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; |
f88fb981 | 50 | spinlock_t endio_lock; |
1da177e4 LT |
51 | }; |
52 | ||
53 | /* | |
8fbf26ad | 54 | * For bio-based dm. |
1da177e4 LT |
55 | * One of these is allocated per target within a bio. Hopefully |
56 | * this will be simplified out one day. | |
57 | */ | |
028867ac | 58 | struct dm_target_io { |
1da177e4 LT |
59 | struct dm_io *io; |
60 | struct dm_target *ti; | |
61 | union map_info info; | |
62 | }; | |
63 | ||
8fbf26ad KU |
64 | /* |
65 | * For request-based dm. | |
66 | * One of these is allocated per request. | |
67 | */ | |
68 | struct dm_rq_target_io { | |
69 | struct mapped_device *md; | |
70 | struct dm_target *ti; | |
71 | struct request *orig, clone; | |
72 | int error; | |
73 | union map_info info; | |
74 | }; | |
75 | ||
76 | /* | |
77 | * For request-based dm. | |
78 | * One of these is allocated per bio. | |
79 | */ | |
80 | struct dm_rq_clone_bio_info { | |
81 | struct bio *orig; | |
cec47e3d | 82 | struct dm_rq_target_io *tio; |
8fbf26ad KU |
83 | }; |
84 | ||
1da177e4 LT |
85 | union map_info *dm_get_mapinfo(struct bio *bio) |
86 | { | |
17b2f66f | 87 | if (bio && bio->bi_private) |
028867ac | 88 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 89 | return NULL; |
1da177e4 LT |
90 | } |
91 | ||
cec47e3d KU |
92 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
93 | { | |
94 | if (rq && rq->end_io_data) | |
95 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
96 | return NULL; | |
97 | } | |
98 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
99 | ||
ba61fdd1 JM |
100 | #define MINOR_ALLOCED ((void *)-1) |
101 | ||
1da177e4 LT |
102 | /* |
103 | * Bits for the md->flags field. | |
104 | */ | |
1eb787ec | 105 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 106 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 107 | #define DMF_FROZEN 2 |
fba9f90e | 108 | #define DMF_FREEING 3 |
5c6bd75d | 109 | #define DMF_DELETING 4 |
2e93ccc1 | 110 | #define DMF_NOFLUSH_SUSPENDING 5 |
1eb787ec | 111 | #define DMF_QUEUE_IO_TO_THREAD 6 |
1da177e4 | 112 | |
304f3f6a MB |
113 | /* |
114 | * Work processed by per-device workqueue. | |
115 | */ | |
1da177e4 | 116 | struct mapped_device { |
2ca3310e | 117 | struct rw_semaphore io_lock; |
e61290a4 | 118 | struct mutex suspend_lock; |
1da177e4 LT |
119 | rwlock_t map_lock; |
120 | atomic_t holders; | |
5c6bd75d | 121 | atomic_t open_count; |
1da177e4 LT |
122 | |
123 | unsigned long flags; | |
124 | ||
165125e1 | 125 | struct request_queue *queue; |
1da177e4 | 126 | struct gendisk *disk; |
7e51f257 | 127 | char name[16]; |
1da177e4 LT |
128 | |
129 | void *interface_ptr; | |
130 | ||
131 | /* | |
132 | * A list of ios that arrived while we were suspended. | |
133 | */ | |
316d315b | 134 | atomic_t pending[2]; |
1da177e4 | 135 | wait_queue_head_t wait; |
53d5914f | 136 | struct work_struct work; |
74859364 | 137 | struct bio_list deferred; |
022c2611 | 138 | spinlock_t deferred_lock; |
1da177e4 | 139 | |
af7e466a MP |
140 | /* |
141 | * An error from the barrier request currently being processed. | |
142 | */ | |
143 | int barrier_error; | |
144 | ||
304f3f6a MB |
145 | /* |
146 | * Processing queue (flush/barriers) | |
147 | */ | |
148 | struct workqueue_struct *wq; | |
149 | ||
1da177e4 LT |
150 | /* |
151 | * The current mapping. | |
152 | */ | |
153 | struct dm_table *map; | |
154 | ||
155 | /* | |
156 | * io objects are allocated from here. | |
157 | */ | |
158 | mempool_t *io_pool; | |
159 | mempool_t *tio_pool; | |
160 | ||
9faf400f SB |
161 | struct bio_set *bs; |
162 | ||
1da177e4 LT |
163 | /* |
164 | * Event handling. | |
165 | */ | |
166 | atomic_t event_nr; | |
167 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
168 | atomic_t uevent_seq; |
169 | struct list_head uevent_list; | |
170 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
171 | |
172 | /* | |
173 | * freeze/thaw support require holding onto a super block | |
174 | */ | |
175 | struct super_block *frozen_sb; | |
db8fef4f | 176 | struct block_device *bdev; |
3ac51e74 DW |
177 | |
178 | /* forced geometry settings */ | |
179 | struct hd_geometry geometry; | |
784aae73 | 180 | |
cec47e3d KU |
181 | /* marker of flush suspend for request-based dm */ |
182 | struct request suspend_rq; | |
183 | ||
184 | /* For saving the address of __make_request for request based dm */ | |
185 | make_request_fn *saved_make_request_fn; | |
186 | ||
784aae73 MB |
187 | /* sysfs handle */ |
188 | struct kobject kobj; | |
52b1fd5a MP |
189 | |
190 | /* zero-length barrier that will be cloned and submitted to targets */ | |
191 | struct bio barrier_bio; | |
1da177e4 LT |
192 | }; |
193 | ||
e6ee8c0b KU |
194 | /* |
195 | * For mempools pre-allocation at the table loading time. | |
196 | */ | |
197 | struct dm_md_mempools { | |
198 | mempool_t *io_pool; | |
199 | mempool_t *tio_pool; | |
200 | struct bio_set *bs; | |
201 | }; | |
202 | ||
1da177e4 | 203 | #define MIN_IOS 256 |
e18b890b CL |
204 | static struct kmem_cache *_io_cache; |
205 | static struct kmem_cache *_tio_cache; | |
8fbf26ad KU |
206 | static struct kmem_cache *_rq_tio_cache; |
207 | static struct kmem_cache *_rq_bio_info_cache; | |
1da177e4 | 208 | |
1da177e4 LT |
209 | static int __init local_init(void) |
210 | { | |
51157b4a | 211 | int r = -ENOMEM; |
1da177e4 | 212 | |
1da177e4 | 213 | /* allocate a slab for the dm_ios */ |
028867ac | 214 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 215 | if (!_io_cache) |
51157b4a | 216 | return r; |
1da177e4 LT |
217 | |
218 | /* allocate a slab for the target ios */ | |
028867ac | 219 | _tio_cache = KMEM_CACHE(dm_target_io, 0); |
51157b4a KU |
220 | if (!_tio_cache) |
221 | goto out_free_io_cache; | |
1da177e4 | 222 | |
8fbf26ad KU |
223 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
224 | if (!_rq_tio_cache) | |
225 | goto out_free_tio_cache; | |
226 | ||
227 | _rq_bio_info_cache = KMEM_CACHE(dm_rq_clone_bio_info, 0); | |
228 | if (!_rq_bio_info_cache) | |
229 | goto out_free_rq_tio_cache; | |
230 | ||
51e5b2bd | 231 | r = dm_uevent_init(); |
51157b4a | 232 | if (r) |
8fbf26ad | 233 | goto out_free_rq_bio_info_cache; |
51e5b2bd | 234 | |
1da177e4 LT |
235 | _major = major; |
236 | r = register_blkdev(_major, _name); | |
51157b4a KU |
237 | if (r < 0) |
238 | goto out_uevent_exit; | |
1da177e4 LT |
239 | |
240 | if (!_major) | |
241 | _major = r; | |
242 | ||
243 | return 0; | |
51157b4a KU |
244 | |
245 | out_uevent_exit: | |
246 | dm_uevent_exit(); | |
8fbf26ad KU |
247 | out_free_rq_bio_info_cache: |
248 | kmem_cache_destroy(_rq_bio_info_cache); | |
249 | out_free_rq_tio_cache: | |
250 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
251 | out_free_tio_cache: |
252 | kmem_cache_destroy(_tio_cache); | |
253 | out_free_io_cache: | |
254 | kmem_cache_destroy(_io_cache); | |
255 | ||
256 | return r; | |
1da177e4 LT |
257 | } |
258 | ||
259 | static void local_exit(void) | |
260 | { | |
8fbf26ad KU |
261 | kmem_cache_destroy(_rq_bio_info_cache); |
262 | kmem_cache_destroy(_rq_tio_cache); | |
1da177e4 LT |
263 | kmem_cache_destroy(_tio_cache); |
264 | kmem_cache_destroy(_io_cache); | |
00d59405 | 265 | unregister_blkdev(_major, _name); |
51e5b2bd | 266 | dm_uevent_exit(); |
1da177e4 LT |
267 | |
268 | _major = 0; | |
269 | ||
270 | DMINFO("cleaned up"); | |
271 | } | |
272 | ||
b9249e55 | 273 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
274 | local_init, |
275 | dm_target_init, | |
276 | dm_linear_init, | |
277 | dm_stripe_init, | |
952b3557 | 278 | dm_io_init, |
945fa4d2 | 279 | dm_kcopyd_init, |
1da177e4 LT |
280 | dm_interface_init, |
281 | }; | |
282 | ||
b9249e55 | 283 | static void (*_exits[])(void) = { |
1da177e4 LT |
284 | local_exit, |
285 | dm_target_exit, | |
286 | dm_linear_exit, | |
287 | dm_stripe_exit, | |
952b3557 | 288 | dm_io_exit, |
945fa4d2 | 289 | dm_kcopyd_exit, |
1da177e4 LT |
290 | dm_interface_exit, |
291 | }; | |
292 | ||
293 | static int __init dm_init(void) | |
294 | { | |
295 | const int count = ARRAY_SIZE(_inits); | |
296 | ||
297 | int r, i; | |
298 | ||
299 | for (i = 0; i < count; i++) { | |
300 | r = _inits[i](); | |
301 | if (r) | |
302 | goto bad; | |
303 | } | |
304 | ||
305 | return 0; | |
306 | ||
307 | bad: | |
308 | while (i--) | |
309 | _exits[i](); | |
310 | ||
311 | return r; | |
312 | } | |
313 | ||
314 | static void __exit dm_exit(void) | |
315 | { | |
316 | int i = ARRAY_SIZE(_exits); | |
317 | ||
318 | while (i--) | |
319 | _exits[i](); | |
320 | } | |
321 | ||
322 | /* | |
323 | * Block device functions | |
324 | */ | |
fe5f9f2c | 325 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
326 | { |
327 | struct mapped_device *md; | |
328 | ||
fba9f90e JM |
329 | spin_lock(&_minor_lock); |
330 | ||
fe5f9f2c | 331 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
332 | if (!md) |
333 | goto out; | |
334 | ||
5c6bd75d AK |
335 | if (test_bit(DMF_FREEING, &md->flags) || |
336 | test_bit(DMF_DELETING, &md->flags)) { | |
fba9f90e JM |
337 | md = NULL; |
338 | goto out; | |
339 | } | |
340 | ||
1da177e4 | 341 | dm_get(md); |
5c6bd75d | 342 | atomic_inc(&md->open_count); |
fba9f90e JM |
343 | |
344 | out: | |
345 | spin_unlock(&_minor_lock); | |
346 | ||
347 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
348 | } |
349 | ||
fe5f9f2c | 350 | static int dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 351 | { |
fe5f9f2c | 352 | struct mapped_device *md = disk->private_data; |
5c6bd75d | 353 | atomic_dec(&md->open_count); |
1da177e4 LT |
354 | dm_put(md); |
355 | return 0; | |
356 | } | |
357 | ||
5c6bd75d AK |
358 | int dm_open_count(struct mapped_device *md) |
359 | { | |
360 | return atomic_read(&md->open_count); | |
361 | } | |
362 | ||
363 | /* | |
364 | * Guarantees nothing is using the device before it's deleted. | |
365 | */ | |
366 | int dm_lock_for_deletion(struct mapped_device *md) | |
367 | { | |
368 | int r = 0; | |
369 | ||
370 | spin_lock(&_minor_lock); | |
371 | ||
372 | if (dm_open_count(md)) | |
373 | r = -EBUSY; | |
374 | else | |
375 | set_bit(DMF_DELETING, &md->flags); | |
376 | ||
377 | spin_unlock(&_minor_lock); | |
378 | ||
379 | return r; | |
380 | } | |
381 | ||
3ac51e74 DW |
382 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
383 | { | |
384 | struct mapped_device *md = bdev->bd_disk->private_data; | |
385 | ||
386 | return dm_get_geometry(md, geo); | |
387 | } | |
388 | ||
fe5f9f2c | 389 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
390 | unsigned int cmd, unsigned long arg) |
391 | { | |
fe5f9f2c AV |
392 | struct mapped_device *md = bdev->bd_disk->private_data; |
393 | struct dm_table *map = dm_get_table(md); | |
aa129a22 MB |
394 | struct dm_target *tgt; |
395 | int r = -ENOTTY; | |
396 | ||
aa129a22 MB |
397 | if (!map || !dm_table_get_size(map)) |
398 | goto out; | |
399 | ||
400 | /* We only support devices that have a single target */ | |
401 | if (dm_table_get_num_targets(map) != 1) | |
402 | goto out; | |
403 | ||
404 | tgt = dm_table_get_target(map, 0); | |
405 | ||
406 | if (dm_suspended(md)) { | |
407 | r = -EAGAIN; | |
408 | goto out; | |
409 | } | |
410 | ||
411 | if (tgt->type->ioctl) | |
647b3d00 | 412 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
413 | |
414 | out: | |
415 | dm_table_put(map); | |
416 | ||
aa129a22 MB |
417 | return r; |
418 | } | |
419 | ||
028867ac | 420 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
421 | { |
422 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
423 | } | |
424 | ||
028867ac | 425 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
426 | { |
427 | mempool_free(io, md->io_pool); | |
428 | } | |
429 | ||
028867ac | 430 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 LT |
431 | { |
432 | mempool_free(tio, md->tio_pool); | |
433 | } | |
434 | ||
08885643 KU |
435 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
436 | gfp_t gfp_mask) | |
cec47e3d | 437 | { |
08885643 | 438 | return mempool_alloc(md->tio_pool, gfp_mask); |
cec47e3d KU |
439 | } |
440 | ||
441 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
442 | { | |
443 | mempool_free(tio, tio->md->tio_pool); | |
444 | } | |
445 | ||
446 | static struct dm_rq_clone_bio_info *alloc_bio_info(struct mapped_device *md) | |
447 | { | |
448 | return mempool_alloc(md->io_pool, GFP_ATOMIC); | |
449 | } | |
450 | ||
451 | static void free_bio_info(struct dm_rq_clone_bio_info *info) | |
452 | { | |
453 | mempool_free(info, info->tio->md->io_pool); | |
454 | } | |
455 | ||
90abb8c4 KU |
456 | static int md_in_flight(struct mapped_device *md) |
457 | { | |
458 | return atomic_read(&md->pending[READ]) + | |
459 | atomic_read(&md->pending[WRITE]); | |
460 | } | |
461 | ||
3eaf840e JNN |
462 | static void start_io_acct(struct dm_io *io) |
463 | { | |
464 | struct mapped_device *md = io->md; | |
c9959059 | 465 | int cpu; |
316d315b | 466 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
467 | |
468 | io->start_time = jiffies; | |
469 | ||
074a7aca TH |
470 | cpu = part_stat_lock(); |
471 | part_round_stats(cpu, &dm_disk(md)->part0); | |
472 | part_stat_unlock(); | |
316d315b | 473 | dm_disk(md)->part0.in_flight[rw] = atomic_inc_return(&md->pending[rw]); |
3eaf840e JNN |
474 | } |
475 | ||
d221d2e7 | 476 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
477 | { |
478 | struct mapped_device *md = io->md; | |
479 | struct bio *bio = io->bio; | |
480 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 481 | int pending, cpu; |
3eaf840e JNN |
482 | int rw = bio_data_dir(bio); |
483 | ||
074a7aca TH |
484 | cpu = part_stat_lock(); |
485 | part_round_stats(cpu, &dm_disk(md)->part0); | |
486 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
487 | part_stat_unlock(); | |
3eaf840e | 488 | |
af7e466a MP |
489 | /* |
490 | * After this is decremented the bio must not be touched if it is | |
491 | * a barrier. | |
492 | */ | |
316d315b NK |
493 | dm_disk(md)->part0.in_flight[rw] = pending = |
494 | atomic_dec_return(&md->pending[rw]); | |
495 | pending += atomic_read(&md->pending[rw^0x1]); | |
3eaf840e | 496 | |
d221d2e7 MP |
497 | /* nudge anyone waiting on suspend queue */ |
498 | if (!pending) | |
499 | wake_up(&md->wait); | |
3eaf840e JNN |
500 | } |
501 | ||
1da177e4 LT |
502 | /* |
503 | * Add the bio to the list of deferred io. | |
504 | */ | |
92c63902 | 505 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 506 | { |
2ca3310e | 507 | down_write(&md->io_lock); |
1da177e4 | 508 | |
022c2611 | 509 | spin_lock_irq(&md->deferred_lock); |
1da177e4 | 510 | bio_list_add(&md->deferred, bio); |
022c2611 | 511 | spin_unlock_irq(&md->deferred_lock); |
1da177e4 | 512 | |
92c63902 MP |
513 | if (!test_and_set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) |
514 | queue_work(md->wq, &md->work); | |
515 | ||
2ca3310e | 516 | up_write(&md->io_lock); |
1da177e4 LT |
517 | } |
518 | ||
519 | /* | |
520 | * Everyone (including functions in this file), should use this | |
521 | * function to access the md->map field, and make sure they call | |
522 | * dm_table_put() when finished. | |
523 | */ | |
524 | struct dm_table *dm_get_table(struct mapped_device *md) | |
525 | { | |
526 | struct dm_table *t; | |
523d9297 | 527 | unsigned long flags; |
1da177e4 | 528 | |
523d9297 | 529 | read_lock_irqsave(&md->map_lock, flags); |
1da177e4 LT |
530 | t = md->map; |
531 | if (t) | |
532 | dm_table_get(t); | |
523d9297 | 533 | read_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 LT |
534 | |
535 | return t; | |
536 | } | |
537 | ||
3ac51e74 DW |
538 | /* |
539 | * Get the geometry associated with a dm device | |
540 | */ | |
541 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
542 | { | |
543 | *geo = md->geometry; | |
544 | ||
545 | return 0; | |
546 | } | |
547 | ||
548 | /* | |
549 | * Set the geometry of a device. | |
550 | */ | |
551 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
552 | { | |
553 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
554 | ||
555 | if (geo->start > sz) { | |
556 | DMWARN("Start sector is beyond the geometry limits."); | |
557 | return -EINVAL; | |
558 | } | |
559 | ||
560 | md->geometry = *geo; | |
561 | ||
562 | return 0; | |
563 | } | |
564 | ||
1da177e4 LT |
565 | /*----------------------------------------------------------------- |
566 | * CRUD START: | |
567 | * A more elegant soln is in the works that uses the queue | |
568 | * merge fn, unfortunately there are a couple of changes to | |
569 | * the block layer that I want to make for this. So in the | |
570 | * interests of getting something for people to use I give | |
571 | * you this clearly demarcated crap. | |
572 | *---------------------------------------------------------------*/ | |
573 | ||
2e93ccc1 KU |
574 | static int __noflush_suspending(struct mapped_device *md) |
575 | { | |
576 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
577 | } | |
578 | ||
1da177e4 LT |
579 | /* |
580 | * Decrements the number of outstanding ios that a bio has been | |
581 | * cloned into, completing the original io if necc. | |
582 | */ | |
858119e1 | 583 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 584 | { |
2e93ccc1 | 585 | unsigned long flags; |
b35f8caa MB |
586 | int io_error; |
587 | struct bio *bio; | |
588 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
589 | |
590 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
591 | if (unlikely(error)) { |
592 | spin_lock_irqsave(&io->endio_lock, flags); | |
593 | if (!(io->error > 0 && __noflush_suspending(md))) | |
594 | io->error = error; | |
595 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
596 | } | |
1da177e4 LT |
597 | |
598 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
599 | if (io->error == DM_ENDIO_REQUEUE) { |
600 | /* | |
601 | * Target requested pushing back the I/O. | |
2e93ccc1 | 602 | */ |
022c2611 | 603 | spin_lock_irqsave(&md->deferred_lock, flags); |
2761e95f | 604 | if (__noflush_suspending(md)) { |
1f98a13f | 605 | if (!bio_rw_flagged(io->bio, BIO_RW_BARRIER)) |
2761e95f MP |
606 | bio_list_add_head(&md->deferred, |
607 | io->bio); | |
608 | } else | |
2e93ccc1 KU |
609 | /* noflush suspend was interrupted. */ |
610 | io->error = -EIO; | |
022c2611 | 611 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
612 | } |
613 | ||
b35f8caa MB |
614 | io_error = io->error; |
615 | bio = io->bio; | |
2e93ccc1 | 616 | |
1f98a13f | 617 | if (bio_rw_flagged(bio, BIO_RW_BARRIER)) { |
af7e466a MP |
618 | /* |
619 | * There can be just one barrier request so we use | |
620 | * a per-device variable for error reporting. | |
621 | * Note that you can't touch the bio after end_io_acct | |
622 | */ | |
fdb9572b | 623 | if (!md->barrier_error && io_error != -EOPNOTSUPP) |
5aa2781d | 624 | md->barrier_error = io_error; |
af7e466a MP |
625 | end_io_acct(io); |
626 | } else { | |
627 | end_io_acct(io); | |
b35f8caa | 628 | |
af7e466a MP |
629 | if (io_error != DM_ENDIO_REQUEUE) { |
630 | trace_block_bio_complete(md->queue, bio); | |
2056a782 | 631 | |
af7e466a MP |
632 | bio_endio(bio, io_error); |
633 | } | |
b35f8caa | 634 | } |
af7e466a MP |
635 | |
636 | free_io(md, io); | |
1da177e4 LT |
637 | } |
638 | } | |
639 | ||
6712ecf8 | 640 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
641 | { |
642 | int r = 0; | |
028867ac | 643 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 644 | struct dm_io *io = tio->io; |
9faf400f | 645 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
646 | dm_endio_fn endio = tio->ti->type->end_io; |
647 | ||
1da177e4 LT |
648 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
649 | error = -EIO; | |
650 | ||
651 | if (endio) { | |
652 | r = endio(tio->ti, bio, error, &tio->info); | |
2e93ccc1 KU |
653 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
654 | /* | |
655 | * error and requeue request are handled | |
656 | * in dec_pending(). | |
657 | */ | |
1da177e4 | 658 | error = r; |
45cbcd79 KU |
659 | else if (r == DM_ENDIO_INCOMPLETE) |
660 | /* The target will handle the io */ | |
6712ecf8 | 661 | return; |
45cbcd79 KU |
662 | else if (r) { |
663 | DMWARN("unimplemented target endio return value: %d", r); | |
664 | BUG(); | |
665 | } | |
1da177e4 LT |
666 | } |
667 | ||
9faf400f SB |
668 | /* |
669 | * Store md for cleanup instead of tio which is about to get freed. | |
670 | */ | |
671 | bio->bi_private = md->bs; | |
672 | ||
9faf400f | 673 | free_tio(md, tio); |
b35f8caa MB |
674 | bio_put(bio); |
675 | dec_pending(io, error); | |
1da177e4 LT |
676 | } |
677 | ||
cec47e3d KU |
678 | /* |
679 | * Partial completion handling for request-based dm | |
680 | */ | |
681 | static void end_clone_bio(struct bio *clone, int error) | |
682 | { | |
683 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
684 | struct dm_rq_target_io *tio = info->tio; | |
685 | struct bio *bio = info->orig; | |
686 | unsigned int nr_bytes = info->orig->bi_size; | |
687 | ||
688 | bio_put(clone); | |
689 | ||
690 | if (tio->error) | |
691 | /* | |
692 | * An error has already been detected on the request. | |
693 | * Once error occurred, just let clone->end_io() handle | |
694 | * the remainder. | |
695 | */ | |
696 | return; | |
697 | else if (error) { | |
698 | /* | |
699 | * Don't notice the error to the upper layer yet. | |
700 | * The error handling decision is made by the target driver, | |
701 | * when the request is completed. | |
702 | */ | |
703 | tio->error = error; | |
704 | return; | |
705 | } | |
706 | ||
707 | /* | |
708 | * I/O for the bio successfully completed. | |
709 | * Notice the data completion to the upper layer. | |
710 | */ | |
711 | ||
712 | /* | |
713 | * bios are processed from the head of the list. | |
714 | * So the completing bio should always be rq->bio. | |
715 | * If it's not, something wrong is happening. | |
716 | */ | |
717 | if (tio->orig->bio != bio) | |
718 | DMERR("bio completion is going in the middle of the request"); | |
719 | ||
720 | /* | |
721 | * Update the original request. | |
722 | * Do not use blk_end_request() here, because it may complete | |
723 | * the original request before the clone, and break the ordering. | |
724 | */ | |
725 | blk_update_request(tio->orig, 0, nr_bytes); | |
726 | } | |
727 | ||
728 | /* | |
729 | * Don't touch any member of the md after calling this function because | |
730 | * the md may be freed in dm_put() at the end of this function. | |
731 | * Or do dm_get() before calling this function and dm_put() later. | |
732 | */ | |
733 | static void rq_completed(struct mapped_device *md, int run_queue) | |
734 | { | |
735 | int wakeup_waiters = 0; | |
736 | struct request_queue *q = md->queue; | |
737 | unsigned long flags; | |
738 | ||
739 | spin_lock_irqsave(q->queue_lock, flags); | |
740 | if (!queue_in_flight(q)) | |
741 | wakeup_waiters = 1; | |
742 | spin_unlock_irqrestore(q->queue_lock, flags); | |
743 | ||
744 | /* nudge anyone waiting on suspend queue */ | |
745 | if (wakeup_waiters) | |
746 | wake_up(&md->wait); | |
747 | ||
748 | if (run_queue) | |
749 | blk_run_queue(q); | |
750 | ||
751 | /* | |
752 | * dm_put() must be at the end of this function. See the comment above | |
753 | */ | |
754 | dm_put(md); | |
755 | } | |
756 | ||
a77e28c7 KU |
757 | static void free_rq_clone(struct request *clone) |
758 | { | |
759 | struct dm_rq_target_io *tio = clone->end_io_data; | |
760 | ||
761 | blk_rq_unprep_clone(clone); | |
762 | free_rq_tio(tio); | |
763 | } | |
764 | ||
cec47e3d KU |
765 | static void dm_unprep_request(struct request *rq) |
766 | { | |
767 | struct request *clone = rq->special; | |
cec47e3d KU |
768 | |
769 | rq->special = NULL; | |
770 | rq->cmd_flags &= ~REQ_DONTPREP; | |
771 | ||
a77e28c7 | 772 | free_rq_clone(clone); |
cec47e3d KU |
773 | } |
774 | ||
775 | /* | |
776 | * Requeue the original request of a clone. | |
777 | */ | |
778 | void dm_requeue_unmapped_request(struct request *clone) | |
779 | { | |
780 | struct dm_rq_target_io *tio = clone->end_io_data; | |
781 | struct mapped_device *md = tio->md; | |
782 | struct request *rq = tio->orig; | |
783 | struct request_queue *q = rq->q; | |
784 | unsigned long flags; | |
785 | ||
786 | dm_unprep_request(rq); | |
787 | ||
788 | spin_lock_irqsave(q->queue_lock, flags); | |
789 | if (elv_queue_empty(q)) | |
790 | blk_plug_device(q); | |
791 | blk_requeue_request(q, rq); | |
792 | spin_unlock_irqrestore(q->queue_lock, flags); | |
793 | ||
794 | rq_completed(md, 0); | |
795 | } | |
796 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
797 | ||
798 | static void __stop_queue(struct request_queue *q) | |
799 | { | |
800 | blk_stop_queue(q); | |
801 | } | |
802 | ||
803 | static void stop_queue(struct request_queue *q) | |
804 | { | |
805 | unsigned long flags; | |
806 | ||
807 | spin_lock_irqsave(q->queue_lock, flags); | |
808 | __stop_queue(q); | |
809 | spin_unlock_irqrestore(q->queue_lock, flags); | |
810 | } | |
811 | ||
812 | static void __start_queue(struct request_queue *q) | |
813 | { | |
814 | if (blk_queue_stopped(q)) | |
815 | blk_start_queue(q); | |
816 | } | |
817 | ||
818 | static void start_queue(struct request_queue *q) | |
819 | { | |
820 | unsigned long flags; | |
821 | ||
822 | spin_lock_irqsave(q->queue_lock, flags); | |
823 | __start_queue(q); | |
824 | spin_unlock_irqrestore(q->queue_lock, flags); | |
825 | } | |
826 | ||
827 | /* | |
828 | * Complete the clone and the original request. | |
829 | * Must be called without queue lock. | |
830 | */ | |
831 | static void dm_end_request(struct request *clone, int error) | |
832 | { | |
833 | struct dm_rq_target_io *tio = clone->end_io_data; | |
834 | struct mapped_device *md = tio->md; | |
835 | struct request *rq = tio->orig; | |
836 | ||
837 | if (blk_pc_request(rq)) { | |
838 | rq->errors = clone->errors; | |
839 | rq->resid_len = clone->resid_len; | |
840 | ||
841 | if (rq->sense) | |
842 | /* | |
843 | * We are using the sense buffer of the original | |
844 | * request. | |
845 | * So setting the length of the sense data is enough. | |
846 | */ | |
847 | rq->sense_len = clone->sense_len; | |
848 | } | |
849 | ||
a77e28c7 | 850 | free_rq_clone(clone); |
cec47e3d KU |
851 | |
852 | blk_end_request_all(rq, error); | |
853 | ||
854 | rq_completed(md, 1); | |
855 | } | |
856 | ||
857 | /* | |
858 | * Request completion handler for request-based dm | |
859 | */ | |
860 | static void dm_softirq_done(struct request *rq) | |
861 | { | |
862 | struct request *clone = rq->completion_data; | |
863 | struct dm_rq_target_io *tio = clone->end_io_data; | |
864 | dm_request_endio_fn rq_end_io = tio->ti->type->rq_end_io; | |
865 | int error = tio->error; | |
866 | ||
867 | if (!(rq->cmd_flags & REQ_FAILED) && rq_end_io) | |
868 | error = rq_end_io(tio->ti, clone, error, &tio->info); | |
869 | ||
870 | if (error <= 0) | |
871 | /* The target wants to complete the I/O */ | |
872 | dm_end_request(clone, error); | |
873 | else if (error == DM_ENDIO_INCOMPLETE) | |
874 | /* The target will handle the I/O */ | |
875 | return; | |
876 | else if (error == DM_ENDIO_REQUEUE) | |
877 | /* The target wants to requeue the I/O */ | |
878 | dm_requeue_unmapped_request(clone); | |
879 | else { | |
880 | DMWARN("unimplemented target endio return value: %d", error); | |
881 | BUG(); | |
882 | } | |
883 | } | |
884 | ||
885 | /* | |
886 | * Complete the clone and the original request with the error status | |
887 | * through softirq context. | |
888 | */ | |
889 | static void dm_complete_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 | tio->error = error; | |
895 | rq->completion_data = clone; | |
896 | blk_complete_request(rq); | |
897 | } | |
898 | ||
899 | /* | |
900 | * Complete the not-mapped clone and the original request with the error status | |
901 | * through softirq context. | |
902 | * Target's rq_end_io() function isn't called. | |
903 | * This may be used when the target's map_rq() function fails. | |
904 | */ | |
905 | void dm_kill_unmapped_request(struct request *clone, int error) | |
906 | { | |
907 | struct dm_rq_target_io *tio = clone->end_io_data; | |
908 | struct request *rq = tio->orig; | |
909 | ||
910 | rq->cmd_flags |= REQ_FAILED; | |
911 | dm_complete_request(clone, error); | |
912 | } | |
913 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
914 | ||
915 | /* | |
916 | * Called with the queue lock held | |
917 | */ | |
918 | static void end_clone_request(struct request *clone, int error) | |
919 | { | |
920 | /* | |
921 | * For just cleaning up the information of the queue in which | |
922 | * the clone was dispatched. | |
923 | * The clone is *NOT* freed actually here because it is alloced from | |
924 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
925 | */ | |
926 | __blk_put_request(clone->q, clone); | |
927 | ||
928 | /* | |
929 | * Actual request completion is done in a softirq context which doesn't | |
930 | * hold the queue lock. Otherwise, deadlock could occur because: | |
931 | * - another request may be submitted by the upper level driver | |
932 | * of the stacking during the completion | |
933 | * - the submission which requires queue lock may be done | |
934 | * against this queue | |
935 | */ | |
936 | dm_complete_request(clone, error); | |
937 | } | |
938 | ||
1da177e4 LT |
939 | static sector_t max_io_len(struct mapped_device *md, |
940 | sector_t sector, struct dm_target *ti) | |
941 | { | |
942 | sector_t offset = sector - ti->begin; | |
943 | sector_t len = ti->len - offset; | |
944 | ||
945 | /* | |
946 | * Does the target need to split even further ? | |
947 | */ | |
948 | if (ti->split_io) { | |
949 | sector_t boundary; | |
950 | boundary = ((offset + ti->split_io) & ~(ti->split_io - 1)) | |
951 | - offset; | |
952 | if (len > boundary) | |
953 | len = boundary; | |
954 | } | |
955 | ||
956 | return len; | |
957 | } | |
958 | ||
959 | static void __map_bio(struct dm_target *ti, struct bio *clone, | |
028867ac | 960 | struct dm_target_io *tio) |
1da177e4 LT |
961 | { |
962 | int r; | |
2056a782 | 963 | sector_t sector; |
9faf400f | 964 | struct mapped_device *md; |
1da177e4 | 965 | |
1da177e4 LT |
966 | clone->bi_end_io = clone_endio; |
967 | clone->bi_private = tio; | |
968 | ||
969 | /* | |
970 | * Map the clone. If r == 0 we don't need to do | |
971 | * anything, the target has assumed ownership of | |
972 | * this io. | |
973 | */ | |
974 | atomic_inc(&tio->io->io_count); | |
2056a782 | 975 | sector = clone->bi_sector; |
1da177e4 | 976 | r = ti->type->map(ti, clone, &tio->info); |
45cbcd79 | 977 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 978 | /* the bio has been remapped so dispatch it */ |
2056a782 | 979 | |
5f3ea37c | 980 | trace_block_remap(bdev_get_queue(clone->bi_bdev), clone, |
22a7c31a | 981 | tio->io->bio->bi_bdev->bd_dev, sector); |
2056a782 | 982 | |
1da177e4 | 983 | generic_make_request(clone); |
2e93ccc1 KU |
984 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
985 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
986 | md = tio->io->md; |
987 | dec_pending(tio->io, r); | |
988 | /* | |
989 | * Store bio_set for cleanup. | |
990 | */ | |
991 | clone->bi_private = md->bs; | |
1da177e4 | 992 | bio_put(clone); |
9faf400f | 993 | free_tio(md, tio); |
45cbcd79 KU |
994 | } else if (r) { |
995 | DMWARN("unimplemented target map return value: %d", r); | |
996 | BUG(); | |
1da177e4 LT |
997 | } |
998 | } | |
999 | ||
1000 | struct clone_info { | |
1001 | struct mapped_device *md; | |
1002 | struct dm_table *map; | |
1003 | struct bio *bio; | |
1004 | struct dm_io *io; | |
1005 | sector_t sector; | |
1006 | sector_t sector_count; | |
1007 | unsigned short idx; | |
1008 | }; | |
1009 | ||
3676347a PO |
1010 | static void dm_bio_destructor(struct bio *bio) |
1011 | { | |
9faf400f SB |
1012 | struct bio_set *bs = bio->bi_private; |
1013 | ||
1014 | bio_free(bio, bs); | |
3676347a PO |
1015 | } |
1016 | ||
1da177e4 LT |
1017 | /* |
1018 | * Creates a little bio that is just does part of a bvec. | |
1019 | */ | |
1020 | static struct bio *split_bvec(struct bio *bio, sector_t sector, | |
1021 | unsigned short idx, unsigned int offset, | |
9faf400f | 1022 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1023 | { |
1024 | struct bio *clone; | |
1025 | struct bio_vec *bv = bio->bi_io_vec + idx; | |
1026 | ||
9faf400f | 1027 | clone = bio_alloc_bioset(GFP_NOIO, 1, bs); |
3676347a | 1028 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1029 | *clone->bi_io_vec = *bv; |
1030 | ||
1031 | clone->bi_sector = sector; | |
1032 | clone->bi_bdev = bio->bi_bdev; | |
af7e466a | 1033 | clone->bi_rw = bio->bi_rw & ~(1 << BIO_RW_BARRIER); |
1da177e4 LT |
1034 | clone->bi_vcnt = 1; |
1035 | clone->bi_size = to_bytes(len); | |
1036 | clone->bi_io_vec->bv_offset = offset; | |
1037 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1038 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1039 | |
9c47008d | 1040 | if (bio_integrity(bio)) { |
7878cba9 | 1041 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1042 | bio_integrity_trim(clone, |
1043 | bio_sector_offset(bio, idx, offset), len); | |
1044 | } | |
1045 | ||
1da177e4 LT |
1046 | return clone; |
1047 | } | |
1048 | ||
1049 | /* | |
1050 | * Creates a bio that consists of range of complete bvecs. | |
1051 | */ | |
1052 | static struct bio *clone_bio(struct bio *bio, sector_t sector, | |
1053 | unsigned short idx, unsigned short bv_count, | |
9faf400f | 1054 | unsigned int len, struct bio_set *bs) |
1da177e4 LT |
1055 | { |
1056 | struct bio *clone; | |
1057 | ||
9faf400f SB |
1058 | clone = bio_alloc_bioset(GFP_NOIO, bio->bi_max_vecs, bs); |
1059 | __bio_clone(clone, bio); | |
af7e466a | 1060 | clone->bi_rw &= ~(1 << BIO_RW_BARRIER); |
9faf400f | 1061 | clone->bi_destructor = dm_bio_destructor; |
1da177e4 LT |
1062 | clone->bi_sector = sector; |
1063 | clone->bi_idx = idx; | |
1064 | clone->bi_vcnt = idx + bv_count; | |
1065 | clone->bi_size = to_bytes(len); | |
1066 | clone->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1067 | ||
9c47008d | 1068 | if (bio_integrity(bio)) { |
7878cba9 | 1069 | bio_integrity_clone(clone, bio, GFP_NOIO, bs); |
9c47008d MP |
1070 | |
1071 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1072 | bio_integrity_trim(clone, | |
1073 | bio_sector_offset(bio, idx, 0), len); | |
1074 | } | |
1075 | ||
1da177e4 LT |
1076 | return clone; |
1077 | } | |
1078 | ||
9015df24 AK |
1079 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
1080 | struct dm_target *ti) | |
f9ab94ce | 1081 | { |
9015df24 | 1082 | struct dm_target_io *tio = mempool_alloc(ci->md->tio_pool, GFP_NOIO); |
f9ab94ce MP |
1083 | |
1084 | tio->io = ci->io; | |
1085 | tio->ti = ti; | |
f9ab94ce | 1086 | memset(&tio->info, 0, sizeof(tio->info)); |
9015df24 AK |
1087 | |
1088 | return tio; | |
1089 | } | |
1090 | ||
1091 | static void __flush_target(struct clone_info *ci, struct dm_target *ti, | |
1092 | unsigned flush_nr) | |
1093 | { | |
1094 | struct dm_target_io *tio = alloc_tio(ci, ti); | |
1095 | struct bio *clone; | |
1096 | ||
f9ab94ce MP |
1097 | tio->info.flush_request = flush_nr; |
1098 | ||
1099 | clone = bio_alloc_bioset(GFP_NOIO, 0, ci->md->bs); | |
1100 | __bio_clone(clone, ci->bio); | |
1101 | clone->bi_destructor = dm_bio_destructor; | |
1102 | ||
1103 | __map_bio(ti, clone, tio); | |
1104 | } | |
1105 | ||
1106 | static int __clone_and_map_empty_barrier(struct clone_info *ci) | |
1107 | { | |
1108 | unsigned target_nr = 0, flush_nr; | |
1109 | struct dm_target *ti; | |
1110 | ||
1111 | while ((ti = dm_table_get_target(ci->map, target_nr++))) | |
1112 | for (flush_nr = 0; flush_nr < ti->num_flush_requests; | |
1113 | flush_nr++) | |
1114 | __flush_target(ci, ti, flush_nr); | |
1115 | ||
1116 | ci->sector_count = 0; | |
1117 | ||
1118 | return 0; | |
1119 | } | |
1120 | ||
512875bd | 1121 | static int __clone_and_map(struct clone_info *ci) |
1da177e4 LT |
1122 | { |
1123 | struct bio *clone, *bio = ci->bio; | |
512875bd JN |
1124 | struct dm_target *ti; |
1125 | sector_t len = 0, max; | |
028867ac | 1126 | struct dm_target_io *tio; |
1da177e4 | 1127 | |
f9ab94ce MP |
1128 | if (unlikely(bio_empty_barrier(bio))) |
1129 | return __clone_and_map_empty_barrier(ci); | |
1130 | ||
512875bd JN |
1131 | ti = dm_table_find_target(ci->map, ci->sector); |
1132 | if (!dm_target_is_valid(ti)) | |
1133 | return -EIO; | |
1134 | ||
1135 | max = max_io_len(ci->md, ci->sector, ti); | |
1136 | ||
1da177e4 LT |
1137 | /* |
1138 | * Allocate a target io object. | |
1139 | */ | |
9015df24 | 1140 | tio = alloc_tio(ci, ti); |
1da177e4 LT |
1141 | |
1142 | if (ci->sector_count <= max) { | |
1143 | /* | |
1144 | * Optimise for the simple case where we can do all of | |
1145 | * the remaining io with a single clone. | |
1146 | */ | |
1147 | clone = clone_bio(bio, ci->sector, ci->idx, | |
9faf400f SB |
1148 | bio->bi_vcnt - ci->idx, ci->sector_count, |
1149 | ci->md->bs); | |
1da177e4 LT |
1150 | __map_bio(ti, clone, tio); |
1151 | ci->sector_count = 0; | |
1152 | ||
1153 | } else if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1154 | /* | |
1155 | * There are some bvecs that don't span targets. | |
1156 | * Do as many of these as possible. | |
1157 | */ | |
1158 | int i; | |
1159 | sector_t remaining = max; | |
1160 | sector_t bv_len; | |
1161 | ||
1162 | for (i = ci->idx; remaining && (i < bio->bi_vcnt); i++) { | |
1163 | bv_len = to_sector(bio->bi_io_vec[i].bv_len); | |
1164 | ||
1165 | if (bv_len > remaining) | |
1166 | break; | |
1167 | ||
1168 | remaining -= bv_len; | |
1169 | len += bv_len; | |
1170 | } | |
1171 | ||
9faf400f SB |
1172 | clone = clone_bio(bio, ci->sector, ci->idx, i - ci->idx, len, |
1173 | ci->md->bs); | |
1da177e4 LT |
1174 | __map_bio(ti, clone, tio); |
1175 | ||
1176 | ci->sector += len; | |
1177 | ci->sector_count -= len; | |
1178 | ci->idx = i; | |
1179 | ||
1180 | } else { | |
1181 | /* | |
d2044a94 | 1182 | * Handle a bvec that must be split between two or more targets. |
1da177e4 LT |
1183 | */ |
1184 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
d2044a94 AK |
1185 | sector_t remaining = to_sector(bv->bv_len); |
1186 | unsigned int offset = 0; | |
1da177e4 | 1187 | |
d2044a94 AK |
1188 | do { |
1189 | if (offset) { | |
1190 | ti = dm_table_find_target(ci->map, ci->sector); | |
512875bd JN |
1191 | if (!dm_target_is_valid(ti)) |
1192 | return -EIO; | |
1193 | ||
d2044a94 | 1194 | max = max_io_len(ci->md, ci->sector, ti); |
1da177e4 | 1195 | |
9015df24 | 1196 | tio = alloc_tio(ci, ti); |
d2044a94 AK |
1197 | } |
1198 | ||
1199 | len = min(remaining, max); | |
1200 | ||
1201 | clone = split_bvec(bio, ci->sector, ci->idx, | |
9faf400f SB |
1202 | bv->bv_offset + offset, len, |
1203 | ci->md->bs); | |
d2044a94 AK |
1204 | |
1205 | __map_bio(ti, clone, tio); | |
1206 | ||
1207 | ci->sector += len; | |
1208 | ci->sector_count -= len; | |
1209 | offset += to_bytes(len); | |
1210 | } while (remaining -= len); | |
1da177e4 | 1211 | |
1da177e4 LT |
1212 | ci->idx++; |
1213 | } | |
512875bd JN |
1214 | |
1215 | return 0; | |
1da177e4 LT |
1216 | } |
1217 | ||
1218 | /* | |
8a53c28d | 1219 | * Split the bio into several clones and submit it to targets. |
1da177e4 | 1220 | */ |
f0b9a450 | 1221 | static void __split_and_process_bio(struct mapped_device *md, struct bio *bio) |
1da177e4 LT |
1222 | { |
1223 | struct clone_info ci; | |
512875bd | 1224 | int error = 0; |
1da177e4 LT |
1225 | |
1226 | ci.map = dm_get_table(md); | |
f0b9a450 | 1227 | if (unlikely(!ci.map)) { |
1f98a13f | 1228 | if (!bio_rw_flagged(bio, BIO_RW_BARRIER)) |
af7e466a MP |
1229 | bio_io_error(bio); |
1230 | else | |
5aa2781d MP |
1231 | if (!md->barrier_error) |
1232 | md->barrier_error = -EIO; | |
f0b9a450 MP |
1233 | return; |
1234 | } | |
692d0eb9 | 1235 | |
1da177e4 LT |
1236 | ci.md = md; |
1237 | ci.bio = bio; | |
1238 | ci.io = alloc_io(md); | |
1239 | ci.io->error = 0; | |
1240 | atomic_set(&ci.io->io_count, 1); | |
1241 | ci.io->bio = bio; | |
1242 | ci.io->md = md; | |
f88fb981 | 1243 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 LT |
1244 | ci.sector = bio->bi_sector; |
1245 | ci.sector_count = bio_sectors(bio); | |
f9ab94ce MP |
1246 | if (unlikely(bio_empty_barrier(bio))) |
1247 | ci.sector_count = 1; | |
1da177e4 LT |
1248 | ci.idx = bio->bi_idx; |
1249 | ||
3eaf840e | 1250 | start_io_acct(ci.io); |
512875bd JN |
1251 | while (ci.sector_count && !error) |
1252 | error = __clone_and_map(&ci); | |
1da177e4 LT |
1253 | |
1254 | /* drop the extra reference count */ | |
512875bd | 1255 | dec_pending(ci.io, error); |
1da177e4 LT |
1256 | dm_table_put(ci.map); |
1257 | } | |
1258 | /*----------------------------------------------------------------- | |
1259 | * CRUD END | |
1260 | *---------------------------------------------------------------*/ | |
1261 | ||
f6fccb12 MB |
1262 | static int dm_merge_bvec(struct request_queue *q, |
1263 | struct bvec_merge_data *bvm, | |
1264 | struct bio_vec *biovec) | |
1265 | { | |
1266 | struct mapped_device *md = q->queuedata; | |
1267 | struct dm_table *map = dm_get_table(md); | |
1268 | struct dm_target *ti; | |
1269 | sector_t max_sectors; | |
5037108a | 1270 | int max_size = 0; |
f6fccb12 MB |
1271 | |
1272 | if (unlikely(!map)) | |
5037108a | 1273 | goto out; |
f6fccb12 MB |
1274 | |
1275 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac MP |
1276 | if (!dm_target_is_valid(ti)) |
1277 | goto out_table; | |
f6fccb12 MB |
1278 | |
1279 | /* | |
1280 | * Find maximum amount of I/O that won't need splitting | |
1281 | */ | |
1282 | max_sectors = min(max_io_len(md, bvm->bi_sector, ti), | |
1283 | (sector_t) BIO_MAX_SECTORS); | |
1284 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1285 | if (max_size < 0) | |
1286 | max_size = 0; | |
1287 | ||
1288 | /* | |
1289 | * merge_bvec_fn() returns number of bytes | |
1290 | * it can accept at this offset | |
1291 | * max is precomputed maximal io size | |
1292 | */ | |
1293 | if (max_size && ti->type->merge) | |
1294 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1295 | /* |
1296 | * If the target doesn't support merge method and some of the devices | |
1297 | * provided their merge_bvec method (we know this by looking at | |
1298 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1299 | * entries. So always set max_size to 0, and the code below allows | |
1300 | * just one page. | |
1301 | */ | |
1302 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1303 | ||
1304 | max_size = 0; | |
f6fccb12 | 1305 | |
b01cd5ac | 1306 | out_table: |
5037108a MP |
1307 | dm_table_put(map); |
1308 | ||
1309 | out: | |
f6fccb12 MB |
1310 | /* |
1311 | * Always allow an entire first page | |
1312 | */ | |
1313 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1314 | max_size = biovec->bv_len; | |
1315 | ||
f6fccb12 MB |
1316 | return max_size; |
1317 | } | |
1318 | ||
1da177e4 LT |
1319 | /* |
1320 | * The request function that just remaps the bio built up by | |
1321 | * dm_merge_bvec. | |
1322 | */ | |
cec47e3d | 1323 | static int _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1324 | { |
12f03a49 | 1325 | int rw = bio_data_dir(bio); |
1da177e4 | 1326 | struct mapped_device *md = q->queuedata; |
c9959059 | 1327 | int cpu; |
1da177e4 | 1328 | |
2ca3310e | 1329 | down_read(&md->io_lock); |
1da177e4 | 1330 | |
074a7aca TH |
1331 | cpu = part_stat_lock(); |
1332 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1333 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1334 | part_stat_unlock(); | |
12f03a49 | 1335 | |
1da177e4 | 1336 | /* |
1eb787ec AK |
1337 | * If we're suspended or the thread is processing barriers |
1338 | * we have to queue this io for later. | |
1da177e4 | 1339 | */ |
af7e466a | 1340 | if (unlikely(test_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags)) || |
1f98a13f | 1341 | unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) { |
2ca3310e | 1342 | up_read(&md->io_lock); |
1da177e4 | 1343 | |
54d9a1b4 AK |
1344 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) && |
1345 | bio_rw(bio) == READA) { | |
1346 | bio_io_error(bio); | |
1347 | return 0; | |
1348 | } | |
1da177e4 | 1349 | |
92c63902 | 1350 | queue_io(md, bio); |
1da177e4 | 1351 | |
92c63902 | 1352 | return 0; |
1da177e4 LT |
1353 | } |
1354 | ||
f0b9a450 | 1355 | __split_and_process_bio(md, bio); |
2ca3310e | 1356 | up_read(&md->io_lock); |
f0b9a450 | 1357 | return 0; |
1da177e4 LT |
1358 | } |
1359 | ||
cec47e3d KU |
1360 | static int dm_make_request(struct request_queue *q, struct bio *bio) |
1361 | { | |
1362 | struct mapped_device *md = q->queuedata; | |
1363 | ||
1f98a13f | 1364 | if (unlikely(bio_rw_flagged(bio, BIO_RW_BARRIER))) { |
cec47e3d KU |
1365 | bio_endio(bio, -EOPNOTSUPP); |
1366 | return 0; | |
1367 | } | |
1368 | ||
1369 | return md->saved_make_request_fn(q, bio); /* call __make_request() */ | |
1370 | } | |
1371 | ||
1372 | static int dm_request_based(struct mapped_device *md) | |
1373 | { | |
1374 | return blk_queue_stackable(md->queue); | |
1375 | } | |
1376 | ||
1377 | static int dm_request(struct request_queue *q, struct bio *bio) | |
1378 | { | |
1379 | struct mapped_device *md = q->queuedata; | |
1380 | ||
1381 | if (dm_request_based(md)) | |
1382 | return dm_make_request(q, bio); | |
1383 | ||
1384 | return _dm_request(q, bio); | |
1385 | } | |
1386 | ||
1387 | void dm_dispatch_request(struct request *rq) | |
1388 | { | |
1389 | int r; | |
1390 | ||
1391 | if (blk_queue_io_stat(rq->q)) | |
1392 | rq->cmd_flags |= REQ_IO_STAT; | |
1393 | ||
1394 | rq->start_time = jiffies; | |
1395 | r = blk_insert_cloned_request(rq->q, rq); | |
1396 | if (r) | |
1397 | dm_complete_request(rq, r); | |
1398 | } | |
1399 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1400 | ||
1401 | static void dm_rq_bio_destructor(struct bio *bio) | |
1402 | { | |
1403 | struct dm_rq_clone_bio_info *info = bio->bi_private; | |
1404 | struct mapped_device *md = info->tio->md; | |
1405 | ||
1406 | free_bio_info(info); | |
1407 | bio_free(bio, md->bs); | |
1408 | } | |
1409 | ||
1410 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, | |
1411 | void *data) | |
1412 | { | |
1413 | struct dm_rq_target_io *tio = data; | |
1414 | struct mapped_device *md = tio->md; | |
1415 | struct dm_rq_clone_bio_info *info = alloc_bio_info(md); | |
1416 | ||
1417 | if (!info) | |
1418 | return -ENOMEM; | |
1419 | ||
1420 | info->orig = bio_orig; | |
1421 | info->tio = tio; | |
1422 | bio->bi_end_io = end_clone_bio; | |
1423 | bio->bi_private = info; | |
1424 | bio->bi_destructor = dm_rq_bio_destructor; | |
1425 | ||
1426 | return 0; | |
1427 | } | |
1428 | ||
1429 | static int setup_clone(struct request *clone, struct request *rq, | |
1430 | struct dm_rq_target_io *tio) | |
1431 | { | |
1432 | int r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, | |
1433 | dm_rq_bio_constructor, tio); | |
1434 | ||
1435 | if (r) | |
1436 | return r; | |
1437 | ||
1438 | clone->cmd = rq->cmd; | |
1439 | clone->cmd_len = rq->cmd_len; | |
1440 | clone->sense = rq->sense; | |
1441 | clone->buffer = rq->buffer; | |
1442 | clone->end_io = end_clone_request; | |
1443 | clone->end_io_data = tio; | |
1444 | ||
1445 | return 0; | |
1446 | } | |
1447 | ||
6facdaff KU |
1448 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1449 | gfp_t gfp_mask) | |
1450 | { | |
1451 | struct request *clone; | |
1452 | struct dm_rq_target_io *tio; | |
1453 | ||
1454 | tio = alloc_rq_tio(md, gfp_mask); | |
1455 | if (!tio) | |
1456 | return NULL; | |
1457 | ||
1458 | tio->md = md; | |
1459 | tio->ti = NULL; | |
1460 | tio->orig = rq; | |
1461 | tio->error = 0; | |
1462 | memset(&tio->info, 0, sizeof(tio->info)); | |
1463 | ||
1464 | clone = &tio->clone; | |
1465 | if (setup_clone(clone, rq, tio)) { | |
1466 | /* -ENOMEM */ | |
1467 | free_rq_tio(tio); | |
1468 | return NULL; | |
1469 | } | |
1470 | ||
1471 | return clone; | |
1472 | } | |
1473 | ||
cec47e3d KU |
1474 | static int dm_rq_flush_suspending(struct mapped_device *md) |
1475 | { | |
1476 | return !md->suspend_rq.special; | |
1477 | } | |
1478 | ||
1479 | /* | |
1480 | * Called with the queue lock held. | |
1481 | */ | |
1482 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1483 | { | |
1484 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1485 | struct request *clone; |
1486 | ||
1487 | if (unlikely(rq == &md->suspend_rq)) { | |
1488 | if (dm_rq_flush_suspending(md)) | |
1489 | return BLKPREP_OK; | |
1490 | else | |
1491 | /* The flush suspend was interrupted */ | |
1492 | return BLKPREP_KILL; | |
1493 | } | |
1494 | ||
1495 | if (unlikely(rq->special)) { | |
1496 | DMWARN("Already has something in rq->special."); | |
1497 | return BLKPREP_KILL; | |
1498 | } | |
1499 | ||
6facdaff KU |
1500 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1501 | if (!clone) | |
cec47e3d | 1502 | return BLKPREP_DEFER; |
cec47e3d KU |
1503 | |
1504 | rq->special = clone; | |
1505 | rq->cmd_flags |= REQ_DONTPREP; | |
1506 | ||
1507 | return BLKPREP_OK; | |
1508 | } | |
1509 | ||
598de409 | 1510 | static void map_request(struct dm_target *ti, struct request *clone, |
cec47e3d KU |
1511 | struct mapped_device *md) |
1512 | { | |
1513 | int r; | |
cec47e3d KU |
1514 | struct dm_rq_target_io *tio = clone->end_io_data; |
1515 | ||
1516 | /* | |
1517 | * Hold the md reference here for the in-flight I/O. | |
1518 | * We can't rely on the reference count by device opener, | |
1519 | * because the device may be closed during the request completion | |
1520 | * when all bios are completed. | |
1521 | * See the comment in rq_completed() too. | |
1522 | */ | |
1523 | dm_get(md); | |
1524 | ||
1525 | tio->ti = ti; | |
1526 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1527 | switch (r) { | |
1528 | case DM_MAPIO_SUBMITTED: | |
1529 | /* The target has taken the I/O to submit by itself later */ | |
1530 | break; | |
1531 | case DM_MAPIO_REMAPPED: | |
1532 | /* The target has remapped the I/O so dispatch it */ | |
1533 | dm_dispatch_request(clone); | |
1534 | break; | |
1535 | case DM_MAPIO_REQUEUE: | |
1536 | /* The target wants to requeue the I/O */ | |
1537 | dm_requeue_unmapped_request(clone); | |
1538 | break; | |
1539 | default: | |
1540 | if (r > 0) { | |
1541 | DMWARN("unimplemented target map return value: %d", r); | |
1542 | BUG(); | |
1543 | } | |
1544 | ||
1545 | /* The target wants to complete the I/O */ | |
1546 | dm_kill_unmapped_request(clone, r); | |
1547 | break; | |
1548 | } | |
1549 | } | |
1550 | ||
1551 | /* | |
1552 | * q->request_fn for request-based dm. | |
1553 | * Called with the queue lock held. | |
1554 | */ | |
1555 | static void dm_request_fn(struct request_queue *q) | |
1556 | { | |
1557 | struct mapped_device *md = q->queuedata; | |
1558 | struct dm_table *map = dm_get_table(md); | |
1559 | struct dm_target *ti; | |
1560 | struct request *rq; | |
1561 | ||
1562 | /* | |
1563 | * For noflush suspend, check blk_queue_stopped() to immediately | |
1564 | * quit I/O dispatching. | |
1565 | */ | |
1566 | while (!blk_queue_plugged(q) && !blk_queue_stopped(q)) { | |
1567 | rq = blk_peek_request(q); | |
1568 | if (!rq) | |
1569 | goto plug_and_out; | |
1570 | ||
1571 | if (unlikely(rq == &md->suspend_rq)) { /* Flush suspend maker */ | |
1572 | if (queue_in_flight(q)) | |
1573 | /* Not quiet yet. Wait more */ | |
1574 | goto plug_and_out; | |
1575 | ||
1576 | /* This device should be quiet now */ | |
1577 | __stop_queue(q); | |
1578 | blk_start_request(rq); | |
1579 | __blk_end_request_all(rq, 0); | |
1580 | wake_up(&md->wait); | |
1581 | goto out; | |
1582 | } | |
1583 | ||
1584 | ti = dm_table_find_target(map, blk_rq_pos(rq)); | |
1585 | if (ti->type->busy && ti->type->busy(ti)) | |
1586 | goto plug_and_out; | |
1587 | ||
1588 | blk_start_request(rq); | |
1589 | spin_unlock(q->queue_lock); | |
598de409 | 1590 | map_request(ti, rq->special, md); |
cec47e3d KU |
1591 | spin_lock_irq(q->queue_lock); |
1592 | } | |
1593 | ||
1594 | goto out; | |
1595 | ||
1596 | plug_and_out: | |
1597 | if (!elv_queue_empty(q)) | |
1598 | /* Some requests still remain, retry later */ | |
1599 | blk_plug_device(q); | |
1600 | ||
1601 | out: | |
1602 | dm_table_put(map); | |
1603 | ||
1604 | return; | |
1605 | } | |
1606 | ||
1607 | int dm_underlying_device_busy(struct request_queue *q) | |
1608 | { | |
1609 | return blk_lld_busy(q); | |
1610 | } | |
1611 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1612 | ||
1613 | static int dm_lld_busy(struct request_queue *q) | |
1614 | { | |
1615 | int r; | |
1616 | struct mapped_device *md = q->queuedata; | |
1617 | struct dm_table *map = dm_get_table(md); | |
1618 | ||
1619 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1620 | r = 1; | |
1621 | else | |
1622 | r = dm_table_any_busy_target(map); | |
1623 | ||
1624 | dm_table_put(map); | |
1625 | ||
1626 | return r; | |
1627 | } | |
1628 | ||
165125e1 | 1629 | static void dm_unplug_all(struct request_queue *q) |
1da177e4 LT |
1630 | { |
1631 | struct mapped_device *md = q->queuedata; | |
1632 | struct dm_table *map = dm_get_table(md); | |
1633 | ||
1634 | if (map) { | |
cec47e3d KU |
1635 | if (dm_request_based(md)) |
1636 | generic_unplug_device(q); | |
1637 | ||
1da177e4 LT |
1638 | dm_table_unplug_all(map); |
1639 | dm_table_put(map); | |
1640 | } | |
1641 | } | |
1642 | ||
1643 | static int dm_any_congested(void *congested_data, int bdi_bits) | |
1644 | { | |
8a57dfc6 CS |
1645 | int r = bdi_bits; |
1646 | struct mapped_device *md = congested_data; | |
1647 | struct dm_table *map; | |
1da177e4 | 1648 | |
1eb787ec | 1649 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
8a57dfc6 CS |
1650 | map = dm_get_table(md); |
1651 | if (map) { | |
cec47e3d KU |
1652 | /* |
1653 | * Request-based dm cares about only own queue for | |
1654 | * the query about congestion status of request_queue | |
1655 | */ | |
1656 | if (dm_request_based(md)) | |
1657 | r = md->queue->backing_dev_info.state & | |
1658 | bdi_bits; | |
1659 | else | |
1660 | r = dm_table_any_congested(map, bdi_bits); | |
1661 | ||
8a57dfc6 CS |
1662 | dm_table_put(map); |
1663 | } | |
1664 | } | |
1665 | ||
1da177e4 LT |
1666 | return r; |
1667 | } | |
1668 | ||
1669 | /*----------------------------------------------------------------- | |
1670 | * An IDR is used to keep track of allocated minor numbers. | |
1671 | *---------------------------------------------------------------*/ | |
1da177e4 LT |
1672 | static DEFINE_IDR(_minor_idr); |
1673 | ||
2b06cfff | 1674 | static void free_minor(int minor) |
1da177e4 | 1675 | { |
f32c10b0 | 1676 | spin_lock(&_minor_lock); |
1da177e4 | 1677 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1678 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1679 | } |
1680 | ||
1681 | /* | |
1682 | * See if the device with a specific minor # is free. | |
1683 | */ | |
cf13ab8e | 1684 | static int specific_minor(int minor) |
1da177e4 LT |
1685 | { |
1686 | int r, m; | |
1687 | ||
1688 | if (minor >= (1 << MINORBITS)) | |
1689 | return -EINVAL; | |
1690 | ||
62f75c2f JM |
1691 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
1692 | if (!r) | |
1693 | return -ENOMEM; | |
1694 | ||
f32c10b0 | 1695 | spin_lock(&_minor_lock); |
1da177e4 LT |
1696 | |
1697 | if (idr_find(&_minor_idr, minor)) { | |
1698 | r = -EBUSY; | |
1699 | goto out; | |
1700 | } | |
1701 | ||
ba61fdd1 | 1702 | r = idr_get_new_above(&_minor_idr, MINOR_ALLOCED, minor, &m); |
62f75c2f | 1703 | if (r) |
1da177e4 | 1704 | goto out; |
1da177e4 LT |
1705 | |
1706 | if (m != minor) { | |
1707 | idr_remove(&_minor_idr, m); | |
1708 | r = -EBUSY; | |
1709 | goto out; | |
1710 | } | |
1711 | ||
1712 | out: | |
f32c10b0 | 1713 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1714 | return r; |
1715 | } | |
1716 | ||
cf13ab8e | 1717 | static int next_free_minor(int *minor) |
1da177e4 | 1718 | { |
2b06cfff | 1719 | int r, m; |
1da177e4 | 1720 | |
1da177e4 | 1721 | r = idr_pre_get(&_minor_idr, GFP_KERNEL); |
62f75c2f JM |
1722 | if (!r) |
1723 | return -ENOMEM; | |
1724 | ||
f32c10b0 | 1725 | spin_lock(&_minor_lock); |
1da177e4 | 1726 | |
ba61fdd1 | 1727 | r = idr_get_new(&_minor_idr, MINOR_ALLOCED, &m); |
cf13ab8e | 1728 | if (r) |
1da177e4 | 1729 | goto out; |
1da177e4 LT |
1730 | |
1731 | if (m >= (1 << MINORBITS)) { | |
1732 | idr_remove(&_minor_idr, m); | |
1733 | r = -ENOSPC; | |
1734 | goto out; | |
1735 | } | |
1736 | ||
1737 | *minor = m; | |
1738 | ||
1739 | out: | |
f32c10b0 | 1740 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1741 | return r; |
1742 | } | |
1743 | ||
83d5cde4 | 1744 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1745 | |
53d5914f MP |
1746 | static void dm_wq_work(struct work_struct *work); |
1747 | ||
1da177e4 LT |
1748 | /* |
1749 | * Allocate and initialise a blank device with a given minor. | |
1750 | */ | |
2b06cfff | 1751 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1752 | { |
1753 | int r; | |
cf13ab8e | 1754 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1755 | void *old_md; |
1da177e4 LT |
1756 | |
1757 | if (!md) { | |
1758 | DMWARN("unable to allocate device, out of memory."); | |
1759 | return NULL; | |
1760 | } | |
1761 | ||
10da4f79 | 1762 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1763 | goto bad_module_get; |
10da4f79 | 1764 | |
1da177e4 | 1765 | /* get a minor number for the dev */ |
2b06cfff | 1766 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1767 | r = next_free_minor(&minor); |
2b06cfff | 1768 | else |
cf13ab8e | 1769 | r = specific_minor(minor); |
1da177e4 | 1770 | if (r < 0) |
6ed7ade8 | 1771 | goto bad_minor; |
1da177e4 | 1772 | |
2ca3310e | 1773 | init_rwsem(&md->io_lock); |
e61290a4 | 1774 | mutex_init(&md->suspend_lock); |
022c2611 | 1775 | spin_lock_init(&md->deferred_lock); |
1da177e4 LT |
1776 | rwlock_init(&md->map_lock); |
1777 | atomic_set(&md->holders, 1); | |
5c6bd75d | 1778 | atomic_set(&md->open_count, 0); |
1da177e4 | 1779 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1780 | atomic_set(&md->uevent_seq, 0); |
1781 | INIT_LIST_HEAD(&md->uevent_list); | |
1782 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1783 | |
e6ee8c0b | 1784 | md->queue = blk_init_queue(dm_request_fn, NULL); |
1da177e4 | 1785 | if (!md->queue) |
6ed7ade8 | 1786 | goto bad_queue; |
1da177e4 | 1787 | |
e6ee8c0b KU |
1788 | /* |
1789 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1790 | * devices. The type of this dm device has not been decided yet, | |
1791 | * although we initialized the queue using blk_init_queue(). | |
1792 | * The type is decided at the first table loading time. | |
1793 | * To prevent problematic device stacking, clear the queue flag | |
1794 | * for request stacking support until then. | |
1795 | * | |
1796 | * This queue is new, so no concurrency on the queue_flags. | |
1797 | */ | |
1798 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1799 | md->saved_make_request_fn = md->queue->make_request_fn; | |
1da177e4 LT |
1800 | md->queue->queuedata = md; |
1801 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1802 | md->queue->backing_dev_info.congested_data = md; | |
1803 | blk_queue_make_request(md->queue, dm_request); | |
daef265f | 1804 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); |
1da177e4 | 1805 | md->queue->unplug_fn = dm_unplug_all; |
f6fccb12 | 1806 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
e6ee8c0b KU |
1807 | blk_queue_softirq_done(md->queue, dm_softirq_done); |
1808 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
1809 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
9faf400f | 1810 | |
1da177e4 LT |
1811 | md->disk = alloc_disk(1); |
1812 | if (!md->disk) | |
6ed7ade8 | 1813 | goto bad_disk; |
1da177e4 | 1814 | |
316d315b NK |
1815 | atomic_set(&md->pending[0], 0); |
1816 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1817 | init_waitqueue_head(&md->wait); |
53d5914f | 1818 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1819 | init_waitqueue_head(&md->eventq); |
1820 | ||
1da177e4 LT |
1821 | md->disk->major = _major; |
1822 | md->disk->first_minor = minor; | |
1823 | md->disk->fops = &dm_blk_dops; | |
1824 | md->disk->queue = md->queue; | |
1825 | md->disk->private_data = md; | |
1826 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1827 | add_disk(md->disk); | |
7e51f257 | 1828 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1829 | |
304f3f6a MB |
1830 | md->wq = create_singlethread_workqueue("kdmflush"); |
1831 | if (!md->wq) | |
1832 | goto bad_thread; | |
1833 | ||
32a926da MP |
1834 | md->bdev = bdget_disk(md->disk, 0); |
1835 | if (!md->bdev) | |
1836 | goto bad_bdev; | |
1837 | ||
ba61fdd1 | 1838 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1839 | spin_lock(&_minor_lock); |
ba61fdd1 | 1840 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1841 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1842 | |
1843 | BUG_ON(old_md != MINOR_ALLOCED); | |
1844 | ||
1da177e4 LT |
1845 | return md; |
1846 | ||
32a926da MP |
1847 | bad_bdev: |
1848 | destroy_workqueue(md->wq); | |
304f3f6a | 1849 | bad_thread: |
03022c54 | 1850 | del_gendisk(md->disk); |
304f3f6a | 1851 | put_disk(md->disk); |
6ed7ade8 | 1852 | bad_disk: |
1312f40e | 1853 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1854 | bad_queue: |
1da177e4 | 1855 | free_minor(minor); |
6ed7ade8 | 1856 | bad_minor: |
10da4f79 | 1857 | module_put(THIS_MODULE); |
6ed7ade8 | 1858 | bad_module_get: |
1da177e4 LT |
1859 | kfree(md); |
1860 | return NULL; | |
1861 | } | |
1862 | ||
ae9da83f JN |
1863 | static void unlock_fs(struct mapped_device *md); |
1864 | ||
1da177e4 LT |
1865 | static void free_dev(struct mapped_device *md) |
1866 | { | |
f331c029 | 1867 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1868 | |
32a926da MP |
1869 | unlock_fs(md); |
1870 | bdput(md->bdev); | |
304f3f6a | 1871 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1872 | if (md->tio_pool) |
1873 | mempool_destroy(md->tio_pool); | |
1874 | if (md->io_pool) | |
1875 | mempool_destroy(md->io_pool); | |
1876 | if (md->bs) | |
1877 | bioset_free(md->bs); | |
9c47008d | 1878 | blk_integrity_unregister(md->disk); |
1da177e4 | 1879 | del_gendisk(md->disk); |
63d94e48 | 1880 | free_minor(minor); |
fba9f90e JM |
1881 | |
1882 | spin_lock(&_minor_lock); | |
1883 | md->disk->private_data = NULL; | |
1884 | spin_unlock(&_minor_lock); | |
1885 | ||
1da177e4 | 1886 | put_disk(md->disk); |
1312f40e | 1887 | blk_cleanup_queue(md->queue); |
10da4f79 | 1888 | module_put(THIS_MODULE); |
1da177e4 LT |
1889 | kfree(md); |
1890 | } | |
1891 | ||
e6ee8c0b KU |
1892 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
1893 | { | |
1894 | struct dm_md_mempools *p; | |
1895 | ||
1896 | if (md->io_pool && md->tio_pool && md->bs) | |
1897 | /* the md already has necessary mempools */ | |
1898 | goto out; | |
1899 | ||
1900 | p = dm_table_get_md_mempools(t); | |
1901 | BUG_ON(!p || md->io_pool || md->tio_pool || md->bs); | |
1902 | ||
1903 | md->io_pool = p->io_pool; | |
1904 | p->io_pool = NULL; | |
1905 | md->tio_pool = p->tio_pool; | |
1906 | p->tio_pool = NULL; | |
1907 | md->bs = p->bs; | |
1908 | p->bs = NULL; | |
1909 | ||
1910 | out: | |
1911 | /* mempool bind completed, now no need any mempools in the table */ | |
1912 | dm_table_free_md_mempools(t); | |
1913 | } | |
1914 | ||
1da177e4 LT |
1915 | /* |
1916 | * Bind a table to the device. | |
1917 | */ | |
1918 | static void event_callback(void *context) | |
1919 | { | |
7a8c3d3b MA |
1920 | unsigned long flags; |
1921 | LIST_HEAD(uevents); | |
1da177e4 LT |
1922 | struct mapped_device *md = (struct mapped_device *) context; |
1923 | ||
7a8c3d3b MA |
1924 | spin_lock_irqsave(&md->uevent_lock, flags); |
1925 | list_splice_init(&md->uevent_list, &uevents); | |
1926 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
1927 | ||
ed9e1982 | 1928 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 1929 | |
1da177e4 LT |
1930 | atomic_inc(&md->event_nr); |
1931 | wake_up(&md->eventq); | |
1932 | } | |
1933 | ||
4e90188b | 1934 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 1935 | { |
4e90188b | 1936 | set_capacity(md->disk, size); |
1da177e4 | 1937 | |
db8fef4f MP |
1938 | mutex_lock(&md->bdev->bd_inode->i_mutex); |
1939 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); | |
1940 | mutex_unlock(&md->bdev->bd_inode->i_mutex); | |
1da177e4 LT |
1941 | } |
1942 | ||
754c5fc7 MS |
1943 | static int __bind(struct mapped_device *md, struct dm_table *t, |
1944 | struct queue_limits *limits) | |
1da177e4 | 1945 | { |
165125e1 | 1946 | struct request_queue *q = md->queue; |
1da177e4 | 1947 | sector_t size; |
523d9297 | 1948 | unsigned long flags; |
1da177e4 LT |
1949 | |
1950 | size = dm_table_get_size(t); | |
3ac51e74 DW |
1951 | |
1952 | /* | |
1953 | * Wipe any geometry if the size of the table changed. | |
1954 | */ | |
1955 | if (size != get_capacity(md->disk)) | |
1956 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
1957 | ||
32a926da | 1958 | __set_size(md, size); |
d5816876 MP |
1959 | |
1960 | if (!size) { | |
1961 | dm_table_destroy(t); | |
1da177e4 | 1962 | return 0; |
d5816876 | 1963 | } |
1da177e4 | 1964 | |
2ca3310e AK |
1965 | dm_table_event_callback(t, event_callback, md); |
1966 | ||
e6ee8c0b KU |
1967 | /* |
1968 | * The queue hasn't been stopped yet, if the old table type wasn't | |
1969 | * for request-based during suspension. So stop it to prevent | |
1970 | * I/O mapping before resume. | |
1971 | * This must be done before setting the queue restrictions, | |
1972 | * because request-based dm may be run just after the setting. | |
1973 | */ | |
1974 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
1975 | stop_queue(q); | |
1976 | ||
1977 | __bind_mempools(md, t); | |
1978 | ||
523d9297 | 1979 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 1980 | md->map = t; |
754c5fc7 | 1981 | dm_table_set_restrictions(t, q, limits); |
523d9297 | 1982 | write_unlock_irqrestore(&md->map_lock, flags); |
1da177e4 | 1983 | |
1da177e4 LT |
1984 | return 0; |
1985 | } | |
1986 | ||
1987 | static void __unbind(struct mapped_device *md) | |
1988 | { | |
1989 | struct dm_table *map = md->map; | |
523d9297 | 1990 | unsigned long flags; |
1da177e4 LT |
1991 | |
1992 | if (!map) | |
1993 | return; | |
1994 | ||
1995 | dm_table_event_callback(map, NULL, NULL); | |
523d9297 | 1996 | write_lock_irqsave(&md->map_lock, flags); |
1da177e4 | 1997 | md->map = NULL; |
523d9297 | 1998 | write_unlock_irqrestore(&md->map_lock, flags); |
d5816876 | 1999 | dm_table_destroy(map); |
1da177e4 LT |
2000 | } |
2001 | ||
2002 | /* | |
2003 | * Constructor for a new device. | |
2004 | */ | |
2b06cfff | 2005 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2006 | { |
2007 | struct mapped_device *md; | |
2008 | ||
2b06cfff | 2009 | md = alloc_dev(minor); |
1da177e4 LT |
2010 | if (!md) |
2011 | return -ENXIO; | |
2012 | ||
784aae73 MB |
2013 | dm_sysfs_init(md); |
2014 | ||
1da177e4 LT |
2015 | *result = md; |
2016 | return 0; | |
2017 | } | |
2018 | ||
637842cf | 2019 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2020 | { |
2021 | struct mapped_device *md; | |
1da177e4 LT |
2022 | unsigned minor = MINOR(dev); |
2023 | ||
2024 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2025 | return NULL; | |
2026 | ||
f32c10b0 | 2027 | spin_lock(&_minor_lock); |
1da177e4 LT |
2028 | |
2029 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2030 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2031 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
17b2f66f | 2032 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2033 | md = NULL; |
fba9f90e JM |
2034 | goto out; |
2035 | } | |
1da177e4 | 2036 | |
fba9f90e | 2037 | out: |
f32c10b0 | 2038 | spin_unlock(&_minor_lock); |
1da177e4 | 2039 | |
637842cf DT |
2040 | return md; |
2041 | } | |
2042 | ||
d229a958 DT |
2043 | struct mapped_device *dm_get_md(dev_t dev) |
2044 | { | |
2045 | struct mapped_device *md = dm_find_md(dev); | |
2046 | ||
2047 | if (md) | |
2048 | dm_get(md); | |
2049 | ||
2050 | return md; | |
2051 | } | |
2052 | ||
9ade92a9 | 2053 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2054 | { |
9ade92a9 | 2055 | return md->interface_ptr; |
1da177e4 LT |
2056 | } |
2057 | ||
2058 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2059 | { | |
2060 | md->interface_ptr = ptr; | |
2061 | } | |
2062 | ||
2063 | void dm_get(struct mapped_device *md) | |
2064 | { | |
2065 | atomic_inc(&md->holders); | |
2066 | } | |
2067 | ||
72d94861 AK |
2068 | const char *dm_device_name(struct mapped_device *md) |
2069 | { | |
2070 | return md->name; | |
2071 | } | |
2072 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2073 | ||
1da177e4 LT |
2074 | void dm_put(struct mapped_device *md) |
2075 | { | |
1134e5ae | 2076 | struct dm_table *map; |
1da177e4 | 2077 | |
fba9f90e JM |
2078 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
2079 | ||
f32c10b0 | 2080 | if (atomic_dec_and_lock(&md->holders, &_minor_lock)) { |
1134e5ae | 2081 | map = dm_get_table(md); |
f331c029 TH |
2082 | idr_replace(&_minor_idr, MINOR_ALLOCED, |
2083 | MINOR(disk_devt(dm_disk(md)))); | |
fba9f90e | 2084 | set_bit(DMF_FREEING, &md->flags); |
f32c10b0 | 2085 | spin_unlock(&_minor_lock); |
cf222b37 | 2086 | if (!dm_suspended(md)) { |
1da177e4 LT |
2087 | dm_table_presuspend_targets(map); |
2088 | dm_table_postsuspend_targets(map); | |
2089 | } | |
784aae73 | 2090 | dm_sysfs_exit(md); |
1134e5ae | 2091 | dm_table_put(map); |
a1b51e98 | 2092 | __unbind(md); |
1da177e4 LT |
2093 | free_dev(md); |
2094 | } | |
1da177e4 | 2095 | } |
79eb885c | 2096 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2097 | |
401600df | 2098 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2099 | { |
2100 | int r = 0; | |
b44ebeb0 | 2101 | DECLARE_WAITQUEUE(wait, current); |
cec47e3d KU |
2102 | struct request_queue *q = md->queue; |
2103 | unsigned long flags; | |
b44ebeb0 MP |
2104 | |
2105 | dm_unplug_all(md->queue); | |
2106 | ||
2107 | add_wait_queue(&md->wait, &wait); | |
46125c1c MB |
2108 | |
2109 | while (1) { | |
401600df | 2110 | set_current_state(interruptible); |
46125c1c MB |
2111 | |
2112 | smp_mb(); | |
cec47e3d KU |
2113 | if (dm_request_based(md)) { |
2114 | spin_lock_irqsave(q->queue_lock, flags); | |
2115 | if (!queue_in_flight(q) && blk_queue_stopped(q)) { | |
2116 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2117 | break; | |
2118 | } | |
2119 | spin_unlock_irqrestore(q->queue_lock, flags); | |
90abb8c4 | 2120 | } else if (!md_in_flight(md)) |
46125c1c MB |
2121 | break; |
2122 | ||
401600df MP |
2123 | if (interruptible == TASK_INTERRUPTIBLE && |
2124 | signal_pending(current)) { | |
46125c1c MB |
2125 | r = -EINTR; |
2126 | break; | |
2127 | } | |
2128 | ||
2129 | io_schedule(); | |
2130 | } | |
2131 | set_current_state(TASK_RUNNING); | |
2132 | ||
b44ebeb0 MP |
2133 | remove_wait_queue(&md->wait, &wait); |
2134 | ||
46125c1c MB |
2135 | return r; |
2136 | } | |
2137 | ||
531fe963 | 2138 | static void dm_flush(struct mapped_device *md) |
af7e466a MP |
2139 | { |
2140 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
52b1fd5a MP |
2141 | |
2142 | bio_init(&md->barrier_bio); | |
2143 | md->barrier_bio.bi_bdev = md->bdev; | |
2144 | md->barrier_bio.bi_rw = WRITE_BARRIER; | |
2145 | __split_and_process_bio(md, &md->barrier_bio); | |
2146 | ||
2147 | dm_wait_for_completion(md, TASK_UNINTERRUPTIBLE); | |
af7e466a MP |
2148 | } |
2149 | ||
2150 | static void process_barrier(struct mapped_device *md, struct bio *bio) | |
2151 | { | |
5aa2781d MP |
2152 | md->barrier_error = 0; |
2153 | ||
531fe963 | 2154 | dm_flush(md); |
af7e466a | 2155 | |
5aa2781d MP |
2156 | if (!bio_empty_barrier(bio)) { |
2157 | __split_and_process_bio(md, bio); | |
2158 | dm_flush(md); | |
af7e466a MP |
2159 | } |
2160 | ||
af7e466a | 2161 | if (md->barrier_error != DM_ENDIO_REQUEUE) |
531fe963 | 2162 | bio_endio(bio, md->barrier_error); |
2761e95f MP |
2163 | else { |
2164 | spin_lock_irq(&md->deferred_lock); | |
2165 | bio_list_add_head(&md->deferred, bio); | |
2166 | spin_unlock_irq(&md->deferred_lock); | |
2167 | } | |
af7e466a MP |
2168 | } |
2169 | ||
1da177e4 LT |
2170 | /* |
2171 | * Process the deferred bios | |
2172 | */ | |
ef208587 | 2173 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2174 | { |
ef208587 MP |
2175 | struct mapped_device *md = container_of(work, struct mapped_device, |
2176 | work); | |
6d6f10df | 2177 | struct bio *c; |
1da177e4 | 2178 | |
ef208587 MP |
2179 | down_write(&md->io_lock); |
2180 | ||
3b00b203 | 2181 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2182 | spin_lock_irq(&md->deferred_lock); |
2183 | c = bio_list_pop(&md->deferred); | |
2184 | spin_unlock_irq(&md->deferred_lock); | |
2185 | ||
2186 | if (!c) { | |
1eb787ec | 2187 | clear_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); |
df12ee99 AK |
2188 | break; |
2189 | } | |
022c2611 | 2190 | |
3b00b203 MP |
2191 | up_write(&md->io_lock); |
2192 | ||
e6ee8c0b KU |
2193 | if (dm_request_based(md)) |
2194 | generic_make_request(c); | |
2195 | else { | |
1f98a13f | 2196 | if (bio_rw_flagged(c, BIO_RW_BARRIER)) |
e6ee8c0b KU |
2197 | process_barrier(md, c); |
2198 | else | |
2199 | __split_and_process_bio(md, c); | |
2200 | } | |
3b00b203 MP |
2201 | |
2202 | down_write(&md->io_lock); | |
022c2611 | 2203 | } |
73d410c0 | 2204 | |
ef208587 | 2205 | up_write(&md->io_lock); |
1da177e4 LT |
2206 | } |
2207 | ||
9a1fb464 | 2208 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2209 | { |
3b00b203 MP |
2210 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2211 | smp_mb__after_clear_bit(); | |
53d5914f | 2212 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2213 | } |
2214 | ||
1da177e4 LT |
2215 | /* |
2216 | * Swap in a new table (destroying old one). | |
2217 | */ | |
2218 | int dm_swap_table(struct mapped_device *md, struct dm_table *table) | |
2219 | { | |
754c5fc7 | 2220 | struct queue_limits limits; |
93c534ae | 2221 | int r = -EINVAL; |
1da177e4 | 2222 | |
e61290a4 | 2223 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2224 | |
2225 | /* device must be suspended */ | |
cf222b37 | 2226 | if (!dm_suspended(md)) |
93c534ae | 2227 | goto out; |
1da177e4 | 2228 | |
754c5fc7 MS |
2229 | r = dm_calculate_queue_limits(table, &limits); |
2230 | if (r) | |
2231 | goto out; | |
2232 | ||
e6ee8c0b KU |
2233 | /* cannot change the device type, once a table is bound */ |
2234 | if (md->map && | |
2235 | (dm_table_get_type(md->map) != dm_table_get_type(table))) { | |
2236 | DMWARN("can't change the device type after a table is bound"); | |
2237 | goto out; | |
2238 | } | |
2239 | ||
1da177e4 | 2240 | __unbind(md); |
754c5fc7 | 2241 | r = __bind(md, table, &limits); |
1da177e4 | 2242 | |
93c534ae | 2243 | out: |
e61290a4 | 2244 | mutex_unlock(&md->suspend_lock); |
93c534ae | 2245 | return r; |
1da177e4 LT |
2246 | } |
2247 | ||
cec47e3d KU |
2248 | static void dm_rq_invalidate_suspend_marker(struct mapped_device *md) |
2249 | { | |
2250 | md->suspend_rq.special = (void *)0x1; | |
2251 | } | |
2252 | ||
2253 | static void dm_rq_abort_suspend(struct mapped_device *md, int noflush) | |
2254 | { | |
2255 | struct request_queue *q = md->queue; | |
2256 | unsigned long flags; | |
2257 | ||
2258 | spin_lock_irqsave(q->queue_lock, flags); | |
2259 | if (!noflush) | |
2260 | dm_rq_invalidate_suspend_marker(md); | |
2261 | __start_queue(q); | |
2262 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2263 | } | |
2264 | ||
2265 | static void dm_rq_start_suspend(struct mapped_device *md, int noflush) | |
2266 | { | |
2267 | struct request *rq = &md->suspend_rq; | |
2268 | struct request_queue *q = md->queue; | |
2269 | ||
2270 | if (noflush) | |
2271 | stop_queue(q); | |
2272 | else { | |
2273 | blk_rq_init(q, rq); | |
2274 | blk_insert_request(q, rq, 0, NULL); | |
2275 | } | |
2276 | } | |
2277 | ||
2278 | static int dm_rq_suspend_available(struct mapped_device *md, int noflush) | |
2279 | { | |
2280 | int r = 1; | |
2281 | struct request *rq = &md->suspend_rq; | |
2282 | struct request_queue *q = md->queue; | |
2283 | unsigned long flags; | |
2284 | ||
2285 | if (noflush) | |
2286 | return r; | |
2287 | ||
2288 | /* The marker must be protected by queue lock if it is in use */ | |
2289 | spin_lock_irqsave(q->queue_lock, flags); | |
2290 | if (unlikely(rq->ref_count)) { | |
2291 | /* | |
2292 | * This can happen, when the previous flush suspend was | |
2293 | * interrupted, the marker is still in the queue and | |
2294 | * this flush suspend has been invoked, because we don't | |
2295 | * remove the marker at the time of suspend interruption. | |
2296 | * We have only one marker per mapped_device, so we can't | |
2297 | * start another flush suspend while it is in use. | |
2298 | */ | |
2299 | BUG_ON(!rq->special); /* The marker should be invalidated */ | |
2300 | DMWARN("Invalidating the previous flush suspend is still in" | |
2301 | " progress. Please retry later."); | |
2302 | r = 0; | |
2303 | } | |
2304 | spin_unlock_irqrestore(q->queue_lock, flags); | |
2305 | ||
2306 | return r; | |
2307 | } | |
2308 | ||
1da177e4 LT |
2309 | /* |
2310 | * Functions to lock and unlock any filesystem running on the | |
2311 | * device. | |
2312 | */ | |
2ca3310e | 2313 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2314 | { |
e39e2e95 | 2315 | int r; |
1da177e4 LT |
2316 | |
2317 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2318 | |
db8fef4f | 2319 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2320 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2321 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2322 | md->frozen_sb = NULL; |
2323 | return r; | |
dfbe03f6 AK |
2324 | } |
2325 | ||
aa8d7c2f AK |
2326 | set_bit(DMF_FROZEN, &md->flags); |
2327 | ||
1da177e4 LT |
2328 | return 0; |
2329 | } | |
2330 | ||
2ca3310e | 2331 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2332 | { |
aa8d7c2f AK |
2333 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2334 | return; | |
2335 | ||
db8fef4f | 2336 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2337 | md->frozen_sb = NULL; |
aa8d7c2f | 2338 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2339 | } |
2340 | ||
2341 | /* | |
2342 | * We need to be able to change a mapping table under a mounted | |
2343 | * filesystem. For example we might want to move some data in | |
2344 | * the background. Before the table can be swapped with | |
2345 | * dm_bind_table, dm_suspend must be called to flush any in | |
2346 | * flight bios and ensure that any further io gets deferred. | |
2347 | */ | |
cec47e3d KU |
2348 | /* |
2349 | * Suspend mechanism in request-based dm. | |
2350 | * | |
2351 | * After the suspend starts, further incoming requests are kept in | |
2352 | * the request_queue and deferred. | |
2353 | * Remaining requests in the request_queue at the start of suspend are flushed | |
2354 | * if it is flush suspend. | |
2355 | * The suspend completes when the following conditions have been satisfied, | |
2356 | * so wait for it: | |
2357 | * 1. q->in_flight is 0 (which means no in_flight request) | |
2358 | * 2. queue has been stopped (which means no request dispatching) | |
2359 | * | |
2360 | * | |
2361 | * Noflush suspend | |
2362 | * --------------- | |
2363 | * Noflush suspend doesn't need to dispatch remaining requests. | |
2364 | * So stop the queue immediately. Then, wait for all in_flight requests | |
2365 | * to be completed or requeued. | |
2366 | * | |
2367 | * To abort noflush suspend, start the queue. | |
2368 | * | |
2369 | * | |
2370 | * Flush suspend | |
2371 | * ------------- | |
2372 | * Flush suspend needs to dispatch remaining requests. So stop the queue | |
2373 | * after the remaining requests are completed. (Requeued request must be also | |
2374 | * re-dispatched and completed. Until then, we can't stop the queue.) | |
2375 | * | |
2376 | * During flushing the remaining requests, further incoming requests are also | |
2377 | * inserted to the same queue. To distinguish which requests are to be | |
2378 | * flushed, we insert a marker request to the queue at the time of starting | |
2379 | * flush suspend, like a barrier. | |
2380 | * The dispatching is blocked when the marker is found on the top of the queue. | |
2381 | * And the queue is stopped when all in_flight requests are completed, since | |
2382 | * that means the remaining requests are completely flushed. | |
2383 | * Then, the marker is removed from the queue. | |
2384 | * | |
2385 | * To abort flush suspend, we also need to take care of the marker, not only | |
2386 | * starting the queue. | |
2387 | * We don't remove the marker forcibly from the queue since it's against | |
2388 | * the block-layer manner. Instead, we put a invalidated mark on the marker. | |
2389 | * When the invalidated marker is found on the top of the queue, it is | |
2390 | * immediately removed from the queue, so it doesn't block dispatching. | |
2391 | * Because we have only one marker per mapped_device, we can't start another | |
2392 | * flush suspend until the invalidated marker is removed from the queue. | |
2393 | * So fail and return with -EBUSY in such a case. | |
2394 | */ | |
a3d77d35 | 2395 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2396 | { |
2ca3310e | 2397 | struct dm_table *map = NULL; |
46125c1c | 2398 | int r = 0; |
a3d77d35 | 2399 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2400 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2401 | |
e61290a4 | 2402 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2403 | |
73d410c0 MB |
2404 | if (dm_suspended(md)) { |
2405 | r = -EINVAL; | |
d287483d | 2406 | goto out_unlock; |
73d410c0 | 2407 | } |
1da177e4 | 2408 | |
cec47e3d KU |
2409 | if (dm_request_based(md) && !dm_rq_suspend_available(md, noflush)) { |
2410 | r = -EBUSY; | |
2411 | goto out_unlock; | |
2412 | } | |
2413 | ||
1da177e4 | 2414 | map = dm_get_table(md); |
1da177e4 | 2415 | |
2e93ccc1 KU |
2416 | /* |
2417 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2418 | * This flag is cleared before dm_suspend returns. | |
2419 | */ | |
2420 | if (noflush) | |
2421 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2422 | ||
cf222b37 AK |
2423 | /* This does not get reverted if there's an error later. */ |
2424 | dm_table_presuspend_targets(map); | |
2425 | ||
32a926da MP |
2426 | /* |
2427 | * Flush I/O to the device. noflush supersedes do_lockfs, | |
2428 | * because lock_fs() needs to flush I/Os. | |
2429 | */ | |
2430 | if (!noflush && do_lockfs) { | |
2431 | r = lock_fs(md); | |
2432 | if (r) | |
f431d966 | 2433 | goto out; |
aa8d7c2f | 2434 | } |
1da177e4 LT |
2435 | |
2436 | /* | |
3b00b203 MP |
2437 | * Here we must make sure that no processes are submitting requests |
2438 | * to target drivers i.e. no one may be executing | |
2439 | * __split_and_process_bio. This is called from dm_request and | |
2440 | * dm_wq_work. | |
2441 | * | |
2442 | * To get all processes out of __split_and_process_bio in dm_request, | |
2443 | * we take the write lock. To prevent any process from reentering | |
2444 | * __split_and_process_bio from dm_request, we set | |
2445 | * DMF_QUEUE_IO_TO_THREAD. | |
2446 | * | |
2447 | * To quiesce the thread (dm_wq_work), we set DMF_BLOCK_IO_FOR_SUSPEND | |
2448 | * and call flush_workqueue(md->wq). flush_workqueue will wait until | |
2449 | * dm_wq_work exits and DMF_BLOCK_IO_FOR_SUSPEND will prevent any | |
2450 | * further calls to __split_and_process_bio from dm_wq_work. | |
1da177e4 | 2451 | */ |
2ca3310e | 2452 | down_write(&md->io_lock); |
1eb787ec AK |
2453 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2454 | set_bit(DMF_QUEUE_IO_TO_THREAD, &md->flags); | |
2ca3310e | 2455 | up_write(&md->io_lock); |
1da177e4 | 2456 | |
3b00b203 MP |
2457 | flush_workqueue(md->wq); |
2458 | ||
cec47e3d KU |
2459 | if (dm_request_based(md)) |
2460 | dm_rq_start_suspend(md, noflush); | |
2461 | ||
1da177e4 | 2462 | /* |
3b00b203 MP |
2463 | * At this point no more requests are entering target request routines. |
2464 | * We call dm_wait_for_completion to wait for all existing requests | |
2465 | * to finish. | |
1da177e4 | 2466 | */ |
401600df | 2467 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2468 | |
2ca3310e | 2469 | down_write(&md->io_lock); |
6d6f10df | 2470 | if (noflush) |
022c2611 | 2471 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
94d6351e | 2472 | up_write(&md->io_lock); |
2e93ccc1 | 2473 | |
1da177e4 | 2474 | /* were we interrupted ? */ |
46125c1c | 2475 | if (r < 0) { |
9a1fb464 | 2476 | dm_queue_flush(md); |
73d410c0 | 2477 | |
cec47e3d KU |
2478 | if (dm_request_based(md)) |
2479 | dm_rq_abort_suspend(md, noflush); | |
2480 | ||
2ca3310e | 2481 | unlock_fs(md); |
2e93ccc1 | 2482 | goto out; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2483 | } |
1da177e4 | 2484 | |
3b00b203 MP |
2485 | /* |
2486 | * If dm_wait_for_completion returned 0, the device is completely | |
2487 | * quiescent now. There is no request-processing activity. All new | |
2488 | * requests are being added to md->deferred list. | |
2489 | */ | |
2490 | ||
cf222b37 | 2491 | dm_table_postsuspend_targets(map); |
1da177e4 | 2492 | |
2ca3310e | 2493 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2494 | |
2ca3310e AK |
2495 | out: |
2496 | dm_table_put(map); | |
d287483d AK |
2497 | |
2498 | out_unlock: | |
e61290a4 | 2499 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2500 | return r; |
1da177e4 LT |
2501 | } |
2502 | ||
2503 | int dm_resume(struct mapped_device *md) | |
2504 | { | |
cf222b37 | 2505 | int r = -EINVAL; |
cf222b37 | 2506 | struct dm_table *map = NULL; |
1da177e4 | 2507 | |
e61290a4 | 2508 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2509 | if (!dm_suspended(md)) |
cf222b37 | 2510 | goto out; |
cf222b37 AK |
2511 | |
2512 | map = dm_get_table(md); | |
2ca3310e | 2513 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2514 | goto out; |
1da177e4 | 2515 | |
8757b776 MB |
2516 | r = dm_table_resume_targets(map); |
2517 | if (r) | |
2518 | goto out; | |
2ca3310e | 2519 | |
9a1fb464 | 2520 | dm_queue_flush(md); |
2ca3310e | 2521 | |
cec47e3d KU |
2522 | /* |
2523 | * Flushing deferred I/Os must be done after targets are resumed | |
2524 | * so that mapping of targets can work correctly. | |
2525 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2526 | */ | |
2527 | if (dm_request_based(md)) | |
2528 | start_queue(md->queue); | |
2529 | ||
2ca3310e AK |
2530 | unlock_fs(md); |
2531 | ||
2532 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2533 | ||
1da177e4 | 2534 | dm_table_unplug_all(map); |
cf222b37 AK |
2535 | r = 0; |
2536 | out: | |
2537 | dm_table_put(map); | |
e61290a4 | 2538 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2539 | |
cf222b37 | 2540 | return r; |
1da177e4 LT |
2541 | } |
2542 | ||
2543 | /*----------------------------------------------------------------- | |
2544 | * Event notification. | |
2545 | *---------------------------------------------------------------*/ | |
60935eb2 MB |
2546 | void dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
2547 | unsigned cookie) | |
69267a30 | 2548 | { |
60935eb2 MB |
2549 | char udev_cookie[DM_COOKIE_LENGTH]; |
2550 | char *envp[] = { udev_cookie, NULL }; | |
2551 | ||
2552 | if (!cookie) | |
2553 | kobject_uevent(&disk_to_dev(md->disk)->kobj, action); | |
2554 | else { | |
2555 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2556 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
2557 | kobject_uevent_env(&disk_to_dev(md->disk)->kobj, action, envp); | |
2558 | } | |
69267a30 AK |
2559 | } |
2560 | ||
7a8c3d3b MA |
2561 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2562 | { | |
2563 | return atomic_add_return(1, &md->uevent_seq); | |
2564 | } | |
2565 | ||
1da177e4 LT |
2566 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2567 | { | |
2568 | return atomic_read(&md->event_nr); | |
2569 | } | |
2570 | ||
2571 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2572 | { | |
2573 | return wait_event_interruptible(md->eventq, | |
2574 | (event_nr != atomic_read(&md->event_nr))); | |
2575 | } | |
2576 | ||
7a8c3d3b MA |
2577 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2578 | { | |
2579 | unsigned long flags; | |
2580 | ||
2581 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2582 | list_add(elist, &md->uevent_list); | |
2583 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2584 | } | |
2585 | ||
1da177e4 LT |
2586 | /* |
2587 | * The gendisk is only valid as long as you have a reference | |
2588 | * count on 'md'. | |
2589 | */ | |
2590 | struct gendisk *dm_disk(struct mapped_device *md) | |
2591 | { | |
2592 | return md->disk; | |
2593 | } | |
2594 | ||
784aae73 MB |
2595 | struct kobject *dm_kobject(struct mapped_device *md) |
2596 | { | |
2597 | return &md->kobj; | |
2598 | } | |
2599 | ||
2600 | /* | |
2601 | * struct mapped_device should not be exported outside of dm.c | |
2602 | * so use this check to verify that kobj is part of md structure | |
2603 | */ | |
2604 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2605 | { | |
2606 | struct mapped_device *md; | |
2607 | ||
2608 | md = container_of(kobj, struct mapped_device, kobj); | |
2609 | if (&md->kobj != kobj) | |
2610 | return NULL; | |
2611 | ||
4d89b7b4 MB |
2612 | if (test_bit(DMF_FREEING, &md->flags) || |
2613 | test_bit(DMF_DELETING, &md->flags)) | |
2614 | return NULL; | |
2615 | ||
784aae73 MB |
2616 | dm_get(md); |
2617 | return md; | |
2618 | } | |
2619 | ||
1da177e4 LT |
2620 | int dm_suspended(struct mapped_device *md) |
2621 | { | |
2622 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2623 | } | |
2624 | ||
2e93ccc1 KU |
2625 | int dm_noflush_suspending(struct dm_target *ti) |
2626 | { | |
2627 | struct mapped_device *md = dm_table_get_md(ti->table); | |
2628 | int r = __noflush_suspending(md); | |
2629 | ||
2630 | dm_put(md); | |
2631 | ||
2632 | return r; | |
2633 | } | |
2634 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2635 | ||
e6ee8c0b KU |
2636 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type) |
2637 | { | |
2638 | struct dm_md_mempools *pools = kmalloc(sizeof(*pools), GFP_KERNEL); | |
2639 | ||
2640 | if (!pools) | |
2641 | return NULL; | |
2642 | ||
2643 | pools->io_pool = (type == DM_TYPE_BIO_BASED) ? | |
2644 | mempool_create_slab_pool(MIN_IOS, _io_cache) : | |
2645 | mempool_create_slab_pool(MIN_IOS, _rq_bio_info_cache); | |
2646 | if (!pools->io_pool) | |
2647 | goto free_pools_and_out; | |
2648 | ||
2649 | pools->tio_pool = (type == DM_TYPE_BIO_BASED) ? | |
2650 | mempool_create_slab_pool(MIN_IOS, _tio_cache) : | |
2651 | mempool_create_slab_pool(MIN_IOS, _rq_tio_cache); | |
2652 | if (!pools->tio_pool) | |
2653 | goto free_io_pool_and_out; | |
2654 | ||
2655 | pools->bs = (type == DM_TYPE_BIO_BASED) ? | |
2656 | bioset_create(16, 0) : bioset_create(MIN_IOS, 0); | |
2657 | if (!pools->bs) | |
2658 | goto free_tio_pool_and_out; | |
2659 | ||
2660 | return pools; | |
2661 | ||
2662 | free_tio_pool_and_out: | |
2663 | mempool_destroy(pools->tio_pool); | |
2664 | ||
2665 | free_io_pool_and_out: | |
2666 | mempool_destroy(pools->io_pool); | |
2667 | ||
2668 | free_pools_and_out: | |
2669 | kfree(pools); | |
2670 | ||
2671 | return NULL; | |
2672 | } | |
2673 | ||
2674 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2675 | { | |
2676 | if (!pools) | |
2677 | return; | |
2678 | ||
2679 | if (pools->io_pool) | |
2680 | mempool_destroy(pools->io_pool); | |
2681 | ||
2682 | if (pools->tio_pool) | |
2683 | mempool_destroy(pools->tio_pool); | |
2684 | ||
2685 | if (pools->bs) | |
2686 | bioset_free(pools->bs); | |
2687 | ||
2688 | kfree(pools); | |
2689 | } | |
2690 | ||
83d5cde4 | 2691 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2692 | .open = dm_blk_open, |
2693 | .release = dm_blk_close, | |
aa129a22 | 2694 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2695 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2696 | .owner = THIS_MODULE |
2697 | }; | |
2698 | ||
2699 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2700 | ||
2701 | /* | |
2702 | * module hooks | |
2703 | */ | |
2704 | module_init(dm_init); | |
2705 | module_exit(dm_exit); | |
2706 | ||
2707 | module_param(major, uint, 0); | |
2708 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2709 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2710 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2711 | MODULE_LICENSE("GPL"); |