Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * Copyright (C) 2001, 2002 Sistina Software (UK) Limited. | |
784aae73 | 3 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
1da177e4 LT |
4 | * |
5 | * This file is released under the GPL. | |
6 | */ | |
7 | ||
8 | #include "dm.h" | |
51e5b2bd | 9 | #include "dm-uevent.h" |
1da177e4 LT |
10 | |
11 | #include <linux/init.h> | |
12 | #include <linux/module.h> | |
48c9c27b | 13 | #include <linux/mutex.h> |
1da177e4 LT |
14 | #include <linux/moduleparam.h> |
15 | #include <linux/blkpg.h> | |
16 | #include <linux/bio.h> | |
1da177e4 LT |
17 | #include <linux/mempool.h> |
18 | #include <linux/slab.h> | |
19 | #include <linux/idr.h> | |
3ac51e74 | 20 | #include <linux/hdreg.h> |
3f77316d | 21 | #include <linux/delay.h> |
55782138 LZ |
22 | |
23 | #include <trace/events/block.h> | |
1da177e4 | 24 | |
72d94861 AK |
25 | #define DM_MSG_PREFIX "core" |
26 | ||
71a16736 NK |
27 | #ifdef CONFIG_PRINTK |
28 | /* | |
29 | * ratelimit state to be used in DMXXX_LIMIT(). | |
30 | */ | |
31 | DEFINE_RATELIMIT_STATE(dm_ratelimit_state, | |
32 | DEFAULT_RATELIMIT_INTERVAL, | |
33 | DEFAULT_RATELIMIT_BURST); | |
34 | EXPORT_SYMBOL(dm_ratelimit_state); | |
35 | #endif | |
36 | ||
60935eb2 MB |
37 | /* |
38 | * Cookies are numeric values sent with CHANGE and REMOVE | |
39 | * uevents while resuming, removing or renaming the device. | |
40 | */ | |
41 | #define DM_COOKIE_ENV_VAR_NAME "DM_COOKIE" | |
42 | #define DM_COOKIE_LENGTH 24 | |
43 | ||
1da177e4 LT |
44 | static const char *_name = DM_NAME; |
45 | ||
46 | static unsigned int major = 0; | |
47 | static unsigned int _major = 0; | |
48 | ||
d15b774c AK |
49 | static DEFINE_IDR(_minor_idr); |
50 | ||
f32c10b0 | 51 | static DEFINE_SPINLOCK(_minor_lock); |
1da177e4 | 52 | /* |
8fbf26ad | 53 | * For bio-based dm. |
1da177e4 LT |
54 | * One of these is allocated per bio. |
55 | */ | |
56 | struct dm_io { | |
57 | struct mapped_device *md; | |
58 | int error; | |
1da177e4 | 59 | atomic_t io_count; |
6ae2fa67 | 60 | struct bio *bio; |
3eaf840e | 61 | unsigned long start_time; |
f88fb981 | 62 | spinlock_t endio_lock; |
1da177e4 LT |
63 | }; |
64 | ||
8fbf26ad KU |
65 | /* |
66 | * For request-based dm. | |
67 | * One of these is allocated per request. | |
68 | */ | |
69 | struct dm_rq_target_io { | |
70 | struct mapped_device *md; | |
71 | struct dm_target *ti; | |
72 | struct request *orig, clone; | |
73 | int error; | |
74 | union map_info info; | |
75 | }; | |
76 | ||
77 | /* | |
94818742 KO |
78 | * For request-based dm - the bio clones we allocate are embedded in these |
79 | * structs. | |
80 | * | |
81 | * We allocate these with bio_alloc_bioset, using the front_pad parameter when | |
82 | * the bioset is created - this means the bio has to come at the end of the | |
83 | * struct. | |
8fbf26ad KU |
84 | */ |
85 | struct dm_rq_clone_bio_info { | |
86 | struct bio *orig; | |
cec47e3d | 87 | struct dm_rq_target_io *tio; |
94818742 | 88 | struct bio clone; |
8fbf26ad KU |
89 | }; |
90 | ||
1da177e4 LT |
91 | union map_info *dm_get_mapinfo(struct bio *bio) |
92 | { | |
17b2f66f | 93 | if (bio && bio->bi_private) |
028867ac | 94 | return &((struct dm_target_io *)bio->bi_private)->info; |
17b2f66f | 95 | return NULL; |
1da177e4 LT |
96 | } |
97 | ||
cec47e3d KU |
98 | union map_info *dm_get_rq_mapinfo(struct request *rq) |
99 | { | |
100 | if (rq && rq->end_io_data) | |
101 | return &((struct dm_rq_target_io *)rq->end_io_data)->info; | |
102 | return NULL; | |
103 | } | |
104 | EXPORT_SYMBOL_GPL(dm_get_rq_mapinfo); | |
105 | ||
ba61fdd1 JM |
106 | #define MINOR_ALLOCED ((void *)-1) |
107 | ||
1da177e4 LT |
108 | /* |
109 | * Bits for the md->flags field. | |
110 | */ | |
1eb787ec | 111 | #define DMF_BLOCK_IO_FOR_SUSPEND 0 |
1da177e4 | 112 | #define DMF_SUSPENDED 1 |
aa8d7c2f | 113 | #define DMF_FROZEN 2 |
fba9f90e | 114 | #define DMF_FREEING 3 |
5c6bd75d | 115 | #define DMF_DELETING 4 |
2e93ccc1 | 116 | #define DMF_NOFLUSH_SUSPENDING 5 |
d5b9dd04 | 117 | #define DMF_MERGE_IS_OPTIONAL 6 |
1da177e4 | 118 | |
83d5e5b0 MP |
119 | /* |
120 | * A dummy definition to make RCU happy. | |
121 | * struct dm_table should never be dereferenced in this file. | |
122 | */ | |
123 | struct dm_table { | |
124 | int undefined__; | |
125 | }; | |
126 | ||
304f3f6a MB |
127 | /* |
128 | * Work processed by per-device workqueue. | |
129 | */ | |
1da177e4 | 130 | struct mapped_device { |
83d5e5b0 | 131 | struct srcu_struct io_barrier; |
e61290a4 | 132 | struct mutex suspend_lock; |
1da177e4 | 133 | atomic_t holders; |
5c6bd75d | 134 | atomic_t open_count; |
1da177e4 | 135 | |
2a7faeb1 MP |
136 | /* |
137 | * The current mapping. | |
138 | * Use dm_get_live_table{_fast} or take suspend_lock for | |
139 | * dereference. | |
140 | */ | |
141 | struct dm_table *map; | |
142 | ||
1da177e4 LT |
143 | unsigned long flags; |
144 | ||
165125e1 | 145 | struct request_queue *queue; |
a5664dad | 146 | unsigned type; |
4a0b4ddf | 147 | /* Protect queue and type against concurrent access. */ |
a5664dad MS |
148 | struct mutex type_lock; |
149 | ||
36a0456f AK |
150 | struct target_type *immutable_target_type; |
151 | ||
1da177e4 | 152 | struct gendisk *disk; |
7e51f257 | 153 | char name[16]; |
1da177e4 LT |
154 | |
155 | void *interface_ptr; | |
156 | ||
157 | /* | |
158 | * A list of ios that arrived while we were suspended. | |
159 | */ | |
316d315b | 160 | atomic_t pending[2]; |
1da177e4 | 161 | wait_queue_head_t wait; |
53d5914f | 162 | struct work_struct work; |
74859364 | 163 | struct bio_list deferred; |
022c2611 | 164 | spinlock_t deferred_lock; |
1da177e4 | 165 | |
af7e466a | 166 | /* |
29e4013d | 167 | * Processing queue (flush) |
304f3f6a MB |
168 | */ |
169 | struct workqueue_struct *wq; | |
170 | ||
1da177e4 LT |
171 | /* |
172 | * io objects are allocated from here. | |
173 | */ | |
174 | mempool_t *io_pool; | |
1da177e4 | 175 | |
9faf400f SB |
176 | struct bio_set *bs; |
177 | ||
1da177e4 LT |
178 | /* |
179 | * Event handling. | |
180 | */ | |
181 | atomic_t event_nr; | |
182 | wait_queue_head_t eventq; | |
7a8c3d3b MA |
183 | atomic_t uevent_seq; |
184 | struct list_head uevent_list; | |
185 | spinlock_t uevent_lock; /* Protect access to uevent_list */ | |
1da177e4 LT |
186 | |
187 | /* | |
188 | * freeze/thaw support require holding onto a super block | |
189 | */ | |
190 | struct super_block *frozen_sb; | |
db8fef4f | 191 | struct block_device *bdev; |
3ac51e74 DW |
192 | |
193 | /* forced geometry settings */ | |
194 | struct hd_geometry geometry; | |
784aae73 MB |
195 | |
196 | /* sysfs handle */ | |
197 | struct kobject kobj; | |
52b1fd5a | 198 | |
d87f4c14 TH |
199 | /* zero-length flush that will be cloned and submitted to targets */ |
200 | struct bio flush_bio; | |
1da177e4 LT |
201 | }; |
202 | ||
e6ee8c0b KU |
203 | /* |
204 | * For mempools pre-allocation at the table loading time. | |
205 | */ | |
206 | struct dm_md_mempools { | |
207 | mempool_t *io_pool; | |
e6ee8c0b KU |
208 | struct bio_set *bs; |
209 | }; | |
210 | ||
1da177e4 | 211 | #define MIN_IOS 256 |
e18b890b | 212 | static struct kmem_cache *_io_cache; |
8fbf26ad | 213 | static struct kmem_cache *_rq_tio_cache; |
94818742 | 214 | |
1da177e4 LT |
215 | static int __init local_init(void) |
216 | { | |
51157b4a | 217 | int r = -ENOMEM; |
1da177e4 | 218 | |
1da177e4 | 219 | /* allocate a slab for the dm_ios */ |
028867ac | 220 | _io_cache = KMEM_CACHE(dm_io, 0); |
1da177e4 | 221 | if (!_io_cache) |
51157b4a | 222 | return r; |
1da177e4 | 223 | |
8fbf26ad KU |
224 | _rq_tio_cache = KMEM_CACHE(dm_rq_target_io, 0); |
225 | if (!_rq_tio_cache) | |
dba14160 | 226 | goto out_free_io_cache; |
8fbf26ad | 227 | |
51e5b2bd | 228 | r = dm_uevent_init(); |
51157b4a | 229 | if (r) |
23e5083b | 230 | goto out_free_rq_tio_cache; |
51e5b2bd | 231 | |
1da177e4 LT |
232 | _major = major; |
233 | r = register_blkdev(_major, _name); | |
51157b4a KU |
234 | if (r < 0) |
235 | goto out_uevent_exit; | |
1da177e4 LT |
236 | |
237 | if (!_major) | |
238 | _major = r; | |
239 | ||
240 | return 0; | |
51157b4a KU |
241 | |
242 | out_uevent_exit: | |
243 | dm_uevent_exit(); | |
8fbf26ad KU |
244 | out_free_rq_tio_cache: |
245 | kmem_cache_destroy(_rq_tio_cache); | |
51157b4a KU |
246 | out_free_io_cache: |
247 | kmem_cache_destroy(_io_cache); | |
248 | ||
249 | return r; | |
1da177e4 LT |
250 | } |
251 | ||
252 | static void local_exit(void) | |
253 | { | |
8fbf26ad | 254 | kmem_cache_destroy(_rq_tio_cache); |
1da177e4 | 255 | kmem_cache_destroy(_io_cache); |
00d59405 | 256 | unregister_blkdev(_major, _name); |
51e5b2bd | 257 | dm_uevent_exit(); |
1da177e4 LT |
258 | |
259 | _major = 0; | |
260 | ||
261 | DMINFO("cleaned up"); | |
262 | } | |
263 | ||
b9249e55 | 264 | static int (*_inits[])(void) __initdata = { |
1da177e4 LT |
265 | local_init, |
266 | dm_target_init, | |
267 | dm_linear_init, | |
268 | dm_stripe_init, | |
952b3557 | 269 | dm_io_init, |
945fa4d2 | 270 | dm_kcopyd_init, |
1da177e4 LT |
271 | dm_interface_init, |
272 | }; | |
273 | ||
b9249e55 | 274 | static void (*_exits[])(void) = { |
1da177e4 LT |
275 | local_exit, |
276 | dm_target_exit, | |
277 | dm_linear_exit, | |
278 | dm_stripe_exit, | |
952b3557 | 279 | dm_io_exit, |
945fa4d2 | 280 | dm_kcopyd_exit, |
1da177e4 LT |
281 | dm_interface_exit, |
282 | }; | |
283 | ||
284 | static int __init dm_init(void) | |
285 | { | |
286 | const int count = ARRAY_SIZE(_inits); | |
287 | ||
288 | int r, i; | |
289 | ||
290 | for (i = 0; i < count; i++) { | |
291 | r = _inits[i](); | |
292 | if (r) | |
293 | goto bad; | |
294 | } | |
295 | ||
296 | return 0; | |
297 | ||
298 | bad: | |
299 | while (i--) | |
300 | _exits[i](); | |
301 | ||
302 | return r; | |
303 | } | |
304 | ||
305 | static void __exit dm_exit(void) | |
306 | { | |
307 | int i = ARRAY_SIZE(_exits); | |
308 | ||
309 | while (i--) | |
310 | _exits[i](); | |
d15b774c AK |
311 | |
312 | /* | |
313 | * Should be empty by this point. | |
314 | */ | |
d15b774c | 315 | idr_destroy(&_minor_idr); |
1da177e4 LT |
316 | } |
317 | ||
318 | /* | |
319 | * Block device functions | |
320 | */ | |
432a212c MA |
321 | int dm_deleting_md(struct mapped_device *md) |
322 | { | |
323 | return test_bit(DMF_DELETING, &md->flags); | |
324 | } | |
325 | ||
fe5f9f2c | 326 | static int dm_blk_open(struct block_device *bdev, fmode_t mode) |
1da177e4 LT |
327 | { |
328 | struct mapped_device *md; | |
329 | ||
fba9f90e JM |
330 | spin_lock(&_minor_lock); |
331 | ||
fe5f9f2c | 332 | md = bdev->bd_disk->private_data; |
fba9f90e JM |
333 | if (!md) |
334 | goto out; | |
335 | ||
5c6bd75d | 336 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 337 | dm_deleting_md(md)) { |
fba9f90e JM |
338 | md = NULL; |
339 | goto out; | |
340 | } | |
341 | ||
1da177e4 | 342 | dm_get(md); |
5c6bd75d | 343 | atomic_inc(&md->open_count); |
fba9f90e JM |
344 | |
345 | out: | |
346 | spin_unlock(&_minor_lock); | |
347 | ||
348 | return md ? 0 : -ENXIO; | |
1da177e4 LT |
349 | } |
350 | ||
db2a144b | 351 | static void dm_blk_close(struct gendisk *disk, fmode_t mode) |
1da177e4 | 352 | { |
fe5f9f2c | 353 | struct mapped_device *md = disk->private_data; |
6e9624b8 | 354 | |
4a1aeb98 MB |
355 | spin_lock(&_minor_lock); |
356 | ||
5c6bd75d | 357 | atomic_dec(&md->open_count); |
1da177e4 | 358 | dm_put(md); |
4a1aeb98 MB |
359 | |
360 | spin_unlock(&_minor_lock); | |
1da177e4 LT |
361 | } |
362 | ||
5c6bd75d AK |
363 | int dm_open_count(struct mapped_device *md) |
364 | { | |
365 | return atomic_read(&md->open_count); | |
366 | } | |
367 | ||
368 | /* | |
369 | * Guarantees nothing is using the device before it's deleted. | |
370 | */ | |
371 | int dm_lock_for_deletion(struct mapped_device *md) | |
372 | { | |
373 | int r = 0; | |
374 | ||
375 | spin_lock(&_minor_lock); | |
376 | ||
377 | if (dm_open_count(md)) | |
378 | r = -EBUSY; | |
379 | else | |
380 | set_bit(DMF_DELETING, &md->flags); | |
381 | ||
382 | spin_unlock(&_minor_lock); | |
383 | ||
384 | return r; | |
385 | } | |
386 | ||
3ac51e74 DW |
387 | static int dm_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo) |
388 | { | |
389 | struct mapped_device *md = bdev->bd_disk->private_data; | |
390 | ||
391 | return dm_get_geometry(md, geo); | |
392 | } | |
393 | ||
fe5f9f2c | 394 | static int dm_blk_ioctl(struct block_device *bdev, fmode_t mode, |
aa129a22 MB |
395 | unsigned int cmd, unsigned long arg) |
396 | { | |
fe5f9f2c | 397 | struct mapped_device *md = bdev->bd_disk->private_data; |
83d5e5b0 | 398 | int srcu_idx; |
6c182cd8 | 399 | struct dm_table *map; |
aa129a22 MB |
400 | struct dm_target *tgt; |
401 | int r = -ENOTTY; | |
402 | ||
6c182cd8 | 403 | retry: |
83d5e5b0 MP |
404 | map = dm_get_live_table(md, &srcu_idx); |
405 | ||
aa129a22 MB |
406 | if (!map || !dm_table_get_size(map)) |
407 | goto out; | |
408 | ||
409 | /* We only support devices that have a single target */ | |
410 | if (dm_table_get_num_targets(map) != 1) | |
411 | goto out; | |
412 | ||
413 | tgt = dm_table_get_target(map, 0); | |
414 | ||
4f186f8b | 415 | if (dm_suspended_md(md)) { |
aa129a22 MB |
416 | r = -EAGAIN; |
417 | goto out; | |
418 | } | |
419 | ||
420 | if (tgt->type->ioctl) | |
647b3d00 | 421 | r = tgt->type->ioctl(tgt, cmd, arg); |
aa129a22 MB |
422 | |
423 | out: | |
83d5e5b0 | 424 | dm_put_live_table(md, srcu_idx); |
aa129a22 | 425 | |
6c182cd8 HR |
426 | if (r == -ENOTCONN) { |
427 | msleep(10); | |
428 | goto retry; | |
429 | } | |
430 | ||
aa129a22 MB |
431 | return r; |
432 | } | |
433 | ||
028867ac | 434 | static struct dm_io *alloc_io(struct mapped_device *md) |
1da177e4 LT |
435 | { |
436 | return mempool_alloc(md->io_pool, GFP_NOIO); | |
437 | } | |
438 | ||
028867ac | 439 | static void free_io(struct mapped_device *md, struct dm_io *io) |
1da177e4 LT |
440 | { |
441 | mempool_free(io, md->io_pool); | |
442 | } | |
443 | ||
028867ac | 444 | static void free_tio(struct mapped_device *md, struct dm_target_io *tio) |
1da177e4 | 445 | { |
dba14160 | 446 | bio_put(&tio->clone); |
1da177e4 LT |
447 | } |
448 | ||
08885643 KU |
449 | static struct dm_rq_target_io *alloc_rq_tio(struct mapped_device *md, |
450 | gfp_t gfp_mask) | |
cec47e3d | 451 | { |
5f015204 | 452 | return mempool_alloc(md->io_pool, gfp_mask); |
cec47e3d KU |
453 | } |
454 | ||
455 | static void free_rq_tio(struct dm_rq_target_io *tio) | |
456 | { | |
5f015204 | 457 | mempool_free(tio, tio->md->io_pool); |
cec47e3d KU |
458 | } |
459 | ||
90abb8c4 KU |
460 | static int md_in_flight(struct mapped_device *md) |
461 | { | |
462 | return atomic_read(&md->pending[READ]) + | |
463 | atomic_read(&md->pending[WRITE]); | |
464 | } | |
465 | ||
3eaf840e JNN |
466 | static void start_io_acct(struct dm_io *io) |
467 | { | |
468 | struct mapped_device *md = io->md; | |
c9959059 | 469 | int cpu; |
316d315b | 470 | int rw = bio_data_dir(io->bio); |
3eaf840e JNN |
471 | |
472 | io->start_time = jiffies; | |
473 | ||
074a7aca TH |
474 | cpu = part_stat_lock(); |
475 | part_round_stats(cpu, &dm_disk(md)->part0); | |
476 | part_stat_unlock(); | |
1e9bb880 SL |
477 | atomic_set(&dm_disk(md)->part0.in_flight[rw], |
478 | atomic_inc_return(&md->pending[rw])); | |
3eaf840e JNN |
479 | } |
480 | ||
d221d2e7 | 481 | static void end_io_acct(struct dm_io *io) |
3eaf840e JNN |
482 | { |
483 | struct mapped_device *md = io->md; | |
484 | struct bio *bio = io->bio; | |
485 | unsigned long duration = jiffies - io->start_time; | |
c9959059 | 486 | int pending, cpu; |
3eaf840e JNN |
487 | int rw = bio_data_dir(bio); |
488 | ||
074a7aca TH |
489 | cpu = part_stat_lock(); |
490 | part_round_stats(cpu, &dm_disk(md)->part0); | |
491 | part_stat_add(cpu, &dm_disk(md)->part0, ticks[rw], duration); | |
492 | part_stat_unlock(); | |
3eaf840e | 493 | |
af7e466a MP |
494 | /* |
495 | * After this is decremented the bio must not be touched if it is | |
d87f4c14 | 496 | * a flush. |
af7e466a | 497 | */ |
1e9bb880 SL |
498 | pending = atomic_dec_return(&md->pending[rw]); |
499 | atomic_set(&dm_disk(md)->part0.in_flight[rw], pending); | |
316d315b | 500 | pending += atomic_read(&md->pending[rw^0x1]); |
3eaf840e | 501 | |
d221d2e7 MP |
502 | /* nudge anyone waiting on suspend queue */ |
503 | if (!pending) | |
504 | wake_up(&md->wait); | |
3eaf840e JNN |
505 | } |
506 | ||
1da177e4 LT |
507 | /* |
508 | * Add the bio to the list of deferred io. | |
509 | */ | |
92c63902 | 510 | static void queue_io(struct mapped_device *md, struct bio *bio) |
1da177e4 | 511 | { |
05447420 | 512 | unsigned long flags; |
1da177e4 | 513 | |
05447420 | 514 | spin_lock_irqsave(&md->deferred_lock, flags); |
1da177e4 | 515 | bio_list_add(&md->deferred, bio); |
05447420 | 516 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
6a8736d1 | 517 | queue_work(md->wq, &md->work); |
1da177e4 LT |
518 | } |
519 | ||
520 | /* | |
521 | * Everyone (including functions in this file), should use this | |
522 | * function to access the md->map field, and make sure they call | |
83d5e5b0 | 523 | * dm_put_live_table() when finished. |
1da177e4 | 524 | */ |
83d5e5b0 | 525 | struct dm_table *dm_get_live_table(struct mapped_device *md, int *srcu_idx) __acquires(md->io_barrier) |
1da177e4 | 526 | { |
83d5e5b0 MP |
527 | *srcu_idx = srcu_read_lock(&md->io_barrier); |
528 | ||
529 | return srcu_dereference(md->map, &md->io_barrier); | |
530 | } | |
1da177e4 | 531 | |
83d5e5b0 MP |
532 | void dm_put_live_table(struct mapped_device *md, int srcu_idx) __releases(md->io_barrier) |
533 | { | |
534 | srcu_read_unlock(&md->io_barrier, srcu_idx); | |
535 | } | |
536 | ||
537 | void dm_sync_table(struct mapped_device *md) | |
538 | { | |
539 | synchronize_srcu(&md->io_barrier); | |
540 | synchronize_rcu_expedited(); | |
541 | } | |
542 | ||
543 | /* | |
544 | * A fast alternative to dm_get_live_table/dm_put_live_table. | |
545 | * The caller must not block between these two functions. | |
546 | */ | |
547 | static struct dm_table *dm_get_live_table_fast(struct mapped_device *md) __acquires(RCU) | |
548 | { | |
549 | rcu_read_lock(); | |
550 | return rcu_dereference(md->map); | |
551 | } | |
1da177e4 | 552 | |
83d5e5b0 MP |
553 | static void dm_put_live_table_fast(struct mapped_device *md) __releases(RCU) |
554 | { | |
555 | rcu_read_unlock(); | |
1da177e4 LT |
556 | } |
557 | ||
3ac51e74 DW |
558 | /* |
559 | * Get the geometry associated with a dm device | |
560 | */ | |
561 | int dm_get_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
562 | { | |
563 | *geo = md->geometry; | |
564 | ||
565 | return 0; | |
566 | } | |
567 | ||
568 | /* | |
569 | * Set the geometry of a device. | |
570 | */ | |
571 | int dm_set_geometry(struct mapped_device *md, struct hd_geometry *geo) | |
572 | { | |
573 | sector_t sz = (sector_t)geo->cylinders * geo->heads * geo->sectors; | |
574 | ||
575 | if (geo->start > sz) { | |
576 | DMWARN("Start sector is beyond the geometry limits."); | |
577 | return -EINVAL; | |
578 | } | |
579 | ||
580 | md->geometry = *geo; | |
581 | ||
582 | return 0; | |
583 | } | |
584 | ||
1da177e4 LT |
585 | /*----------------------------------------------------------------- |
586 | * CRUD START: | |
587 | * A more elegant soln is in the works that uses the queue | |
588 | * merge fn, unfortunately there are a couple of changes to | |
589 | * the block layer that I want to make for this. So in the | |
590 | * interests of getting something for people to use I give | |
591 | * you this clearly demarcated crap. | |
592 | *---------------------------------------------------------------*/ | |
593 | ||
2e93ccc1 KU |
594 | static int __noflush_suspending(struct mapped_device *md) |
595 | { | |
596 | return test_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
597 | } | |
598 | ||
1da177e4 LT |
599 | /* |
600 | * Decrements the number of outstanding ios that a bio has been | |
601 | * cloned into, completing the original io if necc. | |
602 | */ | |
858119e1 | 603 | static void dec_pending(struct dm_io *io, int error) |
1da177e4 | 604 | { |
2e93ccc1 | 605 | unsigned long flags; |
b35f8caa MB |
606 | int io_error; |
607 | struct bio *bio; | |
608 | struct mapped_device *md = io->md; | |
2e93ccc1 KU |
609 | |
610 | /* Push-back supersedes any I/O errors */ | |
f88fb981 KU |
611 | if (unlikely(error)) { |
612 | spin_lock_irqsave(&io->endio_lock, flags); | |
613 | if (!(io->error > 0 && __noflush_suspending(md))) | |
614 | io->error = error; | |
615 | spin_unlock_irqrestore(&io->endio_lock, flags); | |
616 | } | |
1da177e4 LT |
617 | |
618 | if (atomic_dec_and_test(&io->io_count)) { | |
2e93ccc1 KU |
619 | if (io->error == DM_ENDIO_REQUEUE) { |
620 | /* | |
621 | * Target requested pushing back the I/O. | |
2e93ccc1 | 622 | */ |
022c2611 | 623 | spin_lock_irqsave(&md->deferred_lock, flags); |
6a8736d1 TH |
624 | if (__noflush_suspending(md)) |
625 | bio_list_add_head(&md->deferred, io->bio); | |
626 | else | |
2e93ccc1 KU |
627 | /* noflush suspend was interrupted. */ |
628 | io->error = -EIO; | |
022c2611 | 629 | spin_unlock_irqrestore(&md->deferred_lock, flags); |
2e93ccc1 KU |
630 | } |
631 | ||
b35f8caa MB |
632 | io_error = io->error; |
633 | bio = io->bio; | |
6a8736d1 TH |
634 | end_io_acct(io); |
635 | free_io(md, io); | |
636 | ||
637 | if (io_error == DM_ENDIO_REQUEUE) | |
638 | return; | |
2e93ccc1 | 639 | |
b372d360 | 640 | if ((bio->bi_rw & REQ_FLUSH) && bio->bi_size) { |
af7e466a | 641 | /* |
6a8736d1 TH |
642 | * Preflush done for flush with data, reissue |
643 | * without REQ_FLUSH. | |
af7e466a | 644 | */ |
6a8736d1 TH |
645 | bio->bi_rw &= ~REQ_FLUSH; |
646 | queue_io(md, bio); | |
af7e466a | 647 | } else { |
b372d360 | 648 | /* done with normal IO or empty flush */ |
0a82a8d1 | 649 | trace_block_bio_complete(md->queue, bio, io_error); |
b372d360 | 650 | bio_endio(bio, io_error); |
b35f8caa | 651 | } |
1da177e4 LT |
652 | } |
653 | } | |
654 | ||
6712ecf8 | 655 | static void clone_endio(struct bio *bio, int error) |
1da177e4 LT |
656 | { |
657 | int r = 0; | |
028867ac | 658 | struct dm_target_io *tio = bio->bi_private; |
b35f8caa | 659 | struct dm_io *io = tio->io; |
9faf400f | 660 | struct mapped_device *md = tio->io->md; |
1da177e4 LT |
661 | dm_endio_fn endio = tio->ti->type->end_io; |
662 | ||
1da177e4 LT |
663 | if (!bio_flagged(bio, BIO_UPTODATE) && !error) |
664 | error = -EIO; | |
665 | ||
666 | if (endio) { | |
7de3ee57 | 667 | r = endio(tio->ti, bio, error); |
2e93ccc1 KU |
668 | if (r < 0 || r == DM_ENDIO_REQUEUE) |
669 | /* | |
670 | * error and requeue request are handled | |
671 | * in dec_pending(). | |
672 | */ | |
1da177e4 | 673 | error = r; |
45cbcd79 KU |
674 | else if (r == DM_ENDIO_INCOMPLETE) |
675 | /* The target will handle the io */ | |
6712ecf8 | 676 | return; |
45cbcd79 KU |
677 | else if (r) { |
678 | DMWARN("unimplemented target endio return value: %d", r); | |
679 | BUG(); | |
680 | } | |
1da177e4 LT |
681 | } |
682 | ||
9faf400f | 683 | free_tio(md, tio); |
b35f8caa | 684 | dec_pending(io, error); |
1da177e4 LT |
685 | } |
686 | ||
cec47e3d KU |
687 | /* |
688 | * Partial completion handling for request-based dm | |
689 | */ | |
690 | static void end_clone_bio(struct bio *clone, int error) | |
691 | { | |
692 | struct dm_rq_clone_bio_info *info = clone->bi_private; | |
693 | struct dm_rq_target_io *tio = info->tio; | |
694 | struct bio *bio = info->orig; | |
695 | unsigned int nr_bytes = info->orig->bi_size; | |
696 | ||
697 | bio_put(clone); | |
698 | ||
699 | if (tio->error) | |
700 | /* | |
701 | * An error has already been detected on the request. | |
702 | * Once error occurred, just let clone->end_io() handle | |
703 | * the remainder. | |
704 | */ | |
705 | return; | |
706 | else if (error) { | |
707 | /* | |
708 | * Don't notice the error to the upper layer yet. | |
709 | * The error handling decision is made by the target driver, | |
710 | * when the request is completed. | |
711 | */ | |
712 | tio->error = error; | |
713 | return; | |
714 | } | |
715 | ||
716 | /* | |
717 | * I/O for the bio successfully completed. | |
718 | * Notice the data completion to the upper layer. | |
719 | */ | |
720 | ||
721 | /* | |
722 | * bios are processed from the head of the list. | |
723 | * So the completing bio should always be rq->bio. | |
724 | * If it's not, something wrong is happening. | |
725 | */ | |
726 | if (tio->orig->bio != bio) | |
727 | DMERR("bio completion is going in the middle of the request"); | |
728 | ||
729 | /* | |
730 | * Update the original request. | |
731 | * Do not use blk_end_request() here, because it may complete | |
732 | * the original request before the clone, and break the ordering. | |
733 | */ | |
734 | blk_update_request(tio->orig, 0, nr_bytes); | |
735 | } | |
736 | ||
737 | /* | |
738 | * Don't touch any member of the md after calling this function because | |
739 | * the md may be freed in dm_put() at the end of this function. | |
740 | * Or do dm_get() before calling this function and dm_put() later. | |
741 | */ | |
b4324fee | 742 | static void rq_completed(struct mapped_device *md, int rw, int run_queue) |
cec47e3d | 743 | { |
b4324fee | 744 | atomic_dec(&md->pending[rw]); |
cec47e3d KU |
745 | |
746 | /* nudge anyone waiting on suspend queue */ | |
b4324fee | 747 | if (!md_in_flight(md)) |
cec47e3d KU |
748 | wake_up(&md->wait); |
749 | ||
a8c32a5c JA |
750 | /* |
751 | * Run this off this callpath, as drivers could invoke end_io while | |
752 | * inside their request_fn (and holding the queue lock). Calling | |
753 | * back into ->request_fn() could deadlock attempting to grab the | |
754 | * queue lock again. | |
755 | */ | |
cec47e3d | 756 | if (run_queue) |
a8c32a5c | 757 | blk_run_queue_async(md->queue); |
cec47e3d KU |
758 | |
759 | /* | |
760 | * dm_put() must be at the end of this function. See the comment above | |
761 | */ | |
762 | dm_put(md); | |
763 | } | |
764 | ||
a77e28c7 KU |
765 | static void free_rq_clone(struct request *clone) |
766 | { | |
767 | struct dm_rq_target_io *tio = clone->end_io_data; | |
768 | ||
769 | blk_rq_unprep_clone(clone); | |
770 | free_rq_tio(tio); | |
771 | } | |
772 | ||
980691e5 KU |
773 | /* |
774 | * Complete the clone and the original request. | |
775 | * Must be called without queue lock. | |
776 | */ | |
777 | static void dm_end_request(struct request *clone, int error) | |
778 | { | |
779 | int rw = rq_data_dir(clone); | |
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 | ||
29e4013d | 784 | if (rq->cmd_type == REQ_TYPE_BLOCK_PC) { |
980691e5 KU |
785 | rq->errors = clone->errors; |
786 | rq->resid_len = clone->resid_len; | |
787 | ||
788 | if (rq->sense) | |
789 | /* | |
790 | * We are using the sense buffer of the original | |
791 | * request. | |
792 | * So setting the length of the sense data is enough. | |
793 | */ | |
794 | rq->sense_len = clone->sense_len; | |
795 | } | |
796 | ||
797 | free_rq_clone(clone); | |
29e4013d TH |
798 | blk_end_request_all(rq, error); |
799 | rq_completed(md, rw, true); | |
980691e5 KU |
800 | } |
801 | ||
cec47e3d KU |
802 | static void dm_unprep_request(struct request *rq) |
803 | { | |
804 | struct request *clone = rq->special; | |
cec47e3d KU |
805 | |
806 | rq->special = NULL; | |
807 | rq->cmd_flags &= ~REQ_DONTPREP; | |
808 | ||
a77e28c7 | 809 | free_rq_clone(clone); |
cec47e3d KU |
810 | } |
811 | ||
812 | /* | |
813 | * Requeue the original request of a clone. | |
814 | */ | |
815 | void dm_requeue_unmapped_request(struct request *clone) | |
816 | { | |
b4324fee | 817 | int rw = rq_data_dir(clone); |
cec47e3d KU |
818 | struct dm_rq_target_io *tio = clone->end_io_data; |
819 | struct mapped_device *md = tio->md; | |
820 | struct request *rq = tio->orig; | |
821 | struct request_queue *q = rq->q; | |
822 | unsigned long flags; | |
823 | ||
824 | dm_unprep_request(rq); | |
825 | ||
826 | spin_lock_irqsave(q->queue_lock, flags); | |
cec47e3d KU |
827 | blk_requeue_request(q, rq); |
828 | spin_unlock_irqrestore(q->queue_lock, flags); | |
829 | ||
b4324fee | 830 | rq_completed(md, rw, 0); |
cec47e3d KU |
831 | } |
832 | EXPORT_SYMBOL_GPL(dm_requeue_unmapped_request); | |
833 | ||
834 | static void __stop_queue(struct request_queue *q) | |
835 | { | |
836 | blk_stop_queue(q); | |
837 | } | |
838 | ||
839 | static void stop_queue(struct request_queue *q) | |
840 | { | |
841 | unsigned long flags; | |
842 | ||
843 | spin_lock_irqsave(q->queue_lock, flags); | |
844 | __stop_queue(q); | |
845 | spin_unlock_irqrestore(q->queue_lock, flags); | |
846 | } | |
847 | ||
848 | static void __start_queue(struct request_queue *q) | |
849 | { | |
850 | if (blk_queue_stopped(q)) | |
851 | blk_start_queue(q); | |
852 | } | |
853 | ||
854 | static void start_queue(struct request_queue *q) | |
855 | { | |
856 | unsigned long flags; | |
857 | ||
858 | spin_lock_irqsave(q->queue_lock, flags); | |
859 | __start_queue(q); | |
860 | spin_unlock_irqrestore(q->queue_lock, flags); | |
861 | } | |
862 | ||
11a68244 | 863 | static void dm_done(struct request *clone, int error, bool mapped) |
cec47e3d | 864 | { |
11a68244 | 865 | int r = error; |
cec47e3d | 866 | struct dm_rq_target_io *tio = clone->end_io_data; |
ba1cbad9 | 867 | dm_request_endio_fn rq_end_io = NULL; |
cec47e3d | 868 | |
ba1cbad9 MS |
869 | if (tio->ti) { |
870 | rq_end_io = tio->ti->type->rq_end_io; | |
871 | ||
872 | if (mapped && rq_end_io) | |
873 | r = rq_end_io(tio->ti, clone, error, &tio->info); | |
874 | } | |
cec47e3d | 875 | |
11a68244 | 876 | if (r <= 0) |
cec47e3d | 877 | /* The target wants to complete the I/O */ |
11a68244 KU |
878 | dm_end_request(clone, r); |
879 | else if (r == DM_ENDIO_INCOMPLETE) | |
cec47e3d KU |
880 | /* The target will handle the I/O */ |
881 | return; | |
11a68244 | 882 | else if (r == DM_ENDIO_REQUEUE) |
cec47e3d KU |
883 | /* The target wants to requeue the I/O */ |
884 | dm_requeue_unmapped_request(clone); | |
885 | else { | |
11a68244 | 886 | DMWARN("unimplemented target endio return value: %d", r); |
cec47e3d KU |
887 | BUG(); |
888 | } | |
889 | } | |
890 | ||
11a68244 KU |
891 | /* |
892 | * Request completion handler for request-based dm | |
893 | */ | |
894 | static void dm_softirq_done(struct request *rq) | |
895 | { | |
896 | bool mapped = true; | |
897 | struct request *clone = rq->completion_data; | |
898 | struct dm_rq_target_io *tio = clone->end_io_data; | |
899 | ||
900 | if (rq->cmd_flags & REQ_FAILED) | |
901 | mapped = false; | |
902 | ||
903 | dm_done(clone, tio->error, mapped); | |
904 | } | |
905 | ||
cec47e3d KU |
906 | /* |
907 | * Complete the clone and the original request with the error status | |
908 | * through softirq context. | |
909 | */ | |
910 | static void dm_complete_request(struct request *clone, int error) | |
911 | { | |
912 | struct dm_rq_target_io *tio = clone->end_io_data; | |
913 | struct request *rq = tio->orig; | |
914 | ||
915 | tio->error = error; | |
916 | rq->completion_data = clone; | |
917 | blk_complete_request(rq); | |
918 | } | |
919 | ||
920 | /* | |
921 | * Complete the not-mapped clone and the original request with the error status | |
922 | * through softirq context. | |
923 | * Target's rq_end_io() function isn't called. | |
924 | * This may be used when the target's map_rq() function fails. | |
925 | */ | |
926 | void dm_kill_unmapped_request(struct request *clone, int error) | |
927 | { | |
928 | struct dm_rq_target_io *tio = clone->end_io_data; | |
929 | struct request *rq = tio->orig; | |
930 | ||
931 | rq->cmd_flags |= REQ_FAILED; | |
932 | dm_complete_request(clone, error); | |
933 | } | |
934 | EXPORT_SYMBOL_GPL(dm_kill_unmapped_request); | |
935 | ||
936 | /* | |
937 | * Called with the queue lock held | |
938 | */ | |
939 | static void end_clone_request(struct request *clone, int error) | |
940 | { | |
941 | /* | |
942 | * For just cleaning up the information of the queue in which | |
943 | * the clone was dispatched. | |
944 | * The clone is *NOT* freed actually here because it is alloced from | |
945 | * dm own mempool and REQ_ALLOCED isn't set in clone->cmd_flags. | |
946 | */ | |
947 | __blk_put_request(clone->q, clone); | |
948 | ||
949 | /* | |
950 | * Actual request completion is done in a softirq context which doesn't | |
951 | * hold the queue lock. Otherwise, deadlock could occur because: | |
952 | * - another request may be submitted by the upper level driver | |
953 | * of the stacking during the completion | |
954 | * - the submission which requires queue lock may be done | |
955 | * against this queue | |
956 | */ | |
957 | dm_complete_request(clone, error); | |
958 | } | |
959 | ||
56a67df7 MS |
960 | /* |
961 | * Return maximum size of I/O possible at the supplied sector up to the current | |
962 | * target boundary. | |
963 | */ | |
964 | static sector_t max_io_len_target_boundary(sector_t sector, struct dm_target *ti) | |
965 | { | |
966 | sector_t target_offset = dm_target_offset(ti, sector); | |
967 | ||
968 | return ti->len - target_offset; | |
969 | } | |
970 | ||
971 | static sector_t max_io_len(sector_t sector, struct dm_target *ti) | |
1da177e4 | 972 | { |
56a67df7 | 973 | sector_t len = max_io_len_target_boundary(sector, ti); |
542f9038 | 974 | sector_t offset, max_len; |
1da177e4 LT |
975 | |
976 | /* | |
542f9038 | 977 | * Does the target need to split even further? |
1da177e4 | 978 | */ |
542f9038 MS |
979 | if (ti->max_io_len) { |
980 | offset = dm_target_offset(ti, sector); | |
981 | if (unlikely(ti->max_io_len & (ti->max_io_len - 1))) | |
982 | max_len = sector_div(offset, ti->max_io_len); | |
983 | else | |
984 | max_len = offset & (ti->max_io_len - 1); | |
985 | max_len = ti->max_io_len - max_len; | |
986 | ||
987 | if (len > max_len) | |
988 | len = max_len; | |
1da177e4 LT |
989 | } |
990 | ||
991 | return len; | |
992 | } | |
993 | ||
542f9038 MS |
994 | int dm_set_target_max_io_len(struct dm_target *ti, sector_t len) |
995 | { | |
996 | if (len > UINT_MAX) { | |
997 | DMERR("Specified maximum size of target IO (%llu) exceeds limit (%u)", | |
998 | (unsigned long long)len, UINT_MAX); | |
999 | ti->error = "Maximum size of target IO is too large"; | |
1000 | return -EINVAL; | |
1001 | } | |
1002 | ||
1003 | ti->max_io_len = (uint32_t) len; | |
1004 | ||
1005 | return 0; | |
1006 | } | |
1007 | EXPORT_SYMBOL_GPL(dm_set_target_max_io_len); | |
1008 | ||
bd2a49b8 | 1009 | static void __map_bio(struct dm_target_io *tio) |
1da177e4 LT |
1010 | { |
1011 | int r; | |
2056a782 | 1012 | sector_t sector; |
9faf400f | 1013 | struct mapped_device *md; |
dba14160 | 1014 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1015 | struct dm_target *ti = tio->ti; |
1da177e4 | 1016 | |
1da177e4 LT |
1017 | clone->bi_end_io = clone_endio; |
1018 | clone->bi_private = tio; | |
1019 | ||
1020 | /* | |
1021 | * Map the clone. If r == 0 we don't need to do | |
1022 | * anything, the target has assumed ownership of | |
1023 | * this io. | |
1024 | */ | |
1025 | atomic_inc(&tio->io->io_count); | |
2056a782 | 1026 | sector = clone->bi_sector; |
7de3ee57 | 1027 | r = ti->type->map(ti, clone); |
45cbcd79 | 1028 | if (r == DM_MAPIO_REMAPPED) { |
1da177e4 | 1029 | /* the bio has been remapped so dispatch it */ |
2056a782 | 1030 | |
d07335e5 MS |
1031 | trace_block_bio_remap(bdev_get_queue(clone->bi_bdev), clone, |
1032 | tio->io->bio->bi_bdev->bd_dev, sector); | |
2056a782 | 1033 | |
1da177e4 | 1034 | generic_make_request(clone); |
2e93ccc1 KU |
1035 | } else if (r < 0 || r == DM_MAPIO_REQUEUE) { |
1036 | /* error the io and bail out, or requeue it if needed */ | |
9faf400f SB |
1037 | md = tio->io->md; |
1038 | dec_pending(tio->io, r); | |
9faf400f | 1039 | free_tio(md, tio); |
45cbcd79 KU |
1040 | } else if (r) { |
1041 | DMWARN("unimplemented target map return value: %d", r); | |
1042 | BUG(); | |
1da177e4 LT |
1043 | } |
1044 | } | |
1045 | ||
1046 | struct clone_info { | |
1047 | struct mapped_device *md; | |
1048 | struct dm_table *map; | |
1049 | struct bio *bio; | |
1050 | struct dm_io *io; | |
1051 | sector_t sector; | |
1052 | sector_t sector_count; | |
1053 | unsigned short idx; | |
1054 | }; | |
1055 | ||
bd2a49b8 AK |
1056 | static void bio_setup_sector(struct bio *bio, sector_t sector, sector_t len) |
1057 | { | |
1058 | bio->bi_sector = sector; | |
1059 | bio->bi_size = to_bytes(len); | |
1060 | } | |
1061 | ||
1062 | static void bio_setup_bv(struct bio *bio, unsigned short idx, unsigned short bv_count) | |
1063 | { | |
1064 | bio->bi_idx = idx; | |
1065 | bio->bi_vcnt = idx + bv_count; | |
1066 | bio->bi_flags &= ~(1 << BIO_SEG_VALID); | |
1067 | } | |
1068 | ||
1069 | static void clone_bio_integrity(struct bio *bio, struct bio *clone, | |
1070 | unsigned short idx, unsigned len, unsigned offset, | |
1071 | unsigned trim) | |
1072 | { | |
1073 | if (!bio_integrity(bio)) | |
1074 | return; | |
1075 | ||
1076 | bio_integrity_clone(clone, bio, GFP_NOIO); | |
1077 | ||
1078 | if (trim) | |
1079 | bio_integrity_trim(clone, bio_sector_offset(bio, idx, offset), len); | |
1080 | } | |
1081 | ||
1da177e4 | 1082 | /* |
d87f4c14 | 1083 | * Creates a little bio that just does part of a bvec. |
1da177e4 | 1084 | */ |
14fe594d AK |
1085 | static void clone_split_bio(struct dm_target_io *tio, struct bio *bio, |
1086 | sector_t sector, unsigned short idx, | |
1087 | unsigned offset, unsigned len) | |
1da177e4 | 1088 | { |
dba14160 | 1089 | struct bio *clone = &tio->clone; |
1da177e4 LT |
1090 | struct bio_vec *bv = bio->bi_io_vec + idx; |
1091 | ||
1da177e4 LT |
1092 | *clone->bi_io_vec = *bv; |
1093 | ||
bd2a49b8 AK |
1094 | bio_setup_sector(clone, sector, len); |
1095 | ||
1da177e4 | 1096 | clone->bi_bdev = bio->bi_bdev; |
d87f4c14 | 1097 | clone->bi_rw = bio->bi_rw; |
1da177e4 | 1098 | clone->bi_vcnt = 1; |
1da177e4 LT |
1099 | clone->bi_io_vec->bv_offset = offset; |
1100 | clone->bi_io_vec->bv_len = clone->bi_size; | |
f3e1d26e | 1101 | clone->bi_flags |= 1 << BIO_CLONED; |
1da177e4 | 1102 | |
bd2a49b8 | 1103 | clone_bio_integrity(bio, clone, idx, len, offset, 1); |
1da177e4 LT |
1104 | } |
1105 | ||
1106 | /* | |
1107 | * Creates a bio that consists of range of complete bvecs. | |
1108 | */ | |
dba14160 MP |
1109 | static void clone_bio(struct dm_target_io *tio, struct bio *bio, |
1110 | sector_t sector, unsigned short idx, | |
e4c93811 | 1111 | unsigned short bv_count, unsigned len) |
1da177e4 | 1112 | { |
dba14160 | 1113 | struct bio *clone = &tio->clone; |
bd2a49b8 | 1114 | unsigned trim = 0; |
1da177e4 | 1115 | |
9faf400f | 1116 | __bio_clone(clone, bio); |
bd2a49b8 AK |
1117 | bio_setup_sector(clone, sector, len); |
1118 | bio_setup_bv(clone, idx, bv_count); | |
1119 | ||
1120 | if (idx != bio->bi_idx || clone->bi_size < bio->bi_size) | |
1121 | trim = 1; | |
1122 | clone_bio_integrity(bio, clone, idx, len, 0, trim); | |
1da177e4 LT |
1123 | } |
1124 | ||
9015df24 | 1125 | static struct dm_target_io *alloc_tio(struct clone_info *ci, |
bd2a49b8 | 1126 | struct dm_target *ti, int nr_iovecs, |
55a62eef | 1127 | unsigned target_bio_nr) |
f9ab94ce | 1128 | { |
dba14160 MP |
1129 | struct dm_target_io *tio; |
1130 | struct bio *clone; | |
1131 | ||
1132 | clone = bio_alloc_bioset(GFP_NOIO, nr_iovecs, ci->md->bs); | |
1133 | tio = container_of(clone, struct dm_target_io, clone); | |
f9ab94ce MP |
1134 | |
1135 | tio->io = ci->io; | |
1136 | tio->ti = ti; | |
f9ab94ce | 1137 | memset(&tio->info, 0, sizeof(tio->info)); |
55a62eef | 1138 | tio->target_bio_nr = target_bio_nr; |
9015df24 AK |
1139 | |
1140 | return tio; | |
1141 | } | |
1142 | ||
14fe594d AK |
1143 | static void __clone_and_map_simple_bio(struct clone_info *ci, |
1144 | struct dm_target *ti, | |
1145 | unsigned target_bio_nr, sector_t len) | |
9015df24 | 1146 | { |
55a62eef | 1147 | struct dm_target_io *tio = alloc_tio(ci, ti, ci->bio->bi_max_vecs, target_bio_nr); |
dba14160 | 1148 | struct bio *clone = &tio->clone; |
9015df24 | 1149 | |
06a426ce MS |
1150 | /* |
1151 | * Discard requests require the bio's inline iovecs be initialized. | |
1152 | * ci->bio->bi_max_vecs is BIO_INLINE_VECS anyway, for both flush | |
1153 | * and discard, so no need for concern about wasted bvec allocations. | |
1154 | */ | |
dba14160 | 1155 | __bio_clone(clone, ci->bio); |
bd2a49b8 AK |
1156 | if (len) |
1157 | bio_setup_sector(clone, ci->sector, len); | |
f9ab94ce | 1158 | |
bd2a49b8 | 1159 | __map_bio(tio); |
f9ab94ce MP |
1160 | } |
1161 | ||
14fe594d AK |
1162 | static void __send_duplicate_bios(struct clone_info *ci, struct dm_target *ti, |
1163 | unsigned num_bios, sector_t len) | |
06a426ce | 1164 | { |
55a62eef | 1165 | unsigned target_bio_nr; |
06a426ce | 1166 | |
55a62eef | 1167 | for (target_bio_nr = 0; target_bio_nr < num_bios; target_bio_nr++) |
14fe594d | 1168 | __clone_and_map_simple_bio(ci, ti, target_bio_nr, len); |
06a426ce MS |
1169 | } |
1170 | ||
14fe594d | 1171 | static int __send_empty_flush(struct clone_info *ci) |
f9ab94ce | 1172 | { |
06a426ce | 1173 | unsigned target_nr = 0; |
f9ab94ce MP |
1174 | struct dm_target *ti; |
1175 | ||
b372d360 | 1176 | BUG_ON(bio_has_data(ci->bio)); |
f9ab94ce | 1177 | while ((ti = dm_table_get_target(ci->map, target_nr++))) |
14fe594d | 1178 | __send_duplicate_bios(ci, ti, ti->num_flush_bios, 0); |
f9ab94ce | 1179 | |
f9ab94ce MP |
1180 | return 0; |
1181 | } | |
1182 | ||
e4c93811 AK |
1183 | static void __clone_and_map_data_bio(struct clone_info *ci, struct dm_target *ti, |
1184 | sector_t sector, int nr_iovecs, | |
1185 | unsigned short idx, unsigned short bv_count, | |
1186 | unsigned offset, unsigned len, | |
1187 | unsigned split_bvec) | |
5ae89a87 | 1188 | { |
dba14160 | 1189 | struct bio *bio = ci->bio; |
5ae89a87 | 1190 | struct dm_target_io *tio; |
b0d8ed4d AK |
1191 | unsigned target_bio_nr; |
1192 | unsigned num_target_bios = 1; | |
5ae89a87 | 1193 | |
b0d8ed4d AK |
1194 | /* |
1195 | * Does the target want to receive duplicate copies of the bio? | |
1196 | */ | |
1197 | if (bio_data_dir(bio) == WRITE && ti->num_write_bios) | |
1198 | num_target_bios = ti->num_write_bios(ti, bio); | |
e4c93811 | 1199 | |
b0d8ed4d AK |
1200 | for (target_bio_nr = 0; target_bio_nr < num_target_bios; target_bio_nr++) { |
1201 | tio = alloc_tio(ci, ti, nr_iovecs, target_bio_nr); | |
1202 | if (split_bvec) | |
1203 | clone_split_bio(tio, bio, sector, idx, offset, len); | |
1204 | else | |
1205 | clone_bio(tio, bio, sector, idx, bv_count, len); | |
1206 | __map_bio(tio); | |
1207 | } | |
5ae89a87 MS |
1208 | } |
1209 | ||
55a62eef | 1210 | typedef unsigned (*get_num_bios_fn)(struct dm_target *ti); |
23508a96 | 1211 | |
55a62eef | 1212 | static unsigned get_num_discard_bios(struct dm_target *ti) |
23508a96 | 1213 | { |
55a62eef | 1214 | return ti->num_discard_bios; |
23508a96 MS |
1215 | } |
1216 | ||
55a62eef | 1217 | static unsigned get_num_write_same_bios(struct dm_target *ti) |
23508a96 | 1218 | { |
55a62eef | 1219 | return ti->num_write_same_bios; |
23508a96 MS |
1220 | } |
1221 | ||
1222 | typedef bool (*is_split_required_fn)(struct dm_target *ti); | |
1223 | ||
1224 | static bool is_split_required_for_discard(struct dm_target *ti) | |
1225 | { | |
55a62eef | 1226 | return ti->split_discard_bios; |
23508a96 MS |
1227 | } |
1228 | ||
14fe594d AK |
1229 | static int __send_changing_extent_only(struct clone_info *ci, |
1230 | get_num_bios_fn get_num_bios, | |
1231 | is_split_required_fn is_split_required) | |
5ae89a87 MS |
1232 | { |
1233 | struct dm_target *ti; | |
a79245b3 | 1234 | sector_t len; |
55a62eef | 1235 | unsigned num_bios; |
5ae89a87 | 1236 | |
a79245b3 MS |
1237 | do { |
1238 | ti = dm_table_find_target(ci->map, ci->sector); | |
1239 | if (!dm_target_is_valid(ti)) | |
1240 | return -EIO; | |
5ae89a87 | 1241 | |
5ae89a87 | 1242 | /* |
23508a96 MS |
1243 | * Even though the device advertised support for this type of |
1244 | * request, that does not mean every target supports it, and | |
936688d7 | 1245 | * reconfiguration might also have changed that since the |
a79245b3 | 1246 | * check was performed. |
5ae89a87 | 1247 | */ |
55a62eef AK |
1248 | num_bios = get_num_bios ? get_num_bios(ti) : 0; |
1249 | if (!num_bios) | |
a79245b3 | 1250 | return -EOPNOTSUPP; |
5ae89a87 | 1251 | |
23508a96 | 1252 | if (is_split_required && !is_split_required(ti)) |
7acf0277 MP |
1253 | len = min(ci->sector_count, max_io_len_target_boundary(ci->sector, ti)); |
1254 | else | |
1255 | len = min(ci->sector_count, max_io_len(ci->sector, ti)); | |
06a426ce | 1256 | |
14fe594d | 1257 | __send_duplicate_bios(ci, ti, num_bios, len); |
a79245b3 MS |
1258 | |
1259 | ci->sector += len; | |
1260 | } while (ci->sector_count -= len); | |
5ae89a87 MS |
1261 | |
1262 | return 0; | |
1263 | } | |
1264 | ||
14fe594d | 1265 | static int __send_discard(struct clone_info *ci) |
23508a96 | 1266 | { |
14fe594d AK |
1267 | return __send_changing_extent_only(ci, get_num_discard_bios, |
1268 | is_split_required_for_discard); | |
23508a96 MS |
1269 | } |
1270 | ||
14fe594d | 1271 | static int __send_write_same(struct clone_info *ci) |
23508a96 | 1272 | { |
14fe594d | 1273 | return __send_changing_extent_only(ci, get_num_write_same_bios, NULL); |
23508a96 MS |
1274 | } |
1275 | ||
e4c93811 AK |
1276 | /* |
1277 | * Find maximum number of sectors / bvecs we can process with a single bio. | |
1278 | */ | |
1279 | static sector_t __len_within_target(struct clone_info *ci, sector_t max, int *idx) | |
1280 | { | |
1281 | struct bio *bio = ci->bio; | |
1282 | sector_t bv_len, total_len = 0; | |
1283 | ||
1284 | for (*idx = ci->idx; max && (*idx < bio->bi_vcnt); (*idx)++) { | |
1285 | bv_len = to_sector(bio->bi_io_vec[*idx].bv_len); | |
1286 | ||
1287 | if (bv_len > max) | |
1288 | break; | |
1289 | ||
1290 | max -= bv_len; | |
1291 | total_len += bv_len; | |
1292 | } | |
1293 | ||
1294 | return total_len; | |
1295 | } | |
1296 | ||
1297 | static int __split_bvec_across_targets(struct clone_info *ci, | |
1298 | struct dm_target *ti, sector_t max) | |
1299 | { | |
1300 | struct bio *bio = ci->bio; | |
1301 | struct bio_vec *bv = bio->bi_io_vec + ci->idx; | |
1302 | sector_t remaining = to_sector(bv->bv_len); | |
1303 | unsigned offset = 0; | |
1304 | sector_t len; | |
1305 | ||
1306 | do { | |
1307 | if (offset) { | |
1308 | ti = dm_table_find_target(ci->map, ci->sector); | |
1309 | if (!dm_target_is_valid(ti)) | |
1310 | return -EIO; | |
1311 | ||
1312 | max = max_io_len(ci->sector, ti); | |
1313 | } | |
1314 | ||
1315 | len = min(remaining, max); | |
1316 | ||
1317 | __clone_and_map_data_bio(ci, ti, ci->sector, 1, ci->idx, 0, | |
1318 | bv->bv_offset + offset, len, 1); | |
1319 | ||
1320 | ci->sector += len; | |
1321 | ci->sector_count -= len; | |
1322 | offset += to_bytes(len); | |
1323 | } while (remaining -= len); | |
1324 | ||
1325 | ci->idx++; | |
1326 | ||
1327 | return 0; | |
1328 | } | |
1329 | ||
1330 | /* | |
1331 | * Select the correct strategy for processing a non-flush bio. | |
1332 | */ | |
14fe594d | 1333 | static int __split_and_process_non_flush(struct clone_info *ci) |
1da177e4 | 1334 | { |
dba14160 | 1335 | struct bio *bio = ci->bio; |
512875bd | 1336 | struct dm_target *ti; |
e4c93811 AK |
1337 | sector_t len, max; |
1338 | int idx; | |
1da177e4 | 1339 | |
5ae89a87 | 1340 | if (unlikely(bio->bi_rw & REQ_DISCARD)) |
14fe594d | 1341 | return __send_discard(ci); |
23508a96 | 1342 | else if (unlikely(bio->bi_rw & REQ_WRITE_SAME)) |
14fe594d | 1343 | return __send_write_same(ci); |
5ae89a87 | 1344 | |
512875bd JN |
1345 | ti = dm_table_find_target(ci->map, ci->sector); |
1346 | if (!dm_target_is_valid(ti)) | |
1347 | return -EIO; | |
1348 | ||
56a67df7 | 1349 | max = max_io_len(ci->sector, ti); |
512875bd | 1350 | |
e4c93811 AK |
1351 | /* |
1352 | * Optimise for the simple case where we can do all of | |
1353 | * the remaining io with a single clone. | |
1354 | */ | |
1da177e4 | 1355 | if (ci->sector_count <= max) { |
e4c93811 AK |
1356 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1357 | ci->idx, bio->bi_vcnt - ci->idx, 0, | |
1358 | ci->sector_count, 0); | |
1359 | ci->sector_count = 0; | |
1360 | return 0; | |
1361 | } | |
1da177e4 | 1362 | |
e4c93811 AK |
1363 | /* |
1364 | * There are some bvecs that don't span targets. | |
1365 | * Do as many of these as possible. | |
1366 | */ | |
1367 | if (to_sector(bio->bi_io_vec[ci->idx].bv_len) <= max) { | |
1368 | len = __len_within_target(ci, max, &idx); | |
1da177e4 | 1369 | |
e4c93811 AK |
1370 | __clone_and_map_data_bio(ci, ti, ci->sector, bio->bi_max_vecs, |
1371 | ci->idx, idx - ci->idx, 0, len, 0); | |
1da177e4 LT |
1372 | |
1373 | ci->sector += len; | |
1374 | ci->sector_count -= len; | |
e4c93811 | 1375 | ci->idx = idx; |
d2044a94 | 1376 | |
e4c93811 | 1377 | return 0; |
1da177e4 | 1378 | } |
512875bd | 1379 | |
e4c93811 AK |
1380 | /* |
1381 | * Handle a bvec that must be split between two or more targets. | |
1382 | */ | |
1383 | return __split_bvec_across_targets(ci, ti, max); | |
1da177e4 LT |
1384 | } |
1385 | ||
1386 | /* | |
14fe594d | 1387 | * Entry point to split a bio into clones and submit them to the targets. |
1da177e4 | 1388 | */ |
83d5e5b0 MP |
1389 | static void __split_and_process_bio(struct mapped_device *md, |
1390 | struct dm_table *map, struct bio *bio) | |
1da177e4 LT |
1391 | { |
1392 | struct clone_info ci; | |
512875bd | 1393 | int error = 0; |
1da177e4 | 1394 | |
83d5e5b0 | 1395 | if (unlikely(!map)) { |
6a8736d1 | 1396 | bio_io_error(bio); |
f0b9a450 MP |
1397 | return; |
1398 | } | |
692d0eb9 | 1399 | |
83d5e5b0 | 1400 | ci.map = map; |
1da177e4 | 1401 | ci.md = md; |
1da177e4 LT |
1402 | ci.io = alloc_io(md); |
1403 | ci.io->error = 0; | |
1404 | atomic_set(&ci.io->io_count, 1); | |
1405 | ci.io->bio = bio; | |
1406 | ci.io->md = md; | |
f88fb981 | 1407 | spin_lock_init(&ci.io->endio_lock); |
1da177e4 | 1408 | ci.sector = bio->bi_sector; |
1da177e4 LT |
1409 | ci.idx = bio->bi_idx; |
1410 | ||
3eaf840e | 1411 | start_io_acct(ci.io); |
bd2a49b8 | 1412 | |
b372d360 MS |
1413 | if (bio->bi_rw & REQ_FLUSH) { |
1414 | ci.bio = &ci.md->flush_bio; | |
1415 | ci.sector_count = 0; | |
14fe594d | 1416 | error = __send_empty_flush(&ci); |
b372d360 MS |
1417 | /* dec_pending submits any data associated with flush */ |
1418 | } else { | |
6a8736d1 | 1419 | ci.bio = bio; |
d87f4c14 | 1420 | ci.sector_count = bio_sectors(bio); |
b372d360 | 1421 | while (ci.sector_count && !error) |
14fe594d | 1422 | error = __split_and_process_non_flush(&ci); |
d87f4c14 | 1423 | } |
1da177e4 LT |
1424 | |
1425 | /* drop the extra reference count */ | |
512875bd | 1426 | dec_pending(ci.io, error); |
1da177e4 LT |
1427 | } |
1428 | /*----------------------------------------------------------------- | |
1429 | * CRUD END | |
1430 | *---------------------------------------------------------------*/ | |
1431 | ||
f6fccb12 MB |
1432 | static int dm_merge_bvec(struct request_queue *q, |
1433 | struct bvec_merge_data *bvm, | |
1434 | struct bio_vec *biovec) | |
1435 | { | |
1436 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1437 | struct dm_table *map = dm_get_live_table_fast(md); |
f6fccb12 MB |
1438 | struct dm_target *ti; |
1439 | sector_t max_sectors; | |
5037108a | 1440 | int max_size = 0; |
f6fccb12 MB |
1441 | |
1442 | if (unlikely(!map)) | |
5037108a | 1443 | goto out; |
f6fccb12 MB |
1444 | |
1445 | ti = dm_table_find_target(map, bvm->bi_sector); | |
b01cd5ac | 1446 | if (!dm_target_is_valid(ti)) |
83d5e5b0 | 1447 | goto out; |
f6fccb12 MB |
1448 | |
1449 | /* | |
1450 | * Find maximum amount of I/O that won't need splitting | |
1451 | */ | |
56a67df7 | 1452 | max_sectors = min(max_io_len(bvm->bi_sector, ti), |
f6fccb12 MB |
1453 | (sector_t) BIO_MAX_SECTORS); |
1454 | max_size = (max_sectors << SECTOR_SHIFT) - bvm->bi_size; | |
1455 | if (max_size < 0) | |
1456 | max_size = 0; | |
1457 | ||
1458 | /* | |
1459 | * merge_bvec_fn() returns number of bytes | |
1460 | * it can accept at this offset | |
1461 | * max is precomputed maximal io size | |
1462 | */ | |
1463 | if (max_size && ti->type->merge) | |
1464 | max_size = ti->type->merge(ti, bvm, biovec, max_size); | |
8cbeb67a MP |
1465 | /* |
1466 | * If the target doesn't support merge method and some of the devices | |
1467 | * provided their merge_bvec method (we know this by looking at | |
1468 | * queue_max_hw_sectors), then we can't allow bios with multiple vector | |
1469 | * entries. So always set max_size to 0, and the code below allows | |
1470 | * just one page. | |
1471 | */ | |
1472 | else if (queue_max_hw_sectors(q) <= PAGE_SIZE >> 9) | |
1473 | ||
1474 | max_size = 0; | |
f6fccb12 | 1475 | |
5037108a | 1476 | out: |
83d5e5b0 | 1477 | dm_put_live_table_fast(md); |
f6fccb12 MB |
1478 | /* |
1479 | * Always allow an entire first page | |
1480 | */ | |
1481 | if (max_size <= biovec->bv_len && !(bvm->bi_size >> SECTOR_SHIFT)) | |
1482 | max_size = biovec->bv_len; | |
1483 | ||
f6fccb12 MB |
1484 | return max_size; |
1485 | } | |
1486 | ||
1da177e4 LT |
1487 | /* |
1488 | * The request function that just remaps the bio built up by | |
1489 | * dm_merge_bvec. | |
1490 | */ | |
5a7bbad2 | 1491 | static void _dm_request(struct request_queue *q, struct bio *bio) |
1da177e4 | 1492 | { |
12f03a49 | 1493 | int rw = bio_data_dir(bio); |
1da177e4 | 1494 | struct mapped_device *md = q->queuedata; |
c9959059 | 1495 | int cpu; |
83d5e5b0 MP |
1496 | int srcu_idx; |
1497 | struct dm_table *map; | |
1da177e4 | 1498 | |
83d5e5b0 | 1499 | map = dm_get_live_table(md, &srcu_idx); |
1da177e4 | 1500 | |
074a7aca TH |
1501 | cpu = part_stat_lock(); |
1502 | part_stat_inc(cpu, &dm_disk(md)->part0, ios[rw]); | |
1503 | part_stat_add(cpu, &dm_disk(md)->part0, sectors[rw], bio_sectors(bio)); | |
1504 | part_stat_unlock(); | |
12f03a49 | 1505 | |
6a8736d1 TH |
1506 | /* if we're suspended, we have to queue this io for later */ |
1507 | if (unlikely(test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags))) { | |
83d5e5b0 | 1508 | dm_put_live_table(md, srcu_idx); |
1da177e4 | 1509 | |
6a8736d1 TH |
1510 | if (bio_rw(bio) != READA) |
1511 | queue_io(md, bio); | |
1512 | else | |
54d9a1b4 | 1513 | bio_io_error(bio); |
5a7bbad2 | 1514 | return; |
1da177e4 LT |
1515 | } |
1516 | ||
83d5e5b0 MP |
1517 | __split_and_process_bio(md, map, bio); |
1518 | dm_put_live_table(md, srcu_idx); | |
5a7bbad2 | 1519 | return; |
cec47e3d KU |
1520 | } |
1521 | ||
1522 | static int dm_request_based(struct mapped_device *md) | |
1523 | { | |
1524 | return blk_queue_stackable(md->queue); | |
1525 | } | |
1526 | ||
5a7bbad2 | 1527 | static void dm_request(struct request_queue *q, struct bio *bio) |
cec47e3d KU |
1528 | { |
1529 | struct mapped_device *md = q->queuedata; | |
1530 | ||
1531 | if (dm_request_based(md)) | |
5a7bbad2 CH |
1532 | blk_queue_bio(q, bio); |
1533 | else | |
1534 | _dm_request(q, bio); | |
cec47e3d KU |
1535 | } |
1536 | ||
1537 | void dm_dispatch_request(struct request *rq) | |
1538 | { | |
1539 | int r; | |
1540 | ||
1541 | if (blk_queue_io_stat(rq->q)) | |
1542 | rq->cmd_flags |= REQ_IO_STAT; | |
1543 | ||
1544 | rq->start_time = jiffies; | |
1545 | r = blk_insert_cloned_request(rq->q, rq); | |
1546 | if (r) | |
1547 | dm_complete_request(rq, r); | |
1548 | } | |
1549 | EXPORT_SYMBOL_GPL(dm_dispatch_request); | |
1550 | ||
cec47e3d KU |
1551 | static int dm_rq_bio_constructor(struct bio *bio, struct bio *bio_orig, |
1552 | void *data) | |
1553 | { | |
1554 | struct dm_rq_target_io *tio = data; | |
94818742 KO |
1555 | struct dm_rq_clone_bio_info *info = |
1556 | container_of(bio, struct dm_rq_clone_bio_info, clone); | |
cec47e3d KU |
1557 | |
1558 | info->orig = bio_orig; | |
1559 | info->tio = tio; | |
1560 | bio->bi_end_io = end_clone_bio; | |
1561 | bio->bi_private = info; | |
cec47e3d KU |
1562 | |
1563 | return 0; | |
1564 | } | |
1565 | ||
1566 | static int setup_clone(struct request *clone, struct request *rq, | |
1567 | struct dm_rq_target_io *tio) | |
1568 | { | |
d0bcb878 | 1569 | int r; |
cec47e3d | 1570 | |
29e4013d TH |
1571 | r = blk_rq_prep_clone(clone, rq, tio->md->bs, GFP_ATOMIC, |
1572 | dm_rq_bio_constructor, tio); | |
1573 | if (r) | |
1574 | return r; | |
cec47e3d | 1575 | |
29e4013d TH |
1576 | clone->cmd = rq->cmd; |
1577 | clone->cmd_len = rq->cmd_len; | |
1578 | clone->sense = rq->sense; | |
1579 | clone->buffer = rq->buffer; | |
cec47e3d KU |
1580 | clone->end_io = end_clone_request; |
1581 | clone->end_io_data = tio; | |
1582 | ||
1583 | return 0; | |
1584 | } | |
1585 | ||
6facdaff KU |
1586 | static struct request *clone_rq(struct request *rq, struct mapped_device *md, |
1587 | gfp_t gfp_mask) | |
1588 | { | |
1589 | struct request *clone; | |
1590 | struct dm_rq_target_io *tio; | |
1591 | ||
1592 | tio = alloc_rq_tio(md, gfp_mask); | |
1593 | if (!tio) | |
1594 | return NULL; | |
1595 | ||
1596 | tio->md = md; | |
1597 | tio->ti = NULL; | |
1598 | tio->orig = rq; | |
1599 | tio->error = 0; | |
1600 | memset(&tio->info, 0, sizeof(tio->info)); | |
1601 | ||
1602 | clone = &tio->clone; | |
1603 | if (setup_clone(clone, rq, tio)) { | |
1604 | /* -ENOMEM */ | |
1605 | free_rq_tio(tio); | |
1606 | return NULL; | |
1607 | } | |
1608 | ||
1609 | return clone; | |
1610 | } | |
1611 | ||
cec47e3d KU |
1612 | /* |
1613 | * Called with the queue lock held. | |
1614 | */ | |
1615 | static int dm_prep_fn(struct request_queue *q, struct request *rq) | |
1616 | { | |
1617 | struct mapped_device *md = q->queuedata; | |
cec47e3d KU |
1618 | struct request *clone; |
1619 | ||
cec47e3d KU |
1620 | if (unlikely(rq->special)) { |
1621 | DMWARN("Already has something in rq->special."); | |
1622 | return BLKPREP_KILL; | |
1623 | } | |
1624 | ||
6facdaff KU |
1625 | clone = clone_rq(rq, md, GFP_ATOMIC); |
1626 | if (!clone) | |
cec47e3d | 1627 | return BLKPREP_DEFER; |
cec47e3d KU |
1628 | |
1629 | rq->special = clone; | |
1630 | rq->cmd_flags |= REQ_DONTPREP; | |
1631 | ||
1632 | return BLKPREP_OK; | |
1633 | } | |
1634 | ||
9eef87da KU |
1635 | /* |
1636 | * Returns: | |
1637 | * 0 : the request has been processed (not requeued) | |
1638 | * !0 : the request has been requeued | |
1639 | */ | |
1640 | static int map_request(struct dm_target *ti, struct request *clone, | |
1641 | struct mapped_device *md) | |
cec47e3d | 1642 | { |
9eef87da | 1643 | int r, requeued = 0; |
cec47e3d KU |
1644 | struct dm_rq_target_io *tio = clone->end_io_data; |
1645 | ||
cec47e3d KU |
1646 | tio->ti = ti; |
1647 | r = ti->type->map_rq(ti, clone, &tio->info); | |
1648 | switch (r) { | |
1649 | case DM_MAPIO_SUBMITTED: | |
1650 | /* The target has taken the I/O to submit by itself later */ | |
1651 | break; | |
1652 | case DM_MAPIO_REMAPPED: | |
1653 | /* The target has remapped the I/O so dispatch it */ | |
6db4ccd6 JN |
1654 | trace_block_rq_remap(clone->q, clone, disk_devt(dm_disk(md)), |
1655 | blk_rq_pos(tio->orig)); | |
cec47e3d KU |
1656 | dm_dispatch_request(clone); |
1657 | break; | |
1658 | case DM_MAPIO_REQUEUE: | |
1659 | /* The target wants to requeue the I/O */ | |
1660 | dm_requeue_unmapped_request(clone); | |
9eef87da | 1661 | requeued = 1; |
cec47e3d KU |
1662 | break; |
1663 | default: | |
1664 | if (r > 0) { | |
1665 | DMWARN("unimplemented target map return value: %d", r); | |
1666 | BUG(); | |
1667 | } | |
1668 | ||
1669 | /* The target wants to complete the I/O */ | |
1670 | dm_kill_unmapped_request(clone, r); | |
1671 | break; | |
1672 | } | |
9eef87da KU |
1673 | |
1674 | return requeued; | |
cec47e3d KU |
1675 | } |
1676 | ||
ba1cbad9 MS |
1677 | static struct request *dm_start_request(struct mapped_device *md, struct request *orig) |
1678 | { | |
1679 | struct request *clone; | |
1680 | ||
1681 | blk_start_request(orig); | |
1682 | clone = orig->special; | |
1683 | atomic_inc(&md->pending[rq_data_dir(clone)]); | |
1684 | ||
1685 | /* | |
1686 | * Hold the md reference here for the in-flight I/O. | |
1687 | * We can't rely on the reference count by device opener, | |
1688 | * because the device may be closed during the request completion | |
1689 | * when all bios are completed. | |
1690 | * See the comment in rq_completed() too. | |
1691 | */ | |
1692 | dm_get(md); | |
1693 | ||
1694 | return clone; | |
1695 | } | |
1696 | ||
cec47e3d KU |
1697 | /* |
1698 | * q->request_fn for request-based dm. | |
1699 | * Called with the queue lock held. | |
1700 | */ | |
1701 | static void dm_request_fn(struct request_queue *q) | |
1702 | { | |
1703 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 MP |
1704 | int srcu_idx; |
1705 | struct dm_table *map = dm_get_live_table(md, &srcu_idx); | |
cec47e3d | 1706 | struct dm_target *ti; |
b4324fee | 1707 | struct request *rq, *clone; |
29e4013d | 1708 | sector_t pos; |
cec47e3d KU |
1709 | |
1710 | /* | |
b4324fee KU |
1711 | * For suspend, check blk_queue_stopped() and increment |
1712 | * ->pending within a single queue_lock not to increment the | |
1713 | * number of in-flight I/Os after the queue is stopped in | |
1714 | * dm_suspend(). | |
cec47e3d | 1715 | */ |
7eaceacc | 1716 | while (!blk_queue_stopped(q)) { |
cec47e3d KU |
1717 | rq = blk_peek_request(q); |
1718 | if (!rq) | |
7eaceacc | 1719 | goto delay_and_out; |
cec47e3d | 1720 | |
29e4013d TH |
1721 | /* always use block 0 to find the target for flushes for now */ |
1722 | pos = 0; | |
1723 | if (!(rq->cmd_flags & REQ_FLUSH)) | |
1724 | pos = blk_rq_pos(rq); | |
1725 | ||
1726 | ti = dm_table_find_target(map, pos); | |
ba1cbad9 MS |
1727 | if (!dm_target_is_valid(ti)) { |
1728 | /* | |
1729 | * Must perform setup, that dm_done() requires, | |
1730 | * before calling dm_kill_unmapped_request | |
1731 | */ | |
1732 | DMERR_LIMIT("request attempted access beyond the end of device"); | |
1733 | clone = dm_start_request(md, rq); | |
1734 | dm_kill_unmapped_request(clone, -EIO); | |
1735 | continue; | |
1736 | } | |
d0bcb878 | 1737 | |
cec47e3d | 1738 | if (ti->type->busy && ti->type->busy(ti)) |
7eaceacc | 1739 | goto delay_and_out; |
cec47e3d | 1740 | |
ba1cbad9 | 1741 | clone = dm_start_request(md, rq); |
b4324fee | 1742 | |
cec47e3d | 1743 | spin_unlock(q->queue_lock); |
9eef87da KU |
1744 | if (map_request(ti, clone, md)) |
1745 | goto requeued; | |
1746 | ||
052189a2 KU |
1747 | BUG_ON(!irqs_disabled()); |
1748 | spin_lock(q->queue_lock); | |
cec47e3d KU |
1749 | } |
1750 | ||
1751 | goto out; | |
1752 | ||
9eef87da | 1753 | requeued: |
052189a2 KU |
1754 | BUG_ON(!irqs_disabled()); |
1755 | spin_lock(q->queue_lock); | |
9eef87da | 1756 | |
7eaceacc JA |
1757 | delay_and_out: |
1758 | blk_delay_queue(q, HZ / 10); | |
cec47e3d | 1759 | out: |
83d5e5b0 | 1760 | dm_put_live_table(md, srcu_idx); |
cec47e3d KU |
1761 | } |
1762 | ||
1763 | int dm_underlying_device_busy(struct request_queue *q) | |
1764 | { | |
1765 | return blk_lld_busy(q); | |
1766 | } | |
1767 | EXPORT_SYMBOL_GPL(dm_underlying_device_busy); | |
1768 | ||
1769 | static int dm_lld_busy(struct request_queue *q) | |
1770 | { | |
1771 | int r; | |
1772 | struct mapped_device *md = q->queuedata; | |
83d5e5b0 | 1773 | struct dm_table *map = dm_get_live_table_fast(md); |
cec47e3d KU |
1774 | |
1775 | if (!map || test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) | |
1776 | r = 1; | |
1777 | else | |
1778 | r = dm_table_any_busy_target(map); | |
1779 | ||
83d5e5b0 | 1780 | dm_put_live_table_fast(md); |
cec47e3d KU |
1781 | |
1782 | return r; | |
1783 | } | |
1784 | ||
1da177e4 LT |
1785 | static int dm_any_congested(void *congested_data, int bdi_bits) |
1786 | { | |
8a57dfc6 CS |
1787 | int r = bdi_bits; |
1788 | struct mapped_device *md = congested_data; | |
1789 | struct dm_table *map; | |
1da177e4 | 1790 | |
1eb787ec | 1791 | if (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
83d5e5b0 | 1792 | map = dm_get_live_table_fast(md); |
8a57dfc6 | 1793 | if (map) { |
cec47e3d KU |
1794 | /* |
1795 | * Request-based dm cares about only own queue for | |
1796 | * the query about congestion status of request_queue | |
1797 | */ | |
1798 | if (dm_request_based(md)) | |
1799 | r = md->queue->backing_dev_info.state & | |
1800 | bdi_bits; | |
1801 | else | |
1802 | r = dm_table_any_congested(map, bdi_bits); | |
8a57dfc6 | 1803 | } |
83d5e5b0 | 1804 | dm_put_live_table_fast(md); |
8a57dfc6 CS |
1805 | } |
1806 | ||
1da177e4 LT |
1807 | return r; |
1808 | } | |
1809 | ||
1810 | /*----------------------------------------------------------------- | |
1811 | * An IDR is used to keep track of allocated minor numbers. | |
1812 | *---------------------------------------------------------------*/ | |
2b06cfff | 1813 | static void free_minor(int minor) |
1da177e4 | 1814 | { |
f32c10b0 | 1815 | spin_lock(&_minor_lock); |
1da177e4 | 1816 | idr_remove(&_minor_idr, minor); |
f32c10b0 | 1817 | spin_unlock(&_minor_lock); |
1da177e4 LT |
1818 | } |
1819 | ||
1820 | /* | |
1821 | * See if the device with a specific minor # is free. | |
1822 | */ | |
cf13ab8e | 1823 | static int specific_minor(int minor) |
1da177e4 | 1824 | { |
c9d76be6 | 1825 | int r; |
1da177e4 LT |
1826 | |
1827 | if (minor >= (1 << MINORBITS)) | |
1828 | return -EINVAL; | |
1829 | ||
c9d76be6 | 1830 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1831 | spin_lock(&_minor_lock); |
1da177e4 | 1832 | |
c9d76be6 | 1833 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, minor, minor + 1, GFP_NOWAIT); |
1da177e4 | 1834 | |
f32c10b0 | 1835 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1836 | idr_preload_end(); |
1837 | if (r < 0) | |
1838 | return r == -ENOSPC ? -EBUSY : r; | |
1839 | return 0; | |
1da177e4 LT |
1840 | } |
1841 | ||
cf13ab8e | 1842 | static int next_free_minor(int *minor) |
1da177e4 | 1843 | { |
c9d76be6 | 1844 | int r; |
62f75c2f | 1845 | |
c9d76be6 | 1846 | idr_preload(GFP_KERNEL); |
f32c10b0 | 1847 | spin_lock(&_minor_lock); |
1da177e4 | 1848 | |
c9d76be6 | 1849 | r = idr_alloc(&_minor_idr, MINOR_ALLOCED, 0, 1 << MINORBITS, GFP_NOWAIT); |
1da177e4 | 1850 | |
f32c10b0 | 1851 | spin_unlock(&_minor_lock); |
c9d76be6 TH |
1852 | idr_preload_end(); |
1853 | if (r < 0) | |
1854 | return r; | |
1855 | *minor = r; | |
1856 | return 0; | |
1da177e4 LT |
1857 | } |
1858 | ||
83d5cde4 | 1859 | static const struct block_device_operations dm_blk_dops; |
1da177e4 | 1860 | |
53d5914f MP |
1861 | static void dm_wq_work(struct work_struct *work); |
1862 | ||
4a0b4ddf MS |
1863 | static void dm_init_md_queue(struct mapped_device *md) |
1864 | { | |
1865 | /* | |
1866 | * Request-based dm devices cannot be stacked on top of bio-based dm | |
1867 | * devices. The type of this dm device has not been decided yet. | |
1868 | * The type is decided at the first table loading time. | |
1869 | * To prevent problematic device stacking, clear the queue flag | |
1870 | * for request stacking support until then. | |
1871 | * | |
1872 | * This queue is new, so no concurrency on the queue_flags. | |
1873 | */ | |
1874 | queue_flag_clear_unlocked(QUEUE_FLAG_STACKABLE, md->queue); | |
1875 | ||
1876 | md->queue->queuedata = md; | |
1877 | md->queue->backing_dev_info.congested_fn = dm_any_congested; | |
1878 | md->queue->backing_dev_info.congested_data = md; | |
1879 | blk_queue_make_request(md->queue, dm_request); | |
1880 | blk_queue_bounce_limit(md->queue, BLK_BOUNCE_ANY); | |
4a0b4ddf MS |
1881 | blk_queue_merge_bvec(md->queue, dm_merge_bvec); |
1882 | } | |
1883 | ||
1da177e4 LT |
1884 | /* |
1885 | * Allocate and initialise a blank device with a given minor. | |
1886 | */ | |
2b06cfff | 1887 | static struct mapped_device *alloc_dev(int minor) |
1da177e4 LT |
1888 | { |
1889 | int r; | |
cf13ab8e | 1890 | struct mapped_device *md = kzalloc(sizeof(*md), GFP_KERNEL); |
ba61fdd1 | 1891 | void *old_md; |
1da177e4 LT |
1892 | |
1893 | if (!md) { | |
1894 | DMWARN("unable to allocate device, out of memory."); | |
1895 | return NULL; | |
1896 | } | |
1897 | ||
10da4f79 | 1898 | if (!try_module_get(THIS_MODULE)) |
6ed7ade8 | 1899 | goto bad_module_get; |
10da4f79 | 1900 | |
1da177e4 | 1901 | /* get a minor number for the dev */ |
2b06cfff | 1902 | if (minor == DM_ANY_MINOR) |
cf13ab8e | 1903 | r = next_free_minor(&minor); |
2b06cfff | 1904 | else |
cf13ab8e | 1905 | r = specific_minor(minor); |
1da177e4 | 1906 | if (r < 0) |
6ed7ade8 | 1907 | goto bad_minor; |
1da177e4 | 1908 | |
83d5e5b0 MP |
1909 | r = init_srcu_struct(&md->io_barrier); |
1910 | if (r < 0) | |
1911 | goto bad_io_barrier; | |
1912 | ||
a5664dad | 1913 | md->type = DM_TYPE_NONE; |
e61290a4 | 1914 | mutex_init(&md->suspend_lock); |
a5664dad | 1915 | mutex_init(&md->type_lock); |
022c2611 | 1916 | spin_lock_init(&md->deferred_lock); |
1da177e4 | 1917 | atomic_set(&md->holders, 1); |
5c6bd75d | 1918 | atomic_set(&md->open_count, 0); |
1da177e4 | 1919 | atomic_set(&md->event_nr, 0); |
7a8c3d3b MA |
1920 | atomic_set(&md->uevent_seq, 0); |
1921 | INIT_LIST_HEAD(&md->uevent_list); | |
1922 | spin_lock_init(&md->uevent_lock); | |
1da177e4 | 1923 | |
4a0b4ddf | 1924 | md->queue = blk_alloc_queue(GFP_KERNEL); |
1da177e4 | 1925 | if (!md->queue) |
6ed7ade8 | 1926 | goto bad_queue; |
1da177e4 | 1927 | |
4a0b4ddf | 1928 | dm_init_md_queue(md); |
9faf400f | 1929 | |
1da177e4 LT |
1930 | md->disk = alloc_disk(1); |
1931 | if (!md->disk) | |
6ed7ade8 | 1932 | goto bad_disk; |
1da177e4 | 1933 | |
316d315b NK |
1934 | atomic_set(&md->pending[0], 0); |
1935 | atomic_set(&md->pending[1], 0); | |
f0b04115 | 1936 | init_waitqueue_head(&md->wait); |
53d5914f | 1937 | INIT_WORK(&md->work, dm_wq_work); |
f0b04115 JM |
1938 | init_waitqueue_head(&md->eventq); |
1939 | ||
1da177e4 LT |
1940 | md->disk->major = _major; |
1941 | md->disk->first_minor = minor; | |
1942 | md->disk->fops = &dm_blk_dops; | |
1943 | md->disk->queue = md->queue; | |
1944 | md->disk->private_data = md; | |
1945 | sprintf(md->disk->disk_name, "dm-%d", minor); | |
1946 | add_disk(md->disk); | |
7e51f257 | 1947 | format_dev_t(md->name, MKDEV(_major, minor)); |
1da177e4 | 1948 | |
9c4376de TH |
1949 | md->wq = alloc_workqueue("kdmflush", |
1950 | WQ_NON_REENTRANT | WQ_MEM_RECLAIM, 0); | |
304f3f6a MB |
1951 | if (!md->wq) |
1952 | goto bad_thread; | |
1953 | ||
32a926da MP |
1954 | md->bdev = bdget_disk(md->disk, 0); |
1955 | if (!md->bdev) | |
1956 | goto bad_bdev; | |
1957 | ||
6a8736d1 TH |
1958 | bio_init(&md->flush_bio); |
1959 | md->flush_bio.bi_bdev = md->bdev; | |
1960 | md->flush_bio.bi_rw = WRITE_FLUSH; | |
1961 | ||
ba61fdd1 | 1962 | /* Populate the mapping, nobody knows we exist yet */ |
f32c10b0 | 1963 | spin_lock(&_minor_lock); |
ba61fdd1 | 1964 | old_md = idr_replace(&_minor_idr, md, minor); |
f32c10b0 | 1965 | spin_unlock(&_minor_lock); |
ba61fdd1 JM |
1966 | |
1967 | BUG_ON(old_md != MINOR_ALLOCED); | |
1968 | ||
1da177e4 LT |
1969 | return md; |
1970 | ||
32a926da MP |
1971 | bad_bdev: |
1972 | destroy_workqueue(md->wq); | |
304f3f6a | 1973 | bad_thread: |
03022c54 | 1974 | del_gendisk(md->disk); |
304f3f6a | 1975 | put_disk(md->disk); |
6ed7ade8 | 1976 | bad_disk: |
1312f40e | 1977 | blk_cleanup_queue(md->queue); |
6ed7ade8 | 1978 | bad_queue: |
83d5e5b0 MP |
1979 | cleanup_srcu_struct(&md->io_barrier); |
1980 | bad_io_barrier: | |
1da177e4 | 1981 | free_minor(minor); |
6ed7ade8 | 1982 | bad_minor: |
10da4f79 | 1983 | module_put(THIS_MODULE); |
6ed7ade8 | 1984 | bad_module_get: |
1da177e4 LT |
1985 | kfree(md); |
1986 | return NULL; | |
1987 | } | |
1988 | ||
ae9da83f JN |
1989 | static void unlock_fs(struct mapped_device *md); |
1990 | ||
1da177e4 LT |
1991 | static void free_dev(struct mapped_device *md) |
1992 | { | |
f331c029 | 1993 | int minor = MINOR(disk_devt(md->disk)); |
63d94e48 | 1994 | |
32a926da MP |
1995 | unlock_fs(md); |
1996 | bdput(md->bdev); | |
304f3f6a | 1997 | destroy_workqueue(md->wq); |
e6ee8c0b KU |
1998 | if (md->io_pool) |
1999 | mempool_destroy(md->io_pool); | |
2000 | if (md->bs) | |
2001 | bioset_free(md->bs); | |
9c47008d | 2002 | blk_integrity_unregister(md->disk); |
1da177e4 | 2003 | del_gendisk(md->disk); |
83d5e5b0 | 2004 | cleanup_srcu_struct(&md->io_barrier); |
63d94e48 | 2005 | free_minor(minor); |
fba9f90e JM |
2006 | |
2007 | spin_lock(&_minor_lock); | |
2008 | md->disk->private_data = NULL; | |
2009 | spin_unlock(&_minor_lock); | |
2010 | ||
1da177e4 | 2011 | put_disk(md->disk); |
1312f40e | 2012 | blk_cleanup_queue(md->queue); |
10da4f79 | 2013 | module_put(THIS_MODULE); |
1da177e4 LT |
2014 | kfree(md); |
2015 | } | |
2016 | ||
e6ee8c0b KU |
2017 | static void __bind_mempools(struct mapped_device *md, struct dm_table *t) |
2018 | { | |
c0820cf5 | 2019 | struct dm_md_mempools *p = dm_table_get_md_mempools(t); |
e6ee8c0b | 2020 | |
5f015204 | 2021 | if (md->io_pool && md->bs) { |
16245bdc JN |
2022 | /* The md already has necessary mempools. */ |
2023 | if (dm_table_get_type(t) == DM_TYPE_BIO_BASED) { | |
2024 | /* | |
2025 | * Reload bioset because front_pad may have changed | |
2026 | * because a different table was loaded. | |
2027 | */ | |
2028 | bioset_free(md->bs); | |
2029 | md->bs = p->bs; | |
2030 | p->bs = NULL; | |
2031 | } else if (dm_table_get_type(t) == DM_TYPE_REQUEST_BASED) { | |
16245bdc JN |
2032 | /* |
2033 | * There's no need to reload with request-based dm | |
2034 | * because the size of front_pad doesn't change. | |
2035 | * Note for future: If you are to reload bioset, | |
2036 | * prep-ed requests in the queue may refer | |
2037 | * to bio from the old bioset, so you must walk | |
2038 | * through the queue to unprep. | |
2039 | */ | |
2040 | } | |
e6ee8c0b | 2041 | goto out; |
c0820cf5 | 2042 | } |
e6ee8c0b | 2043 | |
5f015204 | 2044 | BUG_ON(!p || md->io_pool || md->bs); |
e6ee8c0b KU |
2045 | |
2046 | md->io_pool = p->io_pool; | |
2047 | p->io_pool = NULL; | |
e6ee8c0b KU |
2048 | md->bs = p->bs; |
2049 | p->bs = NULL; | |
2050 | ||
2051 | out: | |
2052 | /* mempool bind completed, now no need any mempools in the table */ | |
2053 | dm_table_free_md_mempools(t); | |
2054 | } | |
2055 | ||
1da177e4 LT |
2056 | /* |
2057 | * Bind a table to the device. | |
2058 | */ | |
2059 | static void event_callback(void *context) | |
2060 | { | |
7a8c3d3b MA |
2061 | unsigned long flags; |
2062 | LIST_HEAD(uevents); | |
1da177e4 LT |
2063 | struct mapped_device *md = (struct mapped_device *) context; |
2064 | ||
7a8c3d3b MA |
2065 | spin_lock_irqsave(&md->uevent_lock, flags); |
2066 | list_splice_init(&md->uevent_list, &uevents); | |
2067 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2068 | ||
ed9e1982 | 2069 | dm_send_uevents(&uevents, &disk_to_dev(md->disk)->kobj); |
7a8c3d3b | 2070 | |
1da177e4 LT |
2071 | atomic_inc(&md->event_nr); |
2072 | wake_up(&md->eventq); | |
2073 | } | |
2074 | ||
c217649b MS |
2075 | /* |
2076 | * Protected by md->suspend_lock obtained by dm_swap_table(). | |
2077 | */ | |
4e90188b | 2078 | static void __set_size(struct mapped_device *md, sector_t size) |
1da177e4 | 2079 | { |
4e90188b | 2080 | set_capacity(md->disk, size); |
1da177e4 | 2081 | |
db8fef4f | 2082 | i_size_write(md->bdev->bd_inode, (loff_t)size << SECTOR_SHIFT); |
1da177e4 LT |
2083 | } |
2084 | ||
d5b9dd04 MP |
2085 | /* |
2086 | * Return 1 if the queue has a compulsory merge_bvec_fn function. | |
2087 | * | |
2088 | * If this function returns 0, then the device is either a non-dm | |
2089 | * device without a merge_bvec_fn, or it is a dm device that is | |
2090 | * able to split any bios it receives that are too big. | |
2091 | */ | |
2092 | int dm_queue_merge_is_compulsory(struct request_queue *q) | |
2093 | { | |
2094 | struct mapped_device *dev_md; | |
2095 | ||
2096 | if (!q->merge_bvec_fn) | |
2097 | return 0; | |
2098 | ||
2099 | if (q->make_request_fn == dm_request) { | |
2100 | dev_md = q->queuedata; | |
2101 | if (test_bit(DMF_MERGE_IS_OPTIONAL, &dev_md->flags)) | |
2102 | return 0; | |
2103 | } | |
2104 | ||
2105 | return 1; | |
2106 | } | |
2107 | ||
2108 | static int dm_device_merge_is_compulsory(struct dm_target *ti, | |
2109 | struct dm_dev *dev, sector_t start, | |
2110 | sector_t len, void *data) | |
2111 | { | |
2112 | struct block_device *bdev = dev->bdev; | |
2113 | struct request_queue *q = bdev_get_queue(bdev); | |
2114 | ||
2115 | return dm_queue_merge_is_compulsory(q); | |
2116 | } | |
2117 | ||
2118 | /* | |
2119 | * Return 1 if it is acceptable to ignore merge_bvec_fn based | |
2120 | * on the properties of the underlying devices. | |
2121 | */ | |
2122 | static int dm_table_merge_is_optional(struct dm_table *table) | |
2123 | { | |
2124 | unsigned i = 0; | |
2125 | struct dm_target *ti; | |
2126 | ||
2127 | while (i < dm_table_get_num_targets(table)) { | |
2128 | ti = dm_table_get_target(table, i++); | |
2129 | ||
2130 | if (ti->type->iterate_devices && | |
2131 | ti->type->iterate_devices(ti, dm_device_merge_is_compulsory, NULL)) | |
2132 | return 0; | |
2133 | } | |
2134 | ||
2135 | return 1; | |
2136 | } | |
2137 | ||
042d2a9b AK |
2138 | /* |
2139 | * Returns old map, which caller must destroy. | |
2140 | */ | |
2141 | static struct dm_table *__bind(struct mapped_device *md, struct dm_table *t, | |
2142 | struct queue_limits *limits) | |
1da177e4 | 2143 | { |
042d2a9b | 2144 | struct dm_table *old_map; |
165125e1 | 2145 | struct request_queue *q = md->queue; |
1da177e4 | 2146 | sector_t size; |
d5b9dd04 | 2147 | int merge_is_optional; |
1da177e4 LT |
2148 | |
2149 | size = dm_table_get_size(t); | |
3ac51e74 DW |
2150 | |
2151 | /* | |
2152 | * Wipe any geometry if the size of the table changed. | |
2153 | */ | |
2154 | if (size != get_capacity(md->disk)) | |
2155 | memset(&md->geometry, 0, sizeof(md->geometry)); | |
2156 | ||
32a926da | 2157 | __set_size(md, size); |
d5816876 | 2158 | |
2ca3310e AK |
2159 | dm_table_event_callback(t, event_callback, md); |
2160 | ||
e6ee8c0b KU |
2161 | /* |
2162 | * The queue hasn't been stopped yet, if the old table type wasn't | |
2163 | * for request-based during suspension. So stop it to prevent | |
2164 | * I/O mapping before resume. | |
2165 | * This must be done before setting the queue restrictions, | |
2166 | * because request-based dm may be run just after the setting. | |
2167 | */ | |
2168 | if (dm_table_request_based(t) && !blk_queue_stopped(q)) | |
2169 | stop_queue(q); | |
2170 | ||
2171 | __bind_mempools(md, t); | |
2172 | ||
d5b9dd04 MP |
2173 | merge_is_optional = dm_table_merge_is_optional(t); |
2174 | ||
042d2a9b | 2175 | old_map = md->map; |
83d5e5b0 | 2176 | rcu_assign_pointer(md->map, t); |
36a0456f AK |
2177 | md->immutable_target_type = dm_table_get_immutable_target_type(t); |
2178 | ||
754c5fc7 | 2179 | dm_table_set_restrictions(t, q, limits); |
d5b9dd04 MP |
2180 | if (merge_is_optional) |
2181 | set_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
2182 | else | |
2183 | clear_bit(DMF_MERGE_IS_OPTIONAL, &md->flags); | |
83d5e5b0 | 2184 | dm_sync_table(md); |
1da177e4 | 2185 | |
042d2a9b | 2186 | return old_map; |
1da177e4 LT |
2187 | } |
2188 | ||
a7940155 AK |
2189 | /* |
2190 | * Returns unbound table for the caller to free. | |
2191 | */ | |
2192 | static struct dm_table *__unbind(struct mapped_device *md) | |
1da177e4 LT |
2193 | { |
2194 | struct dm_table *map = md->map; | |
2195 | ||
2196 | if (!map) | |
a7940155 | 2197 | return NULL; |
1da177e4 LT |
2198 | |
2199 | dm_table_event_callback(map, NULL, NULL); | |
83d5e5b0 MP |
2200 | rcu_assign_pointer(md->map, NULL); |
2201 | dm_sync_table(md); | |
a7940155 AK |
2202 | |
2203 | return map; | |
1da177e4 LT |
2204 | } |
2205 | ||
2206 | /* | |
2207 | * Constructor for a new device. | |
2208 | */ | |
2b06cfff | 2209 | int dm_create(int minor, struct mapped_device **result) |
1da177e4 LT |
2210 | { |
2211 | struct mapped_device *md; | |
2212 | ||
2b06cfff | 2213 | md = alloc_dev(minor); |
1da177e4 LT |
2214 | if (!md) |
2215 | return -ENXIO; | |
2216 | ||
784aae73 MB |
2217 | dm_sysfs_init(md); |
2218 | ||
1da177e4 LT |
2219 | *result = md; |
2220 | return 0; | |
2221 | } | |
2222 | ||
a5664dad MS |
2223 | /* |
2224 | * Functions to manage md->type. | |
2225 | * All are required to hold md->type_lock. | |
2226 | */ | |
2227 | void dm_lock_md_type(struct mapped_device *md) | |
2228 | { | |
2229 | mutex_lock(&md->type_lock); | |
2230 | } | |
2231 | ||
2232 | void dm_unlock_md_type(struct mapped_device *md) | |
2233 | { | |
2234 | mutex_unlock(&md->type_lock); | |
2235 | } | |
2236 | ||
2237 | void dm_set_md_type(struct mapped_device *md, unsigned type) | |
2238 | { | |
2239 | md->type = type; | |
2240 | } | |
2241 | ||
2242 | unsigned dm_get_md_type(struct mapped_device *md) | |
2243 | { | |
2244 | return md->type; | |
2245 | } | |
2246 | ||
36a0456f AK |
2247 | struct target_type *dm_get_immutable_target_type(struct mapped_device *md) |
2248 | { | |
2249 | return md->immutable_target_type; | |
2250 | } | |
2251 | ||
4a0b4ddf MS |
2252 | /* |
2253 | * Fully initialize a request-based queue (->elevator, ->request_fn, etc). | |
2254 | */ | |
2255 | static int dm_init_request_based_queue(struct mapped_device *md) | |
2256 | { | |
2257 | struct request_queue *q = NULL; | |
2258 | ||
2259 | if (md->queue->elevator) | |
2260 | return 1; | |
2261 | ||
2262 | /* Fully initialize the queue */ | |
2263 | q = blk_init_allocated_queue(md->queue, dm_request_fn, NULL); | |
2264 | if (!q) | |
2265 | return 0; | |
2266 | ||
2267 | md->queue = q; | |
4a0b4ddf MS |
2268 | dm_init_md_queue(md); |
2269 | blk_queue_softirq_done(md->queue, dm_softirq_done); | |
2270 | blk_queue_prep_rq(md->queue, dm_prep_fn); | |
2271 | blk_queue_lld_busy(md->queue, dm_lld_busy); | |
4a0b4ddf MS |
2272 | |
2273 | elv_register_queue(md->queue); | |
2274 | ||
2275 | return 1; | |
2276 | } | |
2277 | ||
2278 | /* | |
2279 | * Setup the DM device's queue based on md's type | |
2280 | */ | |
2281 | int dm_setup_md_queue(struct mapped_device *md) | |
2282 | { | |
2283 | if ((dm_get_md_type(md) == DM_TYPE_REQUEST_BASED) && | |
2284 | !dm_init_request_based_queue(md)) { | |
2285 | DMWARN("Cannot initialize queue for request-based mapped device"); | |
2286 | return -EINVAL; | |
2287 | } | |
2288 | ||
2289 | return 0; | |
2290 | } | |
2291 | ||
637842cf | 2292 | static struct mapped_device *dm_find_md(dev_t dev) |
1da177e4 LT |
2293 | { |
2294 | struct mapped_device *md; | |
1da177e4 LT |
2295 | unsigned minor = MINOR(dev); |
2296 | ||
2297 | if (MAJOR(dev) != _major || minor >= (1 << MINORBITS)) | |
2298 | return NULL; | |
2299 | ||
f32c10b0 | 2300 | spin_lock(&_minor_lock); |
1da177e4 LT |
2301 | |
2302 | md = idr_find(&_minor_idr, minor); | |
fba9f90e | 2303 | if (md && (md == MINOR_ALLOCED || |
f331c029 | 2304 | (MINOR(disk_devt(dm_disk(md))) != minor) || |
abdc568b | 2305 | dm_deleting_md(md) || |
17b2f66f | 2306 | test_bit(DMF_FREEING, &md->flags))) { |
637842cf | 2307 | md = NULL; |
fba9f90e JM |
2308 | goto out; |
2309 | } | |
1da177e4 | 2310 | |
fba9f90e | 2311 | out: |
f32c10b0 | 2312 | spin_unlock(&_minor_lock); |
1da177e4 | 2313 | |
637842cf DT |
2314 | return md; |
2315 | } | |
2316 | ||
d229a958 DT |
2317 | struct mapped_device *dm_get_md(dev_t dev) |
2318 | { | |
2319 | struct mapped_device *md = dm_find_md(dev); | |
2320 | ||
2321 | if (md) | |
2322 | dm_get(md); | |
2323 | ||
2324 | return md; | |
2325 | } | |
3cf2e4ba | 2326 | EXPORT_SYMBOL_GPL(dm_get_md); |
d229a958 | 2327 | |
9ade92a9 | 2328 | void *dm_get_mdptr(struct mapped_device *md) |
637842cf | 2329 | { |
9ade92a9 | 2330 | return md->interface_ptr; |
1da177e4 LT |
2331 | } |
2332 | ||
2333 | void dm_set_mdptr(struct mapped_device *md, void *ptr) | |
2334 | { | |
2335 | md->interface_ptr = ptr; | |
2336 | } | |
2337 | ||
2338 | void dm_get(struct mapped_device *md) | |
2339 | { | |
2340 | atomic_inc(&md->holders); | |
3f77316d | 2341 | BUG_ON(test_bit(DMF_FREEING, &md->flags)); |
1da177e4 LT |
2342 | } |
2343 | ||
72d94861 AK |
2344 | const char *dm_device_name(struct mapped_device *md) |
2345 | { | |
2346 | return md->name; | |
2347 | } | |
2348 | EXPORT_SYMBOL_GPL(dm_device_name); | |
2349 | ||
3f77316d | 2350 | static void __dm_destroy(struct mapped_device *md, bool wait) |
1da177e4 | 2351 | { |
1134e5ae | 2352 | struct dm_table *map; |
83d5e5b0 | 2353 | int srcu_idx; |
1da177e4 | 2354 | |
3f77316d | 2355 | might_sleep(); |
fba9f90e | 2356 | |
3f77316d | 2357 | spin_lock(&_minor_lock); |
83d5e5b0 | 2358 | map = dm_get_live_table(md, &srcu_idx); |
3f77316d KU |
2359 | idr_replace(&_minor_idr, MINOR_ALLOCED, MINOR(disk_devt(dm_disk(md)))); |
2360 | set_bit(DMF_FREEING, &md->flags); | |
2361 | spin_unlock(&_minor_lock); | |
2362 | ||
2363 | if (!dm_suspended_md(md)) { | |
2364 | dm_table_presuspend_targets(map); | |
2365 | dm_table_postsuspend_targets(map); | |
1da177e4 | 2366 | } |
3f77316d | 2367 | |
83d5e5b0 MP |
2368 | /* dm_put_live_table must be before msleep, otherwise deadlock is possible */ |
2369 | dm_put_live_table(md, srcu_idx); | |
2370 | ||
3f77316d KU |
2371 | /* |
2372 | * Rare, but there may be I/O requests still going to complete, | |
2373 | * for example. Wait for all references to disappear. | |
2374 | * No one should increment the reference count of the mapped_device, | |
2375 | * after the mapped_device state becomes DMF_FREEING. | |
2376 | */ | |
2377 | if (wait) | |
2378 | while (atomic_read(&md->holders)) | |
2379 | msleep(1); | |
2380 | else if (atomic_read(&md->holders)) | |
2381 | DMWARN("%s: Forcibly removing mapped_device still in use! (%d users)", | |
2382 | dm_device_name(md), atomic_read(&md->holders)); | |
2383 | ||
2384 | dm_sysfs_exit(md); | |
3f77316d KU |
2385 | dm_table_destroy(__unbind(md)); |
2386 | free_dev(md); | |
2387 | } | |
2388 | ||
2389 | void dm_destroy(struct mapped_device *md) | |
2390 | { | |
2391 | __dm_destroy(md, true); | |
2392 | } | |
2393 | ||
2394 | void dm_destroy_immediate(struct mapped_device *md) | |
2395 | { | |
2396 | __dm_destroy(md, false); | |
2397 | } | |
2398 | ||
2399 | void dm_put(struct mapped_device *md) | |
2400 | { | |
2401 | atomic_dec(&md->holders); | |
1da177e4 | 2402 | } |
79eb885c | 2403 | EXPORT_SYMBOL_GPL(dm_put); |
1da177e4 | 2404 | |
401600df | 2405 | static int dm_wait_for_completion(struct mapped_device *md, int interruptible) |
46125c1c MB |
2406 | { |
2407 | int r = 0; | |
b44ebeb0 MP |
2408 | DECLARE_WAITQUEUE(wait, current); |
2409 | ||
b44ebeb0 | 2410 | add_wait_queue(&md->wait, &wait); |
46125c1c MB |
2411 | |
2412 | while (1) { | |
401600df | 2413 | set_current_state(interruptible); |
46125c1c | 2414 | |
b4324fee | 2415 | if (!md_in_flight(md)) |
46125c1c MB |
2416 | break; |
2417 | ||
401600df MP |
2418 | if (interruptible == TASK_INTERRUPTIBLE && |
2419 | signal_pending(current)) { | |
46125c1c MB |
2420 | r = -EINTR; |
2421 | break; | |
2422 | } | |
2423 | ||
2424 | io_schedule(); | |
2425 | } | |
2426 | set_current_state(TASK_RUNNING); | |
2427 | ||
b44ebeb0 MP |
2428 | remove_wait_queue(&md->wait, &wait); |
2429 | ||
46125c1c MB |
2430 | return r; |
2431 | } | |
2432 | ||
1da177e4 LT |
2433 | /* |
2434 | * Process the deferred bios | |
2435 | */ | |
ef208587 | 2436 | static void dm_wq_work(struct work_struct *work) |
1da177e4 | 2437 | { |
ef208587 MP |
2438 | struct mapped_device *md = container_of(work, struct mapped_device, |
2439 | work); | |
6d6f10df | 2440 | struct bio *c; |
83d5e5b0 MP |
2441 | int srcu_idx; |
2442 | struct dm_table *map; | |
1da177e4 | 2443 | |
83d5e5b0 | 2444 | map = dm_get_live_table(md, &srcu_idx); |
ef208587 | 2445 | |
3b00b203 | 2446 | while (!test_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags)) { |
df12ee99 AK |
2447 | spin_lock_irq(&md->deferred_lock); |
2448 | c = bio_list_pop(&md->deferred); | |
2449 | spin_unlock_irq(&md->deferred_lock); | |
2450 | ||
6a8736d1 | 2451 | if (!c) |
df12ee99 | 2452 | break; |
022c2611 | 2453 | |
e6ee8c0b KU |
2454 | if (dm_request_based(md)) |
2455 | generic_make_request(c); | |
6a8736d1 | 2456 | else |
83d5e5b0 | 2457 | __split_and_process_bio(md, map, c); |
022c2611 | 2458 | } |
73d410c0 | 2459 | |
83d5e5b0 | 2460 | dm_put_live_table(md, srcu_idx); |
1da177e4 LT |
2461 | } |
2462 | ||
9a1fb464 | 2463 | static void dm_queue_flush(struct mapped_device *md) |
304f3f6a | 2464 | { |
3b00b203 MP |
2465 | clear_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
2466 | smp_mb__after_clear_bit(); | |
53d5914f | 2467 | queue_work(md->wq, &md->work); |
304f3f6a MB |
2468 | } |
2469 | ||
1da177e4 | 2470 | /* |
042d2a9b | 2471 | * Swap in a new table, returning the old one for the caller to destroy. |
1da177e4 | 2472 | */ |
042d2a9b | 2473 | struct dm_table *dm_swap_table(struct mapped_device *md, struct dm_table *table) |
1da177e4 | 2474 | { |
87eb5b21 | 2475 | struct dm_table *live_map = NULL, *map = ERR_PTR(-EINVAL); |
754c5fc7 | 2476 | struct queue_limits limits; |
042d2a9b | 2477 | int r; |
1da177e4 | 2478 | |
e61290a4 | 2479 | mutex_lock(&md->suspend_lock); |
1da177e4 LT |
2480 | |
2481 | /* device must be suspended */ | |
4f186f8b | 2482 | if (!dm_suspended_md(md)) |
93c534ae | 2483 | goto out; |
1da177e4 | 2484 | |
3ae70656 MS |
2485 | /* |
2486 | * If the new table has no data devices, retain the existing limits. | |
2487 | * This helps multipath with queue_if_no_path if all paths disappear, | |
2488 | * then new I/O is queued based on these limits, and then some paths | |
2489 | * reappear. | |
2490 | */ | |
2491 | if (dm_table_has_no_data_devices(table)) { | |
83d5e5b0 | 2492 | live_map = dm_get_live_table_fast(md); |
3ae70656 MS |
2493 | if (live_map) |
2494 | limits = md->queue->limits; | |
83d5e5b0 | 2495 | dm_put_live_table_fast(md); |
3ae70656 MS |
2496 | } |
2497 | ||
87eb5b21 MC |
2498 | if (!live_map) { |
2499 | r = dm_calculate_queue_limits(table, &limits); | |
2500 | if (r) { | |
2501 | map = ERR_PTR(r); | |
2502 | goto out; | |
2503 | } | |
042d2a9b | 2504 | } |
754c5fc7 | 2505 | |
042d2a9b | 2506 | map = __bind(md, table, &limits); |
1da177e4 | 2507 | |
93c534ae | 2508 | out: |
e61290a4 | 2509 | mutex_unlock(&md->suspend_lock); |
042d2a9b | 2510 | return map; |
1da177e4 LT |
2511 | } |
2512 | ||
2513 | /* | |
2514 | * Functions to lock and unlock any filesystem running on the | |
2515 | * device. | |
2516 | */ | |
2ca3310e | 2517 | static int lock_fs(struct mapped_device *md) |
1da177e4 | 2518 | { |
e39e2e95 | 2519 | int r; |
1da177e4 LT |
2520 | |
2521 | WARN_ON(md->frozen_sb); | |
dfbe03f6 | 2522 | |
db8fef4f | 2523 | md->frozen_sb = freeze_bdev(md->bdev); |
dfbe03f6 | 2524 | if (IS_ERR(md->frozen_sb)) { |
cf222b37 | 2525 | r = PTR_ERR(md->frozen_sb); |
e39e2e95 AK |
2526 | md->frozen_sb = NULL; |
2527 | return r; | |
dfbe03f6 AK |
2528 | } |
2529 | ||
aa8d7c2f AK |
2530 | set_bit(DMF_FROZEN, &md->flags); |
2531 | ||
1da177e4 LT |
2532 | return 0; |
2533 | } | |
2534 | ||
2ca3310e | 2535 | static void unlock_fs(struct mapped_device *md) |
1da177e4 | 2536 | { |
aa8d7c2f AK |
2537 | if (!test_bit(DMF_FROZEN, &md->flags)) |
2538 | return; | |
2539 | ||
db8fef4f | 2540 | thaw_bdev(md->bdev, md->frozen_sb); |
1da177e4 | 2541 | md->frozen_sb = NULL; |
aa8d7c2f | 2542 | clear_bit(DMF_FROZEN, &md->flags); |
1da177e4 LT |
2543 | } |
2544 | ||
2545 | /* | |
2546 | * We need to be able to change a mapping table under a mounted | |
2547 | * filesystem. For example we might want to move some data in | |
2548 | * the background. Before the table can be swapped with | |
2549 | * dm_bind_table, dm_suspend must be called to flush any in | |
2550 | * flight bios and ensure that any further io gets deferred. | |
2551 | */ | |
cec47e3d KU |
2552 | /* |
2553 | * Suspend mechanism in request-based dm. | |
2554 | * | |
9f518b27 KU |
2555 | * 1. Flush all I/Os by lock_fs() if needed. |
2556 | * 2. Stop dispatching any I/O by stopping the request_queue. | |
2557 | * 3. Wait for all in-flight I/Os to be completed or requeued. | |
cec47e3d | 2558 | * |
9f518b27 | 2559 | * To abort suspend, start the request_queue. |
cec47e3d | 2560 | */ |
a3d77d35 | 2561 | int dm_suspend(struct mapped_device *md, unsigned suspend_flags) |
1da177e4 | 2562 | { |
2ca3310e | 2563 | struct dm_table *map = NULL; |
46125c1c | 2564 | int r = 0; |
a3d77d35 | 2565 | int do_lockfs = suspend_flags & DM_SUSPEND_LOCKFS_FLAG ? 1 : 0; |
2e93ccc1 | 2566 | int noflush = suspend_flags & DM_SUSPEND_NOFLUSH_FLAG ? 1 : 0; |
1da177e4 | 2567 | |
e61290a4 | 2568 | mutex_lock(&md->suspend_lock); |
2ca3310e | 2569 | |
4f186f8b | 2570 | if (dm_suspended_md(md)) { |
73d410c0 | 2571 | r = -EINVAL; |
d287483d | 2572 | goto out_unlock; |
73d410c0 | 2573 | } |
1da177e4 | 2574 | |
83d5e5b0 | 2575 | map = md->map; |
1da177e4 | 2576 | |
2e93ccc1 KU |
2577 | /* |
2578 | * DMF_NOFLUSH_SUSPENDING must be set before presuspend. | |
2579 | * This flag is cleared before dm_suspend returns. | |
2580 | */ | |
2581 | if (noflush) | |
2582 | set_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); | |
2583 | ||
cf222b37 AK |
2584 | /* This does not get reverted if there's an error later. */ |
2585 | dm_table_presuspend_targets(map); | |
2586 | ||
32a926da | 2587 | /* |
9f518b27 KU |
2588 | * Flush I/O to the device. |
2589 | * Any I/O submitted after lock_fs() may not be flushed. | |
2590 | * noflush takes precedence over do_lockfs. | |
2591 | * (lock_fs() flushes I/Os and waits for them to complete.) | |
32a926da MP |
2592 | */ |
2593 | if (!noflush && do_lockfs) { | |
2594 | r = lock_fs(md); | |
2595 | if (r) | |
83d5e5b0 | 2596 | goto out_unlock; |
aa8d7c2f | 2597 | } |
1da177e4 LT |
2598 | |
2599 | /* | |
3b00b203 MP |
2600 | * Here we must make sure that no processes are submitting requests |
2601 | * to target drivers i.e. no one may be executing | |
2602 | * __split_and_process_bio. This is called from dm_request and | |
2603 | * dm_wq_work. | |
2604 | * | |
2605 | * To get all processes out of __split_and_process_bio in dm_request, | |
2606 | * we take the write lock. To prevent any process from reentering | |
6a8736d1 TH |
2607 | * __split_and_process_bio from dm_request and quiesce the thread |
2608 | * (dm_wq_work), we set BMF_BLOCK_IO_FOR_SUSPEND and call | |
2609 | * flush_workqueue(md->wq). | |
1da177e4 | 2610 | */ |
1eb787ec | 2611 | set_bit(DMF_BLOCK_IO_FOR_SUSPEND, &md->flags); |
83d5e5b0 | 2612 | synchronize_srcu(&md->io_barrier); |
1da177e4 | 2613 | |
d0bcb878 | 2614 | /* |
29e4013d TH |
2615 | * Stop md->queue before flushing md->wq in case request-based |
2616 | * dm defers requests to md->wq from md->queue. | |
d0bcb878 | 2617 | */ |
cec47e3d | 2618 | if (dm_request_based(md)) |
9f518b27 | 2619 | stop_queue(md->queue); |
cec47e3d | 2620 | |
d0bcb878 KU |
2621 | flush_workqueue(md->wq); |
2622 | ||
1da177e4 | 2623 | /* |
3b00b203 MP |
2624 | * At this point no more requests are entering target request routines. |
2625 | * We call dm_wait_for_completion to wait for all existing requests | |
2626 | * to finish. | |
1da177e4 | 2627 | */ |
401600df | 2628 | r = dm_wait_for_completion(md, TASK_INTERRUPTIBLE); |
1da177e4 | 2629 | |
6d6f10df | 2630 | if (noflush) |
022c2611 | 2631 | clear_bit(DMF_NOFLUSH_SUSPENDING, &md->flags); |
83d5e5b0 | 2632 | synchronize_srcu(&md->io_barrier); |
2e93ccc1 | 2633 | |
1da177e4 | 2634 | /* were we interrupted ? */ |
46125c1c | 2635 | if (r < 0) { |
9a1fb464 | 2636 | dm_queue_flush(md); |
73d410c0 | 2637 | |
cec47e3d | 2638 | if (dm_request_based(md)) |
9f518b27 | 2639 | start_queue(md->queue); |
cec47e3d | 2640 | |
2ca3310e | 2641 | unlock_fs(md); |
83d5e5b0 | 2642 | goto out_unlock; /* pushback list is already flushed, so skip flush */ |
2ca3310e | 2643 | } |
1da177e4 | 2644 | |
3b00b203 MP |
2645 | /* |
2646 | * If dm_wait_for_completion returned 0, the device is completely | |
2647 | * quiescent now. There is no request-processing activity. All new | |
2648 | * requests are being added to md->deferred list. | |
2649 | */ | |
2650 | ||
2ca3310e | 2651 | set_bit(DMF_SUSPENDED, &md->flags); |
b84b0287 | 2652 | |
4d4471cb KU |
2653 | dm_table_postsuspend_targets(map); |
2654 | ||
d287483d | 2655 | out_unlock: |
e61290a4 | 2656 | mutex_unlock(&md->suspend_lock); |
cf222b37 | 2657 | return r; |
1da177e4 LT |
2658 | } |
2659 | ||
2660 | int dm_resume(struct mapped_device *md) | |
2661 | { | |
cf222b37 | 2662 | int r = -EINVAL; |
cf222b37 | 2663 | struct dm_table *map = NULL; |
1da177e4 | 2664 | |
e61290a4 | 2665 | mutex_lock(&md->suspend_lock); |
4f186f8b | 2666 | if (!dm_suspended_md(md)) |
cf222b37 | 2667 | goto out; |
cf222b37 | 2668 | |
83d5e5b0 | 2669 | map = md->map; |
2ca3310e | 2670 | if (!map || !dm_table_get_size(map)) |
cf222b37 | 2671 | goto out; |
1da177e4 | 2672 | |
8757b776 MB |
2673 | r = dm_table_resume_targets(map); |
2674 | if (r) | |
2675 | goto out; | |
2ca3310e | 2676 | |
9a1fb464 | 2677 | dm_queue_flush(md); |
2ca3310e | 2678 | |
cec47e3d KU |
2679 | /* |
2680 | * Flushing deferred I/Os must be done after targets are resumed | |
2681 | * so that mapping of targets can work correctly. | |
2682 | * Request-based dm is queueing the deferred I/Os in its request_queue. | |
2683 | */ | |
2684 | if (dm_request_based(md)) | |
2685 | start_queue(md->queue); | |
2686 | ||
2ca3310e AK |
2687 | unlock_fs(md); |
2688 | ||
2689 | clear_bit(DMF_SUSPENDED, &md->flags); | |
2690 | ||
cf222b37 AK |
2691 | r = 0; |
2692 | out: | |
e61290a4 | 2693 | mutex_unlock(&md->suspend_lock); |
2ca3310e | 2694 | |
cf222b37 | 2695 | return r; |
1da177e4 LT |
2696 | } |
2697 | ||
2698 | /*----------------------------------------------------------------- | |
2699 | * Event notification. | |
2700 | *---------------------------------------------------------------*/ | |
3abf85b5 | 2701 | int dm_kobject_uevent(struct mapped_device *md, enum kobject_action action, |
60935eb2 | 2702 | unsigned cookie) |
69267a30 | 2703 | { |
60935eb2 MB |
2704 | char udev_cookie[DM_COOKIE_LENGTH]; |
2705 | char *envp[] = { udev_cookie, NULL }; | |
2706 | ||
2707 | if (!cookie) | |
3abf85b5 | 2708 | return kobject_uevent(&disk_to_dev(md->disk)->kobj, action); |
60935eb2 MB |
2709 | else { |
2710 | snprintf(udev_cookie, DM_COOKIE_LENGTH, "%s=%u", | |
2711 | DM_COOKIE_ENV_VAR_NAME, cookie); | |
3abf85b5 PR |
2712 | return kobject_uevent_env(&disk_to_dev(md->disk)->kobj, |
2713 | action, envp); | |
60935eb2 | 2714 | } |
69267a30 AK |
2715 | } |
2716 | ||
7a8c3d3b MA |
2717 | uint32_t dm_next_uevent_seq(struct mapped_device *md) |
2718 | { | |
2719 | return atomic_add_return(1, &md->uevent_seq); | |
2720 | } | |
2721 | ||
1da177e4 LT |
2722 | uint32_t dm_get_event_nr(struct mapped_device *md) |
2723 | { | |
2724 | return atomic_read(&md->event_nr); | |
2725 | } | |
2726 | ||
2727 | int dm_wait_event(struct mapped_device *md, int event_nr) | |
2728 | { | |
2729 | return wait_event_interruptible(md->eventq, | |
2730 | (event_nr != atomic_read(&md->event_nr))); | |
2731 | } | |
2732 | ||
7a8c3d3b MA |
2733 | void dm_uevent_add(struct mapped_device *md, struct list_head *elist) |
2734 | { | |
2735 | unsigned long flags; | |
2736 | ||
2737 | spin_lock_irqsave(&md->uevent_lock, flags); | |
2738 | list_add(elist, &md->uevent_list); | |
2739 | spin_unlock_irqrestore(&md->uevent_lock, flags); | |
2740 | } | |
2741 | ||
1da177e4 LT |
2742 | /* |
2743 | * The gendisk is only valid as long as you have a reference | |
2744 | * count on 'md'. | |
2745 | */ | |
2746 | struct gendisk *dm_disk(struct mapped_device *md) | |
2747 | { | |
2748 | return md->disk; | |
2749 | } | |
2750 | ||
784aae73 MB |
2751 | struct kobject *dm_kobject(struct mapped_device *md) |
2752 | { | |
2753 | return &md->kobj; | |
2754 | } | |
2755 | ||
2756 | /* | |
2757 | * struct mapped_device should not be exported outside of dm.c | |
2758 | * so use this check to verify that kobj is part of md structure | |
2759 | */ | |
2760 | struct mapped_device *dm_get_from_kobject(struct kobject *kobj) | |
2761 | { | |
2762 | struct mapped_device *md; | |
2763 | ||
2764 | md = container_of(kobj, struct mapped_device, kobj); | |
2765 | if (&md->kobj != kobj) | |
2766 | return NULL; | |
2767 | ||
4d89b7b4 | 2768 | if (test_bit(DMF_FREEING, &md->flags) || |
432a212c | 2769 | dm_deleting_md(md)) |
4d89b7b4 MB |
2770 | return NULL; |
2771 | ||
784aae73 MB |
2772 | dm_get(md); |
2773 | return md; | |
2774 | } | |
2775 | ||
4f186f8b | 2776 | int dm_suspended_md(struct mapped_device *md) |
1da177e4 LT |
2777 | { |
2778 | return test_bit(DMF_SUSPENDED, &md->flags); | |
2779 | } | |
2780 | ||
64dbce58 KU |
2781 | int dm_suspended(struct dm_target *ti) |
2782 | { | |
ecdb2e25 | 2783 | return dm_suspended_md(dm_table_get_md(ti->table)); |
64dbce58 KU |
2784 | } |
2785 | EXPORT_SYMBOL_GPL(dm_suspended); | |
2786 | ||
2e93ccc1 KU |
2787 | int dm_noflush_suspending(struct dm_target *ti) |
2788 | { | |
ecdb2e25 | 2789 | return __noflush_suspending(dm_table_get_md(ti->table)); |
2e93ccc1 KU |
2790 | } |
2791 | EXPORT_SYMBOL_GPL(dm_noflush_suspending); | |
2792 | ||
c0820cf5 | 2793 | struct dm_md_mempools *dm_alloc_md_mempools(unsigned type, unsigned integrity, unsigned per_bio_data_size) |
e6ee8c0b | 2794 | { |
5f015204 JN |
2795 | struct dm_md_mempools *pools = kzalloc(sizeof(*pools), GFP_KERNEL); |
2796 | struct kmem_cache *cachep; | |
2797 | unsigned int pool_size; | |
2798 | unsigned int front_pad; | |
e6ee8c0b KU |
2799 | |
2800 | if (!pools) | |
2801 | return NULL; | |
2802 | ||
23e5083b | 2803 | if (type == DM_TYPE_BIO_BASED) { |
5f015204 JN |
2804 | cachep = _io_cache; |
2805 | pool_size = 16; | |
2806 | front_pad = roundup(per_bio_data_size, __alignof__(struct dm_target_io)) + offsetof(struct dm_target_io, clone); | |
2807 | } else if (type == DM_TYPE_REQUEST_BASED) { | |
2808 | cachep = _rq_tio_cache; | |
2809 | pool_size = MIN_IOS; | |
2810 | front_pad = offsetof(struct dm_rq_clone_bio_info, clone); | |
2811 | /* per_bio_data_size is not used. See __bind_mempools(). */ | |
2812 | WARN_ON(per_bio_data_size != 0); | |
2813 | } else | |
2814 | goto out; | |
e6ee8c0b | 2815 | |
5f015204 JN |
2816 | pools->io_pool = mempool_create_slab_pool(MIN_IOS, cachep); |
2817 | if (!pools->io_pool) | |
2818 | goto out; | |
e6ee8c0b | 2819 | |
5f015204 | 2820 | pools->bs = bioset_create(pool_size, front_pad); |
e6ee8c0b | 2821 | if (!pools->bs) |
5f015204 | 2822 | goto out; |
e6ee8c0b | 2823 | |
a91a2785 | 2824 | if (integrity && bioset_integrity_create(pools->bs, pool_size)) |
5f015204 | 2825 | goto out; |
a91a2785 | 2826 | |
e6ee8c0b KU |
2827 | return pools; |
2828 | ||
5f015204 JN |
2829 | out: |
2830 | dm_free_md_mempools(pools); | |
e6ee8c0b KU |
2831 | |
2832 | return NULL; | |
2833 | } | |
2834 | ||
2835 | void dm_free_md_mempools(struct dm_md_mempools *pools) | |
2836 | { | |
2837 | if (!pools) | |
2838 | return; | |
2839 | ||
2840 | if (pools->io_pool) | |
2841 | mempool_destroy(pools->io_pool); | |
2842 | ||
e6ee8c0b KU |
2843 | if (pools->bs) |
2844 | bioset_free(pools->bs); | |
2845 | ||
2846 | kfree(pools); | |
2847 | } | |
2848 | ||
83d5cde4 | 2849 | static const struct block_device_operations dm_blk_dops = { |
1da177e4 LT |
2850 | .open = dm_blk_open, |
2851 | .release = dm_blk_close, | |
aa129a22 | 2852 | .ioctl = dm_blk_ioctl, |
3ac51e74 | 2853 | .getgeo = dm_blk_getgeo, |
1da177e4 LT |
2854 | .owner = THIS_MODULE |
2855 | }; | |
2856 | ||
2857 | EXPORT_SYMBOL(dm_get_mapinfo); | |
2858 | ||
2859 | /* | |
2860 | * module hooks | |
2861 | */ | |
2862 | module_init(dm_init); | |
2863 | module_exit(dm_exit); | |
2864 | ||
2865 | module_param(major, uint, 0); | |
2866 | MODULE_PARM_DESC(major, "The major number of the device mapper"); | |
2867 | MODULE_DESCRIPTION(DM_NAME " driver"); | |
2868 | MODULE_AUTHOR("Joe Thornber <dm-devel@redhat.com>"); | |
2869 | MODULE_LICENSE("GPL"); |