937 | len = boundary; | |
938 | } | |
939 | ||
940 | return len; | |
941 | } | |
942 | ||
943 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 944 | struct dm_target_io *tio) |
1da177e4 LT |
945 | { |
946 | int r; | |
2056a782 | 947 | sector_t sector; |
9faf400f | 948 | struct mapped_device *md; |
1da177e4 | 949 | |
1da177e4 LT |
950 | clone->bi_end_io = clone_endio; |
951 | clone->bi_private = tio; | |
952 | ||
953 | /* | |
954 | * Map the clone. If r == 0 we don't need to do | |
955 | * anything, the target has assumed ownership of | |
956 | * this io. | |
957 | */ | |
958 | atomic_inc(&tio->io->io_count); | |
2056a782 | 959 | sector = clone->bi_sector; |
1da177e4 | 960 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 961 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 962 | /* the bio has been remapped so dispatch it */ |
2056a782 | 963 | |
5f3ea37c | 964 | trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, |
22a7c31a | 965 | tio->io->bio->bi_bdev->bd_dev, sector); |
2056a782 | 966 | |
1da177e4 | 967 | generic_make_request(clone); |
2e93ccc1 KU |
968 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
969 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
970 | md = tio->io->md; |
971 | dec_pending(tio->io, r); | |
972 | /* | |
973 | * Store bio_set for cleanup. | |
974 | */ | |
975 | clone->bi_private = md->bs; | |
1da177e4 | 976 | bio_put(clone); |
9faf400f | 977 | free_tio(md, tio); |
45cbcd79 KU |
978 | } else if (r) { |
979 | DMWARN("unimplemented target map return value: %d", r); | |
980 | BUG(); | |
1da177e4 LT |
981 | } |
982 | } | |
983 | ||
984 | struct clone_info { | |
985 | struct mapped_device *md; | |
986 | struct dm_table *map; | |
987 | struct bio *bio; | |
988 | struct dm_io *io; | |
989 | sector_t sector; | |
990 | sector_t sector_count; | |
991 | unsigned short idx; | |
992 | }; | |
993 | ||
3676347a PO |
994 | static void dm_bio_destructor(struct bio *bio) |
995 | { | |
9faf400f SB |
996 | struct bio_set *bs = bio->bi_private; |
997 | ||
998 | bio_free(bio, bs); | |
3676347a PO |
999 | } |
1000 | ||
1da177e4 LT |
1001 | /* |
1002 | * Creates a little bio that is just does part of a bvec. | |
1003 | */ | |
1004 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1005 | unsigned short idx, unsigned int offset, | |
9faf400f | 1006 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1007 | { |
1008 | struct bio *clone; | |
1009 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1010 | ||
9faf400f | 1011 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1012 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1013 | *clone->bi_io_vec = *bv; |
1014 | ||
1015 | clone->bi_sector = sector; | |
1016 | clone->bi_bdev = bio->bi_bdev; | |
af7e466a | 1017 | clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER); |
1da177e4 LT |
1018 | clone->bi_vcnt = 1; |
1019 | clone->bi_size = to_bytes(len); | |
1020 | clone->bi_io_vec->bv_offset = offset; | |
1021 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1022 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1023 | |
9c47008d | 1024 | if (bio_integrity(bio)) { |
7878cba9 | 1025 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1026 | bio_integrity_trim(clone, |
1027 | bio_sector_offset(bio, idx, offset), len); | |
1028 | } | |
1029 | ||
1da177e4 LT |
1030 | return clone; |
1031 | } | |
1032 | ||
1033 | /* | |
1034 | * Creates a bio that consists of range of complete bvecs. | |
1035 | */ | |
1036 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1037 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1038 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1039 | { |
1040 | struct bio *clone; | |
1041 | ||
9faf400f SB |
1042 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1043 | __bio_clone(clone, bio); | |
af7e466a | 1044 | clone->bi_rw &= ~(1 << BIO_RW_BARRIER); |
9faf400f | 1045 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1046 | clone->bi_sector = sector; |
1047 | clone->bi_idx = idx; | |
1048 | clone->bi_vcnt = idx + bv_count; | |
1049 | clone->bi_size = to_bytes(len); | |
1050 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1051 | ||
9c47008d | 1052 | if (bio_integrity(bio)) { |
7878cba9 | 1053 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1054 | |
1055 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1056 | bio_integrity_trim(clone, | |
1057 | bio_sector_offset(bio, idx, 0), len); | |
1058 | } | |
1059 | ||
1da177e4 LT |
1060 | return clone; |
1061 | } | |
1062 | ||
9015df24 AK |
1063 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1064 | struct dm_target *ti) | |
f9ab94ce | 1065 | { |
9015df24 | 1066 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1067 | |
1068 | tio->io = ci->io; | |
1069 | tio->ti = ti; | |
f9ab94ce | 1070 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1071 | |
1072 | return tio; | |
1073 | } | |
1074 | ||
1075 | static void __flush_target(struct clone_info *ci, struct dm_target *ti, | |
1076 | unsigned flush_nr) | |
1077 | { | |
1078 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1079 | struct bio *clone; | |
1080 | ||
f9ab94ce MP |
1081 | tio->info.flush_request = flush_nr; |
1082 | ||
1083 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); | |
1084 | __bio_clone(clone, ci->bio); | |
1085 | clone->bi_destructor = dm_bio_destructor; | |
1086 | ||
1087 | __map_bio(ti, clone, tio); | |
1088 | } | |
1089 | ||
1090 | static int __clone_and_map_empty_barrier(struct clone_info *ci) | |
1091 | { | |
1092 | unsigned target_nr = 0, flush_nr; | |
1093 | struct dm_target *ti; | |
1094 | ||
1095 | while ((ti = dm_table_get_target(ci->map, target_nr++))) | |
1096 | for (flush_nr = 0; flush_nr < ti->num_flush_requests; | |
1097 | flush_nr++) | |
1098 | __flush_target(ci, ti, flush_nr); | |
1099 | ||
1100 | ci->sector_count = 0; | |
1101 | ||
1102 | return 0; | |
1103 | } | |
1104 | ||
512875bd | 1105 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1106 | { |
1107 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1108 | struct dm_target *ti; |
1109 | sector_t len = 0, max; | |
028867ac | 1110 | struct dm_target_io *tio; |
1da177e4 | 1111 | |
f9ab94ce MP |
1112 | if (unlikely(bio_empty_barrier(bio))) |
1113 | return __clone_and_map_empty_barrier(ci); | |
1114 | ||
512875bd JN |
1115 | ti = dm_table_find_target(ci->map, ci->sector); |
1116 | if (!dm_target_is_valid(ti)) | |
1117 | return -EIO; | |
1118 | ||
1119 | max = max_io_len(ci->md, ci->sector, ti); | |
1120 | ||
1da177e4 LT |
1121 | /* |
1122 | * Allocate a target io object. | |
1123 | */ | |
9015df24 | 1124 | tio = alloc_tio(ci, ti); |
1da177e4 LT |
1125 | |
1126 | if (ci->sector_count <= max) { | |
1127 | /* | |
1128 | * Optimise for the simple case where we can do all of | |
1129 | * the remaining io with a single clone. | |
1130 | */ | |
1131 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
1132 | bio->bi_vcnt - ci->idx, ci->sector_count, |
1133 | ci->md->bs); | |
1da177e4 LT |
1134 | __map_bio(ti, clone, tio); |
1135 | ci->sector_count = 0; | |
1136 | ||
1137 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1138 | /* | |
1139 | * There are some bvecs that don't span targets. | |
1140 | * Do as many of these as possible. | |
1141 | */ | |
1142 | int i; | |
1143 | sector_t remaining = max; | |
1144 | sector_t bv_len; | |
1145 | ||
1146 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1147 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1148 | ||
1149 | if (bv_len > remaining) | |
1150 | break; | |
1151 | ||
1152 | remaining -= bv_len; | |
1153 | len += bv_len; | |
1154 | } | |
1155 | ||
9faf400f SB |
1156 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1157 | ci->md->bs); | |
1da177e4 LT |
1158 | __map_bio(ti, clone, tio); |
1159 | ||
1160 | ci->sector += len; | |
1161 | ci->sector_count -= len; | |
1162 | ci->idx = i; | |
1163 | ||
1164 | } else { | |
1165 | /* | |
d2044a94 | 1166 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1167 | */ |
1168 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1169 | sector_t remaining = to_sector(bv->bv_len); |
1170 | unsigned int offset = 0; | |
1da177e4 | 1171 | |
d2044a94 AK |
1172 | do { |
1173 | if (offset) { | |
1174 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1175 | if (!dm_target_is_valid(ti)) |
1176 | return -EIO; | |
1177 | ||
d2044a94 | 1178 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 1179 | |
9015df24 | 1180 | tio = alloc_tio(ci, ti); |
d2044a94 AK |
1181 | } |
1182 | ||
1183 | len = min(remaining, max); | |
1184 | ||
1185 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
1186 | bv->bv_offset + offset, len, |
1187 | ci->md->bs); | |
d2044a94 AK |
1188 | |
1189 | __map_bio(ti, clone, tio); | |
1190 | ||
1191 | ci->sector += len; | |
1192 | ci->sector_count -= len; | |
1193 | offset += to_bytes(len); | |
1194 | } while (remaining -= len); | |
1da177e4 | 1195 | |
1da177e4 LT |
1196 | ci->idx++; |
1197 | } | |
512875bd JN |
1198 | |
1199 | return 0; | |
1da177e4 LT |
1200 | } |
1201 | ||
1202 | /* | |
8a53c28d | 1203 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1204 | */ |
f0b9a450 | 1205 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1206 | { |
1207 | struct clone_info ci; | |
512875bd | 1208 | int error = 0; |
1da177e4 LT |
1209 | |
1210 | ci.map = dm_get_table(md); | |
f0b9a450 | 1211 | if (unlikely(!ci.map)) { |
af7e466a MP |
1212 | if (!bio_barrier(bio)) |
1213 | bio_io_error(bio); | |
1214 | else | |
5aa2781d MP |
1215 | if (!md->barrier_error) |
1216 | md->barrier_error = -EIO; | |
f0b9a450 MP |
1217 | return; |
1218 | } | |
692d0eb9 | 1219 | |
1da177e4 LT |
1220 | ci.md = md; |
1221 | ci.bio = bio; | |
1222 | ci.io = alloc_io(md); | |
1223 | ci.io->error = 0; | |
1224 | atomic_set(&ci.io->io_count, 1); | |
1225 | ci.io->bio = bio; | |
1226 | ci.io->md = md; | |
1227 | ci.sector = bio->bi_sector; | |
1228 | ci.sector_count = bio_sectors(bio); | |
f9ab94ce MP |
1229 | if (unlikely(bio_empty_barrier(bio))) |
1230 | ci.sector_count = 1; | |
1da177e4 LT |
1231 | ci.idx = bio->bi_idx; |
1232 | ||
3eaf840e | 1233 | start_io_acct(ci.io); |
512875bd JN |
1234 | while (ci.sector_count && !error) |
1235 | error = __clone_and_map(&ci); | |
1da177e4 LT |
1236 | |
1237 | /* drop the extra reference count */ | |
512875bd | 1238 | dec_pending(ci.io, error); |
1da177e4 LT |
1239 | dm_table_put(ci.map); |
1240 | } | |
1241 | /*----------------------------------------------------------------- | |
1242 | * CRUD END | |
1243 | *---------------------------------------------------------------*/ | |
1244 | ||
f6fccb12 MB |
1245 | static int dm_merge_bvec(struct request_queue *q, |
1246 | struct bvec_merge_data *bvm, | |
1247 | struct bio_vec *biovec) | |
1248 | { | |
1249 | struct mapped_device *md = q->queuedata; | |
1250 | struct dm_table *map = dm_get_table(md); | |
1251 | struct dm_target *ti; | |
1252 | sector_t max_sectors; | |
5037108a | 1253 | int max_size = 0; |
f6fccb12 MB |
1254 | |
1255 | if (unlikely(!map)) | |
5037108a | 1256 | goto out; |
f6fccb12 MB |
1257 | |
1258 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1259 | if (!dm_target_is_valid(ti)) |
1260 | goto out_table; | |
f6fccb12 MB |
1261 | |
1262 | /* | |
1263 | * Find maximum amount of I/O that won't need splitting | |
1264 | */ | |
1265 | max_sectors = min(max_io_len(md, bvm->bi_sector, ti), | |
1266 | (sector_t) BIO_MAX_SECTORS); | |
1267 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1268 | if (max_size < 0) | |
1269 | max_size = 0; | |
1270 | ||
1271 | /* | |
1272 | * merge_bvec_fn() returns number of bytes | |
1273 | * it can accept at this offset | |
1274 | * max is precomputed maximal io size | |
1275 | */ | |
1276 | if (max_size && ti->type->merge) | |
1277 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1278 | /* |
1279 | * If the target doesn't support merge method and some of the devices | |
1280 | * provided their merge_bvec method (we know this by looking at | |
1281 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1282 | * entries. So always set max_size to 0, and the code below allows | |
1283 | * just one page. | |
1284 | */ | |
1285 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1286 | ||
1287 | max_size = 0; | |
f6fccb12 | 1288 | |
b01cd5ac | 1289 | out_table: |
5037108a MP |
1290 | dm_table_put(map); |
1291 | ||
1292 | out: | |
f6fccb12 MB |
1293 | /* |
1294 | * Always allow an entire first page | |
1295 | */ | |
1296 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1297 | max_size = biovec->bv_len; | |
1298 | ||
f6fccb12 MB |
1299 | return max_size; |
1300 | } | |
1301 | ||
1da177e4 LT |
1302 | /* |
1303 | * The request function that just remaps the bio built up by | |
1304 | * dm_merge_bvec. | |
1305 | */ | |
cec47e3d | 1306 | static int _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1307 | { |
12f03a49 | 1308 | int rw = bio_data_dir(bio); |
1da177e4 | 1309 | struct mapped_device *md = q->queuedata; |
c9959059 | 1310 | int cpu; |
1da177e4 | 1311 | |
2ca3310e | 1312 | down_read(&md->io_lock); |
1da177e4 | 1313 | |
074a7aca TH |
1314 | cpu = part_stat_lock(); |
1315 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1316 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1317 | part_stat_unlock(); | |
12f03a49 | 1318 | |
1da177e4 | 1319 | /* |
1eb787ec AK |
1320 | * If we're suspended or the thread is processing barriers |
1321 | * we have to queue this io for later. | |
1da177e4 | 1322 | */ |
af7e466a MP |
1323 | if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || |
1324 | unlikely(bio_barrier(bio))) { | |
2ca3310e | 1325 | up_read(&md->io_lock); |
1da177e4 | 1326 | |
54d9a1b4 AK |
1327 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && |
1328 | bio_rw(bio) == READA) { | |
1329 | bio_io_error(bio); | |
1330 | return 0; | |
1331 | } | |
1da177e4 | 1332 | |
92c63902 | 1333 | queue_io(md, bio); |
1da177e4 | 1334 | |
92c63902 | 1335 | return 0; |
1da177e4 LT |
1336 | } |
1337 | ||
f0b9a450 | 1338 | __split_and_process_bio(md, bio); |
2ca3310e | 1339 | up_read(&md->io_lock); |
f0b9a450 | 1340 | return 0; |
1da177e4 LT |
1341 | } |
1342 | ||
cec47e3d KU |
1343 | static int dm_make_request(struct request_queue *q, struct bio *bio) |
1344 | { | |
1345 | struct mapped_device *md = q->queuedata; | |
1346 | ||
1347 | if (unlikely(bio_barrier(bio))) { | |
1348 | bio_endio(bio, -EOPNOTSUPP); | |
1349 | return 0; | |
1350 | } | |
1351 | ||
1352 | return md->saved_make_request_fn(q, bio); /* call __make_request() */ | |
1353 | } | |
1354 | ||
1355 | static int dm_request_based(struct mapped_device *md) | |
1356 | { | |
1357 | return blk_queue_stackable(md->queue); | |
1358 | } | |
1359 | ||
1360 | static int dm_request(struct request_queue *q, struct bio *bio) | |
1361 | { | |
1362 | struct mapped_device *md = q->queuedata; | |
1363 | ||
1364 | if (dm_request_based(md)) | |
1365 | return dm_make_request(q, bio); | |
1366 | ||
1367 | return _dm_request(q, bio); | |
1368 | } | |
1369 | ||
1370 | void dm_dispatch_request(struct request *rq) | |
1371 | { | |
1372 | int r; | |
1373 | ||
1374 | if (blk_queue_io_stat(rq->q)) | |
1375 | rq->cmd_flags |= REQ_IO_STAT; | |
1376 | ||
1377 | rq->start_time = jiffies; | |
1378 | r = blk_insert_cloned_request(rq->q, rq); | |
1379 | if (r) | |
1380 | dm_complete_request(rq, r); | |
1381 | } | |
1382 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1383 | ||
1384 | static void dm_rq_bio_destructor(struct bio *bio) | |
1385 | { | |
1386 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1387 | struct mapped_device *md = info->tio->md; | |
1388 | ||
1389 | free_bio_info(info); | |
1390 | bio_free(bio, md->bs); | |
1391 | } | |
1392 | ||
1393 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1394 | void *data) | |
1395 | { | |
1396 | struct dm_rq_target_io *tio = data; | |
1397 | struct mapped_device *md = tio->md; | |
1398 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1399 | ||
1400 | if (!info) | |
1401 | return -ENOMEM; | |
1402 | ||
1403 | info->orig = bio_orig; | |
1404 | info->tio = tio; | |
1405 | bio->bi_end_io = end_clone_bio; | |
1406 | bio->bi_private = info; | |
1407 | bio->bi_destructor = dm_rq_bio_destructor; | |
1408 | ||
1409 | return 0; | |
1410 | } | |
1411 | ||
1412 | static int setup_clone(struct request *clone, struct request *rq, | |
1413 | struct dm_rq_target_io *tio) | |
1414 | { | |
1415 | int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, | |
1416 | dm_rq_bio_constructor, tio); | |
1417 | ||
1418 | if (r) | |
1419 | return r; | |
1420 | ||
1421 | clone->cmd = rq->cmd; | |
1422 | clone->cmd_len = rq->cmd_len; | |
1423 | clone->sense = rq->sense; | |
1424 | clone->buffer = rq->buffer; | |
1425 | clone->end_io = end_clone_request; | |
1426 | clone->end_io_data = tio; | |
1427 | ||
1428 | return 0; | |
1429 | } | |
1430 | ||
1431 | static int dm_rq_flush_suspending(struct mapped_device *md) | |
1432 | { | |
1433 | return !md->suspend_rq.special; | |
1434 | } | |
1435 | ||
1436 | /* | |
1437 | * Called with the queue lock held. | |
1438 | */ | |
1439 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1440 | { | |
1441 | struct mapped_device *md = q->queuedata; | |
1442 | struct dm_rq_target_io *tio; | |
1443 | struct request *clone; | |
1444 | ||
1445 | if (unlikely(rq == &md->suspend_rq)) { | |
1446 | if (dm_rq_flush_suspending(md)) | |
1447 | return BLKPREP_OK; | |
1448 | else | |
1449 | /* The flush suspend was interrupted */ | |
1450 | return BLKPREP_KILL; | |
1451 | } | |
1452 | ||
1453 | if (unlikely(rq->special)) { | |
1454 | DMWARN("Already has something in rq->special."); | |
1455 | return BLKPREP_KILL; | |
1456 | } | |
1457 | ||
1458 | tio = alloc_rq_tio(md); /* Only one for each original request */ | |
1459 | if (!tio) | |
1460 | /* -ENOMEM */ | |
1461 | return BLKPREP_DEFER; | |
1462 | ||
1463 | tio->md = md; | |
1464 | tio->ti = NULL; | |
1465 | tio->orig = rq; | |
1466 | tio->error = 0; | |
1467 | memset(&tio->info, 0, sizeof(tio->info)); | |
1468 | ||
1469 | clone = &tio->clone; | |
1470 | if (setup_clone(clone, rq, tio)) { | |
1471 | /* -ENOMEM */ | |
1472 | free_rq_tio(tio); | |
1473 | return BLKPREP_DEFER; | |
1474 | } | |
1475 | ||
1476 | rq->special = clone; | |
1477 | rq->cmd_flags |= REQ_DONTPREP; | |
1478 | ||
1479 | return BLKPREP_OK; | |
1480 | } | |
1481 | ||
1482 | static void map_request(struct dm_target *ti, struct request *rq, | |
1483 | struct mapped_device *md) | |
1484 | { | |
1485 | int r; | |
1486 | struct request *clone = rq->special; | |
1487 | struct dm_rq_target_io *tio = clone->end_io_data; | |
1488 | ||
1489 | /* | |
1490 | * Hold the md reference here for the in-flight I/O. | |
1491 | * We can't rely on the reference count by device opener, | |
1492 | * because the device may be closed during the request completion | |
1493 | * when all bios are completed. | |
1494 | * See the comment in rq_completed() too. | |
1495 | */ | |
1496 | dm_get(md); | |
1497 | ||
1498 | tio->ti = ti; | |
1499 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1500 | switch (r) { | |
1501 | case DM_MAPIO_SUBMITTED: | |
1502 | /* The target has taken the I/O to submit by itself later */ | |
1503 | break; | |
1504 | case DM_MAPIO_REMAPPED: | |
1505 | /* The target has remapped the I/O so dispatch it */ | |
1506 | dm_dispatch_request(clone); | |
1507 | break; | |
1508 | case DM_MAPIO_REQUEUE: | |
1509 | /* The target wants to requeue the I/O */ | |
1510 | dm_requeue_unmapped_request(clone); | |
1511 | break; | |
1512 | default: | |
1513 | if (r > 0) { | |
1514 | DMWARN("unimplemented target map return value: %d", r); | |
1515 | BUG(); | |
1516 | } | |
1517 | ||
1518 | /* The target wants to complete the I/O */ | |
1519 | dm_kill_unmapped_request(clone, r); | |
1520 | break; | |
1521 | } | |
1522 | } | |
1523 | ||
1524 | /* | |
1525 | * q->request_fn for request-based dm. | |
1526 | * Called with the queue lock held. | |
1527 | */ | |
1528 | static void dm_request_fn(struct request_queue *q) | |
1529 | { | |
1530 | struct mapped_device *md = q->queuedata; | |
1531 | struct dm_table *map = dm_get_table(md); | |
1532 | struct dm_target *ti; | |
1533 | struct request *rq; | |
1534 | ||
1535 | /* | |
1536 | * For noflush suspend, check blk_queue_stopped() to immediately | |
1537 | * quit I/O dispatching. | |
1538 | */ | |
1539 | while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) { | |
1540 | rq = blk_peek_request(q); | |
1541 | if (!rq) | |
1542 | goto plug_and_out; | |
1543 | ||
1544 | if (unlikely(rq == &md->suspend_rq)) { /* Flush suspend maker */ | |
1545 | if (queue_in_flight(q)) | |
1546 | /* Not quiet yet. Wait more */ | |
1547 | goto plug_and_out; | |
1548 | ||
1549 | /* This device should be quiet now */ | |
1550 | __stop_queue(q); | |
1551 | blk_start_request(rq); | |
1552 | __blk_end_request_all(rq, 0); | |
1553 | wake_up(&md->wait); | |
1554 | goto out; | |
1555 | } | |
1556 | ||
1557 | ti = dm_table_find_target(map, blk_rq_pos(rq)); | |
1558 | if (ti->type->busy && ti->type->busy(ti)) | |
1559 | goto plug_and_out; | |
1560 | ||
1561 | blk_start_request(rq); | |
1562 | spin_unlock(q->queue_lock); | |
1563 | map_request(ti, rq, md); | |
1564 | spin_lock_irq(q->queue_lock); | |
1565 | } | |
1566 | ||
1567 | goto out; | |
1568 | ||
1569 | plug_and_out: | |
1570 | if (!elv_queue_empty(q)) | |
1571 | /* Some requests still remain, retry later */ | |
1572 | blk_plug_device(q); | |
1573 | ||
1574 | out: | |
1575 | dm_table_put(map); | |
1576 | ||
1577 | return; | |
1578 | } | |
1579 | ||
1580 | int dm_underlying_device_busy(struct request_queue *q) | |
1581 | { | |
1582 | return blk_lld_busy(q); | |
1583 | } | |
1584 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1585 | ||
1586 | static int dm_lld_busy(struct request_queue *q) | |
1587 | { | |
1588 | int r; | |
1589 | struct mapped_device *md = q->queuedata; | |
1590 | struct dm_table *map = dm_get_table(md); | |
1591 | ||
1592 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1593 | r = 1; | |
1594 | else | |
1595 | r = dm_table_any_busy_target(map); | |
1596 | ||
1597 | dm_table_put(map); | |
1598 | ||
1599 | return r; | |
1600 | } | |
1601 | ||
165125e1 | 1602 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
1603 | { |
1604 | struct mapped_device *md = q->queuedata; | |
1605 | struct dm_table *map = dm_get_table(md); | |
1606 | ||
1607 | if (map) { | |
cec47e3d KU |
1608 | if (dm_request_based(md)) |
1609 | generic_unplug_device(q); | |
1610 | ||
1da177e4 LT |
1611 | dm_table_unplug_all(map); |
1612 | dm_table_put(map); | |
1613 | } | |
1614 | } | |
1615 | ||
1616 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
1617 | { | |
8a57dfc6 CS |
1618 | int r = bdi_bits; |
1619 | struct mapped_device *md = congested_data; | |
1620 | struct dm_table *map; | |
1da177e4 | 1621 | |
1eb787ec | 1622 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
8a57dfc6 CS |
1623 | map = dm_get_table(md); |
1624 | if (map) { | |
cec47e3d KU |
1625 | /* |
1626 | * Request-based dm cares about only own queue for | |
1627 | * the query about congestion status of request_queue | |
1628 | */ | |
1629 | if (dm_request_based(md)) | |
1630 | r = md->queue->backing_dev_info.state & | |
1631 | bdi_bits; | |
1632 | else | |
1633 | r = dm_table_any_congested(map, bdi_bits); | |
1634 | ||
8a57dfc6 CS |
1635 | dm_table_put(map); |
1636 | } | |
1637 | } | |
1638 | ||
1da177e4 LT |
1639 | return r; |
1640 | } | |
1641 | ||
1642 | /*----------------------------------------------------------------- | |
1643 | * An IDR is used to keep track of allocated minor numbers. | |
1644 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1645 | static DEFINE_IDR(_minor_idr); |
1646 | ||
2b06cfff | 1647 | static void free_minor(int minor) |
1da177e4 | 1648 | { |
f32c10b0 | 1649 | spin_lock(&_minor_lock); |
1da177e4 | 1650 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1651 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1652 | } |
1653 | ||
1654 | /* | |
1655 | * See if the device with a specific minor # is free. | |
1656 | */ | |
cf13ab8e | 1657 | static int specific_minor(int minor) |
1da177e4 LT |
1658 | { |
1659 | int r, m; | |
1660 | ||
1661 | if (minor >= (1 << MINORBITS)) | |
1662 | return -EINVAL; | |
1663 | ||
62f75c2f JM |
1664 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1665 | if (!r) | |
1666 | return -ENOMEM; | |
1667 | ||
f32c10b0 | 1668 | spin_lock(&_minor_lock); |
1da177e4 LT |
1669 | |
1670 | if (idr_find(&_minor_idr, minor)) { | |
1671 | r = -EBUSY; | |
1672 | goto out; | |
1673 | } | |
1674 | ||
ba61fdd1 | 1675 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1676 | if (r) |
1da177e4 | 1677 | goto out; |
1da177e4 LT |
1678 | |
1679 | if (m != minor) { | |
1680 | idr_remove(&_minor_idr, m); | |
1681 | r = -EBUSY; | |
1682 | goto out; | |
1683 | } | |
1684 | ||
1685 | out: | |
f32c10b0 | 1686 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1687 | return r; |
1688 | } | |
1689 | ||
cf13ab8e | 1690 | static int next_free_minor(int *minor) |
1da177e4 | 1691 | { |
2b06cfff | 1692 | int r, m; |
1da177e4 | 1693 | |
1da177e4 | 1694 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1695 | if (!r) |
1696 | return -ENOMEM; | |
1697 | ||
f32c10b0 | 1698 | spin_lock(&_minor_lock); |
1da177e4 | 1699 | |
ba61fdd1 | 1700 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1701 | if (r) |
1da177e4 | 1702 | goto out; |
1da177e4 LT |
1703 | |
1704 | if (m >= (1 << MINORBITS)) { | |
1705 | idr_remove(&_minor_idr, m); | |
1706 | r = -ENOSPC; | |
1707 | goto out; | |
1708 | } | |
1709 | ||
1710 | *minor = m; | |
1711 | ||
1712 | out: | |
f32c10b0 | 1713 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1714 | return r; |
1715 | } | |
1716 | ||
1717 | static struct block_device_operations dm_blk_dops; | |
1718 | ||
53d5914f MP |
1719 | static void dm_wq_work(struct work_struct *work); |
1720 | ||
1da177e4 LT |
1721 | /* |
1722 | * Allocate and initialise a blank device with a given minor. | |
1723 | */ | |
2b06cfff | 1724 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1725 | { |
1726 | int r; | |
cf13ab8e | 1727 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1728 | void *old_md; |
1da177e4 LT |
1729 | |
1730 | if (!md) { | |
1731 | DMWARN("unable to allocate device, out of memory."); | |
1732 | return NULL; | |
1733 | } | |
1734 | ||
10da4f79 | 1735 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1736 | goto bad_module_get; |
10da4f79 | 1737 | |
1da177e4 | 1738 | /* get a minor number for the dev */ |
2b06cfff | 1739 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1740 | r = next_free_minor(&minor); |
2b06cfff | 1741 | else |
cf13ab8e | 1742 | r = specific_minor(minor); |
1da177e4 | 1743 | if (r < 0) |
6ed7ade8 | 1744 | goto bad_minor; |
1da177e4 | 1745 | |
2ca3310e | 1746 | init_rwsem(&md->io_lock); |
e61290a4 | 1747 | mutex_init(&md->suspend_lock); |
022c2611 | 1748 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1749 | rwlock_init(&md->map_lock); |
1750 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1751 | atomic_set(&md->open_count, 0); |
1da177e4 | 1752 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1753 | atomic_set(&md->uevent_seq, 0); |
1754 | INIT_LIST_HEAD(&md->uevent_list); | |
1755 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1756 | |
e6ee8c0b | 1757 | md->queue = blk_init_queue(dm_request_fn, NULL); |
1da177e4 | 1758 | if (!md->queue) |
6ed7ade8 | 1759 | goto bad_queue; |
1da177e4 | 1760 | |
e6ee8c0b KU |
1761 | /* |
1762 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1763 | * devices. The type of this dm device has not been decided yet, | |
1764 | * although we initialized the queue using blk_init_queue(). | |
1765 | * The type is decided at the first table loading time. | |
1766 | * To prevent problematic device stacking, clear the queue flag | |
1767 | * for request stacking support until then. | |
1768 | * | |
1769 | * This queue is new, so no concurrency on the queue_flags. | |
1770 | */ | |
1771 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1772 | md->saved_make_request_fn = md->queue->make_request_fn; | |
1da177e4 LT |
1773 | md->queue->queuedata = md; |
1774 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1775 | md->queue->backing_dev_info.congested_data = md; | |
1776 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1777 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1778 | md->queue->unplug_fn = dm_unplug_all; |
f6fccb12 | 1779 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
e6ee8c0b KU |
1780 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
1781 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
1782 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
9faf400f | 1783 | |
1da177e4 LT |
1784 | md->disk = alloc_disk(1); |
1785 | if (!md->disk) | |
6ed7ade8 | 1786 | goto bad_disk; |
1da177e4 | 1787 | |
f0b04115 JM |
1788 | atomic_set(&md->pending, 0); |
1789 | init_waitqueue_head(&md->wait); | |
53d5914f | 1790 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1791 | init_waitqueue_head(&md->eventq); |
1792 | ||
1da177e4 LT |
1793 | md->disk->major = _major; |
1794 | md->disk->first_minor = minor; | |
1795 | md->disk->fops = &dm_blk_dops; | |
1796 | md->disk->queue = md->queue; | |
1797 | md->disk->private_data = md; | |
1798 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1799 | add_disk(md->disk); | |
7e51f257 | 1800 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1801 | |
304f3f6a MB |
1802 | md->wq = create_singlethread_workqueue("kdmflush"); |
1803 | if (!md->wq) | |
1804 | goto bad_thread; | |
1805 | ||
32a926da MP |
1806 | md->bdev = bdget_disk(md->disk, 0); |
1807 | if (!md->bdev) | |
1808 | goto bad_bdev; | |
1809 | ||
ba61fdd1 | 1810 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1811 | spin_lock(&_minor_lock); |
ba61fdd1 | 1812 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1813 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1814 | |
1815 | BUG_ON(old_md != MINOR_ALLOCED); | |
1816 | ||
1da177e4 LT |
1817 | return md; |
1818 | ||
32a926da MP |
1819 | bad_bdev: |
1820 | destroy_workqueue(md->wq); | |
304f3f6a MB |
1821 | bad_thread: |
1822 | put_disk(md->disk); | |
6ed7ade8 | 1823 | bad_disk: |
1312f40e | 1824 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1825 | bad_queue: |
1da177e4 | 1826 | free_minor(minor); |
6ed7ade8 | 1827 | bad_minor: |
10da4f79 | 1828 | module_put(THIS_MODULE); |
6ed7ade8 | 1829 | bad_module_get: |
1da177e4 LT |
1830 | kfree(md); |
1831 | return NULL; | |
1832 | } | |
1833 | ||
ae9da83f JN |
1834 | static void unlock_fs(struct mapped_device *md); |
1835 | ||
1da177e4 LT |
1836 | static void free_dev(struct mapped_device *md) |
1837 | { | |
f331c029 | 1838 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1839 | |
32a926da MP |
1840 | unlock_fs(md); |
1841 | bdput(md->bdev); | |
304f3f6a | 1842 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1843 | if (md->tio_pool) |
1844 | mempool_destroy(md->tio_pool); | |
1845 | if (md->io_pool) | |
1846 | mempool_destroy(md->io_pool); | |
1847 | if (md->bs) | |
1848 | bioset_free(md->bs); | |
9c47008d | 1849 | blk_integrity_unregister(md->disk); |
1da177e4 | 1850 | del_gendisk(md->disk); |
63d94e48 | 1851 | free_minor(minor); |
fba9f90e JM |
1852 | |
1853 | spin_lock(&_minor_lock); | |
1854 | md->disk->private_data = NULL; | |
1855 | spin_unlock(&_minor_lock); | |
1856 | ||
1da177e4 | 1857 | put_disk(md->disk); |
1312f40e | 1858 | blk_cleanup_queue(md->queue); |
10da4f79 | 1859 | module_put(THIS_MODULE); |
1da177e4 LT |
1860 | kfree(md); |
1861 | } | |
1862 | ||
e6ee8c0b KU |
1863 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1864 | { | |
1865 | struct dm_md_mempools *p; | |
1866 | ||
1867 | if (md->io_pool && md->tio_pool && md->bs) | |
1868 | /* the md already has necessary mempools */ | |
1869 | goto out; | |
1870 | ||
1871 | p = dm_table_get_md_mempools(t); | |
1872 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1873 | ||
1874 | md->io_pool = p->io_pool; | |
1875 | p->io_pool = NULL; | |
1876 | md->tio_pool = p->tio_pool; | |
1877 | p->tio_pool = NULL; | |
1878 | md->bs = p->bs; | |
1879 | p->bs = NULL; | |
1880 | ||
1881 | out: | |
1882 | /* mempool bind completed, now no need any mempools in the table */ | |
1883 | dm_table_free_md_mempools(t); | |
1884 | } | |
1885 | ||
1da177e4 LT |
1886 | /* |
1887 | * Bind a table to the device. | |
1888 | */ | |
1889 | static void event_callback(void *context) | |
1890 | { | |
7a8c3d3b MA |
1891 | unsigned long flags; |
1892 | LIST_HEAD(uevents); | |
1da177e4 LT |
1893 | struct mapped_device *md = (struct mapped_device *) context; |
1894 | ||
7a8c3d3b MA |
1895 | spin_lock_irqsave(&md->uevent_lock, flags); |
1896 | list_splice_init(&md->uevent_list, &uevents); | |
1897 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1898 | ||
ed9e1982 | 1899 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1900 | |
1da177e4 LT |
1901 | atomic_inc(&md->event_nr); |
1902 | wake_up(&md->eventq); | |
1903 | } | |
1904 | ||
4e90188b | 1905 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1906 | { |
4e90188b | 1907 | set_capacity(md->disk, size); |
1da177e4 | 1908 | |
db8fef4f MP |
1909 | mutex_lock(&md->bdev->bd_inode->i_mutex); |
1910 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
1911 | mutex_unlock(&md->bdev->bd_inode->i_mutex); | |
1da177e4 LT |
1912 | } |
1913 | ||
754c5fc7 MS |
1914 | static int __bind(struct mapped_device *md, struct dm_table *t, |
1915 | struct queue_limits *limits) | |
1da177e4 | 1916 | { |
165125e1 | 1917 | struct request_queue *q = md->queue; |
1da177e4 | 1918 | sector_t size; |
523d9297 | 1919 | unsigned long flags; |
1da177e4 LT |
1920 | |
1921 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1922 | |
1923 | /* | |
1924 | * Wipe any geometry if the size of the table changed. | |
1925 | */ | |
1926 | if (size != get_capacity(md->disk)) | |
1927 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1928 | ||
32a926da | 1929 | __set_size(md, size); |
d5816876 MP |
1930 | |
1931 | if (!size) { | |
1932 | dm_table_destroy(t); | |
1da177e4 | 1933 | return 0; |
d5816876 | 1934 | } |
1da177e4 | 1935 | |
2ca3310e AK |
1936 | dm_table_event_callback(t, event_callback, md); |
1937 | ||
e6ee8c0b KU |
1938 | /* |
1939 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1940 | * for request-based during suspension. So stop it to prevent | |
1941 | * I/O mapping before resume. | |
1942 | * This must be done before setting the queue restrictions, | |
1943 | * because request-based dm may be run just after the setting. | |
1944 | */ | |
1945 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
1946 | stop_queue(q); | |
1947 | ||
1948 | __bind_mempools(md, t); | |
1949 | ||
523d9297 | 1950 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 1951 | md->map = t; |
754c5fc7 | 1952 | dm_table_set_restrictions(t, q, limits); |
523d9297 | 1953 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 1954 | |
1da177e4 LT |
1955 | return 0; |
1956 | } | |
1957 | ||
1958 | static void __unbind(struct mapped_device *md) | |
1959 | { | |
1960 | struct dm_table *map = md->map; | |
523d9297 | 1961 | unsigned long flags; |
1da177e4 LT |
1962 | |
1963 | if (!map) | |
1964 | return; | |
1965 | ||
1966 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 1967 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 1968 | md->map = NULL; |
523d9297 | 1969 | write_unlock_irqrestore(&md->map_lock, flags); |
d5816876 | 1970 | dm_table_destroy(map); |
1da177e4 LT |
1971 | } |
1972 | ||
1973 | /* | |
1974 | * Constructor for a new device. | |
1975 | */ | |
2b06cfff | 1976 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
1977 | { |
1978 | struct mapped_device *md; | |
1979 | ||
2b06cfff | 1980 | md = alloc_dev(minor); |
1da177e4 LT |
1981 | if (!md) |
1982 | return -ENXIO; | |
1983 | ||
784aae73 MB |
1984 | dm_sysfs_init(md); |
1985 | ||
1da177e4 LT |
1986 | *result = md; |
1987 | return 0; | |
1988 | } | |
1989 | ||
637842cf | 1990 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
1991 | { |
1992 | struct mapped_device *md; | |
1da177e4 LT |
1993 | unsigned minor = MINOR(dev); |
1994 | ||
1995 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
1996 | return NULL; | |
1997 | ||
f32c10b0 | 1998 | spin_lock(&_minor_lock); |
1da177e4 LT |
1999 | |
2000 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2001 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2002 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
17b2f66f | 2003 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2004 | md = NULL; |
fba9f90e JM |
2005 | goto out; |
2006 | } | |
1da177e4 | 2007 | |
fba9f90e | 2008 | out: |
f32c10b0 | 2009 | spin_unlock(&_minor_lock); |
1da177e4 | 2010 | |
637842cf DT |
2011 | return md; |
2012 | } | |
2013 | ||
d229a958 DT |
2014 | struct mapped_device *dm_get_md(dev_t dev) |
2015 | { | |
2016 | struct mapped_device *md = dm_find_md(dev); | |
2017 | ||
2018 | if (md) | |
2019 | dm_get(md); | |
2020 | ||
2021 | return md; | |
2022 | } | |
2023 | ||
9ade92a9 | 2024 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2025 | { |
9ade92a9 | 2026 | return md->interface_ptr; |
1da177e4 LT |
2027 | } |
2028 | ||
2029 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2030 | { | |
2031 | md->interface_ptr = ptr; | |
2032 | } | |
2033 | ||
2034 | void dm_get(struct mapped_device *md) | |
2035 | { | |
2036 | atomic_inc(&md->holders); | |
2037 | } | |
2038 | ||
72d94861 AK |
2039 | const char *dm_device_name(struct mapped_device *md) |
2040 | { | |
2041 | return md->name; | |
2042 | } | |
2043 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2044 | ||
1da177e4 LT |
2045 | void dm_put(struct mapped_device *md) |
2046 | { | |
1134e5ae | 2047 | struct dm_table *map; |
1da177e4 | 2048 | |
fba9f90e JM |
2049 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
2050 | ||
f32c10b0 | 2051 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 2052 | map = dm_get_table(md); |
f331c029 TH |
2053 | idr_replace(&_minor_idr, MINOR_ALLOCED, |
2054 | MINOR(disk_devt(dm_disk(md)))); | |
fba9f90e | 2055 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 2056 | spin_unlock(&_minor_lock); |
cf222b37 | 2057 | if (!dm_suspended(md)) { |
1da177e4 LT |
2058 | dm_table_presuspend_targets(map); |
2059 | dm_table_postsuspend_targets(map); | |
2060 | } | |
784aae73 | 2061 | dm_sysfs_exit(md); |
1134e5ae | 2062 | dm_table_put(map); |
a1b51e98 | 2063 | __unbind(md); |
1da177e4 LT |
2064 | free_dev(md); |
2065 | } | |
1da177e4 | 2066 | } |
79eb885c | 2067 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2068 | |
401600df | 2069 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2070 | { |
2071 | int r = 0; | |
b44ebeb0 | 2072 | DECLARE_WAITQUEUE(wait, current); |
cec47e3d KU |
2073 | struct request_queue *q = md->queue; |
2074 | unsigned long flags; | |
b44ebeb0 MP |
2075 | |
2076 | dm_unplug_all(md->queue); | |
2077 | ||
2078 | add_wait_queue(&md->wait, &wait); | |
46125c1c MB |
2079 | |
2080 | while (1) { | |
401600df | 2081 | set_current_state(interruptible); |
46125c1c MB |
2082 | |
2083 | smp_mb(); | |
cec47e3d KU |
2084 | if (dm_request_based(md)) { |
2085 | spin_lock_irqsave(q->queue_lock, flags); | |
2086 | if (!queue_in_flight(q) && blk_queue_stopped(q)) { | |
2087 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2088 | break; | |
2089 | } | |
2090 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2091 | } else if (!atomic_read(&md->pending)) | |
46125c1c MB |
2092 | break; |
2093 | ||
401600df MP |
2094 | if (interruptible == TASK_INTERRUPTIBLE && |
2095 | signal_pending(current)) { | |
46125c1c MB |
2096 | r = -EINTR; |
2097 | break; | |
2098 | } | |
2099 | ||
2100 | io_schedule(); | |
2101 | } | |
2102 | set_current_state(TASK_RUNNING); | |
2103 | ||
b44ebeb0 MP |
2104 | remove_wait_queue(&md->wait, &wait); |
2105 | ||
46125c1c MB |
2106 | return r; |
2107 | } | |
2108 | ||
531fe963 | 2109 | static void dm_flush(struct mapped_device *md) |
af7e466a MP |
2110 | { |
2111 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
52b1fd5a MP |
2112 | |
2113 | bio_init(&md->barrier_bio); | |
2114 | md->barrier_bio.bi_bdev = md->bdev; | |
2115 | md->barrier_bio.bi_rw = WRITE_BARRIER; | |
2116 | __split_and_process_bio(md, &md->barrier_bio); | |
2117 | ||
2118 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
af7e466a MP |
2119 | } |
2120 | ||
2121 | static void process_barrier(struct mapped_device *md, struct bio *bio) | |
2122 | { | |
5aa2781d MP |
2123 | md->barrier_error = 0; |
2124 | ||
531fe963 | 2125 | dm_flush(md); |
af7e466a | 2126 | |
5aa2781d MP |
2127 | if (!bio_empty_barrier(bio)) { |
2128 | __split_and_process_bio(md, bio); | |
2129 | dm_flush(md); | |
af7e466a MP |
2130 | } |
2131 | ||
af7e466a | 2132 | if (md->barrier_error != DM_ENDIO_REQUEUE) |
531fe963 | 2133 | bio_endio(bio, md->barrier_error); |
2761e95f MP |
2134 | else { |
2135 | spin_lock_irq(&md->deferred_lock); | |
2136 | bio_list_add_head(&md->deferred, bio); | |
2137 | spin_unlock_irq(&md->deferred_lock); | |
2138 | } | |
af7e466a MP |
2139 | } |
2140 | ||
1da177e4 LT |
2141 | /* |
2142 | * Process the deferred bios | |
2143 | */ | |
ef208587 | 2144 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2145 | { |
ef208587 MP |
2146 | struct mapped_device *md = container_of(work, struct mapped_device, |
2147 | work); | |
6d6f10df | 2148 | struct bio *c; |
1da177e4 | 2149 | |
ef208587 MP |
2150 | down_write(&md->io_lock); |
2151 | ||
3b00b203 | 2152 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2153 | spin_lock_irq(&md->deferred_lock); |
2154 | c = bio_list_pop(&md->deferred); | |
2155 | spin_unlock_irq(&md->deferred_lock); | |
2156 | ||
2157 | if (!c) { | |
1eb787ec | 2158 | clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); |
df12ee99 AK |
2159 | break; |
2160 | } | |
022c2611 | 2161 | |
3b00b203 MP |
2162 | up_write(&md->io_lock); |
2163 | ||
e6ee8c0b KU |
2164 | if (dm_request_based(md)) |
2165 | generic_make_request(c); | |
2166 | else { | |
2167 | if (bio_barrier(c)) | |
2168 | process_barrier(md, c); | |
2169 | else | |
2170 | __split_and_process_bio(md, c); | |
2171 | } | |
3b00b203 MP |
2172 | |
2173 | down_write(&md->io_lock); | |
022c2611 | 2174 | } |
73d410c0 | 2175 | |
ef208587 | 2176 | up_write(&md->io_lock); |
1da177e4 LT |
2177 | } |
2178 | ||
9a1fb464 | 2179 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2180 | { |
3b00b203 MP |
2181 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2182 | smp_mb__after_clear_bit(); | |
53d5914f | 2183 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2184 | } |
2185 | ||
1da177e4 LT |
2186 | /* |
2187 | * Swap in a new table (destroying old one). | |
2188 | */ | |
2189 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
2190 | { | |
754c5fc7 | 2191 | struct queue_limits limits; |
93c534ae | 2192 | int r = -EINVAL; |
1da177e4 | 2193 | |
e61290a4 | 2194 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2195 | |
2196 | /* device must be suspended */ | |
cf222b37 | 2197 | if (!dm_suspended(md)) |
93c534ae | 2198 | goto out; |
1da177e4 | 2199 | |
754c5fc7 MS |
2200 | r = dm_calculate_queue_limits(table, &limits); |
2201 | if (r) | |
2202 | goto out; | |
2203 | ||
e6ee8c0b KU |
2204 | /* cannot change the device type, once a table is bound */ |
2205 | if (md->map && | |
2206 | (dm_table_get_type(md->map) != dm_table_get_type(table))) { | |
2207 | DMWARN("can't change the device type after a table is bound"); | |
2208 | goto out; | |
2209 | } | |
2210 | ||
1da177e4 | 2211 | __unbind(md); |
754c5fc7 | 2212 | r = __bind(md, table, &limits); |
1da177e4 | 2213 | |
93c534ae | 2214 | out: |
e61290a4 | 2215 | mutex_unlock(&md->suspend_lock); |
93c534ae | 2216 | return r; |
1da177e4 LT |
2217 | } |
2218 | ||
cec47e3d KU |
2219 | static void dm_rq_invalidate_suspend_marker(struct mapped_device *md) |
2220 | { | |
2221 | md->suspend_rq.special = (void *)0x1; | |
2222 | } | |
2223 | ||
2224 | static void dm_rq_abort_suspend(struct mapped_device *md, int noflush) | |
2225 | { | |
2226 | struct request_queue *q = md->queue; | |
2227 | unsigned long flags; | |
2228 | ||
2229 | spin_lock_irqsave(q->queue_lock, flags); | |
2230 | if (!noflush) | |
2231 | dm_rq_invalidate_suspend_marker(md); | |
2232 | __start_queue(q); | |
2233 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2234 | } | |
2235 | ||
2236 | static void dm_rq_start_suspend(struct mapped_device *md, int noflush) | |
2237 | { | |
2238 | struct request *rq = &md->suspend_rq; | |
2239 | struct request_queue *q = md->queue; | |
2240 | ||
2241 | if (noflush) | |
2242 | stop_queue(q); | |
2243 | else { | |
2244 | blk_rq_init(q, rq); | |
2245 | blk_insert_request(q, rq, 0, NULL); | |
2246 | } | |
2247 | } | |
2248 | ||
2249 | static int dm_rq_suspend_available(struct mapped_device *md, int noflush) | |
2250 | { | |
2251 | int r = 1; | |
2252 | struct request *rq = &md->suspend_rq; | |
2253 | struct request_queue *q = md->queue; | |
2254 | unsigned long flags; | |
2255 | ||
2256 | if (noflush) | |
2257 | return r; | |
2258 | ||
2259 | /* The marker must be protected by queue lock if it is in use */ | |
2260 | spin_lock_irqsave(q->queue_lock, flags); | |
2261 | if (unlikely(rq->ref_count)) { | |
2262 | /* | |
2263 | * This can happen, when the previous flush suspend was | |
2264 | * interrupted, the marker is still in the queue and | |
2265 | * this flush suspend has been invoked, because we don't | |
2266 | * remove the marker at the time of suspend interruption. | |
2267 | * We have only one marker per mapped_device, so we can't | |
2268 | * start another flush suspend while it is in use. | |
2269 | */ | |
2270 | BUG_ON(!rq->special); /* The marker should be invalidated */ | |
2271 | DMWARN("Invalidating the previous flush suspend is still in" | |
2272 | " progress. Please retry later."); | |
2273 | r = 0; | |
2274 | } | |
2275 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2276 | ||
2277 | return r; | |
2278 | } | |
2279 | ||
1da177e4 LT |
2280 | /* |
2281 | * Functions to lock and unlock any filesystem running on the | |
2282 | * device. | |
2283 | */ | |
2ca3310e | 2284 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2285 | { |
e39e2e95 | 2286 | int r; |
1da177e4 LT |
2287 | |
2288 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2289 | |
db8fef4f | 2290 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2291 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2292 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2293 | md->frozen_sb = NULL; |
2294 | return r; | |
dfbe03f6 AK |
2295 | } |
2296 | ||
aa8d7c2f AK |
2297 | set_bit(DMF_FROZEN, &md->flags); |
2298 | ||
1da177e4 LT |
2299 | return 0; |
2300 | } | |
2301 | ||
2ca3310e | 2302 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2303 | { |
aa8d7c2f AK |
2304 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2305 | return; | |
2306 | ||
db8fef4f | 2307 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2308 | md->frozen_sb = NULL; |
aa8d7c2f | 2309 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2310 | } |
2311 | ||
2312 | /* | |
2313 | * We need to be able to change a mapping table under a mounted | |
2314 | * filesystem. For example we might want to move some data in | |
2315 | * the background. Before the table can be swapped with | |
2316 | * dm_bind_table, dm_suspend must be called to flush any in | |
2317 | * flight bios and ensure that any further io gets deferred. | |
2318 | */ | |
cec47e3d KU |
2319 | /* |
2320 | * Suspend mechanism in request-based dm. | |
2321 | * | |
2322 | * After the suspend starts, further incoming requests are kept in | |
2323 | * the request_queue and deferred. | |
2324 | * Remaining requests in the request_queue at the start of suspend are flushed | |
2325 | * if it is flush suspend. | |
2326 | * The suspend completes when the following conditions have been satisfied, | |
2327 | * so wait for it: | |
2328 | * 1. q->in_flight is 0 (which means no in_flight request) | |
2329 | * 2. queue has been stopped (which means no request dispatching) | |
2330 | * | |
2331 | * | |
2332 | * Noflush suspend | |
2333 | * --------------- | |
2334 | * Noflush suspend doesn't need to dispatch remaining requests. | |
2335 | * So stop the queue immediately. Then, wait for all in_flight requests | |
2336 | * to be completed or requeued. | |
2337 | * | |
2338 | * To abort noflush suspend, start the queue. | |
2339 | * | |
2340 | * | |
2341 | * Flush suspend | |
2342 | * ------------- | |
2343 | * Flush suspend needs to dispatch remaining requests. So stop the queue | |
2344 | * after the remaining requests are completed. (Requeued request must be also | |
2345 | * re-dispatched and completed. Until then, we can't stop the queue.) | |
2346 | * | |
2347 | * During flushing the remaining requests, further incoming requests are also | |
2348 | * inserted to the same queue. To distinguish which requests are to be | |
2349 | * flushed, we insert a marker request to the queue at the time of starting | |
2350 | * flush suspend, like a barrier. | |
2351 | * The dispatching is blocked when the marker is found on the top of the queue. | |
2352 | * And the queue is stopped when all in_flight requests are completed, since | |
2353 | * that means the remaining requests are completely flushed. | |
2354 | * Then, the marker is removed from the queue. | |
2355 | * | |
2356 | * To abort flush suspend, we also need to take care of the marker, not only | |
2357 | * starting the queue. | |
2358 | * We don't remove the marker forcibly from the queue since it's against | |
2359 | * the block-layer manner. Instead, we put a invalidated mark on the marker. | |
2360 | * When the invalidated marker is found on the top of the queue, it is | |
2361 | * immediately removed from the queue, so it doesn't block dispatching. | |
2362 | * Because we have only one marker per mapped_device, we can't start another | |
2363 | * flush suspend until the invalidated marker is removed from the queue. | |
2364 | * So fail and return with -EBUSY in such a case. | |
2365 | */ | |
a3d77d35 | 2366 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2367 | { |
2ca3310e | 2368 | struct dm_table *map = NULL; |
46125c1c | 2369 | int r = 0; |
a3d77d35 | 2370 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2371 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2372 | |
e61290a4 | 2373 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2374 | |
73d410c0 MB |
2375 | if (dm_suspended(md)) { |
2376 | r = -EINVAL; | |
d287483d | 2377 | goto out_unlock; |
73d410c0 | 2378 | } |
1da177e4 | 2379 | |
cec47e3d KU |
2380 | if (dm_request_based(md) && !dm_rq_suspend_available(md, noflush)) { |
2381 | r = -EBUSY; | |
2382 | goto out_unlock; | |
2383 | } | |
2384 | ||
1da177e4 | 2385 | map = dm_get_table(md); |
1da177e4 | 2386 | |
2e93ccc1 KU |
2387 | /* |
2388 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2389 | * This flag is cleared before dm_suspend returns. | |
2390 | */ | |
2391 | if (noflush) | |
2392 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2393 | ||
cf222b37 AK |
2394 | /* This does not get reverted if there's an error later. */ |
2395 | dm_table_presuspend_targets(map); | |
2396 | ||
32a926da MP |
2397 | /* |
2398 | * Flush I/O to the device. noflush supersedes do_lockfs, | |
2399 | * because lock_fs() needs to flush I/Os. | |
2400 | */ | |
2401 | if (!noflush && do_lockfs) { | |
2402 | r = lock_fs(md); | |
2403 | if (r) | |
f431d966 | 2404 | goto out; |
aa8d7c2f | 2405 | } |
1da177e4 LT |
2406 | |
2407 | /* | |
3b00b203 MP |
2408 | * Here we must make sure that no processes are submitting requests |
2409 | * to target drivers i.e. no one may be executing | |
2410 | * __split_and_process_bio. This is called from dm_request and | |
2411 | * dm_wq_work. | |
2412 | * | |
2413 | * To get all processes out of __split_and_process_bio in dm_request, | |
2414 | * we take the write lock. To prevent any process from reentering | |
2415 | * __split_and_process_bio from dm_request, we set | |
2416 | * DMF_QUEUE_IO_TO_THREAD. | |
2417 | * | |
2418 | * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND | |
2419 | * and call flush_workqueue(md->wq). flush_workqueue will wait until | |
2420 | * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any | |
2421 | * further calls to __split_and_process_bio from dm_wq_work. | |
1da177e4 | 2422 | */ |
2ca3310e | 2423 | down_write(&md->io_lock); |
1eb787ec AK |
2424 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2425 | set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); | |
2ca3310e | 2426 | up_write(&md->io_lock); |
1da177e4 | 2427 | |
3b00b203 MP |
2428 | flush_workqueue(md->wq); |
2429 | ||
cec47e3d KU |
2430 | if (dm_request_based(md)) |
2431 | dm_rq_start_suspend(md, noflush); | |
2432 | ||
1da177e4 | 2433 | /* |
3b00b203 MP |
2434 | * At this point no more requests are entering target request routines. |
2435 | * We call dm_wait_for_completion to wait for all existing requests | |
2436 | * to finish. | |
1da177e4 | 2437 | */ |
401600df | 2438 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2439 | |
2ca3310e | 2440 | down_write(&md->io_lock); |
6d6f10df | 2441 | if (noflush) |
022c2611 | 2442 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2443 | up_write(&md->io_lock); |
2e93ccc1 | 2444 | |
1da177e4 | 2445 | /* were we interrupted ? */ |
46125c1c | 2446 | if (r < 0) { |
9a1fb464 | 2447 | dm_queue_flush(md); |
73d410c0 | 2448 | |
cec47e3d KU |
2449 | if (dm_request_based(md)) |
2450 | dm_rq_abort_suspend(md, noflush); | |
2451 | ||
2ca3310e | 2452 | unlock_fs(md); |
2e93ccc1 | 2453 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2454 | } |
1da177e4 | 2455 | |
3b00b203 MP |
2456 | /* |
2457 | * If dm_wait_for_completion returned 0, the device is completely | |
2458 | * quiescent now. There is no request-processing activity. All new | |
2459 | * requests are being added to md->deferred list. | |
2460 | */ | |
2461 | ||
cf222b37 | 2462 | dm_table_postsuspend_targets(map); |
1da177e4 | 2463 | |
2ca3310e | 2464 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2465 | |
2ca3310e AK |
2466 | out: |
2467 | dm_table_put(map); | |
d287483d AK |
2468 | |
2469 | out_unlock: | |
e61290a4 | 2470 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2471 | return r; |
1da177e4 LT |
2472 | } |
2473 | ||
2474 | int dm_resume(struct mapped_device *md) | |
2475 | { | |
cf222b37 | 2476 | int r = -EINVAL; |
cf222b37 | 2477 | struct dm_table *map = NULL; |
1da177e4 | 2478 | |
e61290a4 | 2479 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2480 | if (!dm_suspended(md)) |
cf222b37 | 2481 | goto out; |
cf222b37 AK |
2482 | |
2483 | map = dm_get_table(md); | |
2ca3310e | 2484 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2485 | goto out; |
1da177e4 | 2486 | |
8757b776 MB |
2487 | r = dm_table_resume_targets(map); |
2488 | if (r) | |
2489 | goto out; | |
2ca3310e | 2490 | |
9a1fb464 | 2491 | dm_queue_flush(md); |
2ca3310e | 2492 | |
cec47e3d KU |
2493 | /* |
2494 | * Flushing deferred I/Os must be done after targets are resumed | |
2495 | * so that mapping of targets can work correctly. | |
2496 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2497 | */ | |
2498 | if (dm_request_based(md)) | |
2499 | start_queue(md->queue); | |
2500 | ||
2ca3310e AK |
2501 | unlock_fs(md); |
2502 | ||
2503 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2504 | ||
1da177e4 | 2505 | dm_table_unplug_all(map); |
cf222b37 AK |
2506 | r = 0; |
2507 | out: | |
2508 | dm_table_put(map); | |
e61290a4 | 2509 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2510 | |
cf222b37 | 2511 | return r; |
1da177e4 LT |
2512 | } |
2513 | ||
2514 | /*----------------------------------------------------------------- | |
2515 | * Event notification. | |
2516 | *---------------------------------------------------------------*/ | |
60935eb2 MB |
2517 | void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
2518 | unsigned cookie) | |
69267a30 | 2519 | { |
60935eb2 MB |
2520 | char udev_cookie[DM_COOKIE_LENGTH]; |
2521 | char *envp[] = { udev_cookie, NULL }; | |
2522 | ||
2523 | if (!cookie) | |
2524 | kobject_uevent(&disk_to_dev(md->disk)->kobj, action); | |
2525 | else { | |
2526 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2527 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
2528 | kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp); | |
2529 | } | |
69267a30 AK |
2530 | } |
2531 | ||
7a8c3d3b MA |
2532 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2533 | { | |
2534 | return atomic_add_return(1, &md->uevent_seq); | |
2535 | } | |
2536 | ||
1da177e4 LT |
2537 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2538 | { | |
2539 | return atomic_read(&md->event_nr); | |
2540 | } | |
2541 | ||
2542 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2543 | { | |
2544 | return wait_event_interruptible(md->eventq, | |
2545 | (event_nr != atomic_read(&md->event_nr))); | |
2546 | } | |
2547 | ||
7a8c3d3b MA |
2548 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2549 | { | |
2550 | unsigned long flags; | |
2551 | ||
2552 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2553 | list_add(elist, &md->uevent_list); | |
2554 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2555 | } | |
2556 | ||
1da177e4 LT |
2557 | /* |
2558 | * The gendisk is only valid as long as you have a reference | |
2559 | * count on 'md'. | |
2560 | */ | |
2561 | struct gendisk *dm_disk(struct mapped_device *md) | |
2562 | { | |
2563 | return md->disk; | |
2564 | } | |
2565 | ||
784aae73 MB |
2566 | struct kobject *dm_kobject(struct mapped_device *md) |
2567 | { | |
2568 | return &md->kobj; | |
2569 | } | |
2570 | ||
2571 | /* | |
2572 | * struct mapped_device should not be exported outside of dm.c | |
2573 | * so use this check to verify that kobj is part of md structure | |
2574 | */ | |
2575 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2576 | { | |
2577 | struct mapped_device *md; | |
2578 | ||
2579 | md = container_of(kobj, struct mapped_device, kobj); | |
2580 | if (&md->kobj != kobj) | |
2581 | return NULL; | |
2582 | ||
4d89b7b4 MB |
2583 | if (test_bit(DMF_FREEING, &md->flags) || |
2584 | test_bit(DMF_DELETING, &md->flags)) | |
2585 | return NULL; | |
2586 | ||
784aae73 MB |
2587 | dm_get(md); |
2588 | return md; | |
2589 | } | |
2590 | ||
1da177e4 LT |
2591 | int dm_suspended(struct mapped_device *md) |
2592 | { | |
2593 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2594 | } | |
2595 | ||
2e93ccc1 KU |
2596 | int dm_noflush_suspending(struct dm_target *ti) |
2597 | { | |
2598 | struct mapped_device *md = dm_table_get_md(ti->table); | |
2599 | int r = __noflush_suspending(md); | |
2600 | ||
2601 | dm_put(md); | |
2602 | ||
2603 | return r; | |
2604 | } | |
2605 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2606 | ||
e6ee8c0b KU |
2607 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type) |
2608 | { | |
2609 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
2610 | ||
2611 | if (!pools) | |
2612 | return NULL; | |
2613 | ||
2614 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2615 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2616 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2617 | if (!pools->io_pool) | |
2618 | goto free_pools_and_out; | |
2619 | ||
2620 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2621 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2622 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2623 | if (!pools->tio_pool) | |
2624 | goto free_io_pool_and_out; | |
2625 | ||
2626 | pools->bs = (type == DM_TYPE_BIO_BASED) ? | |
2627 | bioset_create(16, 0) : bioset_create(MIN_IOS, 0); | |
2628 | if (!pools->bs) | |
2629 | goto free_tio_pool_and_out; | |
2630 | ||
2631 | return pools; | |
2632 | ||
2633 | free_tio_pool_and_out: | |
2634 | mempool_destroy(pools->tio_pool); | |
2635 | ||
2636 | free_io_pool_and_out: | |
2637 | mempool_destroy(pools->io_pool); | |
2638 | ||
2639 | free_pools_and_out: | |
2640 | kfree(pools); | |
2641 | ||
2642 | return NULL; | |
2643 | } | |
2644 | ||
2645 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2646 | { | |
2647 | if (!pools) | |
2648 | return; | |
2649 | ||
2650 | if (pools->io_pool) | |
2651 | mempool_destroy(pools->io_pool); | |
2652 | ||
2653 | if (pools->tio_pool) | |
2654 | mempool_destroy(pools->tio_pool); | |
2655 | ||
2656 | if (pools->bs) | |
2657 | bioset_free(pools->bs); | |
2658 | ||
2659 | kfree(pools); | |
2660 | } | |
2661 | ||
1da177e4 LT |
2662 | static struct block_device_operations dm_blk_dops = { |
2663 | .open = dm_blk_open, | |
2664 | .release = dm_blk_close, | |
aa129a22 | 2665 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2666 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2667 | .owner = THIS_MODULE |
2668 | }; | |
2669 | ||
2670 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2671 | ||
2672 | /* | |
2673 | * module hooks | |
2674 | */ | |
2675 | module_init(dm_init); | |
2676 | module_exit(dm_exit); | |
2677 | ||
2678 | module_param(major, uint, 0); | |
2679 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2680 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2681 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2682 | MODULE_LICENSE("GPL"); |